Spartan-3 Generation

FPGA User Guide

Extended Spartan-3A,
Spartan-3E, and Spartan-3
FPGA Families

UG331 (v1.7) August 19, 2010

R
 R

Xilinx is disclosing this user guide, manual, release note, and/or specification (the “Documentation”) to you solely for use in the development
of designs to operate with Xilinx hardware devices. You may not reproduce, distribute, republish, download, display, post, or transmit the
Documentation in any form or by any means including, but not limited to, electronic, mechanical, photocopying, recording, or otherwise,
without the prior written consent of Xilinx. Xilinx expressly disclaims any liability arising out of your use of the Documentation. Xilinx reserves
the right, at its sole discretion, to change the Documentation without notice at any time. Xilinx assumes no obligation to correct any errors
contained in the Documentation, or to advise you of any corrections or updates. Xilinx expressly disclaims any liability in connection with
technical support or assistance that may be provided to you in connection with the Information.
THE DOCUMENTATION IS DISCLOSED TO YOU “AS-IS” WITH NO WARRANTY OF ANY KIND. XILINX MAKES NO OTHER
WARRANTIES, WHETHER EXPRESS, IMPLIED, OR STATUTORY, REGARDING THE DOCUMENTATION, INCLUDING ANY
WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NONINFRINGEMENT OF THIRD-PARTY
RIGHTS. IN NO EVENT WILL XILINX BE LIABLE FOR ANY CONSEQUENTIAL, INDIRECT, EXEMPLARY, SPECIAL, OR INCIDENTAL
DAMAGES, INCLUDING ANY LOSS OF DATA OR LOST PROFITS, ARISING FROM YOUR USE OF THE DOCUMENTATION.
© 2006–2010 Xilinx, Inc. XILINX, the Xilinx logo, Virtex, Spartan, ISE, and other designated brands included herein are trademarks of Xilinx
in the United States and other countries. All other trademarks are the property of their respective owners.

Revision History
The following table shows the revision history for this document.

Date Version Revision

12/05/06 1.0 Initial release.
02/26/07 1.1 Added Spartan-3AN platform.
04/02/07 1.2 Added Spartan-3A DSP platform.
02/14/08 1.3 Updated for latest package offerings. Updated and corrected descriptions and figures
throughout. Updated links for new xilinx.com navigation.
06/25/08 1.4 Added and updated links to design files. Updated banking rules in I/O section. Added
reference to XAPP459 in “Using Large-Swing Signals”.
01/21/09 1.5 Updated document to refer to Extended Spartan-3A family. Corrected column headers
CLKA and CLKB in Table 4-14 on page 183. Clarified DCM “Output Alignment” on
page 131. Added section “Spartan-3A and Spartan-3A DSP FPGA Dual-Range VCCAUX”
on page 355.

Spartan-3 Generation FPGA User Guide www.xilinx.com UG331 (v1.7) August 19, 2010
 Date Version Revision
12/03/09 1.6 Updated “Extended Spartan-3A Family Features,” page 32. Updated “I/O Capabilities,”
page 38. Updated “Global Clock Resources” and Figure 2-1 on page 46 and “Additional
Information,” page 62 to note that local clocking is not recommended. Added Figure 2-6
and described CLK0 to CLK1 switchover in “BUFGMUX Multiplexing Details,” page
54. Updated “Using Clock Buffers/Multiplexers in a Design,” page 55 and added
Figure 2-7. Updated “XST Synthesis of Clock Buffers,” page 56. Added Table 2-8 with
clock quadrant locations. Clarified “Digital Frequency Synthesizer (DFS)”and “Phase
Shift (PS),” page 70. Updated “Output Availability Depends on DLL Frequency Mode,”
page 118. Updated “Fine Phase Shifting,” page 119. Revised Figure 4-1 to show parity
integers on the data paths. Updated “Carry and Floorplanning,” page 303. Added
“Clamp Diodes,” page 336. Removed references to older software versions in
“Specifying an I/O Standard with the IOSTANDARD Attribute,” page 343,
“LVCMOS/LVTTL Slew Rate Control and Drive Strength,” page 345, and “Differential
I/O Standards,” page 351. Clarified BSDL termination in “BLVDS Output Termination,”
page 353. Removed references to older software versions in “IOBs Organized into
Banks,” page 355. Corrected VCC value for MINI_LVDS_33/Input with DIFF_TERM
inTable 10-20. Revised VREF note in Table 10-20, Table 10-21, and Table 10-22 to state that
VREF is not used for the differential I/O standards described in the table. Removed
references to older software versions in“Floorplanning,” page 413 and “Constraints
Editor,” page 423. Updated “Differences in Packages Between Spartan-3 Generation
Families,” page 467. Added footnote in Table 17-1 to indicate that the CP132 and CPG132
packages are being discontinued. Added recommended power-down sequence to “Hot
Swap,” page 491. Updated “Application State Retained during Suspend Mode,” page
507. Updated “Extended Spartan-3A Family FPGA: Turn Off VCCO,” page 516. Updated
the FG900/FGG900 package drawing in Figure 17-12.
08/19/10 1.7 Updated values for FG400 in upper half of Table 3-18. Updated note under “Address
Input”. Added “Timing Parameters” section. Added “Note relevant to Figure 10-1”.
Added last sentence to paragraph immediately following Figure 10-15. Added last
sentence to second paragraph under “ODDR2”. Added ”to VCCO” to last sentence under
“ESD Protection”. Added “Parasitic Leakage” section. Added last paragraph to “Supply
Sequencing”.

UG331 (v1.7) August 19, 2010 www.xilinx.com Spartan-3 Generation FPGA User Guide
Spartan-3 Generation FPGA User Guide www.xilinx.com UG331 (v1.7) August 19, 2010
Table of Contents

Revision History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Preface: About This Guide

Guide Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Additional Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Typographical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24
Online Document . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25

Section I: Designing with Spartan-3 Generation FPGAs

Chapter 1: Overview
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Extended Spartan-3A Family Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Spartan-3AN Platform Additional Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33
Spartan-3A DSP Platform Additional Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34
Spartan-3 Generation Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Architectural Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
I/O Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Package Marking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Ordering Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Chapter 2: Using Global Clock Resources

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Global Clock Resource Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . 45
Global Clock Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Clocking Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .46
Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Extended Spartan-3A Family Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
Spartan-3E FPGA Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
Spartan-3 FPGA Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
Clock Inputs and DCMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
Differential Clocks Using Two Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
Using Dedicated Clock Inputs in a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52
IBUFG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52
IBUFGDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52
Clock Buffers/Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
BUFGMUX Multiplexing Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54

Spartan-3 Generation FPGA User Guide www.xilinx.com 5

UG331 (v1.7) August 19, 2010
 R

Using Clock Buffers/Multiplexers in a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55

BUFGMUX and BUFGMUX_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
BUFG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
BUFGCE and BUFGCE_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .56
XST Synthesis of Clock Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .56
BUFGMUX Connection Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
BUFGMUX Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57
BUFGMUX Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59
Spartan-3 Global Clock Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59
Quadrant Clock Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Choosing Top/Bottom and Left-/Right-Half Global Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
Spartan-3 FPGA Global Clock Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
Other Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Clock Power Consumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
Clock Setup and Hold Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .62
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Additional Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .62

Chapter 3: Using Digital Clock Managers (DCMs)

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Document Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Compatibility and Comparison with Other Xilinx FPGA Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
DCM Locations and Clock Distribution Network Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
DCM Functional Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Delay-Locked Loop (DLL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .70
Digital Frequency Synthesizer (DFS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .70
Phase Shift (PS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .70
Status Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71
DCM Primitive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .72
Connection Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .72
Attributes, Properties, or Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .76
DCM Clock Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Input Clock Frequency Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .80
Output Clock Frequency Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .81
Input Clock and Clock Feedback Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .81
Cycle-to-Cycle Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .82
Period Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .82
DLL Feedback Delay Variance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .83
Spread Spectrum Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .83
Optimal DCM Clock and External Feedback Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .84
Spartan-3E FPGA DCM Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .84
Extended Spartan-3A Family FPGA DCM Clock Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .87
LOCKED Output Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Using the LOCKED Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91
Spartan-3A FPGA DCM Digital Frequency Synthesizer Requires Additional Lock Circuitry . . . . . . . . . .92
RST Input Behavior. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

6 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Invoking Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .93

From Windows Start Button . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .93
From within Project Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .94
General Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .95
Advanced Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .97
Clock Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .98
Clock Frequency Synthesizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .99
Generating HDL Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .100
VHDL and Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Language Templates within Project Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .101
Eliminating Clock Skew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
What is Clock Skew? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .102
Clock Skew: The Performance Thief . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .103
Make It Go Away! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .103
Predicting the Future by Closely Examining the Past . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .104
Locked on Target . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .105
A Stable, Monotonic Clock Input. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .105
Feedback from a Reliable Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .106
Removing Skew from an Internal Clock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .106
Removing Skew from an External Clock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .107
Reset DCM After Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .107
Why Reset? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .108
What is a Delay-Locked Loop? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109
Delay-Locked Loop (DLL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109
Phase-Locked Loop (PLL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110
DLL vs. PLL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110
Skew Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110
System Synchronous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110
Source Synchronous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .111
Timing Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .112
Clock Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Spartan-3E and Extended Spartan-3A Family FPGA Output Clock Conditioning . . . . . . . . . . . . . . . . . . .113
Spartan-3 FPGA Output Clock Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .113
Phase Shifting – Delaying Clock Outputs by a Fraction of a Period . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Half-Period Phase Shifted Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .115
Half-Period Phase Shift Outputs Reduce Duty-Cycle Distortion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116
Dual-Data Rate (DDR) Clocking Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116
Quadrant Phase Shifted Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .117
Output Availability Depends on DLL Frequency Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .118
Spartan-3 FPGA: Optional 50/50 Duty Cycle Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .118
Four Phases, Delayed Clock Edges, Phased Pulses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .119
Fine Phase Shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .119
Fixed Fine Phase Shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .120
Spartan-3 Family Fixed Fine Phase Shift Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .120
Minimum Phase Shift Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122
Other Design Considerations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122
Variable Fine Phase Shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .123
Important Differences Between Spartan-3 Generation FPGA Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .123
Spartan-3E and Extended Spartan-3A Family FPGA Variable Phase Shift Operations . . . . . . . . . . . . . . . . . . .124
Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .125
Variable Fine Phase Shift Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .126

Spartan-3 Generation FPGA User Guide www.xilinx.com 7

UG331 (v1.7) August 19, 2010
 R

Spartan-3 FPGA Family Variable Phase Shift Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .126

Spartan-3E and Extended Spartan-3A Family Variable Phase-Shift Range . . . . . . . . . . . . . . . . . . . . . . . . . . . .127
Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .127
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .129
Example Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .129
Clock Multiplication, Clock Division, and Frequency Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Output Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131
Frequency Synthesis Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131
Input and Output Clock Frequency Restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .132
Clock Doubler (CLK2X, CLK2X180) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .133
Clock Divider (CLKDV) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .134
CLKDV Clock Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .135
CLKDV Jitter Depends on Frequency Mode and Integer or Non-Integer Value . . . . . . . . . . . . . . . . . . . . . . . .136
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .136
Frequency Synthesizer (CLKFX, CLKFX180). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .136
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .138
Clock Forwarding, Mirroring, Rebuffering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Clock Jitter or Phase Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
What is Clock Jitter? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .140
What Causes Clock Jitter? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141
Understanding Clock Jitter Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141
Cycle-to-Cycle Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .141
Period Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .142
Unit Interval (UI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .142
Calculating Total Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143
Adding Input Jitter to DLL Output Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143
Calculating Jitter for Cascaded DCMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .144
Example 1: All DCMs Use DLL Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .144
Example 2: Some DCMs Use the CLKFX or CLKFX180 Outputs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145
Cascaded DCM Design Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145
Jitter Effect on System Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .145
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .146
Recommended Design Practices to Minimize Clock Jitter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .146
Properly Design the Power Distribution System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .147
Properly Design the Printed Circuit Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .147
Obey Simultaneous Switching Output (SSO) Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .147
Optionally Place Virtual Ground Pins Around DCM Input and Output Connections . . . . . . . . . . . . . . . . . . .147
VCCAUX Considerations for Improving Jitter Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .148
Adjusting FACTORY_JF Setting (Spartan-3 FPGA Family Only) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .149
Miscellaneous Advanced Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Bitstream Generation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .149
Setting Bitstream Generation Options in Project Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .149
Setting Bitstream Generation Options via Command Line or Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .150
Setting Configuration Logic to Wait for DCM LOCKED Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .150
Reset DCM During Partial Reconfiguration or During Full Reconfiguration via JTAG . . . . . . . . . . . . . . . . . .151
Momentarily Stopping CLKIN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .152
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

Chapter 4: Using Block RAM

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

8 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

Block RAM Location and Surrounding Neighborhood. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Block RAM/Multiplier Routing Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .160
Data Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Data Inputs and Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .161
Data Input Bus — DI[#:0] (DIA[#:0], DIB[#:0]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .161
Data Output Bus — DO[#:0] (DOA[#:0], DOB[#:0]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .161
Parity Inputs and Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .162
Data Input Parity Bus — DIP[#:0] (DIPA[#:0], DIPB[#:0]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .162
Data Output Parity Bus — DOP[#:0] (DOPA[#:0], DOPB[#:0]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .162
Address Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Address Bus — ADDR[#:0] (ADDRA[#:0], ADDRB[#:0]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Control Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Clock — CLK (CLKA, CLKB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Enable — EN (ENA, ENB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Write Enable — WE (WEA, WEB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .164
Output Register Enable - REGCE (REGCEA, REGCEB) Spartan-3A DSP FPGA Only . . . . . . . . . . . . . . . . . . .166
Output Latch Synchronous Set/Reset — SSR (SSRA, SSRB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .166
Output Latch/Register Synchronous/Asynchronous Set/Reset - RST (RSTA, RSTB)
- Spartan-3A DSP FPGA Only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .166
Global Set/Reset — GSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .167
Inverting Control Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .167
Unused Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .167
Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Number of Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .168
CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .168
VHDL or Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .169
Memory Organization/Aspect Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .169
CORE Generator System — Memory Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .169
VHDL or Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .170
Address and Data Mapping Between Two Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .170
Content Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .171
CORE Generator System — Load Init File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .171
VHDL or Verilog Instantiation — INIT_xx, INITP_xx. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .171
Data Output Latch Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .172
CORE Generator System — Global Init Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .172
VHDL or Verilog Instantiation — INIT (INIT_A and INIT_B) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173
Data Output Latch Synchronous Set/Reset Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173
CORE Generator System — Init Value (SINIT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173
VHDL or Verilog Instantiation — SRVAL (SRVAL_A and SRVAL_B) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173
Read Behavior During Simultaneous Write — WRITE_MODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .173
WRITE_FIRST or Transparent Mode (Default) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .174
READ_FIRST or Read-Before-Write Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .175
NO_CHANGE Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .176
CORE Generator System — Write Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .177
VHDL or Verilog Instantiation — WRITE_MODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .177
Location Constraints (LOC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .177
Block RAM Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
RAM Contents Initialized During Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .179
Global Set/Reset Initializes Data Output Latches Immediately After Configuration or Global Reset. . .179
Enable Input Activates or Disables RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .179

Spartan-3 Generation FPGA User Guide www.xilinx.com 9

UG331 (v1.7) August 19, 2010
 R

Synchronous Set/Reset Initializes Data Output Latches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .180

Simultaneous Write and Synchronous Set/Reset Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .180
Read Operations Occur on Every Clock Edge When Enable is Asserted . . . . . . . . . . . . . . . . . . . . . . . . . . .180
Write Operations Always Have Simultaneous Read Operation, Data Output Latches Affected . . . . . . .180
General Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .180
Functional Compatibility with Other Xilinx FPGA Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .181
Dual-Port RAM Conflicts and Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Timing Violation Conflicts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .181
Simultaneous Writes to Both Ports with Different Data Conflicts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .182
Write Mode Conflicts on Output Latches. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .183
Conflict Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .183
Block RAM Design Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Xilinx CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .183
VHDL and Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .184
Inferring Block RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .184
Instantiation Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .184
Initialization in VHDL or Verilog Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .185
Block RAM Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Creating Larger RAM Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .185
Block RAM as Read-Only Memory (ROM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .185
FIFOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .185
Storage for Embedded Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .186
Updating Block RAM/ROM Content by Directly Modifying Device Bitstream . . . . . . . . . . . . . . . . . . . . .186
Two Independent Single-Port RAMs Using One Block RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .187
A 256x72 Single-Port RAM Using One Block RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .188
Circular Buffers, Shift Registers, and Delay Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .189
Fast Complex State Machines and Microsequencers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .191
Fast, Long Counters Using RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .192
Four-Port Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .194
Content-Addressable Memory (CAM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .194
Implementing Logic Functions Using Block RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .194
Fuzzy Pattern Matching Circuit Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .195
Mapping Logic into Block RAM Using MAP –bp Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .195
Waveform Storage, Function Tables, Direct Digital Synthesis (DDS) Using Block RAM . . . . . . . . . . . . . .196
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Appendix A: VHDL Instantiation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Appendix B: Verilog Instantiation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

Chapter 5: Using Configurable Logic Blocks (CLBs)

CLB Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
CLB Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
CLB Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Slices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
205
Slice Location Designations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .207
Slice Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Logic Cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Slice Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Main Logic Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

10 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Look-Up Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

Wide Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Carry and Arithmetic Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Storage Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .213
Timing Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .213
Distributed RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Shift Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Related Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

Chapter 6: Using Look-Up Tables as Distributed RAM

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Single-Port and Dual-Port RAMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .215
Data Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .215
Write Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .216
Read Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .216
Read During Write . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .216
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Distributed RAM in the CLB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Distributed RAM Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . 218
Compatibility with Other Xilinx FPGA Families. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Library Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Signal Ports. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
Clock — WCLK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Enable — WE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Address — A0, A1, A2, A3 (A4, A5, A6, A7) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Dual-Port Read Address — DPRA0, DPRA1, DPRA2, DPRA3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Data In — D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Data Out — O, SPO, and DPO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Inverting Control Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .222
Global Set/Reset — GSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .222
Global Write Enable — GWE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .222
Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Content Initialization — INIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .222
Placement Location — LOC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .223
Distributed RAM Design Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Xilinx CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .225
VHDL and Verilog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .226
Inferring Distributed RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .226
Instantiation Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .227
Verilog Instantiation Template Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .228
Wider Distributed RAM Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .229
Initialization in VHDL or Verilog Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .229
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

Spartan-3 Generation FPGA User Guide www.xilinx.com 11

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Shift Register Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Shift Register Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
LUT Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .232
Dynamic Length Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .232
Logic Cell Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .232
Registered Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .233
Slice Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .233
CLB Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .234
Library Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Initialization in VHDL and Verilog Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Port Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Clock — CLK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Data In — D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Clock Enable — CE (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Address — A3, A2, A1, A0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
Data Out — Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Data Out — Q15 (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Inverting Control Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
GSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Attributes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Content Initialization — INIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Location Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .238
Shift Register Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .239
Data Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .239
Shift Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .240
Dynamic Read Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .240
Static Read Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .241
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .241
Shift Register Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
VHDL Inference Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .242
Verilog Inference Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .242
Shift Register Submodules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Fully Synchronous Shift Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .244
Static-Length Shift Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .244
VHDL and Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .245
VHDL and Verilog Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .245
CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .247
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
Delay Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .248
Linear Feedback Shift Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .248
Gold Code Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .249
FIFOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .250
Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .250
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

12 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Dedicated Multiplexer Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . 253
Advantages of Dedicated Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .254
CLB Multiplexer Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
F5MUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .255
FiMUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .257
Naming Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .257
Dedicated Local Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .258
Mux Select Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .259
Implementation Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Wide-Input Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .259
Wide-Input Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .261
Timing Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
Programmable Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .264
Floorplanning Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .264
Related Uses of Multiplexers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Multiplexers and Three-State Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .265
Using Memory in Place of Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .265
Other Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .265
Designing with Multiplexers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .266
Verilog Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .267
VHDL Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .267
Library Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .267
Enable Signals in Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .268
Modeling Local Output Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .269
Submodules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .269
Port Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
Data In — DATA_I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
Control In — SELECT_I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
Data Out — DATA_O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
VHDL and Verilog Instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .270
VHDL and Verilog Submodules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .271
CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .274
Related Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

Chapter 9: Using Carry and Arithmetic Logic

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Carry and Arithmetic Logic Differences between Spartan-3 Generation Families . . . . . . . . . . . . . 277
Look-Ahead Carry Addition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Resource Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
MUXCY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281
Carry Chain Bypass and Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281
XORCY. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281

Spartan-3 Generation FPGA User Guide www.xilinx.com 13

UG331 (v1.7) August 19, 2010
 R

Carry Logic Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281

Connections within a Slice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281
Connections between Slices and CLBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .282
Multiplication Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Component and Pin Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
Specifications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Designing with the Carry and Arithmetic Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
Library Elements Using Carry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .294
Primitives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .295
XORCY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .296
MULT_AND . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .297
Emulating Virtex-II FPGA ORCY Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .298
Macros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .298
Using the CORE Generator System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Adder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .299
Accumulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .300
Comparator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .300
Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .300
Logic Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .301
Carry and Synthesis Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
MUX_STYLE Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .301
MULT_STYLE Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .302
Carry and Relative Location Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .302
Carry and Floorplanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
Wide Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .304
Sum of Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .306
Comparators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .306
Adders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .307
Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .308
Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .310
Optimizing Carry-Based Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .311
MULT_AND vs. MULT18X18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .312
MULT_AND vs. CLB Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .312
Other Types of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .313
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

Chapter 10: Using I/O Resources

IOB Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
I/O Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
Number of Resources per Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .318
Input-Only Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .318
Package Footprint Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .319
Summary of Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .319
Design Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
Library Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .320
Registered I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .322
Differential I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .322

14 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

IBUF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .322
IBUFG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .323
IBUFDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .323
OBUF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .323
OBUFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .323
IOBUF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .324
DDR and Adjustable Delay I/O Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .324
HDL Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .324
Architectural Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Input Delay Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .325
Programmable Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .325
Dynamic Delay in the Extended Spartan-3A Family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .326
Storage Element Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .328
Double-Data-Rate Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .330
Register Cascade Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .331
IDDR2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .331
ODDR2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .333
Pull-Up and Pull-Down Resistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .334
FPGA Pull-Up Resistor Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .335
Keeper Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .335
JTAG Boundary-Scan Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .336
SelectIO Signal Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Overview of I/O Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .336
Clamp Diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .336
LVTTL — Low-Voltage TTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
LVCMOS — Low-Voltage CMOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
PCI — Peripheral Component Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
GTL — Gunning Transceiver Logic Terminated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
GTL+ — Gunning Transceiver Logic Plus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
HSTL — High-Speed Transceiver Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
SSTL3 — Stub Series Terminated Logic for 3.3V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
SSTL2 — Stub Series Terminated Logic for 2.5V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
SSTL18 — Stub Series Terminated Logic for 1.8V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
LVDS — Low Voltage Differential Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
BLVDS — Bus LVDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
LVPECL — Low Voltage Positive Emitter Coupled Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
LDT — HyperTransport (formerly known as Lightning Data Transport) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
mini-LVDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
LVDS Extended — Extended Mode LVDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
RSDS — Reduced Swing Differential Signaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
TMDS — Transition Minimized Differential Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
PPDS — Point-to-Point Differential Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
I/O Standard Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .340
Specifying an I/O Standard with the IOSTANDARD Attribute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .343
Timing Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .344
LVCMOS/LVTTL Slew Rate Control and Drive Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .345
Simultaneously Switching Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .347
HSTL/SSTL VREF Reference Voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .348
Single-Ended I/O Termination Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .349
Differential I/O Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .351
On-Chip Differential Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .351
DCI Digitally Controlled Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .354
Supply Voltages for the IOBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355

Spartan-3 Generation FPGA User Guide www.xilinx.com 15

UG331 (v1.7) August 19, 2010
 R

Spartan-3A and Spartan-3A DSP FPGA Dual-Range VCCAUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .355

ESD Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .355
IOBs Organized into Banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .355
Single-Ended I/O Standard Bank Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .356
Differential I/O Standard Bank Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .360
I/O Banking Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .363
Using Large-Swing Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .363
Voltage Translators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .364
Open-Drain Interfacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .364
Voltage Clamps Using Internal Diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .365
Parasitic Leakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .365
I/O and Input-Only Pin Behavior During Power-On, Configuration, and User Mode . . . . . . . . . . . . . . .372
Behavior of Unused I/O Pins After Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .373
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

Chapter 11: Using Embedded Multipliers

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
Embedded Multiplier Resource Differences between Spartan-3 Generation Families . . . . . . . . 376
Two’s-Complement Signed Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
Location Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .377
Multiplier/Block RAM Routing Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .378
Optional Pipeline Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .378
Timing Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .379
Expanding Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
Cascading Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .380
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .381
Two Multipliers in a Single Primitive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Design Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
MULT_STYLE Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .386
Using the CORE Generator System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
System Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .388
MAC Cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .388
Spartan-3 Family Library Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Data Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .390
Multipliers in the Spartan-3 Generation Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .390
Alternative Applications to Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Shifter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .391
Magnitude Return . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .392
Two’s-Complement Return . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .392
Complex Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .393
Time Sharing in Matrix Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .393
Floating-Point Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .393
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Appendix A: Two's-Complement Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395

Chapter 12: Using Interconnect

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

16 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Interconnect Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .397

Switch Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Long Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .399
Hex Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .399
Double Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .400
Direct Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .400
Viewing Interconnect Details with FPGA Editor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Global Controls. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
STARTUP_SPARTAN3 Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .401
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402

Section II: Design Software

Chapter 13: Using ISE Design Tools

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Design Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Design Entry and Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .407
Hierarchical Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .408
Schematic Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .409
HDL Entry and Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .409
Constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .410
Design Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .410
Translating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .411
Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .412
Placing and Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .413
Bitstream Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .413
Design Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .413
Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .414
Static Timing Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .415
In-Circuit Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
ISE Development Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
Introduction to ISE Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
Design Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .416
Device Download . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .417
ISE Versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .417
Project Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .417
Project Navigator Main Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .418
Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .418
Sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
Source Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
ISE Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
Engineering Capture System (ECS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
HDL Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
Xilinx Synthesis Technology (XST) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .419
HDL Advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .420
Partner Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .420

Spartan-3 Generation FPGA User Guide www.xilinx.com 17

UG331 (v1.7) August 19, 2010
 R

Intellectual Property (IP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .420

CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .420
System Generator for DSP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .421
Embedded Development Kit and Platform Studio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .421
Clocking Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .421
Data2MEM Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .421
Automatic Implementation Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .421
Incremental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .422
Modular Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .422
Constraints Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .423
PlanAhead Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .423
FPGA Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .423
Interactive Timing Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .423
ISE Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .424
iMPACT Configuration Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .424
ChipScope Pro Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .424
Power Analysis Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .424
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425

Chapter 14: Using IP Cores

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
The CORE Generator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
Xilinx IP Solutions and the IP Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
LogiCORE Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .428
AllianceCORE Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .428
Candidate Core Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .429
Design Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .429
Xilinx Alliance Program Partner Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .429
SignOnce . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .429
Spartan-3 Generation IP Cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448

Chapter 15: Embedded Processing and Control Solutions

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
PicoBlaze Application Development Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
MicroBlaze Application Development Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Embedded Development Kit (EDK) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
Xilinx Platform Studio (XPS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
GNU Software Development Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
Hardware/Software Development Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
Board Support Packages (BSPs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
Operating Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .453
Processor Peripheral IP Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .454
Processor Peripherals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .454
Serial I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .454
Memory Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .454
Networking Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .454
In-Circuit Hardware Debugger Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .455
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455

18 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Section III: PCB Design Considerations

Chapter 16: Packages and Pinouts

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Differences in Pinouts Between Spartan-3 Generation FPGAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Pin Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
Pin Labeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .462
Differential Pair Labeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .462
Pinout Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .463
Pinout Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .463
Footprint Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .464
PartGen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .465
ISE Development System Pin Assignment Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .465
PlanAhead Design Analysis Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
Packages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
Pb-Free Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .466
Differences in Packages Between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .467
Selecting the Right Package Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .467
Package Thermal Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .467
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468

Chapter 17: Package Drawings

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
VQ100/VQG100 Very Thin QFP Package (pk012) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
CP132/CPG132 Chip Scale BGA Package (pk500) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
TQ144/TQG144 Thin QFP Package (pk009) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
PQ208/PQG208 QFP Package (pk007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
FT256/FTG256 Fine-Pitch Thin BGA Package (pk053) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
FG320/FGG320 Fine-Pitch BGA Package (pk071) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
FG400/FGG400 Fine-Pitch BGA Package (pk083) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
FG456/FGG456 Fine-Pitch BGA Package (pk034) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
FG484/FGG484 Fine-Pitch BGA Package (pk081) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
CS484/CSG484 Chip-Scale BGA Package (pk223) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
FG676/FGG676 Fine-Pitch BGA Package (pk035) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
FG900/FGG900 Fine-Pitch BGA Package (pk038) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
FG1156/FGG1156 Fine-Pitch BGA Package (pk039) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485

Chapter 18: Powering Spartan-3 Generation FPGAs

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
Differences between Spartan-3 Generation Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
Voltage Supplies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
VREF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .489
Power Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Voltage Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Power-On Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490

Spartan-3 Generation FPGA User Guide www.xilinx.com 19

UG331 (v1.7) August 19, 2010
 R

Supply Sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .490

Surplus ICCINT if VCCINT is Applied before VCCAUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .491
Ramp Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .491
Hot Swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
Configuration Data Retention, Brown-Out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
Saving Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
Saving Clock Routing Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .493
Power-Off Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .494
Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .495
Board Design and Signal Integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Simultaneously Switching Outputs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .495
Power Distribution System Design and Decoupling/Bypass Capacitors . . . . . . . . . . . . . . . . . . . . . . . . . . .496
No Internal Charge Pumps or Free-Running Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .496
Large-Swing Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
Related Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496

Chapter 19: Power Management Solutions

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
Extended Spartan-3A Family Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
Suspend Features and Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .499
Design Preparation for Suspend Mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .499
Entering Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .501
Exiting Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .503
PROG_B Programming Pin Always Overrides Suspend Mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .504
Suspend Mode Timing Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .504
Enable the Suspend Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .505
Via User Constraints File (UCF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .505
Via BitGen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .506
Define the I/O Behavior During Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .506
Single-Ended I/O Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .506
Differential I/O Standards. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .506
SUSPEND Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .507
Application State Retained during Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .507
Suspend Mode Wake-Up Timing Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .508
Wake-Up Timing Clock Source (sw_clk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .508
Switch Outputs from Suspend to Normal Behavior (sw_gts_cycle) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .509
Release Write Protect on Clocked Elements (sw_gwe_cycle) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .509
Dedicated Configuration Pins Unaffected During Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .510
SUSPEND Pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .510
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .510
SUSPEND Input Glitch Filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .510
Effect on FPGA Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511
Tie SUSPEND to GND if not Using Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511
AWAKE Pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511
General Behavior (Suspend Feature Disabled) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511
AWAKE Pin Behavior when Suspend Feature Enabled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511
Controlling Wake-Up from an External Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .512
JTAG Operations Allowed During Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .512
Post-Configuration CRC Limitations When Using Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .513
Suspend Mode Bitstream Generator Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .514
FPGA Voltage Requirements During Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .514

20 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Requirements During Suspend Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .515

Hibernate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
Forcing FPGA to Quiescent Current Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .515
Entering Hibernate State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .515
Extended Spartan-3A Family FPGA: Turn Off VCCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .516
Spartan-3E and Spartan-3 FPGAs: Maintain VCCO on I/O Banks Connected to Powered External Devices . . .516
Exiting Hibernate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .517
Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .518

Chapter 20: Using IBIS Models

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
IBIS Advantages over SPICE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
IBIS Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
Xilinx Support of IBIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
IBIS I/V and dV/dt Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
Ramp and dV/dt Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
IBISWriter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522

Chapter 21: Using Boundary-Scan and BSDL Files

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
Boundary-Scan Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
IEEE Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
Boundary-Scan Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
Boundary-Scan Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
BSDL Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
525
BSDL File Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .526
BSDL File Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .528
Using BSDLAnno for Post-Configuration Boundary-Scan Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .528
Software Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
iMPACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .529
SVF Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .529
J Drive Engine for IEEE 1532 Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .530
Using the BSCAN_SPARTAN3A Macro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .530
Related Materials and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531

Spartan-3 Generation FPGA User Guide www.xilinx.com 21

UG331 (v1.7) August 19, 2010
 R

22 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Preface

About This Guide

This user guide provides guidance on how customers can use the architectural features of
each platform in the Spartan®-3 generation: the Extended Spartan-3A family, which
includes the Spartan-3A, Spartan-3AN, and Spartan-3A DSP platforms, and the Spartan-3
and Spartan-3E families. By combining documentation for these families, similarities and
differences are easier to learn, and less material needs to be duplicated in multiple sources.
For an overview of how these platforms compare, see “Section I: Designing with Spartan-
3 Generation FPGAs”.
This user guide includes much of the information previously included in Module 2
(Functional Description) of the Spartan FPGA data sheets and in device application notes.
The data sheets should still be referenced for the platform-specific DC and Switching
Characteristics (located in Module 3) and the pinout information (located in Module 4). All
features of the Spartan-3E and Extended Spartan-3A family are described in this user
guide, but some differences in the Spartan-3 family, such as DCI or the clocking structure,
are discussed in Module 2 of the Spartan-3 FPGA data sheet or in the device application
notes.
Information on the configuration features of the Spartan-3 generation FPGAs is located in
UG332, the Spartan-3 Generation Configuration User Guide. Information on using the internal
SPI flash of the Spartan-3AN FPGAs is located in UG333, Spartan-3AN FPGA In-System
Flash User Guide. Together with the device specifications in the data sheets, these user
guides provide complete documentation on the Spartan-3 generation architecture.
Check for updates on xilinx.com at:
http://www.xilinx.com/support/documentation/spartan-3a.htm. To get an automatic
notification of any updates to this document, click the “Subscribe to Alerts” link on the top
of the page.

Guide Contents
This user guide contains the following chapters:
• “Section I: Designing with Spartan-3 Generation FPGAs”
• Chapter 1, “Overview”
• Chapter 2, “Using Global Clock Resources”
• Chapter 3, “Using Digital Clock Managers (DCMs)”
• Chapter 4, “Using Block RAM”
• Chapter 5, “Using Configurable Logic Blocks (CLBs)”
• Chapter 6, “Using Look-Up Tables as Distributed RAM”
• Chapter 7, “Using Look-Up Tables as Shift Registers (SRL16)”
• Chapter 8, “Using Dedicated Multiplexers”

Spartan-3 Generation FPGA User Guide www.xilinx.com 23

UG331 (v1.7) August 19, 2010
 R

Preface: About This Guide

• Chapter 9, “Using Carry and Arithmetic Logic”

• Chapter 10, “Using I/O Resources”
• Chapter 11, “Using Embedded Multipliers”
• Chapter 12, “Using Interconnect”
• “Section II: Design Software”
• Chapter 13, “Using ISE Design Tools”
• Chapter 14, “Using IP Cores”
• Chapter 15, “Embedded Processing and Control Solutions”
• “Section III: PCB Design Considerations”
• Chapter 16, “Packages and Pinouts”
• Chapter 17, “Package Drawings”
• Chapter 18, “Powering Spartan-3 Generation FPGAs”
• Chapter 19, “Power Management Solutions”
• Chapter 20, “Using IBIS Models”
• Chapter 21, “Using Boundary-Scan and BSDL Files”

Additional Resources
To find additional documentation, see the Xilinx website at:
http://www.xilinx.com/support/documentation/index.htm.
To search the Answer Database of silicon, software, and IP questions and answers, or to
create a technical support WebCase, see the Xilinx website at:
http://www.xilinx.com/support.

Conventions
This document uses the following conventions. An example illustrates each convention.

Typographical
The following typographical conventions are used in this document:

Convention Meaning or Use Example

Messages, prompts, and
Courier font program files that the system speed grade: - 100
displays
Literal commands that you enter
Courier bold ngdbuild design_name
in a syntactical statement
Commands that you select from
File → Open
Helvetica bold a menu
Keyboard shortcuts Ctrl+C

24 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Conventions

Convention Meaning or Use Example

Variables in a syntax statement
for which you must supply ngdbuild design_name
values
See the Development System
Italic font References to other manuals Reference Guide for more
information.
If a wire is drawn so that it
Emphasis in text overlaps the pin of a symbol, the
two nets are not connected.
An optional entry or parameter.
However, in bus specifications, ngdbuild [option_name]
Square brackets [ ]
such as bus[7:0], they are design_name
required.
A list of items from which you
Braces { } lowpwr ={on|off}
must choose one or more
Separates items in a list of
Vertical bar | lowpwr ={on|off}
choices
IOB #1: Name = QOUT’
Vertical ellipsis
IOB #2: Name = CLKIN’
. Repetitive material that has
.
. been omitted
.
.
.
Repetitive material that has allow block block_name loc1
Horizontal ellipsis …
been omitted loc2… locn;

Online Document
The following conventions are used in this document:

Convention Meaning or Use Example

See the section “Additional
Cross-reference link to a location Resources” for details.
Blue text
in the current document Refer to “Title Formats” in
Chapter 1 for details.
Go to http://www.xilinx.com
Blue, underlined text Hyperlink to a website (URL)
for the latest speed files.

Spartan-3 Generation FPGA User Guide www.xilinx.com 25

UG331 (v1.7) August 19, 2010
 R

Preface: About This Guide

26 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Section I: Designing with Spartan-3 Generation

FPGAs

“Overview”

“Using Global Clock Resources”

“Using Digital Clock Managers (DCMs)”

“Using Block RAM”

“Using Configurable Logic Blocks (CLBs)”

“Using Look-Up Tables as Distributed RAM”

“Using Look-Up Tables as Shift Registers (SRL16)”

“Using Dedicated Multiplexers”

“Using Carry and Arithmetic Logic”

“Using I/O Resources”

“Using Embedded Multipliers”

“Using Interconnect”

Spartan-3 Generation FPGA User Guide www.xilinx.com 27

UG331 (v1.7) August 19, 2010
 R

Chapter :

28 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 1

Overview
This chapter provides an overview of the Spartan®-3 generation platforms. Refer to the
links in Table 1-1 for more information.

Table 1-1: Spartan-3 Generation Platforms

Family Platform Product Information Technical Documentation

Spartan-3A DSP FPGAs www.xilinx.com/spartan3adsp www.xilinx.com/support/documentation/spartan-3a_dsp.htm

Extended
Spartan-3AN FPGAs www.xilinx.com/spartan3an www.xilinx.com/support/documentation/spartan-3an.htm
Spartan-3A
Spartan-3A FPGAs www.xilinx.com/spartan3a www.xilinx.com/support/documentation/spartan-3a.htm

Spartan-3E Spartan-3E FPGAs www.xilinx.com/spartan3e www.xilinx.com/support/documentation/spartan-3e.htm

Spartan-3 Spartan-3 FPGAs www.xilinx.com/spartan3 www.xilinx.com/support/documentation/spartan-3.htm

Introduction
The Spartan-3 generation of FPGAs includes the Extended Spartan-3A family
(Spartan-3A, Spartan-3AN, and Spartan-3A DSP platforms), along with the earlier
Spartan-3 and Spartan-3E families. These families of Field Programmable Gate Arrays
(FPGAs) are specifically designed to meet the needs of high volume, cost-sensitive
electronic applications, such as consumer products. The Spartan-3 generation includes 25
devices offering densities ranging from 50,000 to 5 million system gates, as shown in
Table 1-5 through Table 1-7.
The Spartan-3 platform was the industry’s first 90 nm FPGA, delivering more functionality
and bandwidth per dollar than was previously possible, setting new standards in the
programmable logic industry. The Spartan-3E platform builds on the success of the earlier
Spartan-3 platform by adding new features that improve system performance and reduce
the cost of configuration. The Extended Spartan-3A family builds on the success of the
earlier Spartan-3E platform by further enhancing configuration and reducing power to
provide the lowest total cost. The Spartan-3AN platform provides the additional benefits
of non-volatility and large amounts of on-board user flash. The Spartan-3A DSP platform
extends the density range and adds resources often required in digital signal processing
(DSP) applications.
Because of their exceptionally low cost, Spartan-3 generation FPGAs are ideally suited to a
wide range of consumer electronics applications, including broadband access, home
networking, display/projection, and digital television equipment.
The Spartan-3 generation FPGAs provide a superior alternative to mask-programmed
ASICs. FPGAs avoid the high initial cost, the lengthy development cycles, and the inherent
inflexibility of conventional ASICs. Also, FPGA programmability permits design upgrades
in the field with no hardware replacement necessary, an impossibility with ASICs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 29

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

Spartan-3 Generation Families

Extended Spartan-3A family:
• Lowest total cost
• Spartan-3A Platform
- Ideal for bridging, differential signaling, and memory interfacing
• Spartan-3A DSP Platform
- Higher density option in Extended Spartan-3A family
- DSP48A resources for digital signal processing (DSP) applications
• Spartan-3AN Platform
- Non-volatile
- Ideal for space-constrained applications
Spartan-3E family
Spartan-3 family

Table 1-2: Spartan-3 Generation Platform Selection

Spartan-3 Generation Platform
Application/Function Spartan-3 Spartan-3E Spartan-3A Spartan-3AN Spartan-3A DSP
FPGAs FPGAs FPGAs FPGAs FPGAs
Basic Design Characteristics
More than 1.5M system gates + +
500 or less I/O pins + + ++ ++ ++
More than 500 I/O pins +
Embedded Processing
32-bit MicroBlaze Processor + + ++ ++ ++
8-bit PicoBlaze Controller + + + + +
DDR SDRAM Memory Interfaces
DDR SDRAM + + ++ ++ ++
DDR2 SDRAM ++ + ++ ++ ++
Differential I/O
LVDS + ++ +++ +++ +++
RSDS + ++ +++ +++ +++
miniLVDS + ++ ++ ++
TMDS/PPDS + + +
PCI®/PCI Express® Interface
33 MHz PCI Interface + + ++ ++ ++
66 MHz PCI Interface + ++ ++ ++
PCI Express PIPE Interface + + + + +

30 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation Families

Table 1-2: Spartan-3 Generation Platform Selection (Cont’d)

Spartan-3 Generation Platform
Application/Function Spartan-3 Spartan-3E Spartan-3A Spartan-3AN Spartan-3A DSP
FPGAs FPGAs FPGAs FPGAs FPGAs
Power Management + + +
I/O Capabilities
Hot Swap + + ++ ++ ++
High Output Drive Current + ++ ++ ++
Programmable Input Delay + ++ +++ +++ +++
3.3V-only Applications + + ++ ++ ++
Clocking Resources
Digital Clock Managers (DCMs) + ++ ++ ++ ++
Low-Skew Global Clocks + ++ ++ ++ ++
FPGA Configuration
Platform Flash PROM + + + + +
SPI Flash Configuration + + + +
Parallel Flash Configuration + + + +
MultiBoot + + + +
Low-Cost Design Protection + ++ +++ ++
Non-Volatile +
Integrated User Flash +
Digital Signal Processing (DSP)
18x18 Hardware Multipliers + ++ ++ ++ +++
DSP48A +
Block RAM Registers + + ++ ++ +++

Notes:
1. + = supported, ++ = better, +++ = best.

Spartan-3 Generation FPGA User Guide www.xilinx.com 31

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

Extended Spartan-3A Family Features

• Very low cost, high-performance logic solution for high-volume consumer-oriented
applications
• Proven advanced 90-nanometer process technology
• Dual-range VCCAUX supply at 2.5V or 3.3V simplifies the power supply design and
eliminates one power rail
• Suspend mode reduces system power consumption
• Retains all design state and FPGA configuration data
• Activated with SUSPEND pin
• FPGA drops to minimal quiescent power
• I/Os have user-controlled behavior
• Quick wake-up time
• AWAKE pin indicates present status
• 3.3V ±10% supply compatibility
• 4.6V maximum input voltage
• Hot-swap compliance
• FPGA I/O can be driven externally before VCCO powers up without damage to
the device and without disturbing the external bus
• FPGA does not drive out before or during power-up sequence except for
dedicated pins
• Multi-voltage, multi-standard SelectIO™ interface pins
• Up to 519 I/O pins or 227 differential signal pairs
• LVCMOS, LVTTL, HSTL, and SSTL single-ended signal standards
• 3.3V, 2.5V, 1.8V, 1.5V, and 1.2V signaling
• Up to 24 mA output drive
• 622+ Mb/s data transfer rate per I/O
• True LVDS, RSDS, mini-LVDS, PPDS, HSTL/SSTL differential I/O
• Double Data Rate (DDR) support with clock alignment
• DDR/DDR2 SDRAM support up to 400 Mb/s
• Programmable input delays for finer timing control
• Abundant, flexible logic resources
• Densities up to 53,712 logic cells
• Optional SRL16 shift register or distributed RAM support
• Efficient wide multiplexers, wide logic
• Fast look-ahead carry logic
• Dedicated 18 x 18 multipliers with optional pipeline for higher performance
• IEEE 1149.1/1532 JTAG programming/debug port
• Hierarchical SelectRAM™ memory architecture
• Up to 2,268 Kbits of fast block RAM with byte write enables for efficient use in
processor applications
• Up to 373 Kbits of efficient distributed RAM

32 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Features

• Up to eight Digital Clock Managers (DCMs)

• Clock skew elimination (delay locked loop)
• Frequency synthesis, multiplication, division
• High-resolution phase shifting
• Wide frequency range (5 MHz to over 300 MHz)
• Eight global clocks, plus abundant low-skew routing
• Eight additional clocks per each half of the device
• Additional clock inputs for pinout flexibility and differential clocks
• Configuration interface to low-cost Xilinx Platform Flash with JTAG
• Configuration interface to industry-standard PROMs
• Low-cost, space-saving SPI serial Flash PROM
• x8 or x8/x16 parallel NOR Flash PROM
• Configuration watchdog timer automatically recovers from configuration errors
• Unique ID (Device DNA) in each device useful for copy protection algorithms
• Device DNA authentication restricts copying
• MultiBoot automatic reconfiguration between two files
• Complete Xilinx ISE® and WebPACK™ development system support
• Low-cost Starter Kit development systems and advanced demo boards
• 32-bit MicroBlaze™ and 8-bit PicoBlaze™ embedded processor cores
• Fully compliant 32-/64-bit 66 MHz PCI support
• PCI Express PIPE endpoint and other IP cores
• Supported by major EDA partners
• Low-cost QFP and BGA packaging options
• Common footprints support easy density migration within each platform (except
designs using the FT256 package)
• Pb-free (RoHS) packaging options
• Automotive XA platform variants

Spartan-3AN Platform Additional Features

• Integrated robust configuration memory
• Saves board space
• Improves ease-of-use
• Simplifies design
• Reduces support issues
• Plentiful amounts of non-volatile memory available to the user
• Up to 11+ Mb available
• MultiBoot support
• Embedded processing and code shadowing
• Scratchpad memory
• Robust 100K Flash memory program/erase cycles per page
• 20 years Flash memory data retention

Spartan-3 Generation FPGA User Guide www.xilinx.com 33

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

• Security features provide bitstream anti-cloning protection

• Buried configuration interface enhances Device DNA authentication
• Flash memory sector protection and lockdown

Spartan-3A DSP Platform Additional Features

• Optimized for low-cost DSP systems
• High logic capacity, 33K to 47K look-up tables (LUTs)
• Increased block RAM of 85 to 126 blocks and 1.5 to 2.3 Mbits of memory
• High-performance DSP48A blocks
• Based on Virtex®-4 FPGA DSP block architecture
• Full multiply-accumulate functionality
• Integrated 48-bit post adder
• Integrated 18-bit pre-adder for symmetric FIR filters
• Independent routing
• 250 MHz operation
• Improved block RAM
• Internal output register
• 250 MHz operation

Spartan-3 Generation Resources

Table 1-3 through Table 1-7 show the number of resources available in each member of the
Spartan-3A DSP, Spartan-3AN, Spartan-3A, Spartan-3E, and Spartan-3 platforms.
Note: By convention, 1K bits is equivalent to 1,024 bits.
Table 1-3: Summary of Spartan-3A DSP FPGA Attributes
CLB Array
Equivalent (One CLB = Four Slices) Block Maximum
System Distributed Maximum
Device Logic RAM DSP48As DCMs Differential
Gates Total Total RAM Bits User I/O
Cells Rows Columns Bits I/O Pairs
CLBs Slices

XC3SD1800A 1800K 37,440 88 48 4,160 16,640 260K 1,512K 84 8 519 227

XC3SD3400A 3400K 53,712 104 58 5,968 23,872 373K 2,268K 126 8 469 213

Table 1-4: Summary of Spartan-3AN FPGA Attributes

Equivalent Block Maximum
System Distributed Dedicated Maximum In-System
Device Logic CLBs Slices RAM DCMs Differential
Gates RAM Bits Multipliers User I/O Flash bits
Cells Bits I/O Pairs

XC3S50AN 50K 1,584 176 704 11K 54K 3 2 108 50 1M

XC3S200AN 200K 4,032 448 1,792 28K 288K 16 4 195 90 4M

XC3S400AN 400K 8,064 896 3,584 56K 360K 20 4 311 142 4M

XC3S700AN 700K 13,248 1472 5,888 92K 360K 20 8 372 165 8M

XC3S1400AN 1400K 25,344 2816 11,264 176K 576K 32 8 502 227 16M

34 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation Resources

Table 1-5: Summary of Spartan-3A FPGA Attributes

CLB Array
(One CLB = Four Slices) Block Maximum
System Equivalent Distributed Dedicated Maximum
Device RAM DCMs Differential
Gates Logic Cells Total Total RAM Bits Multipliers User I/O
Rows Columns Bits I/O Pairs
CLBs Slices

XC3S50A 50K 1,584 16 12 176 704 11K 54K 3 2 144 64

XC3S200A 200K 4,032 32 16 448 1,792 28K 288K 16 4 248 112

XC3S400A 400K 8,064 40 24 896 3,584 56K 360K 20 4 311 142

XC3S700A 700K 13,248 48 32 1472 5,888 92K 360K 20 8 372 165

XC3S1400A 1400K 25,344 72 40 2816 11,264 176K 576K 32 8 502 227

Table 1-6: Summary of Spartan-3E FPGA Attributes

CLB Array
(One CLB = Four Slices) Block Maximum
System Equivalent Distributed Dedicated Maximum
Device RAM DCMs Differential
Gates Logic Cells Total Total RAM Bits Multipliers User I/O
Rows Columns Bits I/O Pairs
CLBs Slices

XC3S100E 100K 2,160 22 16 240 960 15K 72K 4 2 108 40

XC3S250E 250K 5,508 34 26 612 2,448 38K 216K 12 4 172 68

XC3S500E 500K 10,476 46 34 1,164 4,656 73K 360K 20 4 232 92

XC3S1200E 1200K 19,512 60 46 2,168 8,672 136K 504K 28 8 304 124

XC3S1600E 1600K 33,192 76 58 3,688 14,752 231K 648K 36 8 376 156

Table 1-7: Summary of Spartan-3 FPGA Attributes

CLB Array
(One CLB = Four Slices) Maximum
System Equivalent Distributed Block Dedicated Maximum
Device DCMs Differential
Gates Logic Cells Total Total RAM Bits RAM Bits Multipliers User I/O
Rows Columns I/O Pairs
CLBs Slices

XC3S50 50K 1,728 16 12 192 768 12K 72K 4 2 124 56

XC3S200 200K 4,320 24 20 480 1,920 30K 216K 12 4 173 76

XC3S400 400K 8,064 32 28 896 3,584 56K 288K 16 4 264 116

XC3S1000 1000K 17,280 48 40 1,920 7,680 120K 432K 24 4 391 175

XC3S1500 1500K 29,952 64 52 3,328 13,312 208K 576K 32 4 487 221

XC3S2000 2000K 46,080 80 64 5,120 20,480 320K 720K 40 4 565 270

XC3S4000 4000K 62,208 96 72 6,912 27,648 432K 1,728K 96 4 633 300

XC3S5000 5000K 74,880 104 80 8,320 33,280 520K 1,872K 104 4 633 300

Spartan-3 Generation FPGA User Guide www.xilinx.com 35

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

Architectural Overview
The Spartan-3 generation architecture consists of five fundamental programmable
functional elements:
• Configurable Logic Blocks (CLBs) contain flexible Look-Up Tables (LUTs) that
implement logic plus storage elements used as flip-flops or latches. CLBs perform a
wide variety of logical functions as well as store data.
• Input/Output Blocks (IOBs) control the flow of data between the I/O pins and the
internal logic of the device. IOBs support bidirectional data flow plus 3-state
operation. Supports a variety of signal standards, including several high-performance
differential standards. Double Data-Rate (DDR) registers are included.
• Block RAM provides data storage in the form of 18-Kbit dual-port blocks.
• Multiplier Blocks accept two 18-bit binary numbers as inputs and calculate the
product. The Spartan-3A DSP platform includes special DSP multiply-accumulate
blocks.
• Digital Clock Manager (DCM) Blocks provide self-calibrating, fully digital solutions
for distributing, delaying, multiplying, dividing, and phase-shifting clock signals.
These elements are organized as shown in Figure 1-1, using the Spartan-3A FPGA array as
an example. A dual ring of staggered IOBs surrounds a regular array of CLBs in the
Spartan-3 and Extended Spartan-3A family. The Spartan-3E family has a single ring of
inline IOBs. Each block RAM column consists of several 18-Kbit RAM blocks. Each block
RAM is associated with a dedicated multiplier. The DCMs are positioned with two at the
top and two at the bottom of the device, plus additional DCMs on the sides for the larger
devices.
The Spartan-3 generation features a rich network of traces that interconnect all five
functional elements, transmitting signals among them. Each functional element has an
associated switch matrix that permits multiple connections to the routing.

36 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Configuration

IOBs

CLB

Block RAM

Multiplier
DCM

IOBs
OBs
DCM
Block RAM / Multiplier

CLBs
IOBs

IOBs

DCM

IOBs
DS312-1_01_032606

Figure 1-1: Spartan-3A Platform Architecture

Configuration
Spartan-3 generation FPGAs are programmed by loading configuration data into robust,
reprogrammable, static CMOS configuration latches (CCLs) that collectively control all
functional elements and routing resources. The FPGA’s configuration data is stored
externally in a PROM or some other non-volatile medium, either on or off the board. The
Spartan-3AN platform contains its own internal SPI flash configuration memory. After
applying power, the configuration data is written to the FPGA using one of several
different modes:
• Master Serial from a Xilinx Platform Flash PROM
• Serial Peripheral Interface (SPI) from an industry-standard SPI serial Flash
• Spartan-3E and Extended Spartan-3A family FPGAs only
• Byte Peripheral Interface (BPI) from an industry-standard x8 or x8/x16 parallel NOR
Flash
• Spartan-3E and Extended Spartan-3A family FPGAs only
• Slave Serial, typically downloaded from a processor
• Slave Parallel, typically downloaded from a processor
• Boundary Scan (JTAG), typically downloaded from a processor or system tester

Spartan-3 Generation FPGA User Guide www.xilinx.com 37

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

I/O Capabilities
The Spartan-3 generation SelectIO interface supports many popular single-ended and
differential standards, as shown in Table 1-8 and Table 1-9. Table 1-10 through Table 1-14
show the number of user I/Os as well as the number of differential I/O pairs available for
each device/package combination for the Extended Spartan-3A family, Spartan-3E, and
Spartan-3 families, respectively. Some of the user I/Os are unidirectional input-only pins
as indicated in the tables. See “LVCMOS/LVTTL Slew Rate Control and Drive Strength”
for specific drive strengths supported.

Table 1-8: Single-Ended I/O Standards

Spartan-3 Spartan-3E Extended Spartan-3A
Standard VCCO Class
FPGAs FPGAs FPGAs
1.2V - up to 6 mA 2 mA up to 6 mA
1.5V - up to 12 mA up to 6 mA up to 12 mA
LVCMOS 1.8V - up to 16 mA up to 8 mA up to 16 mA
2.5V - up to 24 mA up to 12 mA up to 24 mA
3.3V - up to 24 mA up to 16 mA up to 24 mA
LVTTL 3.3V - up to 24 mA up to 16 mA up to 24 mA
3.0V - √ √ √
PCI33
3.3V - √ √ √
3.0V - √ √
PCI66
3.3V - √ √
I √ √ √
1.8V
II √ √
I √ √ √
SSTL 2.5V
II √ √
I √
3.3V
II √
I √ √
1.5V
III √ √
HSTL I √ √ √
1.8V II √ √
III √ √ √
- - √
GTL
- Plus √
DCI option - - √

38 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

I/O Capabilities

Table 1-9: Differential I/O Standards

Spartan-3 Spartan-3E Extended Spartan-3A
Standard VCCO
FPGAs FPGAs Family FPGAs
2.5V √ √ √
LVDS
3.3V √
BLVDS 2.5V √ √ √
2.5V √ √
MINI_LVDS
3.3V √
2.5V √ √ √
LVPECL
3.3V √
2.5V √ √ √
RSDS
3.3V √
2.5V
TMDS
3.3V √
2.5V √
PPDS
3.3V √
LDT 2.5V √
LVDSEXT 2.5V √
DIFF_SSTL - √ √ √
DIFF_HSTL - √ √ √
DIFF_TERM - √ √

Table 1-10: Spartan-3A FPGA DSP Available User I/Os and Differential (Diff) I/O Pairs
CS484 FG676
Device CSG484 FGG676

User Diff User Diff

309 140 519 227

XC3SD1800A
(56) (78) (110) (131)

309 140 469 213

XC3SD3400A
(56) (78) (60) (117)

Notes:
1. The number in bold indicates the maximum number of I/O and input-only pins. The number in italics
indicates the number of input-only pins. The differential (Diff) input-only pin count includes both
differential pairs on input-only pins and differential pairs on I/O pins within I/O banks that are
restricted to differential inputs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 39

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

Table 1-11: Spartan-3AN Available User I/Os and Differential (Diff) I/O Pairs
TQG144 FTG256 FGG400 FGG484 FGG676
Device
User Diff User Diff User Diff User Diff User Diff

108 50
XC3S50AN - - - - - - - -
(7) (24)

195 90
XC3S200AN - - - - - - - -
(35) (50)
311 142
XC3S400AN - - - - - - - -
(63) (78)

372 165
XC3S700AN - - - - - - - -
(84) (93)

502 227
XC3S1400AN - - - - - - - -
(94) (131)

Notes:
1. The number in bold indicates the maximum number of I/O and input-only pins. The number in italics indicates the number of
input-only pins. The differential (Diff) input-only pin count includes both differential pairs on input-only pins and differential pairs
on I/O pins within I/O banks that are restricted to differential inputs.
2. Spartan-3AN FPGAs are available in Pb-free packaging options. The Pb-free packages include a ‘G’ character in the ordering code.
Leaded (non-Pb-free) packages might be available for selected devices, with the same pinout and without the ‘G’ in the ordering
code. Contact Xilinx sales for more information.

40 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

I/O Capabilities

Table 1-12: Spartan-3A FPGA Available User I/Os and Differential (Diff) I/O Pairs
VQ100 TQ144 FT256 FG320 FG400 FG484 FG676
Device VQG100 TQG144 FTG256 FGG320 FGG400 FGG484 FGG676

User Diff User Diff User Diff User Diff User Diff User Diff User Diff

68 60 108 50 144 64
XC3S50A - - - - - - - -
(13) (24) (7) (24) (32) (32)

68 60 195 90 248 112

XC3S200A - - - - - - - -
(13) (24) (35) (50) (56) (64)

195 90 251 112 311 142

XC3S400A - - - - - - -
(35) (50) (59) (64) (63) (78)
161 74 311 142 372 165
XC3S700A - - - - - - -
(13) (36) (63) (78) (84) (93)

161 74 375 165 502 227

XC3S1400A - - - - - - -
(13) (36) (87) (93) (94) (131)

Notes:
1. The number in bold indicates the maximum number of I/O and input-only pins. The number in italics indicates the number of
input-only pins. The differential (Diff) input-only pin count includes both differential pairs on input-only pins and differential pairs
on I/O pins within I/O banks that are restricted to differential inputs.

Table 1-13: Spartan-3E FPGA Available User I/Os and Differential (Diff) I/O Pairs
VQ100 CP132 TQ144 PQ208 FT256 FG320 FG400 FG484
Device VQG100 CPG132 TQG144 PQG208 FTG256 FGG320 FGG400 FGG484

User Diff User Diff User Diff User Diff User Diff User Diff User Diff User Diff

66 30 83 35 108 40
XC3S100E - - - - - - - - - -
(7) (2) (11) (2) (28) (4)
66 30 92 41 108 40 158 65 172 68
XC3S250E - - - - - -
(7) (2) (7) (2) (28) (4) (32) (5) (40) (8)

66 30 92 41 158 65 190 77 232 92

XC3S500E - - - - - -
(7) (2) (7) (2) (32) (5) (41) (8) (56) (12)
190 77 250 99 304 124
XC3S1200E - - - - - - - - - -
(40) (8) (56) (12) (72) (20)
250 99 304 124 376 156
XC3S1600E - - - - - - - - - -
(56) (12) (72) (20) (82) (21)

Notes:
1. The number in bold indicates the maximum number of I/O and input-only pins. The number in italics indicates the number of
input-only pins.

Spartan-3 Generation FPGA User Guide www.xilinx.com 41

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

Table 1-14: Spartan-3 FPGA Available User I/Os and Differential (Diff) I/O Pairs
VQ100 TQ144 PQ208 FT256 FG320 FG456 FG676 FG900
Device VQG100 TQG144 PQG208 FTG256 FGG320 FGG456 FGG676 FGG900
User Diff User Diff User Diff User Diff User Diff User Diff User Diff User Diff
XC3S50 63 29 97 46 124 56 - - - - - - - - - -
XC3S200 63 29 97 46 141 62 173 76 - - - - - - - -
XC3S400 - - 97 46 141 62 173 76 221 100 264 116 - - - -
XC3S1000 - - - - - - 173 76 221 100 333 149 391 175 - -
XC3S1500 - - - - - - - - 221 100 333 149 487 221 - -
XC3S2000 - - - - - - - - - - 333 149 489 221 565 270
XC3S4000 - - - - - - - - - - - - 489 221 633 300
XC3S5000 - - - - - - - - - - - - 489 221 633 300

Package Marking
Figure 1-2 provides a top marking example for a Spartan-3A FPGA in the quad-flat
packages. Figure 1-3 shows the top marking for a Spartan-3A FPGA in a BGA package. The
markings for the BGA packages are nearly identical to those for the quad-flat packages,
except that the marking is rotated with respect to the ball A1 indicator.
On Spartan-3E and Extended Spartan-3A family FPGAs, the “5C” and “4I” part
combinations can be dual marked as “5C/4I”.

Mask Revision Code

R
Fabrication Code
R

SPARTAN Process Technology

Device Type XC3S50ATM
Package TQ144AGQ0625 Date Code
D1234567A
Speed Grade 4C Lot Code
Temperature Range

Pin P1 DS529-1_03_080406

Figure 1-2: Spartan-3A FPGA QFP Package Marking Example

Mask Revision Code

BGA Ball A1 R

R
Fabrication Code
SPARTAN Process Code
Device Type XC3S50ATM
Package FT256 AGQ0625 Date Code
D1234567A
4C Lot Code
Speed Grade
Temperature Range
DS529-1_02_021206

Figure 1-3: Spartan-3A FPGA BGA Package Marking Example

42 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Ordering Information

Ordering Information
Spartan-3 generation FPGAs are available in both standard and Pb-free packaging options
for most device/package combinations. The Pb-free packages include a ‘G’ character in the
ordering code. The automotive device part numbers begin with XA instead of XC, and the
automotive temperature ranges include both the I Industrial range and the Q Automotive
range between -40C and +125C.
Figure 1-4 shows an example of the part ordering code. The Industrial Temperature Range
is available exclusively for the Standard (-4) Speed Grade. See Table 1-10 through
Table 1-14 for specific part/package combinations, and see the XA data sheets for specific
automotive ordering codes available.

Example: XC3S50A -4 FT G 256 C

Device Type Temperature Range:
C = Commercial (TJ = 0oC to 85oC)
Speed Grade I = Industrial (TJ = -40oC to 100oC)
-4: Standard Performance Q = Automotive (TJ = -40oC to 125oC)
-5: High Performance (Commercial only) Number of Pins
Package Type Pb-free
UG331-c1_04_122208

Figure 1-4: Spartan-3A FPGA Ordering Example

Spartan-3 Generation FPGA User Guide www.xilinx.com 43

UG331 (v1.7) August 19, 2010
 R

Chapter 1: Overview

44 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 2

Using Global Clock Resources

Summary
This chapter describes how to take advantage of the Spartan®-3 generation global clock
resources, including the dedicated clock inputs, buffers, and routing. The clocking
infrastructure provides a series of low-capacitance, low-skew interconnect lines well-
suited to carrying high-frequency signals throughout the FPGA, minimizing clock skew
and improving performance, and should be used for all clock signals. Third-party
synthesis tools, and Xilinx synthesis and implementation tools, automatically use these
resources for high-fanout clock signals.
This chapter focuses on the global clock resources found in all Spartan-3 generation
platforms, and the quadrant clock resources found in the Spartan-3E and Extended
Spartan-3A families. The clock routing can be used in conjunction with the DCMs, which
are discussed in more detail in Chapter 3, “Using Digital Clock Managers (DCMs).” For
information on the special clock inputs used for configuration (CCLK) and Boundary-Scan
(TCK), see UG332, Spartan-3 Generation Configuration User Guide.

Introduction
Each Spartan-3 generation FPGA offers eight high-speed, low-skew global clock resources
to optimize performance. These resources are used automatically by the Xilinx tools. Even
if the clock rate is relatively slow, it is still important to use the global routing resources to
eliminate any potential for timing hazards. It is important to understand how to define and
best take advantage of these resources.

Global Clock Resource Differences between Spartan-3 Generation

Families
The Spartan-3E and Extended Spartan-3A family of FPGAs have identical global clock
resources, with eight global clock inputs and an additional eight clocks on the left and right
sides of the device. The original Spartan-3 family offers only the eight global clock inputs.
Although the clock resources and routing are similar, there will be timing differences
between each platform and between different densities within a platform. This chapter
focuses on the architecture of the Spartan-3E and Extended Spartan-3 families. The
Spartan-3 family offers a simpler set of dedicated clock inputs and routing – for details, see
the Spartan-3 FPGA Family Data Sheet.

Spartan-3 Generation FPGA User Guide www.xilinx.com 45

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

Global Clock Resources

The global clock resources consist of three connected components: GCLK Global Clock
input pads, BUFGMUX Global Clock Multiplexers, and Global Clock routing. See
Figure 2-1.

GCLK BUFGMUX
Pad Clocks
Global
DCM Routing

Double
Lines
UG331_c2_01_100209

Figure 2-1: Overall View of Clock Connections

The primary clock path is shown with bold lines, with a dedicated clock pad (GCLK)
driving a global clock buffer (BUFGMUX) that connects through global routing resources
to clock inputs on flip-flops and other clocked elements. The GCLK pads can be used as
general-purpose I/O, and include the LHCLK and RHCLK inputs described later. A DCM
can be inserted into the path between the clock pad and clock buffer to manipulate the
clock, or the DCM can acquire the clock signal from general-purpose resources. The
BUFGMUX can multiplex between two clock sources or be used as a simple BUFG clock
buffer. The clock buffer can only drive the clock routing resources, which in turn can only
drive clock inputs. However, clock inputs on flip-flops can also come from general-
purpose routing, although their use is not recommended due to higher skew.

Clocking Infrastructure
The detailed Spartan-3E and Extended Spartan-3A family clocking infrastructure is shown
in Figure 2-2.

46 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Global Clock Resources

Global Clock Inputs

GCLK11 GCLK10 GCLK9 GCLK8
4
GCLK7 GCLK6 GCLK5 GCLK4
4
BUFGMUX Clock Line
pair
X1Y10 X1Y11 X2Y10 X2Y11 in Quadrant
BUFGMUX DCM 4 4 DCM
Top, Left Top, Right
LHCLK6 LHCLK7

H G F E

RHCLK3 RHCLK2
X0Y9

H H

X3Y9 X3Y8
Top Left Top Right
Quadrant (TL) Quadrant (TR)
X0Y8

8 8
4
G G
2 • • 2

Top Spine
2 2
8 8
DCM (1)
DCM
Left, Bottom 8 8 (1)
Right, Bottom
• •
2 2
2
2
LHCLK2 LHCLK3 LHCLK4 LHCLK5

RHCLK1 RHCLK0 RHCLK7 RHCLK6

Left-Half Clock Inputs

X3Y7
X0Y6 X0Y7

F F

Right-Half Clock Inputs

• •

X3Y6
E Note 3 Note 4 E
Left Spine 8 8 Horizontal Spine 8 8 Right Spine
D D
X0Y5

X3Y5 X3Y4
Note 3 Note 4

• •
X0Y4

C C
Bottom Spine

2 8 8 2
2 2
8 8
DCM
Left, Top
(1) • • DCM (1)
Right, Top

2 2
2
• • 2
LHCLK0 LHCLK1

RHCLK5 RHCLK4
X3Y3
B B
X0Y2 X0Y3

X3Y2
Bottom Left 8 4 8 Bottom Right
A Quadrant (BL) Quadrant (BR) A
4 4
D C B A
DCM DCM
Bottom, Left X1Y0 X1Y1 X2Y0 X2Y1 Bottom, Right

4
4 GCLK3 GCLK2 GCLK1 GCLK0

GCLK15 GCLK14 GCLK13 GCLK12

Global Clock Inputs UG331_c4_02_011209

Notes:
1. The diagram presents electrical connectivity. The diagram locations do not necessarily match the physical location on the device,
although the coordinate locations shown are correct.
2. Number of DCMs and locations of these DCM varies for different device densities. See Table 2-1.
3. See Figure 2-13a, which shows how the eight clock lines are multiplexed on the left-hand side of the device.
4. See Figure 2-13b, which shows how the eight clock lines are multiplexed on the right-hand side of the device.
5. For best direct clock inputs to a particular clock buffer, not a DCM, see Table 2-7.
6. For best direct clock inputs to a particular DCM, not a BUFGMUX, see Chapter 3, “Using Digital Clock Managers (DCMs).”

Figure 2-2: Spartan-3E and Extended Spartan-3A Family Internal Quadrant-Based Clock Structure

Spartan-3 Generation FPGA User Guide www.xilinx.com 47

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

Table 2-1: Spartan-3E and Extended Spartan-3A Family DCM Location Designations
Top, Top, Right, Right, Bottom, Bottom, Left, Left,
Left Right Bottom Top Right Left Top Bottom
Spartan-3A DSP FPGAs
XC3SD1800A X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1
XC3SD3400A X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1
Spartan-3A/3AN FPGAs
XC3S50A/AN X0Y0 X1Y0 N/A N/A N/A N/A N/A N/A
XC3S200A/AN X0Y1 X1Y1 N/A N/A X1Y0 X0Y0 N/A N/A
XC3S400A/AN X0Y1 X1Y1 N/A N/A X1Y0 X0Y0 N/A N/A
XC3S700A/AN X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1
XC3S1400A/AN X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1
Spartan-3E FPGAs
XC3S100E N/A X0Y1 N/A N/A X0Y0 N/A N/A N/A
XC3S250E X0Y1 X1Y1 N/A N/A X1Y0 X0Y0 N/A N/A
XC3S500E X0Y1 X1Y1 N/A N/A X1Y0 X0Y0 N/A N/A
XC3S1200E X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1
XC3S1600E X1Y3 X2Y3 X3Y1 X3Y2 X2Y0 X1Y0 X0Y2 X0Y1

Clock Inputs
Clock pins accept external clock signals and connect directly to DCMs and BUFGMUX
elements. Clock pins can also be used as general-purpose I/Os. Each Spartan-3E and
Extended Spartan-3A family FPGA has:
• 16 Global Clock inputs (GCLK0 through GCLK15) located along the top and bottom
edges of the FPGA
• 8 Right-Half Clock inputs (RHCLK0 through RHCLK7) located along the right edge
• 8 Left-Half Clock inputs (LHCLK0 through LHCLK7) located along the left edge
Clock input pins are used automatically when external signals drive clock buffers. The
user can specify a particular pin using a LOC constraint in order to force a clock onto the
left or right regional clocks, or to force a clock into a particular clock buffer and then into a
desired clock routing resource. Table 2-2, page 49 through Table 2-4, page 51 show the
clock inputs for each package with the Extended Spartan-3A family, Spartan-3E, and
Spartan-3 families, respectively.

Extended Spartan-3A Family Clock Inputs

The Extended Spartan-3A family clock inputs are all on bidirectional I/O pins, and none
are shared with configuration functions. The VQ100 package offers only 23 global clock
inputs. The XC3S50A/AN in the TQ144 package has only six global clock inputs on the
bottom edge, with GCLK12 and GCLK13 not available in the package.

48 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Inputs

Table 2-2: Global Clock Input Pads for Extended Spartan-3A Family FPGAs
Pad Bank VQ100 TQ144 FT256 FG320 FG400 CS484 FG484 FG676
GCLK0 2 P43 P57 N9 U10 Y11 U12 AA12 Y14
GCLK1 2 P44 P59 P9 T10 V11 V12 AB12 AA14
GCLK2 2 N/A P58 R9 V11 U11 AB13 V12 AF14
GCLK3 2 N/A P60 T9 U11 V12 AA14 U12 AE14
GCLK4 0 P83 P124 C10 B10 D11 E12 C12 K14
GCLK5 0 P84 P126 D9 C9 E11 F11 E12 J14
GCLK6 0 P85 P125 C9 A10 A10 A9 A12 B14
GCLK7 0 P86 P127 A9 B9 C10 B9 A11 A14
GCLK8 0 P88 P129 C8 A8 D10 F10 B11 F13
GCLK9 0 P89 P131 D8 B7 E10 E11 C11 G13
GCLK10 0 N/A P130 A8 B8 A9 A8 D11 B13
GCLK11 0 P90 P132 B8 C8 A8 B8 E11 C13
GCLK12 2 N/A N/A R7(1) U8 W9 Y11 U11 AA13
GCLK13 2 N/A N/A T7(1) V8 Y9 Y10 V11 Y13
GCLK14 2 P40 P54 P8 U9 V10 AA12 W12 AF13
GCLK15 2 P41 P55 T8 V9 W10 AB12 Y12 AE13
LHCLK0 3 P9 P12 G2 H3 J1 L6 L5 N6
LHCLK1 3 P10 P13 H1 J3 K2 M5 L3 N7
LHCLK2 3 P12 P15 H3 J2 K3 K1 K1 P1
LHCLK3 3 P13 P16 J3 J1 L3 L1 L1 P2
LHCLK4 3 N/A P18 J2 J4 K4 L3 M1 P4
LHCLK5 3 N/A P20 J1 K5 L5 M2 M2 P3
LHCLK6 3 P15 P19 K3 K2 L1 M6 M3 N9
LHCLK7 3 P16 P21 K1 K3 M1 N7 M4 P10
RHCLK0 1 P59 P83 K15 L18 M19 N18 M22 P21
RHCLK1 1 P60 P85 K14 K17 M20 M17 L22 P20
RHCLK2 1 P61 P87 K16 K18 L19 N21 L21 P26
RHCLK3 1 P62 P88 J16 J17 L18 M20 L20 P25
RHCLK4 1 N/A P90 J14 J16 K18 L21 M18 P23
RHCLK5 1 N/A P92 H14 K15 L17 L20 M20 N24
RHCLK6 1 P64 P91 H15 H18 K20 M18 K20 P18
RHCLK7 1 P65 P93 H16 H17 J20 L17 K19 N19

Notes:
1. N/A in XC3S50A.

Spartan-3 Generation FPGA User Guide www.xilinx.com 49

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

Spartan-3E FPGA Clock Inputs

In the Spartan-3E family, avoid using global clock input GCLK1 as it is always shared with
the M2 mode select pin. Global clock inputs GCLK0, GCLK2, GCLK3, GCLK12, GCLK13,
GCLK14, and GCLK15 have shared functionality in some configuration modes, and all the
RHCLK inputs share functionality with address lines for BPI mode. Make sure there is no
conflict between the pin’s use during and after configuration.
Also in the Spartan-3E family, some clock pad pins are input-only pins as indicated in the
“Pinout Descriptions” section of the data sheet. These might be more useful as clock inputs
because using them does not take away the use of an I/O pin.

Table 2-3: Global Clock Input Pads for Spartan-3E FPGAs

Pad Bank VQ100 CP132 TQ144 PQ208 FT256 FG320 FG400 FG484
GCLK0 2 P38 M6 P56 P80 T9 U10 P11 R12
GCLK1 2 P39 N6 P57 P81 R9 T10 P12 P12
GCLK2 2 P40 P6 P58 P82 P9 R10 V10 Y12
GCLK3 2 P41 P7 P59 P83 N9 P10 V11 W12
GCLK4 0 P83 A10 P122 P177 E9 D10 F11 F12
GCLK5 0 P84 C9 P123 P178 F9 E10 G11 E12
GCLK6 0 P85 B9 P125 P180 A10 B10 E11 B12
GCLK7 0 P86 A9 P126 P181 A9 A10 E10 C12
GCLK8 0 P88 B8 P128 P183 A8 B8 H10 H12
GCLK9 0 P89 C8 P129 P184 B8 B9 G10 H11
GCLK10 0 P90 A7 P130 P185 C8 C9 A10 C11
GCLK11 0 P91 B7 P131 P186 D8 D9 A9 B11
GCLK12 2 P32 M4 P50 P74 M8 N9 W9 V11
GCLK13 2 P33 N4 P51 P75 L8 M9 W10 U11
GCLK14 2 P35 M5 P53 P77 N8 U9 R10 R11
GCLK15 2 P36 N5 P54 P78 P8 V9 P10 T11
LHCLK0 3 P9 F3 P14 P22 H5 J5 K3 M5
LHCLK1 3 P10 F2 P15 P23 H6 J4 K2 L5
LHCLK2 3 P11 F1 P16 P24 H3 J1 K7 L8
LHCLK3 3 P12 G1 P17 P25 H4 J2 L7 M8
LHCLK4 3 P15 G3 P20 P28 J2 K3 M1 M1
LHCLK5 3 P16 H1 P21 P29 J3 K4 L1 N1
LHCLK6 3 P17 H2 P22 P30 J5 K6 M31 M3
LHCLK7 3 P18 H3 P23 P31 J4 K5 L3 M4
RHCLK0 1 P60 K14 P85 P126 K16 K13 M16 N22
RHCLK1 1 P61 J12 P86 P127 J16 K12 L16 M22
RHCLK2 1 P62 J13 P87 P128 J14 K15 L15 M15
RHCLK3 1 P63 J14 P88 P129 J13 K14 L14 M16

50 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Inputs

Table 2-3: Global Clock Input Pads for Spartan-3E FPGAs (Cont’d)
Pad Bank VQ100 CP132 TQ144 PQ208 FT256 FG320 FG400 FG484
RHCLK4 1 P65 H13 P91 P132 H15 J17 K13 L20
RHCLK5 1 P66 H12 P92 P133 H14 J16 K14 L21
RHCLK6 1 P67 G14 P93 P134 H12 J15 K20 L18
RHCLK7 1 P68 G13 P94 P135 H11 J14 J20 L19

Spartan-3 FPGA Clock Inputs

Spartan-3 devices have eight Global Clock input pads called GCLK0 through GCLK7.
GCLK0 through GCLK3 are placed at the center of the die’s bottom edge. GCLK4 through
GCLK7 are placed at the center of the die’s top edge. Any of the eight Global Clock inputs
can connect to any resource on the die. There are no restrictions by quadrant, and no
differentiation of primary and secondary clocks, simplifying I/O and logic placement. In
the Spartan-3 family, none of the clock inputs share functionality with configuration pins,
and all are on I/O pins.
The pin locations for the global clock input pads are shown in Table 2-4.

Table 2-4: Global Clock Input Pads for Spartan-3 FPGAs

Pad Bank VQ100 CP132(1) TQ144 PQ208 FT256 FG320 FG456 FG676 FG900 FG1156(2)
GCLK0 4 P38 M7 P55 P79 T9 P10 AB12 AF14 AK16 AP18
GCLK1 4 P39 P8 P56 P80 R9 N10 AA12 AE14 AJ16 AN18
GCLK2 5 P36 P6 P52 P76 N8 P9 Y11 AD13 AH15 AM17
GCLK3 5 P37 P7 P53 P77 P8 N9 AA11 AE13 AJ15 AN17
GCLK4 1 P87 A9 P124 P180 D9 F10 C12 C14 C16 C18
GCLK5 1 P88 A8 P125 P181 C9 E10 B12 B14 B16 B18
GCLK6 0 P89 C8 P128 P183 A8 F9 A11 A13 A15 A17
GCLK7 0 P90 A7 P127 P184 B8 E9 B11 B13 B15 B17

Notes:
1. The CP(G)132 package is being discontinued and is not recommended for new designs. See
http://www.xilinx.com/support/documentation/customer_notices/xcn08011.pdf for details.
2. The FG(G)1156 package is being discontinued and is not recommended for new designs. See
http://www.xilinx.com/support/documentation/customer_notices/xcn07022.pdf for details.

Clock Inputs and DCMs

Clock inputs optionally connect directly to DCMs using dedicated connections. For more
information on the clock inputs that best feed a specific DCM within a given device in each
family, see Chapter 3, “Using Digital Clock Managers (DCMs).”

Differential Clocks Using Two Inputs

A differential clock input requires two global clock inputs. The P and N inputs follow the
same configuration as for standard inputs on those pins. The clock inputs that get paired
together are consecutive pins in clock number, an even clock number and the next higher

Spartan-3 Generation FPGA User Guide www.xilinx.com 51

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

odd value. For example, GCLK0 and GCLK1 are a differential pair as are LHCLK6 and
LHCLK7.
In the Spartan-3E and Extended Spartan-3A families, two clock inputs are available for
each clock buffer, allowing up to twelve differential global clock inputs. In the Spartan-3
family, only four differential clock inputs are allowed.

Using Dedicated Clock Inputs in a Design

All clock input pins, including the LHCLK and RHCLK pins, are represented in a design
by the IBUFG component. In general, an IBUFG is inferred by the synthesis tool on any
top-level clock port. If it is desired to have more control over this process, an IBUFG can be
instantiated. The I port should be connected directly to the top-level port and the O port
should be connected to a DCM, BUFG, or general logic. Most synthesis tools can infer the
BUFG automatically when connecting an IBUFG to the clock resources of the FPGA.

IBUFG
IBUFG (see Figure 2-3) represents the dedicated input buffers for driving the BUFGMUX
or its alternatives, or the DCM.

IBUFG

I O

UG331_c4_03_080906

Figure 2-3: IBUFG Component

IBUFGDS
IBUFGDS (see Figure 2-4) is a dedicated differential signaling input buffer for connection
to the clock buffer (BUFG) or DCM. In IBUFGDS, a design level interface signal is
represented as two distinct ports (I and IB), one called the master and the other called the
slave. The master and the slave are opposite phases of the same logical signal (for example,
MYNET and MYNETB).

52 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Buffers/Multiplexers

I
O
IB
UG331_c4_04_080906

Inputs Outputs
I IB O
0 0 -(1)
0 1 0
1 0 1
1 1 -(1)

Notes:
1. The dash (-) means no change.

Figure 2-4: IBUFGDS Component and Truth Table

The default IBUFGDS I/O standard is LVDS_25.

Clock Buffers/Multiplexers
Clock buffers/multiplexers either drive clock input signals directly onto a clock line
(BUFG) or optionally provide a multiplexer to switch between two unrelated, possibly
asynchronous clock signals (BUFGMUX).
Each BUFGMUX element, shown in Figure 2-5, is a 2-to-1 multiplexer. The select line, S,
chooses which of the two inputs, I0 or I1, drives the BUFGMUX output signal, O, as
described in Table 2-5. As specified in each data sheet’s “DC and Switching
Characteristics” section, the S input has a setup time requirement. It also has
programmable polarity.

BUFGMUX

I0
O

I1

S
UG331_c4_05_080906

Figure 2-5: BUFGMUX Clock Multiplexer

Table 2-5: BUFGMUX Select Mechanism

S Input O Output
0 I0 Input
1 I1 Input

Spartan-3 Generation FPGA User Guide www.xilinx.com 53

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

BUFGMUX Multiplexing Details

The BUFGMUX not only multiplexes two clock signals but does it in a way that eliminates
any timing hazards. This allows switching from one clock source to a completely
asynchronous clock source without glitches. The element guarantees that when the select
line S is toggled to choose the other clock source, the output remains in the inactive state
until the next active clock edge on either input. The output can be either High or Low when
disabled (when toggling between clock inputs). The default is Low. A cross-coupled
register pair ensures the BUFGMUX output does not inadvertently generate a clock edge.
When the S input changes, the BUFGMUX does not drive the new input to the output until
the previous clock input is Low and the new clock input has a High-to-Low transition. By
not toggling on the first Low-to-High transition of the input, the output clock pulse is
never shorter than the shortest input clock pulse.

Table 2-6: BUFGMUX Functionality

Inputs Outputs
I0 I1 S O
I0 X 0 I0
X I1 1 I1
X X ↑ 0
X X ↓ 0

The S input selects clock input I0 when Low and I1 when High, but also has built-in
programmable polarity, equivalent to swapping I0 and I1. Programmable polarity on the
clock signal is available at each flip-flop, which can be rising-edge or falling-edge
triggered, avoiding having to generate and propagate two separate clock signals.
If only one clock input is needed the second clock input and select lines do not need to be
used.
The BUFGMUX is initialized with I0 selected at power-up and after the assertion of the
Global Set/Reset (GSR). Simulation should also start with S = 0 at time 0. If S = 1 at time 0,
the output is unknown until the next falling edge of I1.
The select line can change at almost any time, independent of the clock states or transitions.
The only exception is a short setup time prior to a Low-to-High transition on the selected
clock input, which can result in an undefined runt pulse output.
Figure 2-6 shows a switchover from CLK0 to CLK1.

Wait for Low

CLK0
Switch
CLK1

OUT

UG331_c2_02_111909

Figure 2-6: BUFGMUX Waveform Diagram

54 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Buffers/Multiplexers

• The current clock is CLK0.

• S is activated High.
• If CLK0 is currently High, the multiplexer waits for CLK0 to go Low.
• Once CLK0 is Low, the multiplexer output stays Low until CLK1 transitions High to
Low.
• When CLK1 transitions from High to Low, the output switches to CLK1.
• No glitches or short pulses can appear on the output.

Using Clock Buffers/Multiplexers in a Design

Most synthesis tools infer clock buffers on the highest fanout clock nets, especially if they
have inputs in the top-level design. If there are more clocks than buffers, the most-utilized
clocks get priority for the buffers. The library components are used to specify the buffers
explicitly or to use the multiplexer functionality.

BUFGMUX and BUFGMUX_1

BUFGMUX and BUFGMUX_1 are distinguished by which state the output assumes when
it switches between clocks in response to a change in its select input. BUFGMUX assumes
output state 0 and BUFGMUX_1 assumes output state 1.

BUFG
The BUFGMUX is the physical clock buffer in the device, but it can be used as a simple
single-input clock buffer. The BUFG clock buffer primitive (see Figure 2-7) drives a single
clock signal onto the clock network and is essentially the same element as a BUFGMUX,
just without the clock select mechanism. BUFG is the generic primitive for clock buffers
across multiple architectures.

BUFG

I O

UG331_c4_06_080906

Figure 2-7: BUFG Component

The BUFG is built from the BUFGMUX as shown in Figure 2-8.

I0
O

I1

S
UG331_c4_07_011008

Figure 2-8: BUFG Built from BUFGMUX

The dedicated zero on the select line is actually implemented with a dedicated VCC source
and using the programmable polarity on the S input.

Spartan-3 Generation FPGA User Guide www.xilinx.com 55

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

BUFGCE and BUFGCE_1

The BUFGCE primitive creates an enabled clock buffer using the BUFGMUX select
mechanism. BUFGCE is a global clock buffer with a single gated input. Its O output is "0"
when clock enable (CE) is Low (inactive). When clock enable (CE) is High, the I input is
transferred to the O output.

BUFGCE
CE

I O

UG331_c4_08_080906

Inputs Outputs
I CE O
X 0 0
I 1 I

Figure 2-9: BUFGCE Component and Truth Table

The BUFGCE is built from the BUFGMUX by multiplexing a fixed value for one input. The
default value is Low when disabled. The BUFGCE_1 primitive is similar with VCC
connected to I1, making the output High when disabled. It also uses the BUFGMUX_1
primitive to guarantee there are no glitches during the transition between inputs.
Figure 2-10 shows the equivalent functionality, although the library component truly is a
primitive. The CE inversion is built into the BUFGMUX functionality. The "0" source can be
fed from any convenient unused LUT.

BUFGMUX
I0
O

O
XGND I1

CE_IN S
INV
GND UG331_c4_09_080906

Figure 2-10: Equivalent Functionality of BUFGCE

XST Synthesis of Clock Buffers

XST automatically infers clock buffers on the highest fanout clock nets up to the device or
user limits, but synthesis constraints can be used to control the usage of clock buffers.
BUFFER_TYPE selects the type of buffer to be inserted on the input port. The default is
BUFGP, which is equivalent to a BUFG.
NET “signal_name” buffer_type={bufgdll|ibufg|bufgp|ibuf|bufr|none};
The BUFFER_TYPE parameter can be used on a generic input to make sure that the global
clock buffer is used (= BUFGP). It can also be set to NONE to prevent the automatic usage
of a global clock buffer. This replaces the older constraint CLOCK_BUFFER, which should
not be used in new designs.

56 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

BUFGMUX Connection Details

If a common clock enable is used for all loads on a clock net, the BUFGCE = YES constraint
can be used to move the high-fanout clock enable to a single line on a BUFGCE:
NET “primary_clock_signal” bufgce={yes|no|true|false};
CLOCK_SIGNAL is a synthesis constraint. In the case where a clock signal goes through
combinatorial logic before being connected to the clock input of a flip-flop, XST cannot
identify what input pin is the real clock pin. This constraint can be used to define the clock
pin:
NET “primary_clock_signal” clock_signal={yes|no|true|false};

BUFGMUX Connection Details

BUFGMUX Inputs
The I0 and I1 inputs to a BUFGMUX element originate from clock input pins, DCMs, or
Double-Line interconnect, as shown in Figure 2-11. As shown in Figure 2-2, page 47, there
are 24 BUFGMUX elements distributed around the four edges of the device. Clock signals
from the four BUFGMUX elements at the top edge and the four at the bottom edge are
truly global and connect to all clocking quadrants. The eight left-edge BUFGMUX
elements only connect to the two clock quadrants in the left half of the device. Similarly, the
eight right-edge BUFGMUX elements only connect to the right half of the device.

Left-/Right-Half BUFGMUX Top/Bottom (Global) BUFGMUX

CLK Switch CLK Switch
Matrix Matrix BUFGMUX
BUFGMUX
S S
I0 I0
0 O 0 O
I1 I1
1 1

I0 0 I0
O 0 O
I1 I1
1 1
S S

LHCLK or
RHCLK input 1st GCLK pin

Double Line 1st DCM output

Double Line
DCM output*
2nd DCM output
2nd GCLK pin

*(Only devices with

left/right DCMs) UG331_c4_10_080906

Figure 2-11: Spartan-3E and Extended Spartan-3A Family Clock Switch Matrix for
BUFGMUX Pair Connectivity

BUFGMUX elements are organized in pairs and share I0 and I1 connections with adjacent
BUFGMUX elements from a common clock switch matrix as shown in Figure 2-11. For
example, the input on I0 of one BUFGMUX is also a shared input to I1 of the adjacent
BUFGMUX.
The clock switch matrix for the left- and right-edge BUFGMUX elements receive signals
from any of the three following sources: an LHCLK or RHCLK pin as appropriate, a
Double-Line interconnect, or a DCM in the larger devices. These devices include the

Spartan-3 Generation FPGA User Guide www.xilinx.com 57

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

XC3S1200E and XC3S1600E devices in the Spartan-3E family, and the XC3S700A/AN,
XC3S1400A/AN, XC3SD1800A, and XC3SD3400A devices in the Extended Spartan-3A
family.
By contrast, the clock switch matrixes on the top and bottom edges receive signals from
any of the five following sources: two GCLK pins, two DCM outputs, or one Double-Line
interconnect.
Table 2-7 indicates permissible connections between clock inputs and BUFGMUX
elements. The I0 input provides the best input path to a clock buffer. The I1 input provides
the secondary input for the clock multiplexer function.
Table 2-7: Spartan-3E and Extended Spartan-3A Family Connections from Clock Inputs to BUFGMUX
Elements and Associated Quadrant Clock
Quadrant Left-Half BUFGMUX Top or Bottom BUFGMUX Right-Half BUFGMUX
Clock
Line(1) Location(2) I0 Input I1 Input Location(2) I0 Input I1 Input Location(2) I0 Input I1 Input

Extended Extended
Spartan-3A FPGAs: Spartan-3A FPGAs:
H X0Y9 LHCLK7 LHCLK6 X1Y10 GCLK6 or GCLK10 GCLK7 or GCLK11 X3Y9 RHCLK3 RHCLK2
Spartan-3E FPGAs: Spartan-3E FPGAs:
GCLK7 or GCLK11 GCLK6 or GCLK10

Extended Extended
Spartan-3A FPGAs: Spartan-3A FPGAs:
G X0Y8 LHCLK6 LHCLK7 X1Y11 GCLK7 or GCLK11 GCLK6 or GCLK10 X3Y8 RHCLK2 RHCLK3
Spartan-3E FPGAs: Spartan-3E FPGAs:
GCLK6 or GCLK10 GCLK7 or GCLK11

Extended Extended
Spartan-3A FPGAs: Spartan-3A FPGAs:
F X0Y7 LHCLK5 LHCLK4 X2Y10 GCLK4 or GCLK8 GCLK5 or GCLK9 X3Y7 RHCLK1 RHCLK0
Spartan-3E FPGAs: Spartan-3E FPGAs:
GCLK5 or GCLK9 GCLK4 or GCLK8

Extended Extended
Spartan-3A FPGAs: Spartan-3A FPGAs:
E X0Y6 LHCLK4 LHCLK5 X2Y11 GCLK5 or GCLK9 GCLK4 or GCLK8 X3Y6 RHCLK0 RHCLK1
Spartan-3E FPGAs: Spartan-3E FPGAs:
GCLK4 or GCLK8 GCLK5 or GCLK9

D X0Y5 LHCLK3 LHCLK2 X1Y0 GCLK3 or GCLK15 GCLK2 or GCLK14 X3Y5 RHCLK7 RHCLK6

C X0Y4 LHCLK2 LHCLK3 X1Y1 GCLK2 or GCLK14 GCLK3 or GCLK15 X3Y4 RHCLK6 RHCLK7

B X0Y3 LHCLK1 LHCLK0 X2Y0 GCLK1 or GCLK13 GCLK0 or GCLK12 X3Y3 RHCLK5 RHCLK4

A X0Y2 LHCLK0 LHCLK1 X2Y1 GCLK0 or GCLK12 GCLK1 or GCLK13 X3Y2 RHCLK4 RHCLK5

Notes:
1. See “Quadrant Clock Routing,” page 59 for connectivity details for the eight quadrant clocks.
2. See Figure 2-2 for specific BUFGMUX locations, and Figure 2-13 for information on how BUFGMUX elements drive onto a specific clock line
within a quadrant.

58 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Quadrant Clock Routing

The four BUFGMUX elements on the top edge are paired together and share inputs from
the eight global clock inputs along the top edge. Each BUFGMUX pair connects to four of
the eight global clock inputs, as shown in Figure 2-2, page 47. This optionally allows
differential inputs to the global clock inputs without wasting a BUFGMUX element.
The connections for the bottom-edge BUFGMUX elements are similar to the top-edge
connections (see Figure 2-11). On the left and right edges, only two clock inputs feed each
pair of BUFGMUX elements.

BUFGMUX Outputs
The BUFGMUX drives the global clock routing, which in turn connects to clock inputs on
device resources. The BUFGMUX can also connect to a DCM, typically used for internal
feedback to the DCM CLKFB input, as shown in Figure 2-12.

Global Digital Clock Global

Buffer Input Manager Clock Buffer
IBUFG BUFG
I O I O Low-Skew
GCLK CLKIN Output Global Clock
Network
DCM
CLKFB

UG331_c4_11_080906

Figure 2-12: Using a DCM to Eliminate Clock Skew

For more details on using the DCMs, see Chapter 3, “Using Digital Clock Managers
(DCMs).”

Spartan-3 Global Clock Buffers

The Spartan-3 family has only eight global clock buffers. Four BUFGMUX elements are
placed at the center of the die’s bottom edge, just above the GCLK0 - GCLK3 inputs. The
remaining four BUFGMUX elements are placed at the center of the die’s top edge, just
below the GCLK4 - GCLK7 inputs. Each pair of BUFGMUX elements shares two sources;
each source feeds the I0 input of one BUFGMUX and the I1 input of the adjacent
BUFGMUX. Thus two completely independent pairs of clock inputs to be multiplexed
could be on the same side of the die but not on the adjacent BUFGMUX elements. For more
details, see the Spartan-3 FPGA data sheet.

Quadrant Clock Routing

The clock routing within the Spartan-3E and Extended Spartan-3A family is quadrant-
based, as shown in Figure 2-2, page 47. Each clock quadrant supports eight total clock
signals, labeled A through H in Table 2-7 and Figure 2-13. The clock source for an
individual clock line originates either from a global BUFGMUX element along the top and
bottom edges or from a BUFGMUX element along the associated left/right edge, as shown
in Figure 2-13. The clock lines feed the synchronous resource elements (CLBs, IOBs, block
RAM, multipliers, and DCMs) within the quadrant. Those resources have programmable
polarity on the clock input.

Spartan-3 Generation FPGA User Guide www.xilinx.com 59

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

BUFGMUX Output Clock Line BUFGMUX Output Clock Line

X1Y10 (Global) X1Y10 (Global)
H H
X0Y9 (Left Half) X3Y9 (Right Half)

X1Y11 (Global) X1Y11 (Global)

G G
X0Y8 (Left Half) X3Y8 (Right Half)

X2Y10 (Global) X2Y10 (Global)

F F
X0Y7 (Left Half) X3Y7 (Right Half)

X2Y11 (Global) X2Y11 (Global)

E E
X0Y6 (Left Half) X3Y6 (Right Half)

X1Y0 (Global) X1Y0 (Global)

D D
X0Y5 (Left Half) X3Y5 (Right Half)

X1Y1 (Global) X1Y1 (Global)

C C
X0Y4 (Left Half) X3Y4 (Right Half)

X2Y0 (Global) X2Y0 (Global)

B B
X0Y3 (Left Half) X3Y3 (Right Half)

X2Y1 (Global) X2Y1 (Global)

A A
X0Y2 (Left Half) X3Y2 (Right Half)

a. Left (TL and BL Quadrants) Half of Die b. Right (TR and BR Quadrants) Half of Die
DS312-2_17_103105

Figure 2-13: Spartan-3E and Extended Spartan-3A Family Clock Sources for the Eight Clock Lines within
a Clock Quadrant

The four quadrants of the device are:

• Top Right (TR)
• Bottom Right (BR)
• Bottom Left (BL)
• Top Left (TL)
The quadrant clock notation (TR, BR, BL, TL) is separate from that used for similar IOB
placement constraints.
The outputs of the top or bottom BUFGMUX elements connect to two vertical spines, each
comprising four vertical clock lines as shown in Figure 2-2, page 47. At the center of the
die, these clock signals connect to the eight-line horizontal clock spine. By bringing the
clock to the center of the device and then radiating outward, the skew is minimized across
the device.
Outputs of the left and right BUFGMUX elements are routed onto the left or right
horizontal spines, each comprising eight horizontal clock lines.
Each of the eight clock signals in a clock quadrant derives either from a global clock signal
or a half clock signal. In other words, there are up to 24 total potential clock inputs to the
FPGA, eight of which can connect to clocked elements in a single clock quadrant.
Figure 2-13 shows how the clock lines in each quadrant are selected from associated
BUFGMUX sources. For example, if quadrant clock A in the bottom left (BL) quadrant
originates from BUFGMUX_X2Y1, then the clock signal from BUFGMUX_X0Y2 is

60 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Other Information

unavailable in the bottom left quadrant. However, the top left (TL) quadrant clock A can
still solely use the output from either BUFGMUX_X2Y1 or BUFGMUX_X0Y2 as the source.
To estimate the quadrant location for a particular I/O, see the footprint diagrams in the
device data sheets. For exact quadrant locations, use the PlanAhead floorplanning tool. In
the QFP packages the quadrant borders fall in the middle of each side of the package, at a
GND pin. The clock inputs fall on the quadrant boundaries, as shown in Table 2-8.

Table 2-8: Clock Quadrant Locations

Clock Pins Quadrant
GCLK[3:0] BR
GCLK[7:4] TR
GCLK[11:8] TL
GCLK[15:12] BL
RHCLK[3:0] BR
RHCLK[7:4] TR
LHCLK[3:0] TL
LHCLK[7:4] BL

Choosing Top/Bottom and Left-/Right-Half Global Buffers

The software generally use the top/bottom global buffers as the first choice for high-fanout
clock signals. If there are more than eight clocks in a design, the left-/right-half buffers can
be used. Floorplanning is recommended for designs requiring more than eight clocks,
since the loads on the left-/right-half buffers must be restricted to one half of the device, or
restricted to one quadrant to allow the most freedom for the global input using the same
routing resource.

Spartan-3 FPGA Global Clock Routing

The Spartan-3 FPGA BUFGMUX drives the vertical global clock spine belonging to the
same side of the die — top or bottom — as the BUFGMUX element in use. The two spines
— top and bottom — each comprise four vertical clock lines, each running from one of the
BUFGMUX elements on the same side towards the center of the die. At the center of the
die, clock signals reach the eight-line horizontal spine, which spans the width of the die. In
turn, the horizontal spine branches out into a subsidiary clock interconnect that accesses
the CLBs, IOBs, block RAM, and multipliers. For more details, see the Spartan-3 Family
Data Sheet.

Other Information

Clock Power Consumption

Dynamic power dissipation can be reduced through optimization of the clocks used in a
design.
To minimize the dynamic power dissipation of the clock network, the Xilinx development
software automatically disables all clock segments not in use. To take full advantage of

Spartan-3 Generation FPGA User Guide www.xilinx.com 61

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

this, concentrate logic in the fewest possible clock column regions. Use floorplanning to
reduce the number of clock columns in use. To further reduce clock power, reduce the
number of rows that the clock is driving.
Using a slower clock also reduces power. The DCM can be used to divide clocks, or slow
clocks can be further divided using registers. A design can be organized according to
required clock frequency and then each part clocked at the lowest possible frequency.
Stopping a clock eliminates the power consumed by the clock routing and by the elements
it drives. If possible, stop the clock externally where it enters the FPGA. If you can not stop
the clock externally, then disable it inside the FPGA by using the BUFGMUX or BUFGCE.
Gating a clock through internal CLB logic is not recommended because it introduces route-
dependent skew and makes the design sensitive to lot-to-lot variations, and might require
manual routing.
The alternative is to use the clock enables to disable the clock loads. This is useful when the
clock is still needed in some locations, but it does not reduce the clock distribution power.

Clock Setup and Hold Timing

See the data sheets for detailed clock timing information. Delay parameters are provided
for both pin-to-pin paths through the device and individual component delays. All delay
parameters that begin or end at a device pin are defined for the LVCMOS25 I/O standard,
which is the default. Outputs are defined for 12 mA drive, which is the default, and Fast
slew rate (Slow is the default). Parameters are adjusted by the timing report tools for other
I/O standards. For pin-to-pin setup times, which are calculated as the data delay minus
the clock delay, the clock pin "adder" is actually subtracted from the result. For hold times,
the data pin "adder" is subtracted. A negative hold time means that the data can be
released before the clock edge. This is often considered simply as a zero hold time, allowing
the clock and data to change at the same time. The previous data gets clocked before the
new data arrives.
Delay parameters for an input flip-flop are affected by the IFD_DELAY_VALUE setting.
The default is AUTO, which selects a specific value according to the density of the device.
For exact timing for your design, see the timing reports provided by the ISE® development
tools.

Summary
Global clock inputs, buffers, and routing are automatically used for a design’s highest
fanout clock signals. Implementation reports should be checked to verify the usage of
clock buffers where desired. The user can specify the details of global clock usage in order
to take advantage of special features such as multiplexing and clock enables, or to
maximize the number of clocks using global resources in a design.

Additional Information
For other types of routing resources, see Chapter 12, “Using Interconnect.”
For more details on the DCMs, see Chapter 3, “Using Digital Clock Managers (DCMs).”
This chapter focuses on the Spartan-3E and Extended Spartan-3A family architectures. For
details on Spartan-3 FPGA clocks, see DS099, Spartan-3 FPGA Family Data Sheet.
For more information on input delay elements and IOSTANDARD options, see
Chapter 10, “Using I/O Resources.”

62 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Summary

For information on the clocked resources in the FPGA, such as the CLB flip-flops and the
block RAM, see the appropriate chapters elsewhere in this user guide.
For information on setting clock performance constraints, see the ISE Constraints Guide on
the Xilinx website.

Spartan-3 Generation FPGA User Guide www.xilinx.com 63

UG331 (v1.7) August 19, 2010
 R

Chapter 2: Using Global Clock Resources

64 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 3

Using Digital Clock Managers (DCMs)

Summary
Digital Clock Managers (DCMs) provide advanced clocking capabilities to
Spartan®-3 generation FPGA applications (Spartan-3, Spartan-3E, and Extended
Spartan-3A families). Primarily, DCMs eliminate clock skew, thereby improving system
performance. Similarly, a DCM optionally phase shifts the clock output to delay the
incoming clock by a fraction of the clock period. DCMs optionally multiply or divide the
incoming clock frequency to synthesize a new clock frequency. The DCMs integrate
directly with the FPGA’s global low-skew clock distribution network.

Introduction
DCMs integrate advanced clocking capabilities directly into the FPGA’s global clock
distribution network. Consequently, DCMs solve a variety of common clocking issues,
especially in high-performance, high-frequency applications:
• Eliminate Clock Skew, either within the device or to external components, to
improve overall system performance and to eliminate clock distribution delays.
• Phase Shift a clock signal, either by a fixed fraction of a clock period or by
incremental amounts.
• Multiply or Divide an Incoming Clock Frequency or synthesize a completely new
frequency by a mixture of clock multiplication and division.
• Condition a Clock, ensuring a clean output clock with a 50% duty cycle.
• Mirror, Forward, or Rebuffer a Clock Signal, often to deskew and convert the
incoming clock signal to a different I/O standard—for example, forwarding and
converting an incoming LVTTL clock to LVDS.
• Any or all the above functions, simultaneously.
Table 3-1: Digital Clock Manager Features and Capabilities
Feature Description DCM Signals
Digital Clock Managers (DCMs) per Device Two to eight DCMs, depending on All
array size. See Figure 3-1, page 68.
Clock Input Sources • Global buffer input pad CLKIN
• Global buffer output
• General-purpose I/O (no deskew)
• Internal logic (no deskew)
Frequency Synthesizer Output Multiply CLKIN by the fraction (M/D) • CLKFX
where M = {2..32}, D = {1..32} • CLKFX180

Spartan-3 Generation FPGA User Guide www.xilinx.com 65

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-1: Digital Clock Manager Features and Capabilities (Cont’d)

Feature Description DCM Signals
Clock Divider Output Divide CLKIN by 1.5, 2, 2.5, 3, 3.5, 4, 4.5, CLKDV
5, 5.5, 6, 6.5, 7, 7.5, 8, 9, 10, 11, 12, 13, 14,
15, or 16
Clock Doubler Output Multiply CLKIN frequency by 2 • CLK2X
• CLK2X180
Clock Conditioning, Duty-Cycle Correction Always provided on most outputs. All
Optional on Spartan-3 FPGA outputs
CLK0, CLK90, CLK180, and CLK270.
Quadrant Phase Shift Outputs 0º (no phase shift), • CLK0
90º (¼ period), • CLK90
180º (½ period), • CLK180
270º (¾ period)
• CLK270
Half-Period Phase Shift Outputs Output pairs with 0º and 180º phase • CLK0,
shift, ideal for DDR applications CLK180
• CLK2X,
CLK2X180
• CLKFX,
CLKFX180
Number of DCM Clock Outputs Connected to General- Up to all 9 All
Purpose Interconnect
Number of DCM Clock Outputs Connected to Global Any 4 of 9 All
Clock Network
Number of Clock Outputs Connected to Output Pins Up to all 9 All

Document Overview
This chapter covers an assortment of topics related to Digital Clock Managers, not all of
which are relevant to every specific FPGA application.
The “DCM Functional Overview” section provides a brief introduction to the DCM and its
functions. Similarly the “DCM Primitive” section describes all the connection ports and
attributes or constraints associated with a DCM. Likewise the “Clocking Wizard” and the
“VHDL and Verilog Instantiation” sections demonstrate the various methods to specify a
DCM design.
The “DCM Clock Requirements” and the “Input and Output Clock Frequency
Restrictions” sections explain the frequency requirements on the DCM clock input and the
various DCM clock outputs. Similarly, the “Clock Jitter or Phase Noise” section highlights
the effect jitter has on output clock quality.
Finally, the “Eliminating Clock Skew”, “Clock Conditioning”, “Phase Shifting – Delaying
Clock Outputs by a Fraction of a Period”, “Clock Multiplication, Clock Division, and
Frequency Synthesis”, and “Clock Forwarding, Mirroring, Rebuffering” sections illustrate
various applications using the DCM block.

66 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Compatibility and Comparison with Other Xilinx FPGA Families

Compatibility and Comparison with Other Xilinx FPGA Families

Spartan-3E and Extended Spartan-3A family FPGAs include a fourth-generation DCM
design, incorporating a variety of enhancements and improvements over previous FPGA
families. The Spartan-3 FPGA DCM, a third-generation design, is nearly functionally
identical to the DCM units found in Virtex®-II and Virtex-II Pro FPGA families.
The DCM feature is nearly identical between all Spartan-3 generation families. Table 3-2
summarizes the primary DCM differences between families. Mid- and large-density
Spartan-3E and Extended Spartan-3A family FPGAs have additional DCMs along the left
and right sides of the FPGA. The Spartan-3E and Extended Spartan-3A family DCMs
automatically determine their operating range and, unlike Spartan-3 FPGAs, are not
limited to either a Low or High operating frequency range. Furthermore, Spartan-3E and
Extended Spartan-3A family DCMs support a broader range of input frequencies. There
are also important differences in the way that Spartan-3E and Extended Spartan-3A family
FPGAs implement Variable Phase Shift operations, further described in “Important
Differences Between Spartan-3 Generation FPGA Families,” page 123.

Table 3-2: DCM Differences between Spartan-3 Generation FPGAs

Spartan-3 Spartan-3E Extended Spartan-3A
FPGAs FPGAs Family FPGAs
Design primitive DCM DCM_SP DCM_SP
DCMs per device 2 to 4 global plus 2 to 4 global plus
2 to 4 global
(Figure 3-1, page 68) 0 to 4 side DCMs 0 to 4 side DCMs
DLL minimum input
18 MHz 5 MHz 5 MHz
clock frequency
Distinct DLL
Two:
operating frequency One One
Low and High
ranges
Distinct DFS operating Two:
One One
frequency range Low and High
DFS input clock
1 to 280 MHz 0.2 to 333 MHz 0.2 to 333 MHz
frequency range
Variable Phase Shift 1/
256th of DCM_DELAY_STEP, DCM_DELAY_STEP,
increment or CLKIN Period between 20 to 40 ps between 15 to 35 ps
decrement unit (degrees) (time) (time)
DCM Spartan-3A/3A DSP
VCCAUX voltage 2.5V 2.5V FPGA: 2.5V or 3.3V
supply Spartan-3AN: 3.3V

The Spartan-3 FPGA DCM is a significant enhancement over the Spartan-II/Spartan-IIE

FPGA DLL function. A Spartan-3 FPGA DCM provides all the capabilities of the
Spartan-II/Spartan-IIE FPGA DLL with new capabilities, such as the Frequency
Synthesizer and phase shifting functions. The Spartan-3 FPGA Frequency Synthesizer
multiplies an input clock by up to a factor of 32. The Spartan-II/IIE FPGA DLL has limited
frequency multiplication capabilities—namely, an input clock can be doubled. Similarly,
the Spartan-3 FPGA DCM has a wider divider range compared to Spartan-IIE FPGA DLLs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 67

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

DCM Locations and Clock Distribution Network Interface

Figure 3-1 shows the number and relative location of DCMs on various
Spartan-3 generation FPGAs. The smallest FPGA in each family has two DCMs, although
the physical location varies between families. All larger Spartan-3 FPGAs and the middle-
sized members of the Spartan-3E and Extended Spartan-3A families have four DCM
blocks. The larger members of the Spartan-3E and Extended Spartan-3A families have
eight total DCMs. The two DCMs at the top and the two at the bottom connect into the
FPGA’s global clock network. The DCMs along the left and right edges connect to the clock
network on their respective half of the FPGA.

Spartan-3E Spartan-3A Spartan-3

DCMs FPGAs FPGAs FPGAs

G G G G

2 XC3S100E XC3S50A XC3S50

G G

G G G G G G
All Other
XC3S250E XC3S200A Spartan-3
4 XC3S500E XC3S400A Family
FPGAs
G G G G G G

Global DCM: Connects

G G G G G to global clock network.

XC3S700A Left-Half DCM: Connects

L R L R L to clock network on left
XC3S1200E XC3S1400A half of the FPGA.
8 XC3S1600E XC3SD1800A
L R L R Right-Half DCM: Connects
XC3SD3400A R to clock network on right
half of the FPGA.
G G G G
UG331_c3_03_011008

Figure 3-1: Number and Location of DCMs on Spartan-3 Generation FPGAs

The DCM blocks have dedicated connections to the global buffer inputs and global buffer
multiplexers on the same edge of the device, either top or bottom. They are an integral part
of the FPGA’s global clocking infrastructure. DCMs are an optional element in the clock
distribution network and are available when required by the application. In Figure 3-2a, a
clock input feeds directly into the low-skew, high-fanout global clock network via a global
input buffer and global clock buffer.
If the application requires some or all of the DCM’s advanced clocking features, the DCM
fits neatly between the global buffer input and the buffer itself, as shown in Figure 3-2b.

68 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Functional Overview

Global Buffer Input Global Clock Buffer

IBUFG BUFG
Low-Skew
I O I O Global Clock
GCLK
Network
x462_02a_062403

a. Global Buffer Inputs and Clock Buffers Drive a Low-Skew Global Network in the FPGA

Global Global
Buffer Input Digital Clock Clock Buffer
Manager
IBUFG BUFG
I O I O Low-Skew
GCLK CLKIN Output Global Clock
Network
DCM_SP
CLKFB

UG332_c3_02_113006

b. A Digital Clock Manager (DCM) Inserts Directly into the Global Clock Path

Figure 3-2: DCMs are an Integral Part of the FPGA's Global Clock Network

DCM Functional Overview

The single entity that is generically called a Digital Clock Manager (DCM) consists of four
distinct functional units as depicted in the simplified diagram shown in Figure 3-3 and
described below. These units operate independently or in tandem.

DCM
PSINCDEC Phase
PSEN Shifter PSDONE
PSCLK

CLK0 Clock
CLKIN Distribution
Output Stage

CLK90
Input Stage

Delay
Delay Taps

CLK180
CLK270
CLKFB CLK2X
CLK2X180
CLKDV

CLKFX
DFS
DLL CLKFX180

Status LOCKED
RST 8
Logic STATUS [7:0]

DS099-2_07_021408

Figure 3-3: DCM Functional Block Diagram

Spartan-3 Generation FPGA User Guide www.xilinx.com 69

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Delay-Locked Loop (DLL)

The Delay-Locked Loop (DLL) unit provides an on-chip digital deskew circuit that
effectively generates clock output signals with a net zero delay. The deskew circuit
compensates for the delay on the clock routing network by monitoring an output clock,
from either the DCM’s CLK0 or the CLK2X outputs. The DLL unit effectively eliminates
the delay from the external clock input port to the individual clock loads within the device.
The well-buffered global network minimizes the clock skew on the network caused by
loading differences.
The input signals to the DLL unit are CLKIN and CLKFB. The output signals from the DLL
are CLK0, CLK90, CLK180, CLK270, CLK2X, CLK2X180, and CLKDV.
The DLL unit generates the outputs for the Clock Doubler (CLK2X, CLK2X180), the Clock
Divider (CLKDV) and the Quadrant Phase Shifted Outputs functions.

Digital Frequency Synthesizer (DFS)

The Digital Frequency Synthesizer (DFS) provides a wide and flexible range of output
frequencies based on the ratio of two user-defined integers, a Multiplier
(CLKFX_MULTIPLY) and a Divisor (CLKFX_DIVIDE). The output frequency is derived
from the input clock (CLKIN) by simultaneous frequency division and multiplication. The
DFS feature can be used in conjunction with, or separately from, the DLL feature of the
DCM. If the DLL is not used, the DFS output clocks will not be de-skewed because the DLL
is required to provide the de-skew feedback clock.
The DFS unit generates the Frequency Synthesizer (CLKFX, CLKFX180) outputs.

Phase Shift (PS)

The Phase Shift (PS) unit controls the phase relations of the DCM’s clock outputs to the
CLKIN input.
The Phase Shift unit shifts the phase of all nine DCM clock output signals by a fixed
fraction of the input clock period. The fixed phase shift value is set at design time and
loaded into the DCM during FPGA configuration. If the DLL is not used, the DFS output
clocks will not be de-skewed because the DLL is required to provide the deskew feedback
clock.
The Phase Shift unit also provides a digital interface for the FPGA application to
dynamically advance or retard the current shift value, called Variable Phase Shift. As
shown in Table 3-3, the Spartan-3 FPGA Variable Phase Shift changes by 1/256th of the
CLKIN clock period. On Spartan-3E and Extended Spartan-3A families, the Variable Phase
Shift changes by one DCM_DELAY_STEP, which has a fixed range as defined in the
corresponding data sheet.

Table 3-3: Variable Phase Shift Differences

FPGA Family Smallest Phase Shift Unit
Spartan-3 FPGA 1/256th of the CLKIN clock period, but not less than 30 to 60 ps
Spartan-3E FPGA DCM_DELAY_STEP, which ranges between 20 to 40 ps per step
Extended Spartan-3A DCM_DELAY_STEP, which ranges between 15 to 35 ps per step
FPGA

70 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Primitive

The input signals to the Phase Shift unit are PSINCDEC, PSEN, and PSCLK. The output
signals are PSDONE and the STATUS[0] signal.

Status Logic
The Status Logic indicates the current state of the DCM via the LOCKED and STATUS[0]
(Extended Spartan-3A family FPGAs only), STATUS[1], and STATUS[2] output signals.
The LOCKED output signal indicates whether the DCM outputs are in phase with the
CLKIN input. The STATUS output signals indicate the state of the DLL and PS operations.
The RST input signal resets the DCM logic and returns it to its post-configuration state.
Likewise, a reset forces the DCM to reacquire and lock to the CLKIN input.

DCM Primitive
The DCM design primitive, shown in Figure 3-4, represents all the sub-features within the
Digital Clock Manager. The name of the DCM primitive differs slightly between
Spartan-3 generation FPGA families, as shown in Table 3-4. Spartan-3 FPGAs support the
DCM primitive, while Spartan-3E and Extended Spartan-3A family FPGAs support the
more advanced DCM_SP primitive. The Xilinx ISE® software automatically maps a
Spartan-3 FPGA DCM primitive to the appropriate equivalent in a Spartan-3E or
Extended Spartan-3A family FPGA design.

Table 3-4: Digital Clock Manager Primitive by Spartan-3 Generation FPGA Family
FPGA Family Primitive
Spartan-3E FPGA
DCM_SP
Extended Spartan-3A family FPGA
Spartan-3 FPGA DCM

The DCM’s Connection Ports and Attributes, Properties, or Constraints are summarized
below.

Spartan-3 Generation FPGA User Guide www.xilinx.com 71

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Symbol
Spartan-3 FPGA: DCM
Spartan-3E/3A/3AN/3A DSP FPGAs: DCM_SP
CLKIN CLK0
CLKFB CLK90
CLK180
CLK270
CLK2X
CLK2X180
CLKDV
CLKFX

RST CLKFX180

PSEN STATUS[7:0]
PSINCDEC LOCKED
PSCLK PSDONE
UG331_c3_01_011008

Figure 3-4: DCM Design Primitive

Connection Ports
Table 3-6 lists the various connection ports to the Digital Clock Manager. Each port
connection has a brief description, which includes the signal direction, and which DCM
function units require the connection. Table 3-5 provides the abbreviated name for each
function unit used in Table 3-6.

Table 3-5: Functional Unit Abbreviations for Table 3-6

Abbreviation Functional Unit
DLL Delay-Locked Loop
PS Phase Shifter
DFS Digital Frequency Synthesizer

72 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Primitive

Table 3-6: DCM Connection Ports

Functional Unit
Port Direction Description
DLL PS DFS
CLKIN Clock Clock input to DCM. Always required. The CLKIN frequency and 9 9 9
Input jitter must fall within the limits specified in the data sheet. On
Spartan-3 FPGAs, the frequency limits are further defined by the
DLL_FREQUENCY_MODE and DFS_FREQUENCY_MODE
attributes.
CLKFB Input Clock feedback input to DCM. The feedback input is required 9 Optional 9
unless the Digital Frequency Synthesis outputs, CLKFX or
CLKFX180, are used stand-alone. The source of the CLKFB input
must be the CLK0 or CLK2X output from the DCM and the
CLK_FEEDBACK must be set to 1X or 2X accordingly. The
feedback point ideally includes the delay added by the clock
distribution network, either internally or externally. See
“Feedback from a Reliable Source.”
RST Input Asynchronous reset input. Resets the DCM logic to its post- 9 9 9
configuration state. Causes DCM to reacquire and relock to the
CLKIN input. Invertible within DCM block. Non-inverted
behavior shown below. See “RST Input Behavior.”

0 No effect.
1 Reset DCM block. Hold RST pulse High for at least
three valid CLKIN cycles.

PSEN Input Variable Phase Shift enable. Invertible within DCM block. Non- 9
inverted behavior shown below. See “Variable Fine Phase
Shifting,” page 123.

0 Disable Variable Phase Shifter. Ignore inputs to phase

shifter.
1 Enable Variable Phase Shifter operations on next
rising PSCLK clock edge.

PSINCDEC Input Increment/decrement Variable Phase Shift. Invertible within 9

DCM block. Non-inverted behavior shown below. See “Variable
Fine Phase Shifting,” page 123.

0 Decrement phase shift value on next enabled, rising

PSCLK clock edge.
1 Increment phase shift value on next enabled, rising
PSCLK clock edge.

PSCLK Clock Clock input to Variable Phase Shifter, clocked on rising edge. 9
Input Invertible within DCM block. See “Variable Fine Phase Shifting,”
page 123.

Spartan-3 Generation FPGA User Guide www.xilinx.com 73

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-6: DCM Connection Ports (Cont’d)

Functional Unit
Port Direction Description
DLL PS DFS
CLK0 Clock Same frequency as CLKIN, 0° phase shift (i.e., not phase shifted). 9
Output Always conditioned to a 50% duty cycle on
Extended Spartan-3A family FPGAs or on Spartan-3 FPGAs
when the DUTY_CYCLE_CORRECTION attribute is TRUE.
Either CLK0 or CLK2X is required as a feedback source for DLL
functions. See “Half-Period Phase Shifted Outputs,” and
“Quadrant Phase Shifted Outputs.”
CLK90 Clock Same frequency as CLKIN, 90° phase shifted (quarter period). 9
Output Not available if the DLL_FREQUENCY_MODE attribute is
HIGH. Always conditioned to a 50% duty cycle on
Extended Spartan-3A family FPGAs or on Spartan-3 FPGAs
when the DUTY_CYCLE_CORRECTION attribute is TRUE. See
“Quadrant Phase Shifted Outputs.”
CLK180 Clock Same frequency as CLKIN, 180° phase shifted (half period). 9
Output Always conditioned to a 50% duty cycle on
Extended Spartan-3A family FPGAs or on Spartan-3 FPGAs
when the DUTY_CYCLE_CORRECTION attribute is TRUE. See
“Half-Period Phase Shifted Outputs,” and “Quadrant Phase
Shifted Outputs.”
CLK270 Clock Same frequency as CLKIN, 270° phase shifted (three-quarters 9
Output period). Not available if the DLL_FREQUENCY_MODE attribute
is HIGH. Always conditioned to a 50% duty cycle on
Extended Spartan-3A family FPGAs or on Spartan-3 FPGAs
when the DUTY_CYCLE_CORRECTION attribute is TRUE. See
“Quadrant Phase Shifted Outputs.”
CLK2X Clock Double-frequency clock output, 0° phase shift. Not available if the 9
Output DLL_FREQUENCY_MODE attribute is HIGH. When available,
the CLK2X output always has a 50% duty cycle. Either CLK0 or
CLK2X is required as a feedback source for DLL functions. Clock
Doubler (CLK2X, CLK2X180) output. See “Half-Period Phase
Shifted Outputs.”
CLK2X180 Clock Double-frequency clock output, 180° phase shifted. Not available 9
Output if the DLL_FREQUENCY_MODE attribute is HIGH. When
available, the CLK2X180 output always has a 50% duty cycle.
Clock Doubler (CLK2X, CLK2X180) output. See “Half-Period
Phase Shifted Outputs.”
CLKDV Clock Divided clock output, controlled by the CLKDV_DIVIDE 9
Output attribute. The CLKDV output has a 50% duty cycle unless the
DLL_FREQUENCY_MODE attribute is HIGH and the
CLKDV_DIVIDE attribute is a non-integer value. The locking
time is longer when CLKDV_DIVIDE has a non-integer value.
See the Clock Divider (CLKDV) output.
F CLKIN
F CLKDV = -----------------------------------------------
-
CLKDV_DIVIDE

74 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Primitive

Table 3-6: DCM Connection Ports (Cont’d)

Functional Unit
Port Direction Description
DLL PS DFS
CLKFX Clock Synthesized clock output, controlled by the CLKFX_MULTIPLY 9
Output and CLKFX_DIVIDE attributes. Always has a 50% duty cycle. If
the CLKFX or CLKFX180 clock outputs are used standalone, then
no clock feedback is required. See “Frequency Synthesizer
(CLKFX, CLKFX180),” and “Half-Period Phase Shifted Outputs.”
CLKFX_MULTIPLY
F CLKFX = F CLKIN • ---------------------------------------------------------
CLKFX_DIVIDE

CLKFX180 Clock Synthesized clock output CLKFX, phase shifted by 180° (appears 9
Output to be an inverted version of CLKFX). Always has a 50% duty
cycle. If only CLKFX or CLKFX180 clock outputs are used on the
DCM, then no feedback loop is required. See “Frequency
Synthesizer (CLKFX, CLKFX180),” and “Half-Period Phase
Shifted Outputs.”
STATUS[0] Output Variable Phase Shift Overflow. Control output for “Variable Fine 9
Phase Shifting.” The Variable Phase Shifter has reached its
minimum or maximum limit value. The limit value is either ±255
or a lesser value if the phase shifter reached the end of the delay
line. See “Variable Fine Phase Shifting,” page 123.
Note: This function is not supported in the Spartan-3E family.
In the Spartan-3 family, STATUS[0] also indicates overflow for a
fixed phase shift selection.

0 The Phase Shifter has not yet reached its limit value.
1 The Phase Shifter has reached its limit value.

STATUS[1] Output CLKIN Input Stopped Indicator. Available only when the CLKFB 9 9 9
feedback input is connected. Held in reset until the LOCKED
output is asserted. Requires at least one CLKIN cycle to become
active. Never asserted if CLKIN never toggles.

0 CLKIN input is toggling.

1 CLKIN input is not toggling, even though the
LOCKED output might still be High. See
“Momentarily Stopping CLKIN”.

STATUS[2] Output CLKFX or CLKFX180 Output Stopped Indicator. See Frequency 9

Synthesizer (CLKFX, CLKFX180).

0 CLKFX and CLKFX180 outputs are toggling.

1 CLKFX and CLKFX180 outputs are not toggling,
even though the LOCKED output might still be
High. See “Momentarily Stopping CLKIN”.

STATUS[7:3] Output Reserved

Spartan-3 Generation FPGA User Guide www.xilinx.com 75

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-6: DCM Connection Ports (Cont’d)

Functional Unit
Port Direction Description
DLL PS DFS
LOCKED Output All DCM features have locked onto the CLKIN frequency. Clock 9 9 9
outputs are now valid, assuming CLKIN is within specified limits
(as described in “DCM Clock Requirements”). See “Frequency
Synthesizer (CLKFX, CLKFX180).”

0 DCM is attempting to lock onto CLKIN frequency.

DCM clock outputs are not valid.
1 DCM is locked onto CLKIN frequency. DCM clock
outputs are valid.
1-to-0 DCM lost lock. Reset DCM.

PSDONE Output Variable Phase Shift operation complete. See “Variable Fine Phase 9
Shifting,” page 123.

0 No phase shift operation is active or phase shift

operation is in progress.
1 Requested phase shift operation is complete. Output
High for one PSCLK cycle. Okay to provide next
Variable Phase Shift operation.

Attributes, Properties, or Constraints

Table 3-7 lists the various attributes for the Digital Clock Manager. All attributes are set at
design time and programmed during configuration. Most, except for the Dynamic Fine
Phase Shift function, cannot be changed by the FPGA application at run-time. To set an
attribute, set <ATTRIBUTE>=<SETTING> as appropriate for the design entry tool.

76 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Primitive

Table 3-7: DCM Attributes

Attribute Allowable Settings and Description
DLL_FREQUENCY_MODE Spartan-3 FPGA Family Only. Specifies the allowable frequency range for the
CLKIN input and for the output clocks from the DCM’s Delay-Locked Loop
(DLL) unit. The DLL clock outputs include CLK0, CLK90, CLK180, CLK270,
CLK2X, CLK2X180, CLKDV.

LOW Default. The DLL function unit operates in its low-frequency mode.
All DLL-related outputs are available. The frequency for all clock
inputs and outputs must fall within the low-frequency DLL limits
specified in the Spartan-3 FPGA Data Sheet.
HIGH The DLL function unit operates in its high-frequency mode. The
Clock Doubler (CLK2X, CLK2X180) outputs are not available. The
Quadrant Phase Shifted Outputs CLK90 and CLK270 are not
available. The duty cycle for the CLKDV output is not 50% if the
CLKDV_DIVIDE attribute has a non-integer value. The frequency
for all clock inputs and outputs must fall within the high-frequency
DLL limits specified in the Spartan-3 FPGA Family Data Sheet.

CLKIN_PERIOD Specifies in ns the period of the clock used to drive the CLKIN pin of the DCM.
Optional input, primarily used only for DRC checks. On Spartan-3E and
Extended Spartan-3A family FPGAs, setting CLKIN_PERIOD helps reduce DFS
jitter and results in faster locking time.
CLK_FEEDBACK Defines the frequency of the feedback clock.

1X Default. CLK0 feedback. Same frequency as CLKIN.

2X CLK2X feedback. Double the frequency of CLKIN.
None No feedback. Allowed if using only the CLKFX or CLKFX180
outputs.

DUTY_CYCLE_CORRECTION Spartan-3 FPGAs only. Enables or disables the 50% duty-cycle correction for the
CLK0, CLK90, CLK180, and CLK270 outputs from the DLL unit. The duty cycles
for all outputs on Spartan-3E and Extended Spartan-3A family FPGAs are
always corrected to 50%.

TRUE Default. Enable duty-cycle correction.

FALSE Disable duty-cycle correction.

CLKDV_DIVIDE Defines the frequency of the CLKDV output. Allowable values for
CLKDV_DIVIDE include 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 9, 10, 11, 12,
13, 14, 15, 16.
F CLKIN
F CLKDV = -----------------------------------------------
-
CLKDV_DIVIDE

The locking time is longer, and there is more output jitter when CLKDV_DIVIDE
is a non-integer value.

Spartan-3 Generation FPGA User Guide www.xilinx.com 77

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-7: DCM Attributes (Cont’d)

Attribute Allowable Settings and Description
CLKFX_MULTIPLY Defines the multiplication factor for the frequency of the CLKFX and CLKFX180
outputs. Used in conjunction with the CLKFX_DIVIDE attribute. Allowable
values for CLKFX_MULTIPLY include integers ranging from 2 to 32. Default
value is 4.
CLKFX_MULTIPLY
F CLKFX = F CLKIN • ---------------------------------------------------------
CLKFX_DIVIDE

CLKFX_DIVIDE Defines the division factor for the frequency of the CLKFX and CLKFX180
outputs. Used in conjunction with the CLKFX_MULTIPLY attribute. Allowable
values for CLKFX_DIVIDE include integers ranging from 1 to 32. Default value
is 1.
CLKFX_MULTIPLY
F CLKFX = F CLKIN • ---------------------------------------------------------
CLKFX_DIVIDE

PHASE_SHIFT The PHASE_SHIFT attribute is applicable only if the CLKOUT_PHASE_SHIFT

attribute is set to FIXED or VARIABLE. Defines the rising-edge skew between
CLKIN and all the DCM clock outputs at configuration and consequently phase
shifts the DCM clock outputs.
The skew or phase shift value is specified as an integer that represents a fraction
of the clock period as expressed in the equations in “Fine Phase Shifting.” The
integer value must range from –255 to 255. The default is 0. Actual allowable
values depend on input clock frequency. The actual range is less when TCLKIN >
FINE_SHIFT_RANGE. The FINE_SHIFT_RANGE specification represents the
total delay of all taps in the delay line. See “Fine Phase Shifting,” for more
information.
CLKOUT_PHASE_SHIFT Sets the phase shift mode. Together with the PHASE_SHIFT constraint,
implements the Digital Phase Shifter (DPS) feature of the DCM. Affects all DCM
clock outputs from both the DLL and DFS units. See “Fine Phase Shifting,” for
more information.

NONE Default. CLKIN and CLKFB are in phase (no skew) and phase
relationship cannot be changed. Equivalent to FIXED setting
with a PHASE_SHIFT value of 0.
FIXED Phase relationship is set at configuration by the PHASE_SHIFT
attribute value and cannot be changed by the application.
VARIABLE Phase relationship is set at configuration by the PHASE_SHIFT
attribute value but can be changed by the application using the
Variable Phase Shift controls, PSEN, PSCLK, PSINCDEC, and
PSDONE.

78 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Primitive

Table 3-7: DCM Attributes (Cont’d)

Attribute Allowable Settings and Description
DESKEW_ADJUST Controls the clock delay alignment between the FPGA clock input pin and the
DCM output clocks. See “Skew Adjustment.”

SYSTEM_SYNCHRONOUS Default. All devices clocked by a common,

system-wide clock source.
SOURCE_SYNCHRONOUS Clock is provided by the data source, i.e.,
source-synchronous applications.

Do not use this setting to phase shift DCM clock outputs. Instead, use the
CLKOUT_PHASE_SHIFT and PHASE_SHIFT constraints to achieve accurate
phase shifting.
DFS_FREQUENCY_MODE Spartan-3 FPGA Family Only. Specifies the allowable frequency range for the
CLKFX and CLKFX180 output clocks from the DCM’s Digital Frequency
Synthesizer (DFS). If any DLL clock outputs are used, then the more restrictive
DLL_FREQUENCY_MODE limits the CLKIN input frequency.

LOW Default. The DFS function unit operates in its low-frequency mode.
The frequency for the CLKFX and CLKFX180 outputs must fall
within the low-frequency DFS limits specified in the Spartan-3
FPGA Data Sheet. The frequency limits for the CLKIN input
depend on if any DLL clock outputs are used.
HIGH The DFS function unit operates in its high-frequency mode. The
frequency for the CLKFX and CLKFX180 outputs must fall within
the high-frequency DFS limits specified in the Spartan-3 FPGA Data
Sheet. The frequency limits for the CLKIN input depend on if any
DLL clock outputs are used.

STARTUP_WAIT Controls whether the FPGA configuration signal DONE waits for the DCM to
assert its LOCKED signal before going High.

FALSE Default. DONE is asserted at the end of configuration without

waiting for the DCM to assert LOCKED.
TRUE The DONE signal does not go High until the LOCKED signal goes
HIGH on the associated DCM. STARTUP_WAIT does not prevent
LOCKED from going High. The FPGA startup sequence must also
be modified to insert a LCK (lock) cycle before the postponed cycle
(see “Bitstream Generation Settings”). Either the DONE cycle or
GWE cycle are typical choices.

If more than one DCM is so configured, the FPGA waits until all DCMs are
locked.

Spartan-3 Generation FPGA User Guide www.xilinx.com 79

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-7: DCM Attributes (Cont’d)

Attribute Allowable Settings and Description
CLKIN_DIVIDE_BY_2 Optionally divides the CLKIN in half before entering DCM block. In some
applications, reduces the input clock frequency to within acceptable limits. Can
be used for either DLL or DFS.

FALSE Default. CLKIN input directly feeds the DCM block.

TRUE Divides CLKIN frequency in half and provides roughly a 50% duty-
cycle clock before entering the DCM block. Helpful with high-
frequency clocks to meet the DCM input clock frequency or duty-
cycle requirements. Divides the clock frequency in half when
determining operating frequency modes and calculating phase shift
limits.

FACTORY_JF Spartan-3 FPGA Family Only. Controls how often the DCM’s DLL unit adjusts
its tap settings. The FACTORY_JF setting affects the jitter characteristics of the
DLL element.
The settings are automatically adjusted based on the DLL_FREQUENCY_MODE
attribute.

DLL_FREQUENCY_MODE FACTORY_JF
LOW 0x8080
HIGH 0x8080

Do not change the default values unless otherwise recommended (see

“Adjusting FACTORY_JF Setting (Spartan-3 FPGA Family Only)”).
LOC Specifies the physical location of the DCM.

DCM Clock Requirements

The DCM is built for maximum flexibility, but there are certain requirements on clock
frequency and clock stability, both frequency variation and clock jitter.

Input Clock Frequency Range

The DCM clock input frequency depends on whether the DLL functional unit, the DFS
unit, or both are utilized in the application.

Table 3-8: DFS Unit Clock Input Frequency Requirements (-4 Speed Grade)
Function Minimum Frequency Maximum Frequency Units
Data Sheet Specification CLKIN_FREQ_FX_MIN CLKIN_FREQ_FX_MAX
Spartan-3 FPGA 1 280 MHz
Spartan-3E FPGA 0.200 333 MHz
Extended Spartan-3A family 0.200 333 MHz
FPGAs

Table 3-8 shows the clock input, CLKIN, frequency range for the Digital Frequency
Synthesizer (DFS) unit. The DFS unit, if used stand-alone, has a wider frequency range

80 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Clock Requirements

than the DLL unit. If the application uses both units, then the more restrictive DLL
requirements apply. The table shows the data sheet specification name and an estimated
value. The actual value depends on which speed grade is required for the design and the
value specified in the data sheet takes precedence over the estimate.

Table 3-9 and Table 3-10 show the clock input, CLKIN, frequency range for the Delay-
Locked Loop (DLL) unit. The DLL frequency restrictions apply regardless if the DLL is
used stand-alone or with the DFS unit. The table shows the data sheet specification name
and value. The actual value depends on which speed grade is required for the design, and
the value specified in the data sheet takes precedence over any values shown in this user
guide.
Spartan-3E and Extended Spartan-3A family FPGAs have a single DLL operating range, as
shown in Table 3-9. The frequencies shown for Spartan-3E FPGAs are for the Stepping 1
revision.

Table 3-9: Extended Spartan-3A Family FPGAs: DLL Unit Clock Input Frequency Requirements

Speed Minimum Maximum

FPGA Family Units
Grade CLKIN_FREQ_DLL_MIN CLKIN_FREQ_DLL_MAX
-4 250 MHz
Extended Spartan-3A family FPGAs
-5 280 MHz
5
-4 240 MHz
Spartan-3E FPGAs (Stepping 1)
-5 270 MHz

Table 3-10 shows the frequency range for Spartan-3 FPGAs, where the DLL has two
distinct operating frequency ranges, called Low and High. The operating mode is
controlled by the DLL_FREQUENCY_MODE attribute.

Table 3-10: Spartan-3 FPGAs: DLL Unit Clock Input Frequency Requirements
DLL Frequency Mode Attribute (DLL_FREQUENCY_MODE)

FPGA = LOW = HIGH

Family Minimum Frequency Maximum Frequency Minimum Frequency Maximum Frequency
CLKIN_FREQ_DLL_LF_MIN CLKIN_FREQ_DLL_LF_MAX CLKIN_FREQ_DLL_HF_MIN CLKIN_FREQ_DLL_HF_MIN

Spartan-3
18 MHz 167 MHz 48 MHz 280 MHz
FPGAs

Spartan-3E and Extended Spartan-3A family FPGA DLLs support input clock frequencies
as low as 5 MHz, whereas the Spartan-3 FPGA DLL requires at least 18 MHz.

Output Clock Frequency Range

The various DCM output clocks also have a specified frequency range. See the “Input and
Output Clock Frequency Restrictions” section for more information.

Input Clock and Clock Feedback Variation

As described later in the “A Stable, Monotonic Clock Input” section, the DCM expects a
stable, monotonic clock input. However, for maximum flexibility, the DCM tolerates a

Spartan-3 Generation FPGA User Guide www.xilinx.com 81

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

certain amount of clock jitter on the CLKIN input and a reasonable amount of frequency
variation on both the CLKIN input and the CLKFB clock feedback input.
There are two types of jitter tolerance on the CLKIN input.
• Cycle-to-cycle jitter
• Period jitter

Cycle-to-Cycle Jitter
Cycle-to-cycle jitter indicates how much the CLKIN input period is allowed to change
from one cycle to the next. The maximum allowable cycle-to-cycle change is shown in
Table 3-11, including the data sheet specification name and an estimated value. The table
also indicates when the specification applies. While Spartan-3E and Extended Spartan-3A
family FPGAs have one distinct operating range, the acceptable amount of cycle-to-cycle
jitter decreases at input frequencies about 150 MHz. For Spartan-3 FPGAs, the limits apply
depending on the DLL_FREQUENCY_MODE attribute setting.

Table 3-11: Maximum Allowable Cycle-to-Cycle Jitter

Frequency Mode/Frequency Range
Functional Unit
Low High
DLL CLKIN_CYC_JITT_DLL_LF CLKIN_CYC_JITT_DLL_HF
DFS CLKIN_CYC_JITT_FX_LF CLKIN_CYC_JITT_FX_HF
Cycle-to-cycle jitter ±300 ps ±150 ps
Spartan-3E, When FCLKIN/FX ≤150 MHz When FCLKIN/FX > 150 MHz
Extended Spartan-3A family FPGAs
Spartan-3 FPGAs When When
DLL_FREQUENCY_MODE = LOW DLL_FREQUENCY_MODE = HIGH

Period Jitter
The other applicable type of jitter is called period jitter. Period jitter indicates the maximum
variation in the clock period over millions of clock cycles. Cycle-to-cycle jitter shows the
change from one clock cycle to the next while period jitter indicates the maximum range of
change over time. The maximum allowable period jitter appears in Table 3-12, including
the data sheet specification name and an estimated value.

Table 3-12: Maximum Allowable Period Jitter

Frequency Mode
Functional Unit
Low High
DLL CLKIN_PER_JITT_DLL_LF CLKIN_PER_JITT_DLL_HF
DFS CLKIN_PER_JITT_FX_LF CLKIN_PER_JITT_FX_HF
Period jitter ± 1,000 ps (± 1 ns)

82 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Clock Requirements

DLL Feedback Delay Variance

Another source of stability for the DCM is the clock feedback path used by the DLL unit.
The feedback path delay variance must also be within the limit shown in Table 3-13. This
limit only applies to an external feedback path as any on-chip variance is minimal when
connected to a global clock line.

Table 3-13: External Feedback Path Delay Variation

Description Specification
CLKFB_DELAY_VAR_EXT
Maximum allowable variation in off-chip CLKFB feedback path
±1,000 ps (±1 ns)

Spread Spectrum Clocks

The Spartan-3E and Extended Spartan-3A family FPGA DCMs accept typical spread
spectrum clocks. The DLL part of the DCM tracks the frequency changes created by the
typical spread spectrum clock, to drive the global clocks to the FPGA fabric. The spread
spectrum clock must meet the DLL input requirements as specified in the device data
sheets. See the Input Clock Jitter Tolerance and Delay Path Variation specifications in the
Recommended Operating Conditions for the DLL, CLKIN_CYC_JITT_DLL and
CLKIN_PER_JITT_DLL.
The DFS can track a typical spread spectrum input as long as it meets the input clock
specifications. If phase shift is used, it should be set to FIXED. See XAPP469 for more
details.

Spartan-3 Generation FPGA User Guide www.xilinx.com 83

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Optimal DCM Clock and External Feedback Inputs

Each DCM has multiple optimal inputs for an incoming clock signal or external feedback
signal.

Spartan-3E FPGA DCM Clock Inputs

Table 3-14 through Table 3-16 list the direct inputs to each DCM on Spartan-3E FPGAs.
Each DCM has up to four direct input pins, used for clock or external feedback
connections. Optionally, these pins are also the direct inputs to the global buffers on the
FPGA. Each table shows all four possible direct inputs, the associated pin number by
package, the associated GCLK, RHCLK, or LHCLK clock input, and the BUFGMUX clock
buffers associated with each DCM. Lastly, each table also includes the LOC location
attribute string from the DCM, the associated BUFGMUX buffers, and the direct input
pins.
The pin number is shown for each potential direct input. Two associated pins can be
combined to form a differential clock input.
Table 3-14, page 85 shows the direct connections to the DCMs associated with the global
clock network. These DCMs are the best choice for the highest-speed clocks in the design
and for clocks with the highest fanout. The top DCMs are associated with I/O Bank 0, and
the bottom DCMs are associated with I/O Bank 2. The XC3S100E has only two “global”
DCMs, located in the upper right and lower right. The outputs from a “global” DCM drive
up to four BUFGMUX clock buffers along the same edge. The two DCMs along an edge
share these four clock buffers. Each of these buffers, in turn, connects to one of the eight
global clock lines.
Table 3-15, page 86 and Table 3-16, page 86 show the direct connections to the left- and
right-edge DCMs available on the XC3S1200E and XC3S1600E FPGAs. The output clocks
from these DCMs are available on the associated half of the FPGA. The left-edge DCMs are
associated with I/O Bank 3, and the right-edge DCMs are associated with I/O Bank 1. The
outputs from a left-edge or right-edge DCM each drive up to four BUFGMUX clock buffers
along the same edge, each of which connects to one of the eight clock lines. These
BUFGMUX buffers provide clocks to half of the chip, whereas the “global” DCMs provide
clocks to the entire FPGA.

84 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Clock Requirements

Table 3-14: Spartan-3E FPGA: Direct Input Connections and Optional External Feedback to Associated
DCMs
I/O Bank 0
Differential Pair Differential Pair Differential Pair Differential Pair
N P N P N P N P
Package Pin Number for Single-Ended Input Pin Number for Single-Ended Input
VQ100 P91 P90 P89 P88 P86 P85 P84 P83
CP132 B7 A7 C8 B8 A9 B9 C9 A10
TQ144 P131 P130 P129 P128 P126 P125 P123 P122
PQ208 P186 P185 P184 P183 P181 P180 P178 P177
FT256 D8 C8 B8 A8 A9 A10 F9 E9
FG320 D9 C9 B9 B8 A10 B10 E10 D10
FG400 A9 A10 G10 H10 E10 E11 G11 F11
FG484 B11 C11 H11 H12 C12 B12 E12 F12
Ð Ð Ð Ð Associated Global Buffers Ð Ð Ð Ð
GCLK11 GCLK10 GCLK9 GCLK8 GCLK7 GCLK6 GCLK5 GCLK4
BUFGMUX_X1Y10

BUFGMUX_X1Y11

BUFGMUX_X2Y10

BUFGMUX_X2Y11
Top Left DCM Top Right DCM
XC3S100E: N/A XC3S100E: DCM_X0Y1
XC3S250E, XC3S500E: DCM_X0Y1 XC3S250E, XC3S500E: DCM_X1Y1
XC3S1200E, XC3S1600E: DCM_X1Y3 XC3S1200E, XC3S1600E: DCM_X2Y3

Ð Ð Ð Ð
H G F E
Global Clock Line
D C B A
Ï Ï Ï Ï
BUFGMUX_X1Y0

BUFGMUX_X1Y1

BUFGMUX_X2Y0

BUFGMUX_X2Y1

Bottom Left DCM Bottom Right DCM

XC3S100E: N/A XC3S100E: DCM_X0Y0
XC3S250E, XC3S500E: DCM_X0Y0 XC3S250E, XC3S500E: DCM_X1Y0
XC3S1200E, XC3S1600E: DCM_X1Y0 XC3S1200E, XC3S1600E: DCM_X2Y0

GCLK12 GCLK13 GCLK14 GCLK15 GCLK0 GCLK1 GCLK2 GCLK3

Ï Ï Ï Ï Associated Global Buffers Ï Ï Ï Ï
Differential Pair Differential Pair Differential Pair Differential Pair
Package P N P N P N P N
Pin Number for Single-Ended Input Pin Number for Single-Ended Input
VQ100 P32 P33 P35 P36 P38 P39 P40 P41
CP132 M4 N4 M5 N5 M6 N6 P6 P7
TQ144 P50 P51 P53 P54 P56 P57 P58 P59
PQ208 P74 P75 P77 P78 P80 P81 P82 P83
FT256 M8 L8 N8 P8 T9 R9 P9 N9
FG320 N9 M9 U9 V9 U10 T10 R10 P10
FG400 W9 W10 R10 P10 P11 P12 V10 V11
FG484 V11 U11 R11 T11 R12 P12 Y12 W12
I/O Bank 2

Spartan-3 Generation FPGA User Guide www.xilinx.com 85

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-15: Spartan-3E FPGA: Direct Input and Optional External Feedback to Left-Edge DCMs
(XC3S1200E and XC3S1600E)
Diff. Single-Ended Pin Number by Package Type Left Edge
Clock VQ100 CP132 TQ144 PQ208 FT256 FG320 FG400 FG484 LHCLK DCM/BUFGMUX
BUFGMUX_X0Y5 Î D
BUFGMUX_X0Y4 Î C
P P9 F3 P14 P22 H5 J5 K3 M5 Î LHCLK0

Clock Lines
Pair

N P10 F2 P15 P23 H6 J4 K2 L5 Î LHCLK1

DCM_X0Y2
P P11 F1 P16 P24 H3 J1 K7 L8 Î LHCLK2
Pair

N P12 G1 P17 P25 H4 J2 L7 M8 Î LHCLK3

I/O Bank 3

BUFGMUX_X0Y3 Î B
BUFGMUX_X0Y2 Î A

BUFGMUX_X0Y9 Î H
BUFGMUX_X0Y8 Î G
P P15 G3 P20 P28 J2 K3 M1 M1 Î LHCLK4

Clock Lines
Pair

N P16 H1 P21 P29 J3 K4 L1 N1 Î LHCLK5

DCM_X0Y1
P P17 H2 P22 P30 J5 K6 M3 M3 Î LHCLK6
Pair

N P18 H3 P23 P31 J4 K5 L3 M4 Î LHCLK7

BUFGMUX_X0Y7 Î F
BUFGMUX_X0Y6 Î E

Table 3-16: Spartan-3E FPGA: Direct Input and Optional External Feedback to Right-Edge DCMs
(XC3S1200E and XC3S1600E)
Right Edge Single-Ended Pin Number by Package Type Diff.
DCM/BUFGMUX RHCLK VQ100 CP132 TQ144 PQ208 FT256 FG320 FG400 FG484 Clock
D Í BUFGMUX_X3Y5
C Í BUFGMUX_X3Y4
RHCLK7 Í P68 G13 P94 P135 H11 J14 J20 L19 N
Clock Lines

Pair
RHCLK6 Í P67 G14 P93 P134 H12 J15 K20 L18 P
DCM_X3Y2
RHCLK5 Í P66 H12 P92 P133 H14 J16 K14 L21 N
Pair
RHCLK4 Í P65 H13 P91 P132 H15 J17 K13 L20 P
I/O Bank 1
B Í BUFGMUX_X3Y3
A Í BUFGMUX_X3Y2

H Í BUFGMUX_X3Y9
G Í BUFGMUX_X3Y8
RHCLK3 Í P63 J14 P88 P129 J13 K14 L14 M16 N
Clock Lines

Pair

RHCLK2 Í P62 J13 P87 P128 J14 K15 L15 M15 P

DCM_X3Y1
RHCLK1 Í P61 J12 P86 P127 J16 K12 L16 M22 N
Pair

RHCLK0 Í P60 K14 P85 P126 K16 K13 M16 N22 P

F Í BUFGMUX_X3Y7
E Í BUFGMUX_X3Y6

86 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Clock Requirements

Extended Spartan-3A Family FPGA DCM Clock Inputs

Table 3-18 through Table 3-20 list the direct inputs to each DCM on Extended Spartan-3A
family FPGAs. References to Spartan-3A platform part numbers also apply to the Spartan-
3AN platform. Each DCM has up to four direct input pins, used for clock or external
feedback connections. Optionally, these pins are also the direct inputs to the global buffers
on the FPGA. Each table shows all four possible direct inputs, the associated pin number
by package, the associated GCLK, RHCLK, or LHCLK clock input, and the BUFGMUX
clock buffers associated with each DCM. Lastly, each table also includes the LOC location
attribute string from the DCM, the associated BUFGMUX buffers, and the direct input
pins.
The pin number is shown for each potential direct input. Two associated pins can be
combined to form a differential clock input.
Table 3-18, page 88 shows the direct connections to the DCMs associated with the global
clock network. These DCMs are the best choice for the highest-speed clocks in the design
and for clocks with the highest fanout. The top DCMs are associated with I/O Bank 0, and
the bottom DCMs are associated with I/O Bank 2. The XC3S50A has only two “global”
DCMs, those located in the upper left and upper right. The outputs from a “global” DCM
drive up to four BUFGMUX clock buffers along the same edge. The two DCMs along an
edge share these four clock buffers. Each of these buffers, in turn, connects to one of the
eight global clock lines.
Table 3-19, page 89 and Table 3-20, page 89 show the direct connections to the left- and
right-edge DCMs available on the XC3S700A and XC3S1400A and Spartan-3A DSP
FPGAs. The output clocks from these DCMs are available on the associated half of the
FPGA. The left-edge DCMs are associated with I/O Bank 3, and the right-edge DCMs are
associated with I/O Bank 1. The outputs from a left-edge or right-edge DCM each drive up
to four BUFGMUX clock buffers along the same edge, each of which connects to one of the
eight clock lines. These BUFGMUX buffers provide clocks to half of the chip, whereas the
“global” DCMs provide clocks to the entire FPGA.
When using the DCM to generate high speed clocks to drive the double data rate ODDR2,
a specific BUFGMUX is recommended for CLKFX and another BUFGMUX is
recommended for CLKFX180 to minimize period jitter. See Table 3-17.

Table 3-17: Recommended DCM/BUFG Connections

DCM Recommended BUFGMUX
XC3S700A/AN
XC3S200A/AN XC3S1400A/AN
XC3S50A/AN CLKFX CLKFX180
XC3S400A/AN XC3SD1800A
XC3SD3400A
- X0Y0 X1Y0 X2Y1 X1Y0
- X1Y0 X2Y0 X2Y1 X1Y0
- - X0Y1 X0Y6 X0Y9
- - X0Y2 X0Y2 X0Y5
X0Y0 X0Y1 X1Y3 X2Y11 X1Y10
X1Y0 X1Y1 X2Y3 X2Y11 X1Y10
- - X3Y2 X3Y2 X3Y5
- - X3Y1 X3Y6 X3Y9

Spartan-3 Generation FPGA User Guide www.xilinx.com 87

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-18: Extended Spartan-3A Family: Direct Input Connections and Optional External DCM Feedback
I/O Bank 0
Differential Pair Differential Pair Differential Pair Differential Pair
N P N P N P N P
Package Pin Number for Single-Ended Input Pin Number for Single-Ended Input
VQ100 P90 N/A P89 P88 P86 P85 P84 P83
TQ144 P132 P130 P131 P129 P127 P125 P126 P124
FT256 B8 A8 D8 C8 A9 C9 D9 C10
FG320 C8 B8 B7 A8 B9 A10 C9 B10
FG400 A8 A9 E10 D10 C10 A10 E11 D11
CS484 B8 A8 E11 F10 B9 A9 F11 E12
FG484 E11 D11 C11 B11 A11 A12 E12 C12
FG676 C13 B13 G13 F13 A14 B14 J14 K14
Ð Ð Ð Ð Associated Global Buffers Ð Ð Ð Ð
GCLK11 GCLK10 GCLK9 GCLK8 BUFGMUX_X1Y10 GCLK7 GCLK6 GCLK5 GCLK4

BUFGMUX_X2Y10
BUFGMUX_X1Y11

BUFGMUX_X2Y11
Top Left DCM Top Right DCM
XC3S50A: DCM_X0Y0 XC3S50A: DCM_X1Y0
XC3S200A, XC3S400A: DCM_X0Y1 XC3S200A, XC3S400A: DCM_X1Y1
XC3S700A/1400A, Spartan-3A DSP XC3S700A/1400A, Spartan-3A DSP
FPGA: DCM_X1Y3 FPGA: DCM_X2Y3
Ð Ð Ð Ð
H G F E
Global Clock Line
D C B A
Ï Ï Ï Ï
Bottom Left DCM Bottom Right DCM
BUFGMUX_X1Y0

BUFGMUX_X1Y1

BUFGMUX_X2Y0

BUFGMUX_X2Y1

XC3S50A: N/A XC3S50A: N/A

XC3S200A, XC3S400A: DCM_X0Y0 XC3S200A, XC3S400A: DCM_X1Y0
XC3S700A/1400A, Spartan-3A DSP XC3S700A/1400A, Spartan-3A DSP
FPGA: DCM_X1Y0 FPGA: DCM_X2Y0
GCLK12 GCLK13 GCLK14 GCLK15 GCLK0 GCLK1 GCLK2 GCLK3
Ï Ï Ï Ï Associated Global Buffers Ï Ï Ï Ï
Differential Pair Differential Pair Differential Pair Differential Pair
Package P N P N P N P N
Pin Number for Single-Ended Input Pin Number for Single-Ended Input
VQ100 N/A N/A P40 P41 P43 P44 N/A N/A
TQ144 N/A N/A P54 P55 P57 P59 P58 P60
FT256 R7(1) T7(1) P8 T8 N9 P9 R9 T9
FG320 U8 V8 U9 V9 U10 T10 V11 U11
FG400 W9 Y9 V10 W10 Y11 V11 U11 V12
CS484 Y11 Y10 AA12 AB12 U12 V12 AB13 AA14
FG484 U11 V11 W12 Y12 AA12 AB12 V12 U12
FG676 AA13 Y13 AF13 AE13 Y14 AA14 AF14 AE14
I/O Bank 2
1. N/A in XC3S50A

88 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

DCM Clock Requirements

Table 3-19: Extended Spartan-3A Family FPGA: Direct Clock Input and Optional External Feedback to
Left-Edge DCMs (XC3S700A/AN, XC3S1400A/AN, and Spartan-3A DSP FPGAs)
Diff. Single-Ended Pin Number by Package Type Left Edge
Clock FT256 FG400 FG484 CS484 FG676 LHCLK DCM/BUFGMUX
BUFGMUX_X0Y5 Î D
BUFGMUX_X0Y4 Î C
P G2 J1 L5 L6 N6 Î LHCLK0

Clock Lines
Pair

N H1 K2 L3 M5 N7 Î LHCLK1
DCM_X0Y2
P H3 K3 K1 K1 P1 Î LHCLK2
Pair

N J3 L3 L1 L1 P2 Î LHCLK3
I/O Bank 3

BUFGMUX_X0Y3 Î B
BUFGMUX_X0Y2 Î A

BUFGMUX_X0Y9 Î H
BUFGMUX_X0Y8 Î G
P J2 K4 M1 L3 P4 Î LHCLK4

Clock Lines
Pair

N J1 L5 M2 M2 P3 Î LHCLK5
DCM_X0Y1
P K3 L1 M3 M6 N9 Î LHCLK6
Pair

N K1 M1 M4 N7 P10 Î LHCLK7
BUFGMUX_X0Y7 Î F
BUFGMUX_X0Y6 Î E

Table 3-20: Extended Spartan-3A Family FPGA: Direct Clock Input and Optional External Feedback to
Right-Edge DCMs (XC3S700A/AN, XC3S1400A/AN, and Spartan-3A DSP FPGAs)
Right Edge Single-Ended Pin Number by Package Type Diff.
DCM/BUFGMUX RHCLK FT256 FG400 FG484 CS484 FG676 Clock
D Í BUFGMUX_X3Y5
C Í BUFGMUX_X3Y4
RHCLK7 Í H16 J20 K19 L17 N19 N
Clock Lines

Pair
RHCLK6 Í H15 K20 K20 M18 P18 P
DCM_X3Y2
RHCLK5 Í H14 L17 M20 L20 N24 N Pair
RHCLK4 Í J14 K18 M18 L21 P23 P
I/O Bank 1
B Í BUFGMUX_X3Y3
A Í BUFGMUX_X3Y2

H Í BUFGMUX_X3Y9
G Í BUFGMUX_X3Y8
RHCLK3 Í J16 L18 L20 M20 P25 N
Clock Lines

Pair

RHCLK2 Í K16 L19 L21 N21 P26 P

DCM_X3Y1
RHCLK1 Í K14 M20 L22 M17 P20 N
Pair

RHCLK0 Í K15 M19 M22 N18 P21 P

F Í BUFGMUX_X3Y7
E Í BUFGMUX_X3Y6

Spartan-3 Generation FPGA User Guide www.xilinx.com 89

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

LOCKED Output Behavior

The DCM’s LOCKED output indicates when all the enabled DCM functions have locked to
the CLKIN input. When the DCM asserts LOCKED, the output clocks are valid for use
within the FPGA application.
Figure 3-5 shows the behavior of the LOCKED output. The LOCKED output is Low
immediately after the FPGA finishes its configuration process and is Low whenever the
RST input is asserted.
After configuration, the DCM always attempts to lock, whether the CLKIN signal is valid
yet or not. If the input clock changes, assert the RST input until the CLKIN input stabilizes.
Once the RST input is released, the DCM again relocks to the new CLKIN input frequency.
The DLL unit uses both the CLKIN input and the CLKFB feedback input to determine
when locking is complete, that is, when the rising edges of CLKIN and CLKFB are phase-
aligned. The DFS unit monitors the CLKIN input to determine if a valid frequency is
present on CLKIN. To achieve lock, the DCM might need to sample several thousand clock
cycles.
The DCM asserts its LOCKED output High when its internal state machine has locked onto
the CLKIN input. The DCM clock outputs are then valid and available for use within the
FPGA application. The DCM timing section of the data sheet provides worst-case locking
times. In general, the DLL unit outputs lock faster with increasing clock frequency. The
DFS unit outputs require significantly longer to lock, depending on the multiply and
divide factors. Smaller multiply and divide factors result in faster lock times.
To guarantee that the system clock is established before the FPGA completes its
configuration process, the DCM can optionally delay the completion of the configuration
process until after the DCM locks. The STARTUP_WAIT attribute activates this feature.
Until LOCKED is High, there is no guarantee how the DCM clock outputs behave. The
DCM output clocks are not valid until LOCKED is High and before that time can exhibit
glitches, spikes, or other spurious behavior.
The LOCKED signal might stay High when CLKIN stops - see “Momentarily Stopping
CLKIN”. The LOCKED signal might also stay High when CLKIN varies considerably - see
“A Stable, Monotonic Clock Input”.
In the Extended Spartan-3A family, when two adjacent DCMs are used, the outputs
should be considered valid once both DCMs are LOCKED. Adjacent DCMs should share
the same reset signal.

90 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

LOCKED Output Behavior

FPGA
Configuration
Startup Phase
RST Input
asserted

LOCKED output
is LOW

If CLKIN not yet stable,

assert RST input until
CLKIN stabilizes
Is
CLKIN stable? N FPGA application
Within specified asserts RST input
limits?

Y If lock is lost, assert RST

input to force DCM to
reacquire lock
Phase Lost lock.
aligned? N LOCKED output
Output clocks is LOW
good?

LOCKED output
is HIGH

x462_05_120206

Figure 3-5: Functional Behavior of LOCKED Output

While the CLKIN input stays within the specified limits, the DCM continues to adjust its
internal delay taps to maintain lock. However, if the CLKIN input strays well beyond the
specified limits, then the DCM potentially loses lock and deasserts the LOCKED output.
Once the DCM loses lock, it does not automatically attempt to reacquire lock. When the
DCM loses lock—i.e., LOCKED was High, then goes Low—the FPGA application must
take the appropriate action. For example, once lock is lost, resetting the DCM via the RST
input forces the DCM to reacquire lock.

Using the LOCKED Signal

To operate properly, the DCM requires a stable, monotonic clock input. Once locked, the
DCM tolerates clock period variations up to the value specified in the specific FPGA data
sheet. If the input clock stays within the specified limits, then the output clocks always are
valid when the LOCKED output is High. However, it is possible for the clock to stray well
outside the limits, for the LOCKED output to stay High, and for the DCM outputs to be
invalid. It is good design practice to monitor both LOCKED and the STATUS signals.
Monitoring STATUS[1] is recommended as this will indicate when CLKIN has stopped
(moved outside the acceptable CLKIN tolerances). STATUS[1] will go High after one
missed CLKIN cycle. However, the DCM might not lose LOCKED unless CLKIN is
stopped for more than 100 ms. STATUS[1] is not a sticky bit; it will go Low once CLKIN has
returned. For the most robust indicator of the status of your DCM's output clock,
monitoring both the LOCKED and STATUS[1] bits is recommended.

Spartan-3 Generation FPGA User Guide www.xilinx.com 91

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Spartan-3A FPGA DCM Digital Frequency Synthesizer Requires

Additional Lock Circuitry
To help guarantee DFS lock, a circuit is automatically inserted by the ISE development
software starting with version 9.1i for the Extended Spartan-3A family FPGAs
(Figure 3-6). Using FPGA logic, the circuit monitors both the LOCKED output from the
DCM_SP function and the STATUS[2] bit, which indicates that the DFS output CLKFX has
stopped. If LOCKED = 0 and STATUS[2] = 1, then the circuit asserts the DCM RESET input.
If the FPGA application also resets the DCM, then OR the reset signal from the FPGA
application with the monitored output signals.

DCM_SP
(from FPGA application)
DCM_RESET STATUS[2]
OR RESET LOCKED

AND

DCM_LOCKED
CLKFX_STOPPED
EN035_02_101806

Figure 3-6: Spartan-3A FPGA DCM DFS Lock Logic

RST Input Behavior

The asynchronous RST input forces the DCM to its post-configuration state. Use the RST
pin when changing the input clock frequency beyond the allowable range. The active-High
RST pin either must connect to a dynamic signal or must be tied to ground. The RST input
must be asserted for three valid CLKIN cycles or longer.
If the input clock frequency is not yet stable after configuration, assert RST until the clock
stabilizes. When using external feedback, hold the DCM in reset immediately after
configuration. Figure 3-20, page 107 shows an example reset technique using an SRL16
shift register primitive.
If the DCM loses lock—i.e., the LOCKED output was High then goes Low—then the FPGA
application must assert RST to force the DCM to reacquire the input clock frequency.
If the DCM LOCKED output is High, then the LOCKED signal deactivates within four
source clock cycles after RST is asserted. Asserting RST forces the DCM to reacquire lock.
Asserting RST also resets the DCM’s delay tap position to zero. Due to the tap position
changes, glitches might occur on the DCM clock output pins. Similarly, the duty cycle on
the clock outputs might be affected when RST is asserted.
Asserting RST also resets the present variable phase shift value back to the value specified
by the PHASE_SHIFT attribute.

92 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clocking Wizard

Clocking Wizard
To simplify applications using DCMs, the Xilinx ISE development software includes a
software wizard that provides step-by-step instructions for configuring a DCM. As shown
in Figure 3-7, Clocking Wizard generates a vendor-specific logic synthesis file instantiating
the DCM in either VHDL or Verilog syntax. Similarly, Clocking Wizard generates a user
constraints file (UCF) for the specific implementation. Finally, all the user specifications are
saved in a Xilinx Architecture Wizard (XAW) settings file.

Clocking Wizard
Graphically configure a
Spartan-3 Digital Clock
Manager (DCM)

Xilinx Architecture
Vendor-specific Wizard (XAW)
VHDL or Verilog settings file

User contraints
file (UCF)
UG331_c3_28_022407

Figure 3-7: Clocking Wizard Provides a Graphical Interface for Configuring Digital
Clock Managers

Invoking Clocking Wizard

There are multiple methods to invoke Clocking Wizard, either from the Windows Start
button or from within the Xilinx ISE Project Navigator software.

From Windows Start Button

To invoke Clocking Wizard from the Windows Start button, click Start Æ Programs Æ
Xilinx ISE Æ Accessories Æ Architecture Wizard. The setup window shown in Figure 3-8
appears.
• Specify the name of the Xilinx Architecture Wizard (.xaw) file that holds the option
settings for this DCM.
• Optionally, click Browse and select a directory location for the *.xaw file.
• Select the logic synthesis language for the output file, either VHDL or Verilog.
• Choose the targeted logic synthesis tool. Clocking Wizard creates vendor-specific
output for the specified synthesis tool.
• Select the targeted Spartan-3 generation FPGA.

Spartan-3 Generation FPGA User Guide www.xilinx.com 93

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Enter the filename to

save the settings for
this DCM module
Click here to select
the directory for the
filename

Choose the
language for the
generated output

Choose the targeted

logic synthesis
package
Click here to select
the Spartan-3
Generation FPGA
part number for your
application

Click OK
when finished UG331_c3_05_120206

Figure 3-8: Set Up the Architecture Wizard

From within Project Navigator

Optionally, invoke Clocking Wizard from within Project Navigator, either from the menu
bar or from within the “Sources in Project” window. From the menu bar, select Project Æ
New Source. Alternatively, right-click in the “Sources in Project” window and choose New
Source.
Select IP (Coregen & Architecture Wizard) from the available list, as shown in Figure 3-9.
Enter the file name for the Xilinx Architecture Wizard (*.xaw) file, and select the directory
where the file will be saved. Click Next > to continue.

Select Clocking
Wizard

Choose a
specific wizard

Click OK
when finished UG332_c3_06_120206

Figure 3-9: Configuring a New Architecture Wizard in the Project Navigator

94 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clocking Wizard

General Setup
Specify most of the DCM’s options using the Xilinx Clocking Wizard General Setup panel,
as shown in Figure 3-10. The text in blue ovals shows the DCM primitive attribute name
for the corresponding setting.
• To select the outputs and functions used in the final application, check the option
boxes next to the desired DCM clock outputs. Checking the output boxes enables
related option settings below.
• Enter the frequency of the CLKIN clock input. Either specify the frequency in MHz, or
specify the clock period in nanoseconds. The specified value also sets the DCM’s
DLL_FREQUENCY_MODE attribute for Spartan-3 FPGA designs.
• Specify whether the CLKIN source is internal or external to the FPGA. If External,
then Clocking Wizard automatically inserts a global buffer input (IBUFG) primitive. If
Internal, then the source signal is provided as a top-level input within the generated
HDL source file.
• If the CLKDV output box is checked, then specify the Divide by Value for the Clock
Divider circuit. This setting defines the DCM’s CLKDV_DIVIDE attribute.
• Specify the feedback path to the DCM. If only the CLKFX or CLKFX180 outputs are
used, then select None. Otherwise, feedback is required. If the feedback is from within
the FPGA, choose Internal. If the feedback loop is from outside the FPGA, choose
External. Furthermore, specify the source of the DCM feedback, either from CLK0
(1X) or from CLK2X (2X). This setting defines the DCM’s CLK_FEEDBACK attribute.

Spartan-3 Generation FPGA User Guide www.xilinx.com 95

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Check CLKDV to
enable the Clock
Divider options

Check CLKFX or
DCM attribute name CLKFX180 to
enable the
Frequency
Synthesizer
options
Select Fixed to
phase shift all
outputs by the
value defined
below. Select
Enter input clock Variable mode to
frequency, with dynamically
full accuracy, in adjust phase
MHz or ns shifting using the
PSEN,
DLL_FREQUENCY_MODE PSINCDEC, and
PSCLK inputs.
PHASE_SHIFT

Selecting External automatically

connects CLKIN to an IBUFG Sets the Fixed phase shift
global buffer input primitive. value or the initial value
Select Internal to connect CLKIN for Variable phase shift
to another source. CLKOUT_PHASE_SHIFT
mode, measured as
256ths of a clock period

If clock or Indicate whether the

feedback is feedback is from an
External, Internal or External source
choose or None
whether the
input is
Single-ended
or Differential Set the divider
for the Clock If clock feedback is
Divider output, required, is it from the CLK0
CLKDV output (1X) or the CLK2X
CLKDV_DIVIDE CLK_FEEDBACK output (2X)?

DUTY_CYCLE_CORRECTION

Click for Click Next to

Advanced options continue
UG331_c3_07_120206

Figure 3-10: A Majority of DCM Options are Set in the General Setup Panel

• Specify whether to phase shift all DCM outputs. By default, there is no phase shifting
(None). If phase shifting is required by the application, choose whether the phase shift
value is Fixed or Variable. Selecting Variable also enables the Variable Phase Shift
controls, PSEN, PSINCDEC, PSCLK, and PSDONE. This setting defines the DCM’s
CLKOUT_PHASE_SHIFT attribute. For both Fixed and Variable modes, specify the
related Phase Shift Value, which provides either the fixed phase shift value or the

96 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clocking Wizard

initial value for the Variable Phase Shift. This setting defines the DCM’s
PHASE_SHIFT attribute.
• To open the Advanced Options window, click Advanced.
• When finished, click Next > to continue to the Clock Buffers panel.

Advanced Options
Various advanced DCM options are grouped together in the Advanced Options window,
shown in Figure 3-11:
• By default, the DCM has no effect on the FPGA’s configuration process. Click Wait for
DCM lock before DONE signal goes high to have the FPGA wait for the DCM to
assert its LOCKED output before asserting the DONE signal at the end of
configuration. This setting defines the DCM’s STARTUP_WAIT attribute. If checked,
additional bitstream generation option changes are required, as described in the
“Setting Configuration Logic to Wait for DCM LOCKED Output” section.
• If the CLKIN input frequency is too high for a particular DCM feature, check Divide
Input Clock by 2 to reduce the input frequency by half with nearly ideal 50% duty
cycle before entering the DCM block. This setting defines the DCM’s
CLKIN_DIVIDE_BY_2 attribute.
• If required for source-synchronous data transfer applications, modify the DCM
Deskew Adjust value to SOURCE_SYNCHRONOUS. Do not use any values other
than SOURCE_SYNCHRONOUS or SYSTEM_SYNCHRONOUS without first
consulting Xilinx. This setting defines the DCM’s DESKEW_ADJUST attribute. See
“Skew Adjustment.”
• Click OK when finished to apply any changes and return to the General Setup
window.

Controls how much

skew is purposely
added to the DCM
clock path

Controls whether the

FPGA configuration DESKEW_ADJUST
process waits for the
DCM to lock before
asserting the DONE STARTUP_WAIT
signal
CLKIN_DIVIDE_BY_2
Optionally divides CLKIN
frequency by 2

Automatically insert reset

logic when using external
feedback

Click OK to apply
DCM attribute name changes and
close this window UG331_c3_08_022407

Figure 3-11: DCM Advanced Options Panel

Spartan-3 Generation FPGA User Guide www.xilinx.com 97

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Clock Buffers
Define the clock buffer output type for each DCM clock output, shown in Figure 3-12. By
default, Clocking Wizard automatically assigns all outputs to a global buffer (BUFG).
However, there are only four global buffers along each the top or bottom edge of the
device, shared by two DCMs. In the XC3S50, there is a single DCM along the top or bottom
edge that optionally connects to all four global buffers along the edge.
• To assign clock buffer types for each DCM clock output, click Customize under Clock
Buffer Settings.
• For each DCM clock output, select a Clock Buffer output type using the drop-down
list. Table 3-21 lists the available Clock Buffer options.
• If using an Enabled Buffer output type, either specify a signal name for the buffer
enable (CE) input or use the automatically generated name.
• If using a Clock Mux output type, either specify a signal name for the select (S) input
or use the automatically generated name.
• When finished, click Next > or Finish to continue. The Next > option only appears if
the CLKFX or CLKFX180 outputs were selected in the General Setup panel.
Otherwise, click Finish to generate the HDL output (see “Generating HDL Output”).

By default, Clock
Wizard places
global buffers
(BUFG) on all the
selected DCM
clock outputs

Optionally ,
customize how
the DCM clock
outputs connect to
the other FPGA
logic using the
buttons below

For each clock

output , select the
type of buffer
connecting the
signal to the
FPGA

Click Next to
continue UG331_c3_09_120206

Figure 3-12: Clocking Wizard Provides a Variety of Buffer Options for each DCM Output

98 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clocking Wizard

Table 3-21: Settings for Clock Buffer Output Types

Clock Buffer
Diagram Description
Selection
Global Buffer Connect to one of four global buffers (BUFG) along the same edge as the DCM.
BUFG
I0 O

Enabled Buffer BUFGCE Connect to one of the four global buffers configured as an enable clock buffer
I0 O (BUFGCE). The CE input enables the buffer when High. When CE is Low, the
buffer output is zero.
CE
CE O
0 0
1 I0

Clock Mux BUFGMUX Connect to one of the four global buffers configured as a clock multiplexer
I0 (BUFGMUX). The S input selects the clock source.
O
I1
S O
S
0 I0
1 I1

Lowskewline Connect to low-skew programmable interconnect.

I0

Local Routing I0 Connect to local interconnect, skew not critical.

Clock Frequency Synthesizer

The Clock Frequency Synthesizer panel, shown in Figure 3-13, only appears if the CLKFX
or CLKFX180 outputs were selected in the General Setup panel.
Here, specify either the desired output frequency or enter the specific values for the
Multiply and Divide factors. The frequency limits—or delay limits if CLKIN was specified
in ns—appear under Valid Ranges for Selected Speed Grade. The range is displayed for
possible values of the DFS_FREQUENCY_MODE attribute, which only applies to
Spartan-3 FPGAs. The range is tighter if the DCM uses any of the DLL-related clock
outputs.
• Click Use output frequency and enter the requested value, in as much precision as
possible, either in megahertz (MHz) or in nanoseconds (ns). Click Calculate to
compute the values for the CLKFX_MULTIPLY and CLKFX_DIVIDE attributes. If no
solution is available using the possible multiply and divide values, Clocking Wizard
issues an error message asking for another output frequency value. If a solution exists,
then the multiply and divide values, plus the resulting jitter values (see “Clock Jitter
or Phase Noise”) appear under Generated Output.

Spartan-3 Generation FPGA User Guide www.xilinx.com 99

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

• Optionally, click Use Multiply (M) and Divide (D) values and enter the desired
values. Click Calculate to calculate the resulting output frequency and jitter,
displayed under Generated Output.
• Finally, click Next to generate the HDL output (see “Generating HDL Output”).

DFS_FREQUENCY_MODE

Displays the incoming

clock frequency or
period , specified earlier Displays the frequency
limits for the Frequency
Enter the desired output Synthesizer. Spartan-3
frequency, in MHz or ns, FPGAs have both low -
then click Calculate. and high -frequency limits
Clocking Wizard calculates
the best multiply (M) and CLKFX_MULTIPLY
divide (D) values
possible. CLKFX_DIVIDE

Optionally , enter the After entering the desired output

specific values for the frequency or multiply and divide
multiply (M) and divide values, click Calculate to compute the
(D) values, then click resulting jitter for the Frequency
Calculate Synthesizer output

CLKFX_MULTIPLY CLKFX_DIVIDE

Click Next Displays the calculated

DCM attribute name when finished output jitter values based
on the settings
UG331_c3_10_022607

Figure 3-13: Set the Multiply and Divide Values for the Digital Frequency Synthesizer and Calculate the
Resulting Jitter

Generating HDL Output

After reviewing that all the parameters are correct, as shown in Figure 3-14, click Finish.
Clocking Wizard then generates the requested VHDL or Verilog HDL output file. Clocking
Wizard also generates a User Constraints File (UCF) based on the settings.

100 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

VHDL and Verilog Instantiation

Review options

Click Finish to
generate output file UG331_c3_11_120206

Figure 3-14: Review Settings, Then Click Finish

VHDL and Verilog Instantiation

Clocking Wizard is the easiest method to create a VHDL or Verilog HDL description of a
DCM. However, Verilog and VHDL source examples are also available.

Language Templates within Project Navigator

There are DCM language templates available within the ISE Project Navigator. To select a
DCM template, select Edit Æ Language Templates from the Project Navigator menu. From
the Templates tree shown in Figure 3-15, expand either the Verilog or VHDL folder, then
the Device Primitive Instantiation folder, then FPGA Æ Clock Components Æ Digital
Clock Manager (DCM) folder. Under the DCM folder, select the desired DCM source file.
The source file for the selected DCM appears in the adjacent window.

Spartan-3 Generation FPGA User Guide www.xilinx.com 101

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Click the “light bulb” icon

for language templates

Choose a language

Cut and paste the

template into the
design source file

Choose the
appropriate DCM
function

UG331_c3_12_120206

Figure 3-15: DCM Coding Examples in Project Navigator Language Templates

Use the file either as a reference or cut the content of the window into a new source file.

Eliminating Clock Skew

One of the fundamental functions of a DCM is to eliminate clock skew. Eliminating clock
skew is especially important for higher-performance designs operating at 50 MHz or more.
Furthermore, the concepts involved in clock skew elimination also apply to many of the
other applications of a DCM.

What is Clock Skew?

Clock skew inherently exists in every synchronous system. A pristine clock edge generated
by the clock source actually arrives at different times at different points in the system—
either within a single device or on the clock inputs to the different devices connected to the
clock. This difference in arrival times is called clock skew.
Figure 3-16 illustrates clock skew in an example system. A clock source drives the clock
input to an FPGA. The clock enters through an input pin on the FPGA, is distributed
within the FPGA using the internal low-skew global clock network, and arrives at a flip-
flop within the FPGA. Each element in the clock path delays the arrival of the clock edge at
the flip-flop. Consequently, the clock input at the flip-flop—Point (B)—is delayed, or
skewed compared to the original clock source at Point (A). In this example, this clock skew
or difference in arrival time for this path is called Δb.

102 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Eliminating Clock Skew

FPGA
B
Other
A C Device on
Board

C
Δb
Δc X462_16_123105

Figure 3-16: Clock Skew Inherently Exists in Every Synchronous System

Similarly, the clock source is rebuffered in the FPGA and drives another device on the
board. In this case, again the clock source enters the FPGA via an input pin, is distributed
via the global clock network, feeds an output pin on the FPGA, and finally connects to the
other device via a trace on the printed circuit board (PCB). Because there is more total delay
in this clock path, the resulting skew, Δc, is also larger.

Clock Skew: The Performance Thief

Clock skew potentially reduces the overall performance of the design by increasing setup
times and lengthening clock-to-output delays—both of which increase the clock cycle time.
Similarly, clock skew might require lengthy hold times on some devices. Otherwise,
unreliable operation might result.

Make It Go Away!
Is there a way to eliminate clock skew? Fortunately, a DCM provides such capabilities.
Figure 3-17 shows the same example design as Figure 3-16, except this time implemented
in a Spartan-3 generation FPGA. Two DCMs eliminate the clock skew: one DCM
eliminates the skew for clocked items within the FPGA, the other DCM eliminates the
skew when clocking the other device on the board. The result is practically ideal alignment
between the clock at Points (A), (B), and (C)!
How is clock skew elimination accomplished? Remember, clock skew is caused by the
delay in the clock path. In Figure 3-17, the clock at Point (B) was skewed by Δb and the
clock at Point (C) was skewed by Δc. What if there was a way to provide Point (B) with an
early version of the clock, advanced by Δb and a way to provide Point (C) with an early
version of the clock, advanced by Δc? The result would be that all clocks would arrive at
their destinations with perfect clock edge alignment. Such perfect alignment reduces setup
times, shortens clock-to-output delays, and increases overall system performance.

Spartan-3 Generation FPGA User Guide www.xilinx.com 103

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Spartan-3 Generation FPGA

B

DCM
Other
A C Device on

DCM
Board

Ideal Clock Alignment

Early Clocks Eliminate Skew

UG331_c3_22_1203

Figure 3-17: Eliminating Clock Skew in a Spartan-3 Generation FPGA Design

Predicting the Future by Closely Examining the Past

Even though Spartan-3 generation FPGAs employ highly advanced digital logic, they
cannot predict the future. However, a DCM applies its knowledge of the past behavior of
the clock to predict the future. Most input clocks to a system have a never-changing,
monotonic frequency. Consequently, the input clock has a nearly constant period, T.
Because it is impossible to insert a negative delay to counteract the clock skew, the DCM
actually delays the output clocks enough so that they appear to be advanced in time. How
is this accomplished? The clock cycle is repetitive and has a fixed period, T. As shown in
Figure 3-18, the clock at Point (B) appears to be advanced in time by the delay Δb. In reality
however, the clock is delayed by (T – Δb). Similarly, the clock at Point (C) is delayed by
(T – Δc).

104 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Eliminating Clock Skew

Clock Period (T)

A
Delay=T-Δb
B Δb

Delay=T-Δc
C Δc

x462_18_123105

Figure 3-18: Delaying a Fixed Frequency Clock Appears to Predict the Future

The clock period, T, is easy to derive knowing the frequency of the incoming monotonic
clock signal. But what are the clock skew delays Δb and Δc? With careful analysis, they can
be determined after examining the behavior of multiple systems under different
conditions. In reality, this is impractical. Furthermore, the values of Δb and Δc are different
between devices and vary with temperature and voltage on the same device.
Instead of attempting to determine the Δb and Δc delays in advance, the Spartan-3 FPGA
DCM employs a DLL that constantly monitors the delay via a feedback loop, as shown in
Figure 3-17. In this particular example, two DCMs are required—one to compensate for the
clock skew to internal signals and another to compensate for the skew to external devices,
each with their own clock feedback loop. The DLL constantly adapts to subtle changes
caused by temperature and voltage.

Locked on Target
In order to determine and insert the correct delay, the DCM samples up to several
thousand clock cycles. Once the DCM inserts the correct delay, the DCM asserts its
LOCKED output signal.
Do not use the DCM clock outputs until the DCM asserts its LOCKED signal. Until the
DCM locks onto the input clock signal, the output clocks are invalid. While the DCM
attempts to lock onto the clock signal, the output clocks can exhibit glitches, spikes, or
other spurious movements.
In an application, the LOCKED signal qualifies the output clock. Think of LOCKED as a
“clock signal good” indicator.

A Stable, Monotonic Clock Input

To operate properly, the DCM requires a stable, monotonic clock input. Consequently, the
DCM can predict future clock periods and adjust the output clock timing appropriately.
Once locked, the DCM tolerates clock period variations up to the value specified in the
specific FPGA data sheet. See the “DCM Clock Requirements” section.
Should the input clock vary well outside the specified limits, the DCM loses lock and the
LOCKED output switches Low. If the DCM loses lock, reset the DCM to reacquire lock. If
the input clock stays within the specified limits, then the output clocks always are valid
when the LOCKED output is High. However, it is possible for the clock to stray well
outside the limits, for the LOCKED output to stay High, and for either the CLKDV or
CLKFX outputs to be invalid. In short, a stable, monotonic clock input guarantees
problem-free designs.
The recommended input path to a DCM’s CLKIN input is via one of the four global buffer
inputs (IBUFG) along the same half of the device. Using the IBUFG path, the delay from

Spartan-3 Generation FPGA User Guide www.xilinx.com 105

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

the pad, through the global buffer, to the DCM is eliminated from the deskewed output.
Other paths are possible, however, as shown in Table 3-22. The signal driving the CLKIN
input can also originate a general-purpose input pin (IBUF primitive) via general-purpose
interconnect, from a global buffer input (IBUFG), or from a global buffer multiplexer
(BUFGMUX, BUFGCE). Similarly, an LVDS clock input can provide the clock signal. The
deskew logic is characterized for a single-ended clock input such as LVCMOS or LVTTL.
Differential signals might incur a slight amount of phase error due to I/O timing. See the
corresponding FPGA data sheet for specific I/O timing differences.

Table 3-22: CLKIN Input Sources

CLKIN Source Description
Via global buffer input A global buffer input, IBUFG, is the preferred source for an external clock to the DCM.
IBUFG The delay from the pad, through the global buffer, to the CLKIN input is characterized,
and this delay is removed from the deskewed clock output.
I O

Global Clock Buffer A global clock buffer, using either a BUFG, BUFGCE, or BUFGMUX primitive, is a
preferred source for an internally generated clock to the DCM. The delay through the
BUFG
I O global buffer is characterized, and this delay is removed from the deskewed clock
output.
BUFGCE When using BUFGCE or BUFGMUX, the input clock might change frequency or stop,
I O depending on the design. The DCM should be reset after enabling a BUFGCE or
changing inputs on a BUFGMUX. Also see “Momentarily Stopping CLKIN,” page 152.
CE

BUFGMUX
I0
O
I1

Via general-purpose I/O Any user-I/O pin, IBUF, becomes an alternate source for an external clock. The pad-to-
IBUF DCM delay cannot be predetermined due to the numerous potential input paths, and
consequently, the delay is not compensated by the DCM.
I O

Derived from internal logic Logic within the FPGA also can be the clock source. Again, the logic-to-DCM delay
cannot be predetermined and it is not compensated by the DCM.
Internal
Logic

Feedback from a Reliable Source

In order to lock in on the proper delay, the DCM monitors both the incoming clock and a
feedback clock, tapped after the clock distribution delay. There are no restrictions on the
total delay in the clock feedback path. If required, the DLL effectively delays the output
clock by multiple clock periods. Consequently, a DCM can compensate for either internal
or external delays, but the clock feedback must connect to the correct feedback point.

Removing Skew from an Internal Clock

To eliminate skew within the FPGA, the feedback tap is the same clock as that seen by the
clocked elements within the FPGA, shown in Figure 3-19. The feedback clock is typically
the CLK0 output (no phase shift) from the DCM, connected to the output of a global clock

106 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Eliminating Clock Skew

buffer (BUFG) or a global clock multiplexer (BUFGMUX or BUFGCE primitive) on the

same edge of the device. If a BUFGMUX or BUFGCE global clock multiplexer is used, the
DCM should be reset after the clock is switched or enabled. Alternatively, the DCM’s
CLK2X output (no phase shift, frequency doubled) can be used instead of the CLK0
output.

(or BUFGMUX,
or BUFGCE)
IBUFG BUFG
Clock to
I O I O internal
CLKIN CLK0
(or CLK2X) FPGA logic
(alternate clock inputs DCM or DCM_SP
possible, but not fully
skew adjusted) CLKFB LOCKED “Clock Good”

(Internal Feedback)
UG331_c3_23_022407

Figure 3-19: Eliminating Skew on Internal Clock Signals

Removing Skew from an External Clock

Constructing the DCM feedback for an external clock is slightly more complex. Ideally, the
clock feedback originates from the point where the signal feeds any external clocked
inputs, after any long printed-circuit board traces or external clock rebuffering, as shown
in Figure 3-20.

Circuit-board trace Other

FPGA delay, additional Device(s)
clock buffers, etc. on Board
IBUFG OBUF
I O I O
CLKIN CLK0 CLK
(or CLK2X)
IBUFG DCM or DCM_SP OBUF
I O I O
CLKFB LOCKED ENABLE

SRL16
D Q RESET Feedback path delay
WCLK must match the
forward path delay to
A[3:0] guarantee skew
Recommended elimination
INIT=000F

(External Feedback Trace) UG331_c3_24_120306

Figure 3-20: Eliminating Skew on External Clock Signals

The LOCKED signal indicates when the DCM achieves lock, qualifying the clock signal.
The LOCKED signal can enable external devices or an inverted version can connect to an
active-Low chip enable.

Reset DCM After Configuration

When using external feedback, apply a reset pulse to the DCM immediately after
configuration to ensure consistent locking. An SRL16 primitive, initialized with 0x000F,
supplies the necessary reset pulse, as shown in Figure 3-20. See “RST Input Behavior.”

Spartan-3 Generation FPGA User Guide www.xilinx.com 107

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Why Reset?
Why is this extra reset pulse required? For an optimum locking process, a DCM configured
with external feedback requires both the CLKIN and either the CLK0 or CLK2X signals to
be present and stable when the DCM begins to lock. During the configuration process, the
external feedback, CLKFB, is not available because the FPGA’s I/O buffers are not yet
active.
At the end of configuration, the DCM begins the capture process once the device enters the
startup sequence. Because the FPGA’s global 3-state signal (GTS) still is asserted at this
time, any output pins remain in a 3-state (high-impedance, floating) condition.
Consequently, the CLKFB signal is in an unknown logic state.
When CLKFB eventually appears after the GTS is deasserted, the DCM proceeds to
capture. However, without the reset pulse, the DCM might not lock at the optimal point,
which potentially introduces slightly more jitter and greater clock cycle latency through
the DCM.
Without the reset, another possible issue might occur if the CLKFB signal, while in the
3-state condition, cross-couples with another signal on the board due to a printed-circuit
board signal integrity problem. The DCM might sense this invalid cross-coupled signal as
CLKFB and use it to proceed with a lock. This possibly prevents the DCM from properly
locking once the GTS signal deasserts and the true CLKFB signal appears.

108 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Eliminating Clock Skew

What is a Delay-Locked Loop?

Two basic types of circuits remove clock delay:
• Delay-Locked Loops (DLLs) and
• Phase-Locked Loops (PLLs)
In addition to their primary function of removing clock distribution delay, DLLs and PLLs typically provide
additional functionality such as frequency synthesis, clock conditioning, and phase shifting.

Delay-Locked Loop (DLL)

As shown in Figure 3-21, a DLL in its simplest form consists of a tapped delay line and control logic. The delay line
produces a delayed version of the input clock CLKIN. The clock distribution network routes the clock to all
internal registers and to the clock feedback CLKFB pin. The control logic continuously samples the input clock as
well as the feedback clock to properly adjust the delay line. Delay lines are constructed either using a voltage
controlled delay or as a series of discrete delay elements. For best, ruggedly stable performance, the Spartan-
3 FPGA DLL uses an all-digital delay line.

CLKOUT Clock
CLKIN Variable Distribution
Delay Line Network

Control

CLKFB
x462_21_061903

Figure 3-21: Delay-Locked Loop (DLL) Block Diagram

A DLL works by inserting delay between the input clock and the feedback clock until the two rising edges align,
effectively delaying the feedback clock by almost an entire period—minus the clock distribution delay, of course. In
DLL and PLL parlance, the feedback clock is 360° out of phase, which means that they appear to be exactly in
phase again.
After the edges from the input clock line up with the edges from the feedback clock, the DLL “locks”, and the two
clocks have no discernible difference. Thus, the DLL output clock compensates for the delay in the clock
distribution network, effectively removing the delay between the source clock and its loads. Voila!

Spartan-3 Generation FPGA User Guide www.xilinx.com 109

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Phase-Locked Loop (PLL)

While designed for the same basic function, a PLL uses a different architecture to accomplish the task. As shown in
Figure 3-22, the fundamental difference between the PLL and DLL is that instead of a delay line, the PLL uses a
voltage-controlled oscillator, which generates a clock signal that approximates the input clock CLKIN. The control
logic, consisting of a phase detector and filter, adjusts the oscillator frequency and phase to compensate for the
clock distribution delay. The PLL control logic compares the input clock to the feedback clock CLKFB and adjusts
the oscillator clock until the rising edge of the input clock aligns with the feedback clock. The PLL then “locks.”

CLKOUT Clock
Voltage Controlled Distribution
Oscillator Network

CLKIN Control

CLKFB
x462_22_061903

Figure 3-22: Phase-Locked Loop (PLL) Block Diagram

Implementation
A DLL or PLL is assembled using either analog or digital circuitry; each approach has its own advantages. An
analog implementation with careful circuit design produces a DLL or PLL with a finer timing resolution.
Additionally, analog implementations sometimes consume less silicon area.
Conversely, digital implementations offer advantages in noise immunity, lower power consumption and better
jitter performance. Digital implementations also provide the ability to stop the clock, facilitating power
management. Analog implementations can require additional power supplies, require close control of the power
supply, and pose problems in migrating to new process technologies.

DLL vs. PLL

When choosing between a PLL or a DLL for a particular application, understand the differences in the
architectures. The oscillator used in the PLL inherently introduces some instability, which degrades the
performance of the PLL when attempting to compensate for the delay of the clock distribution network.
Conversely, the unconditionally stable DLL architecture excels at delay compensation and clock conditioning. On
the other hand, the PLL typically has more flexibility when synthesizing a new clock frequency.

Skew Adjustment
Most of this section discusses how to remove skew and how to phase align an internal or
external clock to the clock source. In actuality, the DCM purposely adds a small amount of
skew via an advanced attribute called DESKEW_ADJUST. In Clocking Wizard, the
DESKEW_ADJUST attribute is controlled via the Advanced Options window.
There are two primary applications for this attribute, SYSTEM_SYNCHRONOUS and
SOURCE_SYNCHRONOUS. The overwhelming majority of applications use the default
SYSTEM_SYNCHRONOUS setting. The purpose of each mode is described below.

System Synchronous
In a System Synchronous design, all devices within a data path share a common clock
source, as shown in Figure 3-23. This is the traditional and most-common system
configuration. The SYSTEM_SYNCHRONOUS option, which is the default value, adds a

110 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Eliminating Clock Skew

small amount of clock delay so that there is zero hold time when capturing data. Hold time
is essentially the timing difference between the best-case data path and the worst-case
clock path. The DCM’s clock skew elimination function advances the clock, essentially
dramatically shortening the worst-case clock path. However, if the clock path is advanced
so far that the clock appears before the data, then hold time results. The
SYSTEM_SYNCHRONOUS setting injects enough additional skew on the clock path to
guarantee zero hold times, but at the expense of a slightly longer clock-to-output time.

DATA_OUT DATA_IN

Clock
Source
x462_23_061903

Figure 3-23: System-Synchronous Applications are Clocked by a Single, System-

Wide Clock Source

The extra delay is injected in the SYSTEM_SYNCHRONOUS setting by adding an internal

delay on the feedback path. However, there are some situations where the DCM does not
add this extra delay, and therefore the DESKEW_ADJUST parameter has no affect. These
situations include DCMs that are cascaded, have external feedback, or have an external
CLKIN that does not come from a clock input.

Source Synchronous
SOURCE_SYNCHRONOUS mode is an advanced setting, used primarily in high-speed
data communications interfaces. In Source Synchronous applications, both the data and
the clock are derived from the same clock source, as shown in Figure 3-24. The transmitting
devices sends both data and clock to the receiving device. The receiving device then
adjusts the clock timing for best data reception. High-speed Dual-Data Rate (DDR) and
LVDS connections are examples of such systems.

DATA_OUT DATA_IN

Clock
Source DATA_CLK

x462_24_061903

Figure 3-24: In Source-Synchronous Applications, the Data Clock is Provided by

the Data Source

The SOURCE_SYNCHRONOUS setting essentially zeros out any phase difference

between the incoming clock and the deskewed output clock from the DCM. The FPGA
application must then adjust the clock timing using either the Fixed or Dynamic Fine Phase
Shift mode. The following application notes provide additional information on Source
Synchronous design and using dynamic phase alignment:
• XAPP268: Dynamic Phase Alignment
http://www.xilinx.com/support/documentation/application_notes/xapp268.pdf
• XAPP622: SDR LVDS Transmitter/Receiver
http://www.xilinx.com/support/documentation/application_notes/xapp622.pdf

Spartan-3 Generation FPGA User Guide www.xilinx.com 111

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Similarly, the following application note delves into more details on system-level timing.
Although the application note is written for the Virtex-II and Virtex-II Pro FPGA
architectures, most of the concepts apply directly to Spartan-3 FPGAs.
• XAPP259: System Interface Timing Parameters
http://www.xilinx.com/support/documentation/application_notes/xapp259.pdf

Timing Comparisons
Figure 3-25 compares the effect of both SYSTEM_SYNCHRONOUS and
SOURCE_SYNCHRONOUS settings using a Dual-Data Rate (DDR) application. In DDR
applications, two data bits appear on each data line—one during the first half-period of the
clock, the second during the second half-period.
In SYSTEM_SYNCHRONOUS mode, a small amount of skew is purposely added to the
DCM clock path so that there is zero hold time.
In SOURCE_SYNCHRONOUS mode, no additional skew is inserted to the DCM clock
path. However, the FPGA application must insert additional skew or phase shifting so that
the clock appears at the ideal location in the data window.

Data capture window

or data “eye”

DATA_IN

SYSTEM_SYNCHRONOUS

SOURCE_SYNCHRONOUS

SOURCE_SYNCHRONOUS
+ Fixed or Dynamic Phase Shift
x462_25_061903

Figure 3-25: Comparing SYSTEM_SYNCHRONOUS and

SOURCE_SYNCHRONOUS Timing in a Dual-Data Rate (DDR) Application

Clock Conditioning
Clock conditioning is a function where an incoming clock with a duty cycle other than 50%
is reshaped to have a 50% duty cycle. Figure 3-26 shows an example where an incoming
clock, with roughly a 45% High time and a 55% Low time (45%/55% duty cycle), is
reshaped into a nearly perfect 50% duty cycle—nearly perfect because there is some
residual duty-cycle distortion specified by the CLKOUT_DUTY_CYCLE_DLL and
CLKOUT_DUTY_CYCLE_FX values in the applicable FPGA family data sheet. The DCM
itself adds little to no distortion. Most of the distortion is caused by the difference in rise
and fall times in the internal routing and clock networks. The distortion is estimated at
100 ps to 400 ps, depending on the device.

112 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Conditioning

45% 55%

CLKIN

Conditioned
Clock Output

50% 50%
UG331_c3_04_120206

Figure 3-26: DCM Duty-Cycle Correction Feature Provides 50% Duty Cycle Outputs

Clocks with 50% duty cycle are mandatory for high-speed communications interfaces such
as LVDS or Dual-Data Rate (DDR) and for clock forwarding or clock mirroring
applications. See “Dual-Data Rate (DDR) Clocking Example.”

Spartan-3E and Extended Spartan-3A Family FPGA Output Clock

Conditioning
The DCM automatically conditions all clock outputs on Spartan-3E and
Extended Spartan-3A family FPGAs so that they have a 50% duty cycle.

Spartan-3 FPGA Output Clock Conditioning

On Spartan-3 FPGAs, most of the of the output clocks are conditioned to a 50% duty cycle,
although other outputs are optionally conditioned, depending either on the operating
conditions or on attribute settings, as shown in Table 3-23.

Table 3-23: Spartan-3 FPGA Family: Clock Outputs with Conditioned 50% Duty Cycle
DCM Clock
50% Duty Cycle Output
Output
CLK0
When DUTY_CYCLE_CORRECTION attribute set to TRUE
CLK180
CLK90
CLK270 DLL_FREQUENCY_MODE Attribute
LOW HIGH
When DUTY_CYCLE_CORRECTION attribute set to TRUE Outputs not available

CLK2X
CLK2X180 DLL_FREQUENCY_MODE Attribute
LOW HIGH
Always Outputs not available

Spartan-3 Generation FPGA User Guide www.xilinx.com 113

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-23: Spartan-3 FPGA Family: Clock Outputs with Conditioned 50% Duty Cycle (Cont’d)
DCM Clock
50% Duty Cycle Output
Output
CLKDV
DLL_FREQUENCY_MODE Attribute
LOW HIGH
Always When CLKDV_DIVIDE attribute is an integer value

CLKFX
Always
CLKFX180

The Quadrant Phase Shifted Outputs, CLK0, CLK90, CLK180, and CLK270 have optional
clock conditioning, controlled by the DUTY_CYCLE_CORRECTION attribute. By default,
the DUTY_CYCLE_CORRECTION attribute is set to TRUE, meaning that these outputs
are conditioned to a 50% duty cycle. Setting this attribute to FALSE disables the clock-
conditioning feature, in which case the effected clock outputs have roughly the same duty
cycle as the incoming clock. Exact replication of the CLKIN duty cycle is not guaranteed.

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

A DCM also optionally phase shifts an incoming clock, effectively delaying the clock by a
fraction of the clock period.
The DCM supports four different types of phase shifting. Each type can be used
independently, or in conjunction with other phase shifting modes. The phase shift
capabilities for each clock output appear in Table 3-24.
1. Half-Period Phase Shifted Outputs provide a pair of outputs, one with a rising edge at
0° phase shift and the other at 180° phase shift, at the half-period point during the clock
period.
2. Quadrant Phase Shifted Outputs of 0° (CLK0), 90° (CLK90), 180° (CLK180), and 270°
(CLK270).
3. Fixed Fine Phase Shifting of all DCM clock outputs with a resolution of 1/256th of a
clock cycle.
4. Variable Fine Phase Shifting of all DCM clock outputs from within the FPGA
application. For variable phase shifting, there are significant differences between the
Spartan-3 FPGA family and the Spartan-3E and Extended Spartan-3A families.
Spartan-3 FPGAs provide variable phase shift with a step size of 1/256th of a CLKIN
clock cycle. The size of the step varies depending on the CLKIN input frequency. On
Spartan-3E and Extended Spartan-3A family FPGAs, the step size, called
DCM_DELAY_STEP, is independent of the CLKIN clock frequency.
Table 3-24: Phase Shift Capabilities by Clock Output
Clock Output Half-Period Quadrant Fixed or Dynamic
CLK0 9 9 9
CLK90 9 9
CLK180 9 9 9
CLK270 9 9
CLK2X 9 9

114 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

Table 3-24: Phase Shift Capabilities by Clock Output (Cont’d)

Clock Output Half-Period Quadrant Fixed or Dynamic
CLK2X180 9 9
CLKDV 9
CLKFX 9 9
CLKFX180 9 9

Half-Period Phase Shifted Outputs

The Half-Period Phase Shift outputs provide a non-shifted clock output, and the
equivalent clock output but shifted by half a period (180° phase shift). The Half-Period
Phase Shift outputs appear in pairs, as shown in Table 3-25.
Table 3-25: Half-Period Phase Shifted Outputs
Output Pairs
Comment
No Phase Shift 180° Phase Shift
CLK0 CLK180 Same frequency as CLKIN input. Spartan-3E and Extended Spartan-3A family
FPGAs always have a 50% duty cycle. On Spartan-3 FPGAs, duty cycles for
outputs are corrected to 50% by default, controlled by the
DUTY_CYCLE_CORRECTION attribute.
CLK2X CLK2X180 Outputs from the Clock Doubler (CLK2X, CLK2X180). Twice the frequency of
the CLKIN input, always has a 50% duty cycle.
CLKFX CLKFX180 Outputs from the Frequency Synthesizer (CLKFX, CLKFX180). Output
frequency depends on Frequency Synthesizer attributes. Always has a 50%
duty cycle.

The Half-Period Phase Shift outputs are ideal for duty-cycle critical applications such as
high-speed Dual-Data Rate (DDR) designs and clock mirrors. The Half-Period Phase Shift
output pairs provide two clocks, one with a rising edge at the beginning of the clock
period, and another rising edge precisely aligned at half the clock period, as shown in
Figure 3-27.

Delay (fraction of
clock period) 0 ½T 1T
Phase Shift (degrees) 0˚ 180˚ 360˚

CLKx

CLKx180

Clock Period (T)

x462_27_061903

Figure 3-27: Half-Period Phase Shift Outputs

Spartan-3 Generation FPGA User Guide www.xilinx.com 115

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Half-Period Phase Shift Outputs Reduce Duty-Cycle Distortion

When the DCM clock outputs are duty-cycle corrected to 50%, it appears that the 180°
phase-shifted clock is just an inverted version on the non-shifted clock. For low-frequency
applications, this is essentially true.
However, at very high operating frequencies, duty-cycle distortion—due to differences in
rise and fall times of individual transistors—becomes relevant within the FPGA device.
Starting with a 50% clock cycle, such distortion causes differences between the clock High
and clock Low times, which is consistent from cycle to cycle.

Dual-Data Rate (DDR) Clocking Example

In Figure 3-28, a single DCM clock output, CLKx, drives both clocks on a Dual-Data Rate
(DDR) output flip-flop. One DDR clock input uses the clock output as is, the other input
inverts the clock within the DDR flip-flop. The CLKx output from the DCM has a 50% duty
cycle, but after traveling through the FPGA’s clock network, the duty cycle becomes
slightly distorted. In this exaggerated example, the distortion truncates the clock High time
and elongates the clock Low time. Consequently, the C1 clock input triggers slightly before
half the clock period. At lower frequencies, this distortion is usually negligible. However,
high-performance DDR-based systems require precise clocking due to the extremely short
half-period timing.

ODDR2

D0 Q
D1
DCM or DCM_SP
BUFG CE
CLKIN CLKx C0
C1

CLKx
(50% duty cycle)

Duty-cycle distortion

CLKx at Flip-Flop
(with duty-cycle distortion)

Factor in distortion
when using a single,
inverted clock
UG331_c3_25_120306

Figure 3-28: Dual-Data Rate (DDR) Output Using Both Edges of a Single Clock
Induces Duty-Cycle Distortion

Figure 3-29 shows a slightly modified circuit compared to Figure 3-28. In this case, the
DCM provides both a non-shifted and a 180° phase-shifted output to the DDR output flip-
flop. The CLKx clock signal precisely triggers the DDR flip-flop’s C0 input at the start of
the clock period. Similarly, the CLKx180 clock signal precisely triggers the DDR flip-flop’s
C1 input halfway through the clock period. The cost of this approach is an additional

116 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

global buffer and global clock line, but it potentially reduces the potential duty-cycle
distortion by approximately 300 ps.

ODDR2

D0 Q
D1
DCM or DCM_SP
BUFG CE
CLKIN CLKx C0
CLKx180 C1
BUFG

CLKx
(50% duty cycle)

CLKx at Flip-Flop
(with duty-cycle distortion)
180 °
Phase Shift
CLKx180 at Flip-Flop
(with duty-cycle distortion)

UG331_c3_24_120

Figure 3-29: Using Half-Period Phase Shift Outputs Reduces Potential Duty-Cycle
Distortion

Table 3-26 shows the specified duty-cycle distortion values as measured using DDR output
flip-flops and LVDS outputs. There might be additional distortion on other output types
caused by asymmetrical rise and fall times, which can be simulated using IBIS.
When using the DCM to generate high speed clocks to drive the double data rate ODDR2,
BUFGMUX_X1Y1 is recommended for CLKFX and BUFGMUX_X2Y0 is recommended for
CLKFX180 to minimize period jitter.

Table 3-26: Duty-Cycle Distortion Parameters

Estimated
Parameter Description
Value
TDCD_CLK0 Duty-cycle distortion when local inversion provides negative-edge clock to DDR ~400 ps
element in an I/O block. See Figure 3-28.
TDCD_CLK180 Duty-cycle distortion when DCM CLKx180 output provides clock to DDR element in ~60 ps
an I/O block. See Figure 3-29.

Quadrant Phase Shifted Outputs

The Quadrant Phase Shift outputs shift the CLKIN input, each by a quarter period, as
shown in Figure 3-30 and Table 3-28. Because the Quadrant Phase Shift outputs require a
feedback path back to the CLKFB input, the CLK0 output is phase aligned to the rising
edge of the CLKIN input. The CLK90 output is phase shifted 90° from the CLKIN input,
and so forth.

Spartan-3 Generation FPGA User Guide www.xilinx.com 117

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Delay (fraction of
clock period) 0 ¼T ½T ¾T 1T
Phase Shift (degrees) 0˚ 90˚ 180˚ 270˚ 360˚

CLK0

CLK90

CLK180

CLK270

Clock Period (T)

x462_30_061903

Figure 3-30: Quadrant Phase Shift Outputs Shift CLKIN, Each by a Quarter Period
(Shown with Duty-Cycle Correction Enabled)

Output Availability Depends on DLL Frequency Mode

The availability of the Quadrant Phase Shift outputs depends on the DLL’s frequency
mode. On Spartan-3 FPGAs, the range is controlled by the DLL_FREQUENCY_MODE
attribute. On Spartan-3E and Extended Spartan-3A family FPGAs, the outputs depend on
whether the CLKIN input frequency is above or below a certain frequency, typically
200 MHz.
All four Quadrant Phase Shift outputs are available in low-frequency mode
(DLL_FREQUENCY_MODE = LOW), as shown in Table 3-27. Only the CLK0 and CLK180
outputs are available in both modes.

Table 3-27: Quadrant Phase Shift Output Availability by DLL Frequency Mode
Spartan-3 FPGAs
DLL_FREQUENCY_MODE = LOW DLL_FREQUENCY_MODE = HIGH
Output
Spartan-3E and Extended Spartan-3A family FPGAs
CLKIN ≤167 MHz CLKIN > 167 MHz

CLK0 9 9
CLK90 9
CLK180 9 9
CLK270 9

Spartan-3 FPGA: Optional 50/50 Duty Cycle Correction

On Spartan-3E and Extended Spartan-3A family FPGAs, the quadrant outputs are always
conditioned to a 50% duty cycle. On Spartan-3 FPGAs, the outputs are optionally

118 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

conditioned to a 50% duty cycle, controlled by the DUTY_CYCLE_CORRECTION

attribute. When TRUE, which is the default, all four outputs have a 50% duty cycle. When
FALSE, the outputs do not necessarily have the same duty cycle as the CLKIN input. See
the “Clock Conditioning” section for more information.

Four Phases, Delayed Clock Edges, Phased Pulses

One view of the Quadrant Phase Shift outputs is that each provides a rising clock that is
delayed one quarter period from the preceding pulse, as shown in Table 3-28. These
outputs provide flexible timing for such applications as memory interfaces and peripheral
control.
When these outputs are conditioned with a 50% duty cycle, there are other ways to view
these signals. For example, the outputs also provide falling-edge clocks separated by a
quarter phase. Again, see Table 3-28. Similarly, each output produces a High-going pulse,
and a Low-going pulse, both half a period wide. For example, the CLK90 output shown in
Figure 3-30 produces a High-going pulse, centered within the CLK0 clock period.
Table 3-28: Quadrant Phase Shift Outputs and Characteristics (DUTY_CYCLE_CORRECTION=TRUE)
Delayed by
Phase Rising Falling
DCM Output Period Comment
Shift Edge Edge
Fraction
CLK0 0° 0 0 ½T Deskewed input clock, no phase shift
CLK90 90° ¼T ¼T ¾T High-going pulse, ½T wide, in middle of period
CLK180 180° ½T ½T 0T Inverted CLK0, rising clock edge in middle of
period
CLK270 270° ¾T ¾T ¼T Low-going pulse, ½T wide, in middle of period

Fine Phase Shifting

The DCM provides additional controls over clock skew using fine phase shifting. Fine-
phase adjustment affects all nine DCM output clocks simultaneously. The fine phase shift
capability requires the DCM’s DLL functional unit. Consequently, clock feedback via the
CLKFB input is required. Phase Shifter operation in the Spartan-3 family is only supported
when DLL_FREQUENCY_MODE = LOW.
Physically, the fine phase shift control adjusts the phase relationship between the rising
edges of the CLKIN and CLKFB inputs. The net effect, however, is that all DCM outputs
are phase shifted with relation to the CLKIN input.
By default, fine phase shifting is disabled (CLKOUT_PHASE_SHIFT = NONE), meaning
that the clock outputs are phase aligned with the CLKIN input clock. In this case, there is
no skew between the input clock, CLKIN, and the feedback clock, measured at the
appropriate feedback point (see “Feedback from a Reliable Source” section). When fine
phase shifting is enabled, the output clock edges can be phase shifted so that they are
advanced or are delayed compared to the CLKIN input, as shown in Figure 3-32.
There are two fine phase shift modes as described below. Both are commonly used in high-
speed data communications applications. See the “Source Synchronous” section.
1. Fixed Fine Phase Shift mode sets the phase shift value at design time. The phase shift
value is loaded into the FPGA during configuration and cannot be changed by the
application. The Fixed phase shift feature is identical among Spartan-3 generation
FPGAs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 119

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

2. Variable Fine Phase Shift mode has an initial phase shift value, similar to Fixed Fine
Phase Shift, which is set during FPGA configuration. However, the phase shift value
can be changed by the application after the DCM’s LOCKED output goes High.
Note: There are important differences between the Variable phase shift feature on Spartan-3
FPGAs and that found on Spartan-3E and Extended Spartan-3A family FPGAs. See “Important
Differences Between Spartan-3 Generation FPGA Families,” page 123.

Fixed Fine Phase Shifting

In Fixed Fine Phase Shift mode, the phase shift value is specified at design time and set
during the FPGA configuration process. The application cannot change the value during
run time.
Note: Fixed Fine Phase Shift in the Spartan-3 family should be implemented using the latest
available software update.
Two attributes control this mode. The CLKOUT_PHASE_SHIFT attribute is set to FIXED,
and the PHASE_SHIFT attribute controls the amount of phase shift. If PHASE_SHIFT is 0,
then the output clocks and the CLKIN input are phase aligned, as shown in Figure 3-32. If
PHASE_SHIFT is a negative integer, then the clock output(s) are phase shifted before
CLKIN. If PHASE_SHIFT is a positive integer, then the clock output(s) are phase shifted
after CLKIN.
The size of each phase shift unit is always the same at 1/256th of the CLKIN clock period, as
shown in Figure 3-31, which equates to 1.40625° per step. The physical delay of each step
depends on the CLKIN input clock frequency, as shown in Equation 3-5.

0°

270° 90°
1
of CLKIN clock period
256
(=1.40625°)
180°
0° 90° 180° 270° 0°
0 64 128 192 255
PHASE_SHIFT =

CLKIN Period (TCLKIN)

Phase = Phase Shift 360o = Phase Shift 1.40625o

256

Time = Phase Shift TCLKIN

256 UG331_c3_19_022407

Figure 3-31: Each PHASE_SHIFT Unit is 1/256th of the CLKIN Period

Spartan-3 Family Fixed Fine Phase Shift Range

The PHASE_SHIFT attribute is always an integer value, ranging between –255 and +255.
However, the actual limits for the Spartan-3 FPGA family can be lower depending on the
CLKIN input frequency, as described below.

120 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

CLKIN

Clock Outputs

Fixed Phase Shift Fixed Phase Shift

- Limit + Limit
0
The PHASE_SHIFT attribute, set at design
time, controls the amount of phase shift on the
DCM clock outputs relative to the CLKIN input. x462_31_061903

Figure 3-32: Fixed-Value Fine Phase Shift Control

The minimum and maximum limits of the PHASE_SHIFT attribute depend on two values.
1. The period of the CLKIN input, TCLKIN, measured in nanoseconds.
2. For Spartan-3 family FPGAs, FINE_SHIFT_RANGE defines the maximum guaranteed
delay achievable by the phase shift delay line. The actual delay line within a given
device can be longer, but only the delay up to FINE_SHIFT_RANGE is guaranteed.
The Extended Spartan-3A family does not have a FINE_SHIFT_RANGE limit for fixed
phase shifting.
Using these two values, calculate the SHIFT_DELAY_RATIO using Equation 3-1. The
limits for the PHASE_SHIFT attribute are different, depending on whether the result is less
than or if it is greater than or equal to one.
FINE_SHIFT_RANGE-
SHIFT_DELAY_RATIO = -------------------------------------------------------------- Equation 3-1
T CLKIN

SHIFT_DELAY_RATIO < 1
If the Spartan-3 FPGA clock period is longer than the specified FINE_SHIFT_RANGE, then
the SHIFT_DELAY_RATIO < 1, meaning that maximum fine phase shift is limited by
FINE_SHIFT_RANGE. When SHIFT_DELAY_RATIO < 1, then the PHASE_SHIFT limits
are set according to Equation 3-2:
FINE_SHIFT_RANGE
PHASE_SHIFT LIMITS = ± INTEGER ⎛ 256 • ---------------------------------------------------------------⎞ Equation 3-2
⎝ T CLKIN ⎠

For example, assume that FCLKIN is 75 MHz (TCLKIN = 13.33 ns) and FINE_SHIFT_RANGE
is 10.00 ns. In this case, the PHASE_SHIFT value is limited to ±191.
Consequently, the phase shift value when SHIFT_DELAY_RATIO < 1 is shown by
Equation 3-3. To determine the phase shift resolution, set PHASE_SHIFT = 1.
PHASE_SHIFT
T PhaseShift = ⎛ ---------------------------------------------------------------⎞ • FINE_SHIFT_RANGE Equation 3-3
⎝ PHASE_SHIFT LIMITS ⎠

SHIFT_DELAY_RATIO ≥ 1
By contrast, if the Spartan-3 FPGA clock period is shorter than the specified
FINE_SHIFT_RANGE, then the SHIFT_DELAY_RATIO ≥ 1, meaning that maximum fine
phase shift is limited to ±255.

Spartan-3 Generation FPGA User Guide www.xilinx.com 121

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

PHASE_SHIFTLIMITS = ±255 Equation 3-4

Consequently, the phase shift value when SHIFT_DELAY_RATIO ≥ 1 is shown by

Equation 3-5. To determine the phase shift resolution, set PHASE_SHIFT = 1.
PHASE_SHIFT
T PhaseShift = ⎛ -------------------------------------------⎞ • T CLKIN Equation 3-5
⎝ 256 ⎠

Minimum Phase Shift Size

The minimum phase shift size is controlled by the greater of two limiting factors.
1. 1/ th of the CLKIN clock period. However, the phase shift delay is physically
256
implemented using delay elements.
2. The smallest phase shift amount must be at least as large as the minimum delay
element resolution, listed by FPGA family in Table 3-29.
Table 3-29: Delay Element Step Size
Delay Element
FPGA Family Delay Element Value
Specification Symbol
Spartan-3 FPGA DCM_TAP 30 to 60 ps
Spartan-3E FPGA DCM_DELAY_STEP 20 to 40 ps, 25 ps typical
Extended Spartan-3A family FPGA DCM_DELAY_STEP 15 to 35 ps, 23 ps typical

Other Design Considerations

In Fixed Phase Shift mode, the Variable Phase Shift control inputs must be tied to GND,
which Clocking Wizard and the ISE software do automatically.

Clocking Wizard
To use Fixed Phase Shift mode, select Fixed in the Phase Shift section of Clocking Wizard’s
General Setup panel, shown in Figure 3-33. This action sets the CLKOUT_PHASE_SHIFT
attribute to FIXED.
Enter the phase shift Value, which must be an integer within the limits described above.
This action sets the PHASE_SHIFT attribute value. Clocking Wizard checks that the phase
shift value is within the limits.

Select FIXED

Enter Fixed
phase shift
Value Allowable
phase shift
range at
frequency

Equivalent phase
shift for specified
Value
UG331_c3_13_120206

Figure 3-33: Selecting Fixed Fine Shift Mode

122 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

Variable Fine Phase Shifting

In Variable Fine Phase Shift mode, the initial skew or phase shift is still controlled by the
PHASE_SHIFT attribute during configuration, just as it is for Fixed Fine Shift mode.
However, in dynamic mode, the FPGA application can adjust the current phase shift
location after the DCM’s LOCKED output goes High using the Dynamic Fine Phase Shift
control inputs, PSEN, PSCLK, and PSINCDEC.
The total resulting phase shift is the sum of the initial Fixed phase shift plus any Variable
phase shift adjustments, as shown in Equation 3-6, assuming the same units.

Phase Shift TOTAL = FIXED_PHASE_SHIFT + VARIABLE_PHASE_SHIFT Equation 3-6

Important Differences Between Spartan-3 Generation FPGA Families

For Variable Phase Shift mode, there are important differences between
Spartan-3 generation FPGA families. Spartan-3E and Extended Spartan-3A family FPGAs
both use a silicon-efficient delay-based variable phase shifting method. Spartan-3 FPGAs
use a more elaborate method based on the fraction of the clock period. Although both
methods perform phase shifting, they are completely different.
Table 3-30 summarizes the differences. All Spartan-3 generation FPGAs perform Fixed
Phase Shift identically, as illustrated in Figure 3-31, page 120. The resulting phase shift is
always 1/256th of the CLKIN period. Similarly, all Spartan-3 generation FPGAs use the
identical Variable phase shift control mechanism using the PSEN, PSINCDEC, PSCLK, and
PSDONE connections to the DCM.
The major difference is the result of each Variable Phase Shift operation. For Spartan-3
FPGAs, a Variable phase shift operation is similar to a Fixed phase shift operation. The
operation always results in a phase change measured in degrees, as shown in Figure 3-31,
page 120. The phase shift measured in degrees never changes; the phase shift measured in
time depends on the CLKIN input frequency.

Table 3-30: FIXED and VARIABLE Phase Shift Implementations by Spartan-3 Generation FPGA Family
FPGA Family
Extended Spartan-3A
Spartan-3 FPGA Spartan-3E FPGA
Family FPGA
FIXED Phase Shift unit increment or 1/ th
256 of CLKIN Period
decrement unit
FIXED Phase Shift measurement
Degrees
unit
VARIABLE Phase Shift control PSEN, PSINCDEC, PSCLK, and PSDONE signals
mechanism on the DCM
VARIABLE Phase Shift increment or 1/ thof CLKIN
256
decrement unit Period DCM_DELAY_STEP, DCM_DELAY_STEP,
between 20 to 40 ps between 15 to 35 ps
(1.4065°)
Figure showing VARIABLE Phase Figure 3-31,
Figure 3-34, page 124
Shift logic page 120
VARIABLE Phase Shift equation Equation 3-5 Equation 3-7, Equation 3-8

Spartan-3 Generation FPGA User Guide www.xilinx.com 123

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-30: FIXED and VARIABLE Phase Shift Implementations by Spartan-3 Generation FPGA Family
FPGA Family
Extended Spartan-3A
Spartan-3 FPGA Spartan-3E FPGA
Family FPGA
VARIABLE Phase Shift
Degrees Time
measurement unit
Does Phase Shift, measured in
degrees, change with CLKIN input No Yes
frequency?
Does Phase Shift, measured in time,
change with CLKIN input Yes No
frequency?

On Spartan-3E and Extended Spartan-3A family FPGAs, however, a Variable Phase Shift
operation results in a delay change, not a phase change. The phase shift is implemented by
cascaded delay elements, as shown in Figure 3-34. Each DCM_DELAY_STEP element
ranges from the minimum and maximum values shown in Table 3-29, page 122.
Consequently, the actual amount of phase shift time added to the clock outputs ranges
between the cumulative minimum and maximum delay through all the selected elements.
This time is relatively constant and does not change with the CLKIN frequency. The
corresponding phase shift, measured in degrees, does change with frequency.

MAX_STEPS
Time
DCM_DELAY_STEP

Phase Shift
Time > Phase Shift DCM_DELAY_STEP_MIN
< Phase Shift DCM_DELAY_STEP_MAX

Time
Phase = 360°
T CLKIN UG331_c3_20_120306

Figure 3-34: Spartan-3E and Extended Spartan-3A Family FPGA Variable Phase
Shift Logic

Spartan-3E and Extended Spartan-3A Family FPGA Variable Phase Shift

Operations
The results of a Variable phase shift operation on a Spartan-3E or Extended Spartan-3A
family FPGA is always measured in time, as shown in Equation 3-7 and Equation 3-8. The
resulting phase shift has minimum and maximum values due to the variation of the delay
in each DCM_DELAY_STEP, as shown in Table 3-29, page 122.

T MAX ( VariableShift ) = Variable • DCM_DELAY_STEP_MAX Equation 3-7

124 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

T MIN ( VariableShift ) = Variable • DCM_DELAY_STEP_MIN Equation 3-8

Based on the results from Equation 3-7 and Equation 3-8, the resulting phase shift,
measured in degrees, is determined from Equation 3-9. TCLKIN is the period of the CLKIN
input.
T VariableShift o
PHASE SHIFT = --------------------------------- • 360 Equation 3-9
T CLKIN

Operation
Use the phase shift control inputs to adjust the current phase shift value, as shown in
Figure 3-35. The rising edge of PSCLK synchronizes all Variable Phase Shift operations. A
valid operation starts by asserting the PSEN enable input for one and only one PSCLK
clock period. Asserting PSEN for more than one rising PSCLK clock edge might cause
undesired behavior.

Start new phase shift The timing to complete a phase

operation. Shift by shift operation varies. PSDONE
one phase increment. indicates operation is complete.

PSCLK

PSEN

PSINCDEC
0 = Decrement phase shift
1 = Increment phase shift
PSDONE
Operation complete.
Okay to start new
operation.
STATUS[0]
(Variable Phase
Shift Overflow)
If phase shift incremented or
decremented to limit value,
STATUS[0] stays High until new
operation shifts away from limit.
UG331_c3_29_100509

Figure 3-35: Dynamic Fine Phase Shift Control Interface

The value on the PSINCDEC increment/decrement control input determines the phase
shift direction. When PSINCDEC is High, the present Variable Phase Shift value is
incremented by one unit. Similarly, when PSINCDEC is Low, the present Variable Phase
Shift value is decremented by one unit.
The actual phase shift operation timing varies and the operation completes when the DCM
asserts the PSDONE output High for a single PSCLK clock period. Between enabling PSEN
until PSDONE is asserted, the DCM output clocks slide, bit by bit, from their original
phase shift value to their new phase shift value. During this time, the DCM remains locked
on the incoming clock and continues to assert its LOCKED output.

Spartan-3 Generation FPGA User Guide www.xilinx.com 125

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

The phase adjustment might require as many as 100 CLKIN cycles plus 3 PSCLK cycles to
take effect, at which point the DCM’s PSDONE output goes High for one PSCLK cycle.
This pulse indicates that the PS unit completed the previous adjustment and is now ready
for the next request.
To enable Dynamic Fine Phase Shift mode, set the CLKOUT_PHASE_SHIFT attribute to
VARIABLE. The PHASE_SHIFT attribute value sets the initial phase shift location,
established after FPGA configuration. The FPGA application can the dynamically adjust
the skew or phase shift on the DCM’s output clocks after the DCM's LOCKED output goes
High. If the DCM is reset, the PHASE_SHIFT value reverts to its initial configuration value.

CLKIN The PHASE_SHIFT attribute determines

the initial phase shift position. DCM
initially asserts LOCKED with this phase
shift value. DCM returns to this value
upon RESET.
Clock Outputs

Fixed Phase Shift Fixed Phase Shift

- Limit 0 + Limit

Dynamic Phase Shift Dynamic Phase Shift

- Limit 0 + Limit
Decrem ent Phase DCM Variable Phase Increm ent Phase
Shift Value Shift Control Shift Value

Enable PSEN
Increment/Decrement PSINCDEC PSDONE Phase Shift Done
Phase Shift Clock PSCLK STATUS[0] Variable Phase
Shift Overflow

After the DCM asserts LOCKED, the FPGA

application can increment or decrement the
present phase shift value using the Dynamic
Phase Shift Control logic. x462_34_061903

Figure 3-36: Variable Phase Shift Controls

Variable Fine Phase Shift Range

Just as the PHASE_SHIFT attribute has minimum and maximum phase shift limits, so does
the Variable Phase Shift, as shown in Figure 3-36. Due to the differences between Spartan-3
FPGAs and Spartan-3E and Extended Spartan-3A family FPGAs, the limits are also
different.

Spartan-3 FPGA Family Variable Phase Shift Range

For Spartan-3 FPGAs, the limits again depend on the ratio of the FINE_SHIFT_RANGE
versus the input clock period, as calculated by the SHIFT_DELAY_RATIO equation above.
However, since the Spartan-3 FPGA FINE_SHIFT_RANGE is 10 ns, and Phase Shift is only
supported in the Low Frequency Mode (up to 167 MHz), the SHIFT_DELAY_RATIO will
always be <2.

126 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Phase Shifting – Delaying Clock Outputs by a Fraction of a Period

The maximum dynamic fine phase shift value is limited by FINE_SHIFT_RANGE, the
maximum delay tap value. The Variable Phase Shift limits are set according to
Equation 3-10.
FINE_SHIFT_RANGE
DynamicPhaseShift LIMITS = ± INTEGER ⎛ 128 • ---------------------------------------------------------------⎞ Equation 3-10
⎝ T CLKIN ⎠

For example, assume that FCLKIN is 75 MHz (TCLKIN = 13.33 ns) and FINE_SHIFT_RANGE
is 10.00 ns. In this case, the Variable Phase Shift value is limited to ±96.
The Variable Phase Shift value is shown by Equation 3-11. To determine the Variable Phase
Shift resolution, set Variable Phase Shift = 1.
DynamicPhaseShift
T PhaseShift = ⎛ --------------------------------------------------------------------------⎞ • FINE_SHIFT_RANGE Equation 3-11
⎝ DynamicPhaseShift LIMITS ⎠

Spartan-3E and Extended Spartan-3A Family Variable Phase-Shift Range

For Spartan-3E and Extended Spartan-3A family FPGAs, variable phase shifting is
performed using delay elements. There is a physical maximum for the number of delay
steps, depending on the CLKIN input period, TCLKIN, as shown in Table 3-31.

Table 3-31: Maximum Number of DCM Delay Steps

CLKIN CLKIN Period
Maximum Number of DCM Delay Steps Unit
Frequency TCLKIN
< 60 MHz > 16.67 ns ±[INTEGER(10 • (TCLKIN – 3 ns))]
Steps
≥ 60 MHz < 16.67 ns ±[INTEGER(15 • (TCLKIN – 3 ns))]

For example, assume that the CLKIN clock entering the DCM is 100 MHz, which equates
to a clock period of TCLKIN = 10 ns. Using the equation in Table 3-31, the Variable Phase
Shifter is limited to phase shift operations of ±105 steps. On a Spartan-3E FPGA, this
equates to a maximum variable phase shift measured in time of up to ±2.1 ns to ±4.2 ns.
Measured in degrees, this equates to a maximum between ±75.6° and 151.2°.

Controls
As shown in Figure 3-35, page 125 and Figure 3-36, page 126, the DCM’s Variable Phase
Shift control signals allow the FPGA application to adjust the present phase relationship
between the CLKIN input and the DCM clock outputs. Table 3-32 shows the detailed
relationship between control inputs, the current and next phase relationship, how the
operation affects the delay tap, and the control outputs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 127

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-32: Variable Phase Shifter Control (assumes no internal inversion)

STATUS[0]
(Overflow - Not
Current Next
PSEN PSINC-DEC PSCLK Delay Line PSDONE Available in Operation
Phase Shift Phase Shift
Spartan-3E
FPGAs)
0 X X X No change No change ? ? Variable Phase Shift disabled.
1 0 · > -Limit Current – 1 Current – 1 1* 0 Decrement phase shift and phase
pointer.
1 0 · ≤ -Limit and Current – 1 No Change 1* 1 End of delay line. No phase shift
> –255 change. Phase pointer decremented.
1 0 · –255 –255 No Change 1* 1 End of delay line. No phase shift
change. Phase pointer at limit.
1 1 · < +Limit Current + 1 Current + 1 1* 0 Increment phase shift and phase
pointer.
1 1 · ≥ +Limit and Current + 1 No Change 1* 1 End of delay line. No phase shift
< +255 change. Phase pointer incremented.
1 1 · +255 +255 No Change 1* 1 End of delay line. No phase shift
change. Phase pointer at limit.

Notes:
X = don’t care.
? = indeterminate, depends on current application state.
1* = PSDONE asserted High for one PSCLK period.
-Limit = minimum delay line position.
+Limit = maximum delay line position.
Assert PSEN for only one PSCLK cycle.

When PSEN is Low, the Variable Phase Shifter is disabled and all other inputs are ignored.
All present shift values and the delay line position remain unchanged.
If the delay line has not reached its limits (-Limit or –255 when decrementing, +Limit or
+255 when incrementing), then the FPGA application can change the existing phase shift
value by asserting PSEN High and the appropriate increment/decrement value on
PSINCDEC before the next rising edge of PSCLK. The phase shift value increments or
decrements as instructed. At the end of the operation, PSDONE goes High for a single
PSCLK period indicating that the phase shift operation is complete. STATUS[0] remains
Low because no phase shift overflow condition occurred.
When the DCM is incremented beyond +255 or below –255, the delay line position remains
unchanged at its limit value of +255 or –255 and no phase change occurs. STATUS[0] goes
High, indicating a Variable Phase Shift overflow (not available in Spartan-3E FPGAs).
When a new phase shift operation changes the value in the opposition direction—i.e.,
away from the limit value—STATUS[0] returns Low.
If the phase shift does not reach +255 or –255, but the phase shift exceeds the delay-line
range—indicated by +Limit and –Limit in Table 3-32—then no phase change occurs.
However, STATUS[0] again goes High. In the Spartan-3 and Extended Spartan-3A families
only, the STATUS[0] output indicates when the delay tap reaches the end of the delay line.
In the FPGA application, however, use the limit value calculated using Equation 3-10. The
calculated delay limit is a guaranteed value. A specific device, due to processing, voltage,
or temperature, might have a longer line delay, but this cannot be guaranteed from device
to device. The phase shift value—but not the delay line positions—continues to increment
or decrement until it reaches its +255 or –255 limit. When a new phase shift operation
changes the value in the opposition direction—i.e., away from the limit value—the

128 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Multiplication, Clock Division, and Frequency Synthesis

STATUS[0] signal returns Low. The phase shift value is incremented or decremented back
to a value that corresponds to a valid absolute delay in the delay line.

Clocking Wizard
The Variable Phase Shift options are part of the Clocking Wizard’s General Setup panel,
shown in Figure 3-37. To enable dynamic fine phase shifting, select VARIABLE, as shown
in Figure 3-37. Enter an initial Phase Shift Value in the text box provided. The initial value
behaves exactly like the Fixed Fine Phase Shifting mode described above.

Check STATUS output to

access STATUS[0], the
Dynamic Phase Shift
Overflow status bit

Select
VARIABLE

Selecting VARIABLE
enables the phase shift
controls PSEN,
PSINCDEC, PSCLK, and
PSDONE

UG331_c3_14_120206

Figure 3-37: Selecting Variable Fine Phase Shift Mode in Clocking Wizard

Choosing Variable mode also enables the Variable Phase Shift control signals, PSEN,
PSINCDEC, PSCLK, and PSDONE. For the Spartan-3 family, check the STATUS output box
to enable the STATUS[0] signal. STATUS[0] indicates when the Variable Phase Shifter
reaches its maximum or minimum limit value (not available in Spartan-3E family).

Example Applications
See application note XAPP268 for an example of how to use the Variable Phase Shift
function to perform dynamic phase alignment.
• XAPP268: Dynamic Phase Alignment
http://www.xilinx.com/support/documentation/application_notes/xapp268.pdf

Clock Multiplication, Clock Division, and Frequency Synthesis

A DCM provides flexible methods for generating new clock frequencies—one of the most
common DCM applications. Spartan-3 generation DCMs provide up to three independent
frequency synthesis functions, listed below, and in Figure 3-38, and summarized in
Table 3-33. An application can use one or all three functions simultaneously. Detailed
descriptions for each function follows.
1. A Clock Doubler (CLK2X, CLK2X180) that doubles the frequency of the input clock.
2. A Clock Divider (CLKDV) that reduces the input frequency by a fixed divider value.
3. A Frequency Synthesizer (CLKFX, CLKFX180) for generating a completely new
frequency from an incoming clock frequency.

Spartan-3 Generation FPGA User Guide www.xilinx.com 129

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Output clocks are phase-aligned

when using clock feedback via
the CLKFB input.

Deskewed Clock
CLKIN CLK0
F = FCLKIN
DCM or
Clock DCM_SP
CLK0
or Distribution CLKFB
CLK2X Delay CLK2X
Clock Doubler
F = 2 FCLKIN
Clock Feedback Loop CLK2X180
A clock feedback loop to CLKFB
is required when using the
CLK0, CLK2X, CLK2X180, or Clock Divider
CLKDV outputs. Use only CLK0 FCLKIN
CLKDV F=
or CLK2X as the feedback CLKDV_DIVIDE
source. Feedback is not Usually 50% duty cycle,
required when using only the depending on conditions
CLKFX or CLKFX180 outputs.
Frequency Synthesizer
CLKFX
CLKFX_MULTIPLY
F = FCLKIN
CLKFX_DIVIDE
CLKFX180
50% duty cycle
UG331_c3_022407

Figure 3-38: Clock Synthesis Options

All the frequency synthesis outputs, except CLKDV, always have a 50/50 duty cycle.
CLKDV usually has a 50% duty cycle except when dividing by a non-integer value at high
frequency, as shown in Table 3-37. The Clock Doubler (CLK2X, CLK2X180) circuit is not
available at high frequencies.
All the DCM clock outputs, except CLKFX and CLKFX180, are generated by the DCM’s
Delay-Locked Loop (DLL) unit and consequently require some form of clock feedback to
the CLKFB pin. The DCM’s Digital Frequency Synthesizer (DFS) unit generates the CLKFX
and CLKFX180 clock outputs. If the application uses only the CLKFX or CLKFX180
outputs, then the feedback path can be eliminated, which also extends the DCM’s
operating range. The Frequency Synthesizer has a feedback path within the DCM, based
on CLKIN.

Table 3-33: DCM Frequency Synthesis Options

DCM
DCM Feedback 50%
Function Frequency Functional
Output(s) Required? Duty Cycle?
Unit

Deskewed CLK0 F CLKIN DLL Yes When DUTY_CYCLE_

Clock CORRECTION = TRUE

Clock CLK2X 2 • F CLKIN DLL Yes Always

Doubler CLK2X180

Clock CLKDV F CLKIN DLL Yes Always except when

Divider -----------------------------------------------
- dividing by non-integer value
CLKDV_DIVIDE
in high-frequency mode
Frequency CLKFX DFS Optional. No Always
CLKFX_MULTIPLY
Synthesizer CLKFX180 F CLKIN • ⎛ ---------------------------------------------------------⎞ feedback extends
⎝ CLKFX_DIVIDE ⎠ clock input
frequency limits.

130 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Multiplication, Clock Division, and Frequency Synthesis

Output Alignment
If clock feedback is used, then all the output clocks are phase aligned. Obviously, full clock-
edge alignment across all the DCM outputs occurs only occasionally because some of the
outputs are divided clock values. For example, the CLKDV output is aligned to CLKIN
and CLK0 every CLKDV_DIVIDE cycles. Similarly, the CLK2X output is aligned to CLK0
every other clock cycle. The CLKFX output is aligned to CLKIN every CLKFX_DIVIDE
cycles of CLKIN and every CLKFX_MULTIPLY cycles of CLKFX.
Individual outputs are aligned to CLKIN, but when using divided clocks the DCM
arbitrarily picks a rising edge to align to; therefore, the rising edge of the CLKFX output
might not be aligned to the other outputs. For example, a divide-by-two function on
CLKDV and a divide-by-four function on CLKFX could be aligned on a falling edge
instead of a rising edge. To align the rising edges in this case, use CLKIN_DIVIDE_BY_2 on
the input, and use the CLK0 output for the divide-by-two and the CLKDV output (with D
= 2) for the divide-by-four. If this is not possible, the CLKDV output of one DCM can be
cascaded to a second DCM and CLKDV, with D = 2 for both. Also note that the first rising
edge of CLKFX after LOCKED is High is not always the one aligned to the rising edge of
CLK0. For example, if CLKFX is set to a 1.5X multiple of CLK0, the first rising edge of
CLK0 after LOCKED is achieved might be aligned to the falling edge of CLKFX, or it might
be aligned to the rising edge of CLKFX. In this case, you will have alignment on rising
edges at every other CLK0, but not for the very first CLK0 after LOCKED is High.

Frequency Synthesis Applications

The potential applications for frequency synthesis are almost boundless. Some example
applications include the following.
• Generating a completely new clock frequency for the FPGA and external logic using
an available clock frequency on the board.
• Generate a high-frequency internal clock from a slower external clock source to
reduce system EMI.
• Dividing a high-speed serial data clock to process data in parallel within the FPGA, as
shown in Figure 3-39.
• Multiplying a parallel data clock before converting to a high-speed serial data format,
also shown in Figure 3-39.
• Multiplying an input clock to overclock internal logic to reduce resources by time-
sharing logic when implementing moderately fast functions.

Spartan-3 Generation FPGA User Guide www.xilinx.com 131

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

F F

FPGA

DCM DCM

F F F F.m
n

High-speed serial data Slower parallel data

down-converted to n-bits m-bits up-converted to high-
slower parallel data wide wide speed serial data

Overclocked,
time-shared logic

F F.x

DCM

F x462_38_070903

Figure 3-39: Common Applications of Frequency Synthesis

Input and Output Clock Frequency Restrictions

The input and output clock frequency restrictions for frequency synthesis depend on
which DCM clock outputs are used. For example, the CLKFX and CLKFX180 outputs only
use the DCM’s DFS unit. All the other clock outputs use the DCM’s DLL unit. The DLL
unit has tighter frequency restrictions than the DFS. Consequently, operating the DFS unit
without the DLL allows a wider frequency operating range. When using both the DFS and
DLL units, the DLL frequency range limits the application.
Also, for the Spartan-3 FPGA family, both the DLL and DFS have a low- and a high-
frequency operating mode and the mode settings determine the allowable frequency
operating range.
A valid DCM design requires that the CLKIN frequency be within the operating range
specified in the FPGA data sheet, summarized in Table 3-9, page 81 and Table 3-10,
page 81. Likewise, the output frequency for any of the clock outputs used must fall within
their respective specified operating range.
The example shown in Figure 3-40 uses a Spartan-3 family FPGA because of the extra
restrictions imposed by the High and Low operating frequency modes. Figure 3-40 shows
how the various clock input and clock output specifications line up by frequency range.
Only the low-frequency operating modes are shown. The Spartan-3 FPGA family data
sheet specification for each name is shown within the shaded boxes. Table 3-34, page 133
provides example DCM applications and how the frequency restrictions apply.

132 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Multiplication, Clock Division, and Frequency Synthesis

1 MHz 280 MHz

DFS
(CLKIN_FREQ_FX_{MIN,MAX})
FCLKIN 18 MHz 167 MHz
DLL (Low-Frequency Mode)
(CLKIN_FREQ_DLL_LF_{MIN,MAX})

18 MHz 210 MHz

FCLKFX DFS (Low-Frequency Mode)
FCLKFX180 (CLKOUT_FREQ_FX_LF_{MIN,MAX})

1.125 MHz 110 MHz

FCLKDV DLL, CLKDV (Low-Frequency Mode) Data sheet
(CLKOUT_FREQ_DV_LF_{MIN,MAX}) specification name
36 MHz 334 MHz
FCLK2X DLL, CLK2X (Low-Frequency Mode)
FCLK2X180 (CLKOUT_FREQ_2X_LF_{MIN,MAX})

Frequency
X462_39_011008

See Module 3 of DS099, Spartan-3 FPGA Family: Complete Data Sheet for details.

Figure 3-40: Input and Output Clock Frequency Restrictions (Spartan-3 FPGA Family, Low-Frequency
Mode Example)

Table 3-34: DCM Frequency Restriction Examples (Spartan-3 FPGA Family, Low-Frequency Mode Example)
Input Output
Comments
Frequency Frequency
1.2 MHz 12.8 MHz Not possible in a single DCM. FCLKIN is within acceptable range for DFS unit, but FCLKFX
requires at least an 18 MHz output frequency.
1.2 MHz 32.4 MHz Possible in a single DCM using DFS unit. Set CLKFX_MULTIPLY = 27. FCLKFX is within
the DFS output frequency range.
25 MHz 2.5 MHz Possible in a single DCM using both the DFS and DLL units. Use the CLKDV output for a
30 MHz 2.5 MHz signal, setting CLKDV_DIVIDE=10. Use the CLKFX output for a 30 MHz signal,
setting CLKFX_MULTIPLY = 6 and CLKFX_DIVIDE = 5. All input and output
frequencies are within appropriate ranges.

Clock Doubler (CLK2X, CLK2X180)

The Clock Doubler unit doubles the frequency of the incoming CLKIN input, as
summarized in Table 3-35. The Clock Doubler is part of the DLL functional unit and
requires a clock feedback path back to CLKFB from either the CLK0 or CLK2X output. The
outputs from the Clock Doubler are CLK2X and CLK2X180. Both outputs are always
conditioned to a 50% duty cycle. Both have the same output frequency but CLK2X180 is
180° phase shifted from CLK2X, essentially inverting the CLK2X output. Having both
phases is essential for high-performance Dual-Data Rate (DDR) or clock forwarding
applications.
The CLK2X and CLK2X180 outputs are available in the Spartan-3 family only when the
DLL_FREQUENCY_MODE attribute is LOW. If required by the application, reduce the
CLKIN input frequency using the optional divide-by-two feature (see “Advanced
Options,” page 97).

Spartan-3 Generation FPGA User Guide www.xilinx.com 133

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-35: Clock Doubler Summary

DCM Output(s) CLK2X
CLK2X180
Output Frequency
2 • F CLKIN

DCM Functional Unit Delay-Locked Loop (DLL)

Feedback Required? Yes
50% Duty Cycle? Yes
Controlling Attributes
Spartan-3 FPGAs only: On Spartan-3 FPGAs, the CLK2X and CLK2X180 outputs are only valid when
DLL_FREQUENCY_MODE DLL_FREQUENCY_MODE = LOW.
CLKIN Generally, the CLK2X and CLK2X180 outputs are only available up to a CLKIN of 167
MHz, primarily because the output frequency is limited to 334 MHz. On Spartan-3
FPGAs, the CLKIN frequency limits are determined by the DLL_FREQUENCY_MODE
attribute.
FPGA Family Minimum Frequency Maximum Frequency
CLKIN_FREQ_DLL_LF_MIN CLKIN_FREQ_DLL_LF_MAX
Spartan-3 FPGA
18 MHz 167 MHz
CLKIN_FREQ_DLL_MIN Limited to half maximum
Spartan-3E FPGA 5 MHz CLK2X frequency
(Stepping 1) -4: 155.5 MHz
-5: 167 MHz
CLKIN_FREQ_DLL_MIN Limited to half maximum
Extended Spartan-3A
5 MHz CLK2X frequency
FPGA
167 MHz

CLK2X On Spartan-3 FPGAs, the CLK2X frequency limits are determined by the
CLK2X180 DLL_FREQUENCY_MODE attribute.
FPGA Family Minimum Frequency Maximum Frequency
CLKOUT_FREQ_2X_LF_MIN CLKOUT_FREQ_2X_LF_MAX
Spartan-3 FPGA
36 MHz 334 MHz
CLKOUT_FREQ_2X_MIN CLKOUT_FREQ_2X_MAX
Spartan-3E FPGA
(Stepping 1) 10 MHz -4: 311 MHz
-5: 334 MHz

Extended Spartan-3A CLKOUT_FREQ_2X_MIN CLKOUT_FREQ_2X_MAX

FPGA 10 MHz 334 MHz

Clock Divider (CLKDV)

The Clock Divider unit, summarized in Table 3-36, divides the incoming CLKIN frequency
by the value specified by the CLKDV_DIVIDE attribute, set at design time. The Clock
Divider unit is part of the DLL functional unit and requires a clock feedback path back to
CLKFB from either the CLK0 or CLK2X output.

134 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Multiplication, Clock Division, and Frequency Synthesis

Table 3-36: Clock Divider Summary

DCM Output(s) CLKDV
Output Frequency F CLKIN
-----------------------------------------------
-
CLKDV_DIVIDE
DCM Functional Unit Delay-Locked Loop (DLL)
Feedback Required? Yes, using either CLK0 or CLK2X output from DCM
50% Duty Cycle? Yes, except when DLL_FREQUENCY_MODE=HIGH and CLKDV_DIVIDE is a non-
integer value
Controlling Attributes
Spartan-3 FPGAs only: CLKDV is available in both modes. Potentially affects duty cycle of output (see “CLKDV
DLL_FREQUENCY_MODE Clock Conditioning”), depending on divider value.
CLKDV_DIVIDE Controls the output frequency per the equation above. Legal values include 1.5, 2, 2.5, 3,
3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 9, 10, 11, 12, 13, 14, 15, and 16. The DLL locks faster on
integer values than on non-integer values. Likewise, integer values result in lower
output jitter.
Frequency Constraints
CLKIN The CLKIN frequency limits are listed in the following tables.
Spartan-3E and Extended Spartan-3A family FPGAs: Table 3-9, page 81
Spartan-3 FPGAs: Table 3-10, page 81
CLKDV On Spartan-3 FPGAs, the CLKDV frequency limits are determined by the
DLL_FREQUENCY_MODE attribute.
FPGA Family Minimum Frequency Maximum Frequency

CLKOUT_FREQ_DV_LF_MIN CLKOUT_FREQ_DV_LF_MAX
1.125 MHz 110 MHz
Spartan-3 FPGA
CLKOUT_FREQ_DV_HF_MIN CLKOUT_FREQ_DV_HF_MAX
3.0 MHz 185 MHz

CLKOUT_FREQ_DV_MIN CLKOUT_FREQ_DV_MAX
Spartan-3E FPGA
0.3125 MHz (312.5 kHz) -4: 160 MHz
(Stepping 1) -5: 183 MHz
CLKOUT_FREQ_DV_MIN CLKOUT_FREQ_DV_MAX
Extended Spartan-3A
0.3125 MHz (312.5 kHz) -4: 166 MHz
FPGA
-5: 186 MHz

CLKDV Clock Conditioning

The CLKDV output is conditioned to a 50% duty cycle unless the
DLL_FREQUENCY_MODE attribute is set to HIGH and CLKDV_DIVIDE is a non-integer
value. Under these conditions, the CLKDV duty cycle is shown in Table 3-37. A Spartan-3,
Spartan-3E, or Extended Spartan-3A family DCM requires CLKIN to have at least a

Spartan-3 Generation FPGA User Guide www.xilinx.com 135

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

60%/40% (or 40%/60%) or better duty cycle. Consequently, the CLKDV output, divided
by 1.5 in high-frequency mode cannot provide a clock input to a second cascaded DCM.
Table 3-37: CLKDV Duty Cycle with DLL_FREQUENCY_MODE=HIGH
High Time/
CLKDV_DIVIDE Attribute Duty Cycle
Total Cycle
Integer 50.000% 1/2
1.5 33.333% 1/3
2.5 40.000% 2/5
3.5 42.857% 3/7
4.5 44.444% 4/9
5.5 45.454% 5/11
6.5 46.154% 6/13
7.5 46.667% 7/15

CLKDV Jitter Depends on Frequency Mode and Integer or Non-Integer Value

Similarly, integer values for the CLKDV_DIVIDE attribute result in lower output jitter and
faster DLL locking times.
Table 3-38: CLKDV Output Jitter
Spartan-3 Generation CLKDV Output
CLKDV_DIVIDE Data Sheet Symbol
FPGA Family Period Jitter
Integer Value CLKOUT_PER_JITT_DV1 All ±150 ps
Non-Integer CLKOUT_PER_JITT_DV2 Spartan-3 FPGA ±300 ps
Value
Spartan-3E FPGA
±[1% of CLKIN
Extended Spartan-3A
period + 150] ps
FPGA

Clocking Wizard
The Clock Divider controls are in Clocking Wizard’s General Setup window. Check the
CLKDV output box, shown in Figure 3-41a. Then, choose the Clock Divider’s Divide by
Value using the drop-down list, shown in Figure 3-41b.

Divide By Value
CLKDV 2
UG331_c3_21_120306

x462_40b_061903

a. Check the CLKDV Output Box b. Select the Divide by Value from the Drop-Down List
Figure 3-41: Specifying the Clock Divider in Clocking Wizard

Frequency Synthesizer (CLKFX, CLKFX180)

The Frequency Synthesizer provides the most flexible means to multiply, divide, or
multiply and divide an input frequency. As shown in Table 3-39, the two Frequency
Synthesizer outputs are CLKFX and CLKFX180. The CLKFX180 output has the same

136 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Multiplication, Clock Division, and Frequency Synthesis

frequency as CLKFX but is phase shifted 180°, or half a clock period. Because both
Frequency Synthesizer outputs have 50% duty cycles, CLKFX180 appears to be an inverted
version of CLKFX.
Two attributes, set at design time, control the synthesized output frequency, as shown in
the equation in Table 3-39. The CLKIN clock input is multiplied the fraction formed by
C LKFX_M U LTIPLY as the numerator and CLKFX_DIVIDE as the denominator. For
example, to create a 155 MHz output using a 75 MHz CLKIN input, the Frequency
Synthesizer multiplies CLKIN by the fraction 31/15. Note that it does not multiply CLKIN
by 31 first, then divide by the result by 15. Multiplying CLKIN by 31 would result in a
2.325 GHz output frequency—well outside the frequency range of the Spartan-3 FPGA
DCM.
The multiplier and divider values should be reduced to their simplest form, which results
in faster lock times. For example, reduce the fraction 6/8 to 3/4.
Frequency synthesis always requires some form of clock feedback. However, the DFS unit
has an internal feedback loop based on CLKIN and does not require a separate loop on
CLKFB if used without the DLL unit.
The CLKFX output is phase aligned with the CLKIN input every CLKFX_DIVIDE cycles of
CLKIN and every CLKFX_MULTIPLY cycles of CLKFX. For example, if
CLKFX_MULTIPLY = 3 and CLKFX_DIVIDE = 5, then the CLKFX output is phase aligned
with the CLKIN input every five CLKIN cycles and every three CLKFX cycles. After the
DCM asserts its LOCKED output, the DFS unit is resynchronized to the CLKIN input at
each concurrence and phase alignment is nearly perfect at these edges.
Table 3-39: Frequency Synthesizer Summary
DCM Output(s) CLKFX
CLKFX180 (same as CLKFX, phase shifted 180°)
Output Frequency CLKFX_MULTIPLY
F CLKIN • ---------------------------------------------------------
CLKFX_DIVIDE

DCM Functional Unit Digital Frequency Synthesizer (DFS)

Feedback Required? No. Uses internal feedback based on CLKIN. Optionally can use CLKFB input if required
for Delay-Locked Loop (DLL) functions.
50% Duty Cycle? Yes, always.
Controlling Attributes
Spartan-3 FPGAs only: Affects frequency limits on CLKIN and the CLKFX, CLKFX180 outputs.
DFS_FREQUENCY_MODE
Spartan-3 FPGAs only: Only affects the Frequency Synthesizer if the application uses any DLL outputs.
DLL_FREQUENCY_MODE Potentially reduces the CLKIN frequency to the more restrictive DLL limits. If only the
CLKFX or CLKFX180 outputs are used, then DFS_FREQUENCY_MODE alone defines
the frequency limits.
CLKFX_MULTIPLY Controls the output frequency per the equation above. Legal values include integer
values ranging from 2 to 32. Default value is 4.
CLKFX_DIVIDE Controls the output frequency per the equation above. Legal values include integer
values ranging from 1 to 32. Default value is 1.

Spartan-3 Generation FPGA User Guide www.xilinx.com 137

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Table 3-39: Frequency Synthesizer Summary (Cont’d)

Frequency Constraints
CLKIN The CLKIN frequency limits depend on whether the application uses any outputs from
the Delay-Locked Loop (DLL) unit. If the DLL unit is used, then the more restrictive DLL
clock limits apply.
DFS Alone: Table 3-10, page 81
DFS Used with DLL:
Spartan-3 FPGAs: Table 3-10, page 81
Spartan-3E and Extended Spartan-3A family FPGAs: Table 3-9, page 81.
CLKFX The CLKFX and CLKFX180 output frequency limits are determined by the
CLKFX180 DFS_FREQUENCY_MODE attribute.

FPGA Family Minimum Frequency Maximum Frequency

CLKOUT_FREQ_FX_LF_MIN CLKOUT_FREQ_FX_LF_MAX
18 MHz 210 MHz
Spartan-3 FPGA
(Mask Rev ‘E’) CLKOUT_FREQ_FX_HF_MIN CLKOUT_FREQ_FX_HF_MAX
210 MHz -4: 307 MHz
-5: 326 MHz

Spartan-3E FPGA CLKOUT_FREQ_FX_MIN CLKOUT_FREQ_FX_MAX

5 MHz -4: 311 MHz
(Stepping 1) -5: 333 MHz
CLKOUT_FREQ_FX_MIN CLKOUT_FREQ_FX_MAX
Extended
5 MHz -4: 320 MHz
Spartan-3A FPGA
-5: 350 MHz

Clocking Wizard
To enable the Frequency Synthesizer in Clocking Wizard, check the CLKFX, CLKFX180, or
both clock outputs in the General Setup window, as shown in Figure 3-42.

CLK0

Check CLKFX or
CLKFX CLKFX180 to enable
the Frequency
CLKFX180 Synthesizer options
UG331_c3_15_061708

Figure 3-42: Enabling Frequency Synthesizer in Clocking Wizard

If using the CLKFX or CLKFX180 clock outputs stand-alone, then optionally extend the
frequency limits by disabling any DLL clock outputs and any feedback.
• Disable DCM feedback by selecting None, as shown in Figure 3-43. Without feedback,
the CLKFX and CLKFX180 frequency range is extended to both lower and higher
frequencies and disables the CLK0 and other DLL outputs.

138 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Forwarding, Mirroring, Rebuffering

If only using the CLKFX or CLKFX180

clock ouputs, optionally click None to
extend the DCM frequency limits.

Feedback
Source: Internal External None
Value: 1X 2X
x462_42_070903

Figure 3-43: Select No Feedback (None) to Extend Frequency Synthesizer

Frequency Limits

Finally, enter the desired output frequency or the Multiply and Divide values, as described
in the Clocking Wizard Clock Frequency Synthesizer panel section.

Clock Forwarding, Mirroring, Rebuffering

Because DCMs provide advanced clock control features and Spartan-3 generation I/O
pins support a variety of I/O voltage standards, Spartan-3 generation FPGAs commonly
are used to rebuffer or mirror clock signals, often changing the input clock from one
voltage standard to another. Likewise, the DCM conditions an incoming clock signal so
that it has a 50% duty cycle.
Figure 3-20 shows a simple example where a DCM conditions an incoming clock to a 50%
duty cycle, and then either forwards the clock at the same frequency using the CLK0
output, or doubles the frequency using the DCM CLK2X output. Similarly, the input and
output clocks are phase aligned once the DCM asserts its LOCKED output. The clock
feedback path to CLKFB monitors and eliminates the clock distribution delay at the
external clock feedback point.
If a 50/50 duty cycle is important on the output clock, make sure that the output I/O
standard can switch fast enough to preserve the 50% duty cycle. Verify the duty cycle
performance using IBIS simulation on the output signal. Some I/O standards have
asymmetric rise and fall times that distort the duty cycle higher frequencies. On the
Spartan-3 FPGA family, the DCI versions of HSTL, SSTL, and LVCMOS I/O standards
have better symmetry. Generally, differential I/Os also have less distortion.
To guarantee a 50/50 duty cycle above 100 MHz, the DCM’s duty cycle correction
capability is mandatory for the Spartan-3 FPGA family, even if the CLKIN source provides
a clean 50% duty cycle. Consequently, the DUTY_CYCLE_CORRECTION attribute must
equal TRUE when using the CLK0, CLK90, CLK180, or CLK270 outputs for clock
forwarding. The other DCM clock outputs are normally always clock corrected to a 50%
duty cycle (see “Clock Conditioning”).
For best duty-cycle performance—especially at 200 MHz and greater—use a circuit similar
to that shown in Figure 3-44. Use both the CLKx and CLKx180 outputs from the DCM to
drive the C0 and C1 inputs, respectively, on a Dual-Data Rate (DDR) output flip-flop. The
Spartan-3 family provides the OFDDRCPE and variations, while the Spartan-3E and
Extended Spartan-3A families provide the superset ODDR2 component. Connect the D0
input of the DDR flip-flop to VCC and the D1 input to GND. Each DCM output drives a
separate global buffer, which minimizes duty-cycle distortion. At higher frequencies, it is
best not to distribute just one clock and invert one phase locally within the DDR flip-flop,
as this adds approximately 400 ps of duty-cycle distortion.
At frequencies of 250 MHz or higher, distribute clocks using a differential signaling
standard, such as LVDS. In Figure 3-44, for example, both the CLKIN clock input and the
clock output use LVDS. Additionally, the clock feedback path uses LVDS. For optimal

Spartan-3 Generation FPGA User Guide www.xilinx.com 139

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

performance, both the clock input and the clock feedback paths require differential global
buffer inputs (IBUFGDS), which unfortunately consumes all the global buffer inputs along
one edge of the device. However, this solution provides the best-quality clock forwarding
solution at high frequencies.

FPGA
OFDDRCPE/ODDR2 OBUFDS
IBUFGDS VCC D0 Q
GND D1
DCM
BUFG CE
CLKIN CLKx C0
CLKFB CLKx180 C1
IBUFGDS BUFG

External Clock Feedback (LVDS) x462_43_061908

Figure 3-44: High-Frequency (250+ MHz) LVDS Clock Forwarding Circuit with 50% Duty Cycle

Clock Jitter or Phase Noise

All clocks, including the most expensive, high-precision sources, exhibit some amount of
clock jitter or phase noise. The Spartan-3 FPGA Digital Clock Managers have their own
jitter characteristics, as described in this section. When operating at low frequencies—20
MHz, for example—the effects of jitter usually can be ignored. However, when operating
at high frequencies—200 MHz, for example, especially in dual-data rate (DDR)
applications—clock jitter becomes a relevant design factor. Clock jitter directly subtracts
from the time available to the FPGA application by effectively reducing the available time
between active clock edges.

What is Clock Jitter?

Clock jitter is the variation of a clock edge from its ideal position in time, as illustrated in
Figure 3-45. The heavy line shows the ideal position on the clock signal. On each clock
edge, there is some amount of variation between the actual clock edge and its ideal
location. The difference between the maximum and minimum variations is called peak-to-
peak jitter. Jitter is only relevant on the active clock edge. For example, in single-data rate
(SDR) applications, data is clocked at each rising clock edge and the specified jitter only
subtracts from the total clock period. In dual-data rate (DDR) application, data is clocked at
the start of each period and halfway into the period. Therefore, jitter affects each half
period.

140 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Jitter or Phase Noise

Ideal Clock Peak-to-Peak

Edge Location Jitter

Clock Edge
with Jitter

Clock Period X462_44_060606

Figure 3-45: Jitter in Clock Signals

What Causes Clock Jitter?

Clock jitter is unavoidable and exists in all systems. Clock jitter is caused by the various
sources of noise or by signal imperfections within the system. In fact, jitter is the
manifestation of noise in the time domain. The incoming clock source, for example, has its
own jitter characteristics due to random thermal or mechanical vibration noise from the
crystal. A large number of simultaneous switching outputs (SSOs) adds substrate noise
that slightly changes internal switching thresholds and therefore adds jitter. Similarly, an
improperly designed power supply or insufficient decoupling also contributes to jitter.
Other sources of clock jitter include cross talk from adjacent signals, poor termination,
ground bounce, and electromagnetic interference (EMI).
This chapter only discusses the jitter behavior of Spartan-3 FPGA Digital Clock Managers
(DCMs) and how to improve overall jitter performance within the FPGA.

Understanding Clock Jitter Specifications

Clock jitter is specified in a variety of manners, and the various specifications show
different aspects of the same phenomenon.

Cycle-to-Cycle Jitter
Cycle-to-cycle jitter, also called adjacent cycle jitter, indicates the maximum clock period
variance from one clock cycle to the next, as shown in Figure 3-46. In this simple example,
the maximum change from one cycle to the next is +100 ps and –100 ps, or put simply,
±100 ps. Although the clock period can change by larger absolute amounts when measured
over millions of clock cycles, the clock period never changes by more than ±100 ps from one
clock cycle to the next.

Cycle-to-cycle jitter is the

difference in clock period
from one cycle to the next.

T0 T1 = T0 + 100 ps T2 = T1 - 100 ps

x462_45_062203

Figure 3-46: Cycle-to-Cycle Jitter Example

Spartan-3 Generation FPGA User Guide www.xilinx.com 141

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Cycle-to-cycle is an important measure of the quality of a clock output or oscillator but has
little use in analyzing the timing of an application.

Period Jitter
Period jitter is the summation of all the cycle-to-cycle jitter values over millions of clock
cycles. Peak jitter indicates the earliest and the latest transition times compared to the ideal
clock transition time over consecutive clocks.
Period jitter for Digital Clock Mangers is random and is expressed as peak-to-peak jitter.
Conceptually, the position of the clock transition is a probabilistic distribution or
histogram, centered around the ideal, desired clock position, as shown in Figure 3-47. The
actual distribution might not appear purely Gaussian and can be bimodal. Regardless,
most actual clock transitions occur near the desired ideal position. However, measured
over millions of clock cycles, some clock transitions occur far from the desired position.
The statistical distance from the desired position is measured in standard deviations, also
called σ (sigma). Because the DCM is an all-digital design, it is highly stable and Xilinx
specifies jitter deviation to ±7σ or peak-to-peak jitter to 14σ. As a point of reference, ±7σ
guarantees that 99.99999999974% of the jitter values are less than the specified worst-case
jitter value. A 14σ peak-to-peak jitter, ±7σ jitter deviation, equates to a maximum bit error
rate (BER) of 1.28 x 10-12.

Desired clock period

Number of samples

-7σ +7σ

Measured clock period

Peak-to-peak Period Jitter

(14σ) x462_46_061903

Figure 3-47: Peak-to-Peak Period Jitter Example

Unit Interval (UI)

Another method to specify jitter is as a fraction of the Unit Interval (UI). One UI represents
the time equivalent to one bit time, irrespective of frequency. In single-data rate (SDR)
applications where either the rising or the falling clock edge captures data, one UI equals
one clock period. In dual-data rate (DDR) applications where data is clocked at twice the
clock rate, one UI equals half the clock period.
The peak-to-peak jitter amplitude, quantified in UIs, is the fraction of the peak-to-peak
jitter value compared to the total bit period time.

142 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Jitter or Phase Noise

Peak-to-peak
Period Jitter

Bit Period

Unit Interval (UI) Peak-to-peak period jitter,

represented as fraction of
Unit Interval.
x462_47_061903

Figure 3-48: Period Jitter Specified as a Fraction of a Unit Interval

Calculating Total Jitter

The FPGA family data sheet specifies the output jitter from the DCM clock outputs, except
for the CLKFX/CLKFX180 outputs, which sometime use a separate jitter calculator. The
DFS jitter is calculated based on the multiplier and divider settings.
The clock outputs from the DLL unit—i.e., every clock output except CLKFX and
CLKFX180—have a worst-case specified jitter listed in the data sheet. This specified value
includes the jitter added by the DLL unit. The DLL unit does not remove jitter, so the total
jitter on the DLL clock output includes the jitter on the input clock, CLKIN, plus the
specified value from the data sheet.
The DFS clock outputs, CLKFX and CLKFX180, remove some amount of incoming clock
jitter, so the calculated output jitter is the total jitter.

Adding Input Jitter to DLL Output Jitter

When adding the input jitter and the DLL output jitter, use a root-mean-square (RMS)
calculation, similar to noise calculations.

Peak-to-Peak
2 2
JITTER PK – PK = ( JITTER INPUT ) + ( JITTER SPEC ) Equation 3-12

Peak-to-Peak Deviation
2 2
( JITTER INPUT ) + ( JITTER SPEC )
JITTER PK = ± ----------------------------------------------------------------------------------------------- Equation 3-13
2

where

JITTERINPUT = The input period jitter, measured at the clock input pin of the FPGA

JITTERSPEC = The DLL clock output period jitter, as specified in the FPGA family
data sheet for the associated output port

Example
Assume that an input clock has 150 ps peak-to-peak period jitter, optionally expressed as
±75 ps. The incoming clock is duty-cycle corrected, using the same frequency, on the CLK0
DCM output.

Spartan-3 Generation FPGA User Guide www.xilinx.com 143

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

In this case, JITTERINPUT = 150 ps. The value for JITTERSPEC is the Spartan-3 FPGA Data
Sheet specification called CLKOUT_JITT_PER_0, which is estimated here as ±100 ps, or
200 ps peak-to-peak.
2 2
JITTER PK – PK = ( 150 ps ) + ( 200 ps ) = 250 ps Equation 3-14

Consequently, the total jitter on the DCM output is 250 ps peak-to-peak or ±125 ps.

Calculating Jitter for Cascaded DCMs

Figure 3-49 shows an example application where multiple DCMs are cascaded together to
create various output frequencies. The jitter at any point depends on:
• the incoming jitter from the previous sources and
• which DCM output is used.
Each DCM output has slightly different jitter characteristics, as specified in the data sheet.
Also, the CLKFX and CLKFX180 outputs from the DFS unit remove some amount of input
jitter and Clocking Wizard calculates their jitter values (see “Clock Frequency
Synthesizer”).

DCM A DCM B DCM C

CLKIN CLKx CLKIN CLKx CLKIN CLKx

RST LOCKED RST LOCKED RST LOCKED

x462_48_061903

Figure 3-49: Calculating Jitter for Cascaded DCMs Depends on which DCM Outputs are Used

Consequently, the jitter at any point in the cascaded DCM chain depends on the factors
described above. The following examples illustrate how to calculate total jitter at the
various points in the circuit.

Example 1: All DCMs Use DLL Outputs

In this example, assume that all the FPGAs are from the Spartan-3 family and that the input
clock has 150 ps (±75 ps) of period jitter.
Assume that DCM (A) uses the CLK2X output. Use the Spartan-3 Data Sheet specification
called CLKOUT_PER_JITT_2X for the DCM output jitter, estimated here as 400 ps (±200
ps). Calculate the total period jitter on clock (A) using Equation 3-12.
2 2
JITTER PK – PK ( A ) = ( 150ps ) + ( 400ps ) = 427ps = ±214ps Equation 3-15

Assume that DCM (B) uses the CLKDV output with an integer divider value. Use the
Spartan-3 Data Sheet specification called CLKOUT_PER_JITT_DV1 for the DCM output
jitter, estimated here as 300 ps (±150 ps). Calculate the total period jitter on clock (B) using
Equation 3-12. Because there are now three elements involved—the input jitter, the jitter
from DCM (A), and the jitter from DCM (B)—expand the RMS equation appropriately.
2 2 2
JITTER PK – PK ( B ) = ( 150ps ) + ( 400ps ) + ( 300ps ) = 522ps = ±261ps Equation 3-16

144 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Jitter or Phase Noise

Finally, assume that DCM (C) phase shifts the output from DCM (B) by 90°. Use the
Spartan-3 Data Sheet specification called CLKOUT_PER_JITT_90 for the DCM output jitter,
estimated here as 300 ps (±150 ps). Calculate the total period jitter on clock (C) using
Equation 3-12. Because there are now four elements involved—the input jitter, the jitter
from DCM (A), the jitter from DCM (B), and the jitter from DCM (C)—expand the RMS
equation appropriately.

2 2 2 2
JITTER PK – PK ( C ) = ( 150ps ) + ( 400ps ) + ( 300ps ) + ( 300ps ) Equation 3-17
= 602ps = ±301ps

Example 2: Some DCMs Use the CLKFX or CLKFX180 Outputs

This example is similar to Example 1: All DCMs Use DLL Outputs above except that some
DCMs use the CLKFX or CLKFX180 outputs from the DCM’s DFS unit.
In this example, assume that the 75 MHz input clock has 150 ps (±75 ps) of period jitter.
As in Example 1, assume again that DCM (A) uses the CLK2X output. The resulting output
jitter is the same as that shown in the following equation.
In this example, assume that DCM (B) synthesizes a 90 MHz clock using the 150 MHz
clock generated by DCM (A). Per the Clocking Wizard (see “Clock Frequency
Synthesizer”), set the attributes CLKFX_MULTIPLY = 3 and CLKFX_DIVIDE = 5.
Clocking Wizard also specifies the worst-case output period jitter as 700 ps.

JITTER PK – PK ( B ) = 700ps = ±350ps Equation 3-18

Finally, assume again that DCM (C) phase shifts the output from DCM (B) by 90°. Use the
Spartan-3 Data Sheet specification called CLKOUT_PER_JITT_90 for the DCM output jitter,
estimated here as 300 ps (±150 ps). Calculate the total period jitter on clock (C) using the
following equation. Because the preceding DCM used the CLKFX output, the total
incoming jitter is set at 700 ps, worst-case. Use the RMS equation to calculate the resulting
output jitter as shown below.
2 2
JITTER PK – PK ( C ) = ( 700ps ) + ( 300ps ) = 762ps = ±381ps Equation 3-19

Cascaded DCM Design Recommendations

When cascading DCMs, be sure that the LOCKED output of the preceding DCM controls
the cascaded DCM’s RST input, as shown in Figure 3-49. The cascaded DCM should not
attempt to lock to the input clock until the preceding DCM asserts its LOCKED output,
indicating that the clock is stable.
When cascading DCMs, place the most jitter-critical clock output on the first DCM in the
cascaded chain.

Jitter Effect on System Performance

Clock jitter, along with other effects, adversely affects system performance by reducing the
effective bit period. The bit period available to the FPGA application is the total bit period,
TBIT, minus the following effects, as shown in the following equation. In single-data rate

Spartan-3 Generation FPGA User Guide www.xilinx.com 145

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

(SDR) applications, the clock period and the bit period are equal. However, in dual-data-
rate (DDR) applications, the bit period is half the clock period.

T AVAILABLE = T BIT – t TOTAL_JITTER – t DUTY_CYCLE_DISTORTION Equation 3-20

where

TBIT = Bit period time

tTOTAL_JITTER = Total clock jitter. Includes the clock input jitter plus any DCM
output jitter or cascaded DCM output jitter.
tDUTY_CYCLE_DISTORTION = Duty cycle distortion specification. Only required for dual-
data rate (DDR) applications; otherwise zero. Either data
sheet specification CLKOUT_DUTY_CYCLE_DLL or
CLKOUT_DUTY_CYCLE_FX depending on which DCM
clock output is used.

If the total jitter is specified as a positive value instead of a deviation from the clock
period—e.g., 200 ps instead of ±100 ps—subtract half the positive value—i.e., 100 ps. The
bit period is only shortened by the negative deviation. The positive deviation adds to the
bit period, adding more timing slack.

Example
Assume that an incoming clock signal enters the FPGA at 75 MHz and that the clock source
has ±100 ps of jitter. The application clocks data on the rising edge of an internally
generated 150 MHz clock, or a total bit period, TBIT, of 6.67 ns. How long is the available bit
period, TAVAILABLE, after considering the effects of jitter?
The CLK2X output from the Clock Doubler generates a 150 MHz clock from the 75 MHz
clock input. The Clock Doubler output, CLK2X, has ±200 ps of worst-case jitter according to
the CLKOUT_PER_JITT_2X specification in the Spartan-3 Data Sheet. Adding the DCM’s
±200 ps of jitter to the clock source’s ±100 ps of jitter using root-mean square (RMS), the
total jitter, tTOTAL_JITTER, is ±0.223 ns.
2 2
t TOTAL_JITTER = ( ±100ps ) + ( ±200ps ) = ±223.60ps = ±0.223ns Equation 3-21

Because data is only clocked on the rising clock edge, there are no duty-cycle distortion
effects and tDUTY_CYCLE_DISTORTION = 0.
Therefore, the total available clock period, TAVAILABLE is reduced down to 6.444 ns from a
total bit period of 6.667 ns. Effectively, this forces the logic to operate at 155.1831 MHz
instead of 150 MHz.

T AVAILABLE = 6.667ns – 0.223ns = 6.444ns Equation 3-22

Recommended Design Practices to Minimize Clock Jitter

In higher-performance applications, clock jitter steals valuable bit period time. Adhere to
the following recommendations to minimize the amount of system-wide clock jitter.

146 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Clock Jitter or Phase Noise

Properly Design the Power Distribution System

A properly designed power distribution system (PDS), including proper power-plane
decoupling, reduces system jitter by creating a stable power environment. Application
note XAPP623 discusses recommended design practices for PDS design.
• XAPP623: Power Distribution System (PDS) Design: Using Bypass/Decoupling Capacitors
http://www.xilinx.com/support/documentation/application_notes/xapp623.pdf

Properly Design the Printed Circuit Board

Design the printed circuit board for expected operating frequency range and application
environment.
• WP174: Methodologies for Efficient FPGA Integration into PCBs
http://www.xilinx.com/support/documentation/white_papers/wp174.pdf
• PCB Checklist
www.xilinx.com/products/design_resources/signal_integrity/si_pcbcheck.htm

Obey Simultaneous Switching Output (SSO) Recommendations

To avoid signal-related corruption of clock inputs to or clock outputs from a DCM, be sure
to follow the Simultaneous Switching Output (SSO) recommendations outlined in the
associated FPGA family data sheet.
Whenever possible, avoid placing DCM inputs or outputs near heavily switching I/Os,
especially those with large output voltage swings or with high current drive.

Optionally Place Virtual Ground Pins Around DCM Input and Output
Connections
On sensitive, high frequency DCM inputs or outputs, use additional user-I/O pins to
create extra connections to the PCB ground—i.e., create virtual ground pins. Place these
virtual ground pins on the I/O pads adjacent to the sensitive DCM signal. Make sure that
the I/O pads are on adjacent pads on the FPGA die level, not just on adjacent pins or balls
on the package. Adjacent balls on BGA packages do not necessarily connect to adjacent
pads on the FPGA. These techniques reduce the internal voltage drop and improve the
jitter.
To create a “virtual ground”, configure an IOB as a high-drive output driving GND (Low
logic level) and connect the IOB externally directly to the ground plane, as shown in
Figure 3-50.

Spartan-3 Generation FPGA User Guide www.xilinx.com 147

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

“Virtual” FPGA
ground pin
Direct connection OBUF
to PCB ground GND
plane

IBUFG
Sensitive DCM To DCM
Clock Input CLKIN Input

OBUF
GND

x462_49_061903

Figure 3-50: Place Virtual Ground Pins Adjacent to Sensitive DCM Input or Output
Clock Signals

The same technique can be used to provide a virtual VCC rail connection. Turning I/O into
virtual GND or virtual VCC can not only help with sensitive signals, but also help with pin
migration. For more information on virtual grounds, see white paper
WP323: Signal Integrity Tips and Tricks.

VCCAUX Considerations for Improving Jitter Performance

The Digital Clock Managers are powered by the VCCAUX supply input. Any excessive
noise on the VCCAUX supply input to the FPGA adversely affects the DCM’s
characteristics, especially its jitter performance. For best DCM performance, please follow
these recommendations.
Note: Extended Spartan-3A family FPGAs optionally support VCCAUX = 3.3V, making it possible to
eliminate the 2.5V supply rail in a 3.3-volt only application. Isolate the VCCAUX inputs from possible
switching noise originating from the 3.3V supply connected to VCCO inputs. Spartan-3AN FPGAs
require VCCAUX = 3.3V.
1. Limit changes on the VCCAUX power supply or ground potentials to less than 10 mV
total or 10 mV in any 1 ms interval, as shown in Figure 3-51. This recommendation
allows the DCM to properly track out the change.
2. Limit the noise at the power supply to be within 200 mV peak-to-peak, as shown in
Figure 3-51.

VCCAUX
Supply Keep VCCAUX noise
(at FPGA) envelope to < 200 mV,
peak-to-peak

dV < 10 mV
Avoid sudden changes from
one DC level to another.
Keep dV/dt < 10mV/mS. dt < 1 mS x462_50_061903

Figure 3-51: Recommended VCCAUX Supply Considerations Avoid Voltage Droop

3. If VCCAUX and VCCO are of the same power plane, every VCCAUX/VCCO pin must be
properly decoupled or bypassed (see “Properly Design the Power Distribution

148 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Miscellaneous Advanced Topics

System”). Separate the VCCAUX supply from any VCCO supplies if Guidelines 1 and 2
above cannot be maintained.
4. The CLK2X output is especially affected by the power or ground shift. Consequently,
the CLKFX output, using CLKFX_MULTIPLY =2 and CLKFX_DIVIDE=1, might
provide a better quality output when all IOBs and CLBs are switching. The CLKFX
circuitry updates the tap every three input clocks in the DFS mode, as opposed to the
slower update rate for the CLK2X output.

Adjusting FACTORY_JF Setting (Spartan-3 FPGA Family Only)

Note: The FACTORY_JF attribute only applies for the Spartan-3 FPGA family, not to Spartan-3E or
Extended Spartan-3A family FPGAs.
A well-designed, stable, properly decoupled power supply is the best overall solution to
reducing clock skew and jitter within the FPGA. However, increasing the FACTORY_JF
attribute setting to 0xFFFF might improve jitter performance on a problem board. When
FACTORY_JF=FFFF, the DCM updates its tap settings approximately every twenty input
clocks. The frequency-based default settings update the tap settings much more slowly.
Increasing the FACTORY_JF setting might introduce a small amount of jitter (~30 ps)
because the DCM frequently updates its delay line, which is why FACTORY_JF is not set to
the maximum value by default. If the power supply is unstable, the phase error introduced
can be much bigger than the extra jitter introduced; therefore, increasing the FACTORY_JF
setting might improve the design.

Miscellaneous Advanced Topics

Bitstream Generation Settings

There are two bitstream generation (BitGen) options related to the DCM. Also see
UG332: Spartan-3 Generation Configuration User Guide for more information.
• -g lck_cycle: This option causes the FPGA configuration startup sequence to wait
until all instantiated DCMs assert their LOCKED outputs.
• -g DCMShutdown: This option resets the DCM logic if the "SHUTDOWN"
configuration command is loaded into the configuration logic, as during either partial
reconfiguration or during full reconfiguration via the JTAG port.

Setting Bitstream Generation Options in Project Navigator

If using the ISE Project Navigator graphical interface, set the bitstream generation options
by right-mouse clicking Generate Programming File in the Processes for Current Source
panel, as shown in Figure 3-52. Select Properties from the resulting menu.

Spartan-3 Generation FPGA User Guide www.xilinx.com 149

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

UG332_c1_04_120306

Figure 3-52: Setting Bitstream Generator (BitGen) Options within Project

Navigator

See the "Configuration Bitstream Generator (BitGen) Settings" chapter in UG332: Spartan-3
Generation Configuration User Guide for more information.

Setting Bitstream Generation Options via Command Line or Script

To see the available options, type the following in a command window:
bitgen -help spartan3 (or spartan3e or spartan3a)

Setting Configuration Logic to Wait for DCM LOCKED Output

The DCM’s STARTUP_WAIT attribute signals the FPGA’s configuration start-up logic to
wait for the DCM to assert its LOCKED output before the FPGA asserts its DONE output.
Two actions are required at design time, however. First, set the STARTUP_WAIT attribute
to TRUE on each of the DCMs that must be locked before configuration completes. Then,
modify the bitstream generation options so that the events shown in Figure 3-53 happen
within the six-clock Startup cycle. Sufficient configuration clock cycles must be provided
after the DCM locks to allow the device to complete the configuration start-up sequence.

Start-up Cycles 0 1 2 3 4 5 6

Start-up CLK

GTS_cycle A

LCK_cycle B

GWE_cycle C

DONE_cycle D

x462_52_062403

Figure 3-53: Start-up Logic Interaction with DCM LOCKED Output

a. If using External Feedback, release the FPGA’s internal Global Three-State

(GTS_cycle) signal, enabling all I/O signals.

150 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Miscellaneous Advanced Topics

b. Set the cycle where the start-up logic waits for the DCM(s) to assert LOCKED after
the GTS_cycle. The DCMs require some form of external input—a clock and
possibly a feedback signal—before the DCM can lock on the clock signal.
c. After achieving valid DCM lock, assert the FPGA's internal Global Write Enable
(GWE_cycle) signal.
d. Finally, assert the DONE signal.
Figure 3-54 shows these same option settings from within Project Navigator.

Click Startup Options

Set Done cycle to be If using external

after Release DLL , feedback , set Enable
DCM wait cycle Outputs (GTS _cycle)
(LCK_cycle). to any cycle earlier
than Release DLL ,
Release DLL cycle is DCM wait cycle
the DCM wait cycle (LCK_cycle).
(LCK_cycle) where the
Startup sequencer
waits for DCM(s) to
assert LOCKED
before continuing .

UG331_c3_17_022407

Figure 3-54: Startup Sequencer Options

The specific start-up phase timing and the timing of both the GWE_cycle and DONE_cycle
are flexible. However, if using the STARTUP_WAIT attribute on a DCM, the GTS_cycle
must always happen before the LCK_cycle. Otherwise, the DCM never locks and
configuration never completes! Similarly, if using External Feedback, the FPGA’s outputs
must first be enabled (GTS_cycle) so that the external feedback signal can propagate back
to the DCM.

Reset DCM During Partial Reconfiguration or During Full Reconfiguration via

JTAG
Another bitstream option resets all the DCMs in the FPGA application during
reconfiguration via the SelectMAP interface or during full or partial reconfiguration via
the JTAG port. If the option is enabled, the DCMs are reset when the AGHIGH
configuration command is issued during the SHUTDOWN command sequence. It is
imperative to reset the DCMs when reconfiguring through JTAG. Change the bitstream
generator options in Project Navigator (see “Setting Bitstream Generation Options in
Project Navigator”). Click Configuration options, then check the Reset DCM if
SHUTDOWN & AGHIGH performed option as shown in Figure 3-55.

Spartan-3 Generation FPGA User Guide www.xilinx.com 151

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

Click Configuration
Options

Check to reset all

DCMs in the FPGA if
an AHIGH command
is issued during a
SHUTDOWN
sequence .

UG331_c3_18_120306

Figure 3-55: Configuration Option Allows DCM Reset During Reconfiguration Process

Momentarily Stopping CLKIN

To reduce overall system noise while taking precision analog measurements, it is possible
to momentarily stop the clock inputs to the DCM without adversely affecting the
remainder of the FPGA application. This is possible, in part, because the DCM is an all-
digital, stable system. The DCM must first lock to the input clock and assert the LOCKED
output. If the DCM is not reset, it is possible to momentarily stop the CLKIN input clock
with little impact to the deskew circuit, provided that these guidelines are followed:
• The clock must not be stopped for more than 100 ms to minimize the effect of device
cooling, which would change the tap delays.
• The clock should be stopped during a Low phase, and when restored, must generate a
full High half-period.
Although the above conditions do technically violate the clock input jitter specifications,
the DCM LOCKED output stays High and remains High when the clock is restored.
Consequently, the High on LOCKED does not necessarily mean that a valid clock is
available. The above conditions technically do violate the clock input jitter specifications
but work within the limits described above.
When CLKIN is stopped, an additional one to eight output clock cycles are still generated
as the DCM’s digital delay line is flushed. Similarly, once CLKIN is restarted, output clocks
are not generated for one to four clocks cycles as the delay line is filled. The delay line
usually fills within two or three clocks.
Likewise, it is also possible to phase shift the input clock. This phase shift propagates to the
output one to four clocks after the original shift with no disruption to the DCM control.
Figure 3-56 shows an example where the CLKIN input clock is momentarily stopped. The
figure also illustrates the corresponding effect on the CLK2X clock output.

152 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

Clock input must be When restarted, the clock

stopped in Low phase and input must generate a full
for no more than 100 ms. High half-period.

CLKIN

CLK2X

x462_55_062403

Figure 3-56: Momentarily Stopping CLKIN Clock Input

Related Materials and References

• DS099: Spartan-3 FPGA Family Data Sheet
DCM description and specifications.
http://www.xilinx.com/support/documentation/data_sheets/ds099.pdf
• DS312: Spartan-3E FPGA Family Data Sheet
DCM specifications.
http://www.xilinx.com/support/documentation/data_sheets/ds312.pdf
• DS529: Spartan-3A FPGA Family Data Sheet
DCM specifications.
http://www.xilinx.com/support/documentation/data_sheets/ds529.pdf
• DS557: Spartan-3AN FPGA Family Data Sheet
DCM specifications.
http://www.xilinx.com/support/documentation/data_sheets/ds557.pdf
• DS610: Spartan-3A DSP FPGA Family Data Sheet
DCM specifications.
http://www.xilinx.com/support/documentation/data_sheets/ds610.pdf
• Spartan-3A/3AN/3A DSP CLKFX Jitter Calculator
Excel file to calculate DFS output jitter based on input and output clock frequencies.
http://www.xilinx.com/support/documentation/data_sheets/s3a_jitter_calc.zip
• Libraries Guide (DCM primitive description) and Development System Reference
Guide (BitGen bitstream generation program and options)
http://www.xilinx.com/support/documentation/dt_ise.htm
• XAPP259: System Interface Timing Parameters
http://www.xilinx.com/support/documentation/application_notes/xapp259.pdf
• XAPP268: Dynamic Phase Alignment
http://www.xilinx.com/support/documentation/application_notes/xapp268.pdf
• XAPP485: 1:7 Deserialization in Spartan-3E/3A FPGAs at Speeds Up to 666 Mbps
http://www.xilinx.com/support/documentation/application_notes/xapp485.pdf
• XAPP486: 7:1 Serialization in Spartan-3E FPGAs at Speeds Up to 666 Mbps
http://www.xilinx.com/support/documentation/application_notes/xapp486.pdf
• XAPP622: SDR LVDS Transmitter/Receiver
http://www.xilinx.com/support/documentation/application_notes/xapp622.pdf

Spartan-3 Generation FPGA User Guide www.xilinx.com 153

UG331 (v1.7) August 19, 2010
 R

Chapter 3: Using Digital Clock Managers (DCMs)

154 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 4

Using Block RAM

Summary
For applications requiring large, on-chip memories, Spartan®-3 generation FPGAs
provide plentiful, efficient SelectRAM memory blocks. Using various configuration
options, SelectRAM blocks create RAM, ROM, FIFOs, large look-up tables, data width
converters, circular buffers, and shift registers, each supporting various data widths and
depths. This chapter describes the features and capabilities of block SelectRAM and
illustrates how to specify the various options using the Xilinx CORE Generator™ system
or via VHDL or Verilog instantiation. Various non-obvious block RAM applications are
discussed with references to additional tools, application notes, and documentation.

Introduction
All Spartan-3 generation FPGAs feature multiple block RAMs, organized in columns. The
total amount of block RAM depends on the size of the Spartan-3 generation FPGA as
shown in Table 4-1.

Table 4-1: Block RAM Available in Spartan-3 Generation FPGAs

RAM RAM Blocks Total RAM Total RAM Total RAM
Family Device Columns Per Column Blocks Bits Kbits
Extended XC3SD1800A 4 20-22 84 1,548,288 1,512K
Spartan-3A FPGAs
XC3SD3400A 5 24-26 126 2,322,432 2,268K
XC3S50A/AN 1 3 3 55,296 54K
XC3S200A/AN 2 8 16 294,912 288K
XC3S400A/AN 2 10 20 368,640 360K
XC3S700A/AN 2 10 20 368,640 360K
XC3S1400A/AN 2 16 32 589,824 576K
Spartan-3E FPGAs XC3S100E 1 4 4 73,728 72K
XC3S250E 2 6 12 221,184 216K
XC3S500E 2 10 20 368,640 360K
XC3S1200E 2 14 28 516,096 504K
XC3S1600E 2 18 36 663,552 648K

Spartan-3 Generation FPGA User Guide www.xilinx.com 155

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Table 4-1: Block RAM Available in Spartan-3 Generation FPGAs (Cont’d)

RAM RAM Blocks Total RAM Total RAM Total RAM
Family Device Columns Per Column Blocks Bits Kbits
Spartan-3 FPGAs XC3S50 1 4 4 73,728 72K
XC3S200 2 6 12 221,184 216K
XC3S400 2 8 16 294,912 288K
XC3S1000 2 12 24 442,368 432K
XC3S1500 2 16 32 589,824 576K
XC3S2000 2 20 40 737,280 720K
XC3S4000 4 24 96 1,769,472 1,728K
XC3S5000 4 26 104 1,916,928 1,872K

Notes:
1. 1Kbit = 1,024 bits, per memory conventions.

Each block RAM contains 18,432 bits of fast static RAM, 16K bits of which is allocated to
data storage and, in some memory configurations, an additional 2K bits allocated to parity
or additional "plus" data bits. Physically, the block RAM has two completely independent
access ports, labeled Port A and Port B. The structure is fully symmetrical, and both ports
are interchangeable and support data read and write operations. Each memory port is
synchronous with its own clock, clock enable, and write enable. Read operations are also
synchronous and require a clock edge and clock enable.
Though physically a dual-port memory, block RAM simulates single-port memory in an
application, as shown in Figure 4-1. Furthermore, each block memory supports multiple
configurations or aspect ratios. Table 4-2 summarizes the essential SelectRAM features.
Cascade multiple block RAMs to create deeper and wider memory organizations with a
minimal timing penalty incurred through specialized routing resources.
The block RAMs in the Spartan-3A DSP platform include an optional output register
similar to the block RAM output register of the Virtex®-4 FPGA. The output register
enables full-speed operation at over 250 MHz for all data widths.

156 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Introduction

WEA RAMB16_SwA_SwB
ENA
SSRA
DOPA[pA–1:0]
CLKA
ADDRA[rA–1:0] DOA[wA–pA–1:0]
DIA[wA–pA–1:0]
DIPA[pA–1:0]

WEB WE RAMB16_Sw
ENB EN
SSRB DOPB[pB–1:0] SSR
DOP[p–1:0]
CLKB CLK
DOB[wB–pB–1:0] DO[w–p–1:0]
ADDRB[rB–1:0] ADDR[r–1:0]
DIB[wB–pB–1:0] DI[w–p–1:0]
DIPB[pB–1:0] DIP[p–1:0]

(a) Dual-Port (b) Single-Port

X463_01_112009
Notes:
1. wA and wB are integers representing the total data path width (i.e., data bits plus parity bits) at ports A and B, respectively.
See Table 4-8 and Table 4-9.
2. pA and pB are integers that indicate the number of data path lines serving as parity bits.
3. rA and rB are integers representing the address bus width at ports A and B, respectively.
4. The control signals CLK, WE, EN, and SSR on both ports have the option of inverted polarity.

Figure 4-1: SelectRAM 18K Blocks Perform as Dual-Port (a) and Single-Port (b) Memory

Table 4-2: SelectRAM 18K Block Memory Features and Applications

Total RAM bits, including 18,432 (16K data + 2K parity)
parity
Memory Organizations 16Kx1
8Kx2
4Kx4
2Kx8 (no parity)
2Kx9 (x8 + parity)
1Kx16 (no parity)
1Kx18 (x16 + 2 parity)
512x32 (no parity)
512x36 (x32 + 4 parity)
256x72 (single-port only)
Parity Available and optional only for organizations byte-wide or
greater. Parity bits optionally available as extra data bits.
Performance 240+ MHz (refer to individual FPGA family data sheet)
Timing Interface Simple synchronous interface. Similar to reading and writing
from a register with a setup time for write operations and
clock-to-output delay for read operations.

Spartan-3 Generation FPGA User Guide www.xilinx.com 157

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Table 4-2: SelectRAM 18K Block Memory Features and Applications (Cont’d)
Single-Port Yes
True Dual-Port Yes
ROM, Initial RAM Contents Yes
Mixed Data Port Widths Yes
Power-Up Condition User-defined data, defaults to zero
Potential Applications Local data storage, FIFOs, elastic stores, register files, buffers,
stacks, circular buffers, shift registers, delay lines, waveform
storage and generation, direct digital synthesis, CAMs,
associative memories, function tables, function generators,
wide logic functions, code converters, encoders, decoders,
counters, state machines, microsequencers, program storage
for embedded processor(s)

Block RAM Differences between Spartan-3 Generation Families

Overall, block RAM is similar in all Spartan-3 generation FPGAs. However,
Extended Spartan-3A family FPGAs have some subtle but significant block RAM
enhancements over Spartan-3E and Spartan-3 family FPGAs, as summarized in Table 4-3.
Extended Spartan-3A family FPGAs have byte-level write enable controls, supported by
the RAMB16BWE design primitive. However, Extended Spartan-3A family FPGA designs
continue to support the RAMB16 design primitive that is used for Spartan-3 or Spartan-3E
FPGA designs (see Table 4-8 and Table 4-9). Timing parameters are similar in functionality
between the Spartan-3, Spartan-3E and Extended Spartan-3A family, but have different
names. Spartan-3A DSP FPGAs add an output register, supported by the RAMB16BWER
primitive.

Table 4-3: Comparison Between Spartan-3/3E, Spartan-3A/3AN, and Spartan-3A

DSP FPGA Block RAMs
Spartan-3/3E Spartan-3A/AN Spartan-3A DSP
FPGA FPGA FPGA
Feature Block RAM Block RAM Block RAM
Individual write-enables for each byte No
lane in x9, x18, or x36 configurations (single write- Yes Yes
enable only)
Special routing resources between block
General
RAM and multiplier for x36 No Yes
Purpose
configurations
Output register No No Yes
Supported by RAMB16 primitive Yes Yes Yes
Supported by RAMB16BWE primitive
No Yes Yes
(RAMB16 with byte-level write enable)
Supported by RAMB16BWER primitive
No No Yes
(RAMB16BWE with output register)

The Xilinx CORE Generator system supports various modules containing block RAM for
Spartan-3 devices including:

158 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Location and Surrounding Neighborhood

• Embedded dual- or single-port RAM modules

• ROM modules
• Synchronous and asynchronous FIFO modules
• Content-Addressable Memory (CAM) modules
Furthermore, block RAM can be instantiated in any synthesis-based design using the
appropriate RAMB16 module from the Xilinx design library (see Table 4-8 and Table 4-9).
This chapter describes the signals and attributes of the Spartan-3 FPGA block RAM
feature, including details on the various attributes and applications for block RAM.

Block RAM Location and Surrounding Neighborhood

As mentioned previously, block RAM is organized in columns. Figure 4-2 shows the block
RAM column arrangement for the XC3S200A. The XC3S50A has a single column of block
RAM, located two CLB columns from the left edge of the device. Spartan-3 generation
FPGAs larger than the XC3S50 have at least two columns of block RAM, adjacent to the left
and right edges of the die, located two columns of CLBs from the I/Os at the edge. In
addition to the block RAM columns at the edge, the XC3S4000, XC3S5000, and
XC3SD1800A have two additional columns—a total of four columns—nearly equally
distributed between the two edge columns. The XC3SD3400A adds a fifth block RAM
column, located two CLB columns to the left of the center DCMs. In some devices, the
block RAM column is interrupted by DCMs or CLBs. Table 4-1 describes the number of
columns and the total amount of block RAM on Spartan-3 generation FPGAs. The edge
columns make block RAM particularly useful in buffering or resynchronizing buses
entering or leaving the FPGA.

Block RAM Columns

XC3S50/A/AN
XC3S100E

XC3S200/A/AN XC3S4000
XC3S400/A/AN XC3S5000
XC3S700A/AN
XC3S1400A/AN XC3SD1800A
XC3S250E
XC3S500E
XC3S1200E
XC3S1600E
XC3S1000
XC3S1500 XC3SD3400A
XC3S2000
Embedded Multipliers
2 CLBs 2 CLBs UG332_c4_12_011008

Figure 4-2: Block RAMs Arranged in Columns with Detailed Floorplan of XC3S200

Immediately adjacent to each block RAM is an embedded 18x18 hardware multiplier.

Co-locating block RAM and the embedded multipliers improves the performance of some

Spartan-3 Generation FPGA User Guide www.xilinx.com 159

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

digital signal processing functions. In the Spartan-3A DSP platform, the multiplier is
extended into the DSP48A block.
Special interconnect surrounding the block RAM provides efficient signal distribution for
address and data. Furthermore, special provisions allow multiple block RAMs to be
cascaded to create wider or deeper memories.

Block RAM/Multiplier Routing Interaction

Each multiplier is located adjacent to an 18 Kbit block RAM and shares some interconnect
resources. In the Spartan-3 and Spartan-3E families, configuring an 18 Kbit block RAM for
32/36-bit wide data (512 x 36 mode) prevents use of the associated dedicated multiplier
because the lower 16 bits of the A multiplicand input are shared with the upper 16 bits of
the block RAM’s Port A Data input. Similarly, the lower 16 bits of the B multiplicand input
are shared with Port B’s Data input.
For more details, see “Multiplier/Block RAM Routing Interaction” in Chapter 11.

Data Flows
Spartan-3 generation block RAM is constructed of true dual-port memory and
simultaneously supports all the data flows and operations shown in Figure 4-3. Both ports
access the same set of memory bits but with two potentially different address schemes
depending on the port’s data width.
1. Port A behaves as an independent single-port RAM supporting simultaneous read and
write operations using a single set of address lines.
2. Port B behaves as an independent single-port RAM supporting simultaneous read and
write operations using a single set of address lines.
3. Port A is the write port with a separate write address, and Port B is the read port with
a separate read address. The data widths for Port A and Port B can be different also.
4. Port B is the write port with a separate write address, and Port A is the read port with
a separate read address. The data widths for Port B and Port A can be different also.

Write Read 3
4 Read Write
Port A

Port B

Spartan-3
Dual Port
Block RAM
Write Write
1 2
Read Read
X463_03_060606

Figure 4-3: Block RAM Support Single- and Dual-Port Data Transfers

Signals
The signals connected to a block RAM primitive divide into four categories, as listed
below. Table 4-4 lists the block RAM interface signals, the signal names for both single-port
and dual-port memories, and signal direction.

160 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Signals

1. Data Inputs and Outputs

2. Parity Inputs and Outputs, available when a data port is byte-wide or wider
3. Address inputs to select a specific memory location
4. Various control signals that manage read, write, or set/reset operations
Table 4-4: Block RAM Interface Signals
Dual Port
Signal Description Single Port Port A Port B Direction
Data Input Bus DI DIA DIB Input
Parity Data Input Bus (available only DIP DIPA DIPB Input
for byte-wide and wider organizations)
Data Output Bus DO DOA DOB Output
Parity Data Output (available only for DOP DOPA DOPB Output
byte-wide and wider organizations)
Address Bus ADDR ADDRA ADDRB Input
Write Enable WE WEA WEB Input
Clock Enable EN ENA ENB Input
Synchronous Set/Reset SSR SSRA SSRB Input
Clock CLK CLKA CLKB Input
Synchronous/Asynchronous Set/Reset N/A RSTA RSTB Input
(Spartan-3A DSP FPGA only)
Output Register N/A REGCEA REGCEB Input
(Spartan-3A DSP FPGA only)

Data Inputs and Outputs

The total width of a port’s data port includes both the data bus and the parity bus, when
applicable, as shown in Figure 4-4. In the 512x36 organization, for example, the 36-bit data
port width includes four parity bits as the more significant bits followed by the 32 data bits
as the less significant bits.
The data and parity input and output signals are always buses; that is, in a 1-bit width
configuration, the data input signal is DI[0] and the data output signal is DO[0].

Data Input Bus — DI[#:0] (DIA[#:0], DIB[#:0])

The Data Input bus is the source of data to be written into RAM.
Data at the DI input bus is written to the RAM location specified by the address input bus,
ADDR, during a Low-to-High transition on the CLK input, when the clock enable EN and
write enable WE inputs are High.

Data Output Bus — DO[#:0] (DOA[#:0], DOB[#:0])

The data output bus, DO, presents the contents of memory cells referenced by the address
bus, ADDR, at the active clock edge during a read operation. During a simultaneous write

Spartan-3 Generation FPGA User Guide www.xilinx.com 161

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

operation, the behavior of the data output latches is controlled by the WRITE_MODE
attribute (see “Read Behavior During Simultaneous Write — WRITE_MODE,” page 173).

Parity Inputs and Outputs

Parity is only supported for data paths byte wide and wider.
Although referred to herein as parity bits, the parity inputs and outputs have no special
functionality and can be used as additional data bits. For example, the parity bits could be
used to hold additional information about a data word, tagging the data as code or data,
positive or negative values, old or new data, etc.
Block RAM does not contain any special circuitry for generating or checking parity. These
functions, if required by the application, are created using CLB logic resources.

Data Input Parity Bus — DIP[#:0] (DIPA[#:0], DIPB[#:0])

Data at the DIP input bus is written to the RAM location specified by the address input
bus, ADDR, during a Low-to-High transition on the CLK input, when the clock enable EN
and write enable WE inputs are High.

Data Output Parity Bus — DOP[#:0] (DOPA[#:0], DOPB[#:0])

The data output bus, DOP, presents the contents of memory cells referenced by the address
bus, ADDR, at the active clock edge during a read operation. During a simultaneous write
operation, the behavior of the data output latches is controlled by the WRITE_MODE
attribute (see “Read Behavior During Simultaneous Write — WRITE_MODE,” page 173).

162 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Signals

Parity Data Address

35 34 33 32 31 24 23 16 15 8 7 0

512x36 P3 P2 P1 P0 Byte 3 Byte 2 Byte 1 Byte 0 0

17 16 15 8 7 0
P3 P2 Byte 3 Byte 2 1
1Kx18
P1 P0 Byte 1 Byte 0 0
Pa
r
(16 ity O
2K bits ption
K 8 7 0
bits d a
pa ata, l P3 Byte 3 3
rity
) P2 Byte 2 2
2Kx9
P1 Byte 1 1
P0 Byte 0 0
3 2 1 0
7 6 5 34 7
y te1 0
3B 2 6
4Kx4
7 6 5 04 1
te
3B2y 1 0 0

1 0
7 6 F

Byte 3
5 4 E
3 2 D
(16Kbits data) 1 0 C
8Kx2
No Parity

7 6 3

Byte 0
5 4 2
3 2 1
1 0 0

0
7 1F
Byte 3

6 1E
5 1D
4 1C
16Kx1
3 3
Byte 0

2 2
1 1
0 0
X463_04_062503

Figure 4-4: Data Organization and Mapping Between Modes

Spartan-3 Generation FPGA User Guide www.xilinx.com 163

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Address Input
As dual-port RAM, both ports operate independently while accessing the same set of
18 Kbit memory cells.
Note: Whenever a block RAM port is enabled (ENA or ENB = High), all address transitions must
meet the data sheet setup and hold times with respect to the port clock (CLKA or CLKB). This
requirement must be met even if the RAM read output is of no interest, or WE is deasserted, including
ROM mode. There are some instances in which these requirements might not be able to be met; for
instance, if there is a multi-cycle path on the address input signals, or while the clock is stabilizing.
Work around this by disabling the port via ENA/ENB during the time that the address inputs do not
meet setup and hold requirements. Deasserting ENA/ENB disables the port so that violating the
address input setup and hold requirements does not affect block RAM contents. Assert ENA/ENB
again when resuming normal read/write functionality.

Address Bus — ADDR[#:0] (ADDRA[#:0], ADDRB[#:0])

The address bus selects the memory cells for read or write operations. The memory depth
determines the required address bus width, as shown in Table 4-8.

Control Inputs
Clock — CLK (CLKA, CLKB)
Each port is fully synchronous with independent clock pins. All port input pins have setup
time referenced to the port CLK pin. The data bus has a clock-to-out time referenced to the
CLK pin. Clock polarity is configurable and is rising edge triggered by default.
With default polarity, a Low-to-High transition on the clock (CLK) input controls read,
write, and reset operations.

Enable — EN (ENA, ENB)

The enable input, EN, controls read, write, and set/reset operations. When EN is Low, no
data is written and the outputs DO and DOP retain the last state. The polarity of EN is
configurable and is active High by default.
When EN is asserted, minus an active synchronous set/reset input or write-enable input,
block RAM always reads the memory location specified by the address bus, ADDR, at the
rising clock edge.

Write Enable — WE (WEA, WEB)

The write enable input, WE, controls when data is written to RAM. When both EN and WE
are asserted at the rising clock edge, the value on the data and parity input buses is written
to memory location selected by the address bus.
The data output latches are loaded or not loaded according to the WRITE_MODE attribute.
The polarity of WE is configurable and is active High by default.
All Spartan-3 generation FPGAs support the RAMB16 block RAM primitive that has a
single write-enable input that controls write operations regardless of the data width for the
configured data organization. See Figure 4-4, page 163 for a diagram of all supported data
organizations. Table 4-5, page 165 shows the write-enable behavior for the RAMB16
primitive.

164 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Signals

Spartan-3A/3AN FPGAs introduce a new block RAM primitive called RAMB16BWE,

essentially a RAMB16 primitive with four independent byte-level write enable inputs. The
Spartan-3A DSP FPGA primitive RAMB16BWER has the same byte-level write enable
function. As shown in Table 4-6, page 165, the independent write-enable inputs allow an
application to write an individual byte or select bytes from a multi-byte data word without
affecting the unselected RAM locations. This feature is useful for a variety of applications,
especially MicroBlaze processor designs. For 1Kx18 data organizations, connect WE0 with
WE2 to select the lower 9 bits and connect WE1 with WE3 to select the upper 9 bits.

Table 4-5: RAMB16 Write Operations (All Spartan-3 Generation FPGAs)

Data
EN WE CLK
Organization Function
0 X X Block RAM disabled. No operation.
1 0 Ç Block RAM enabled but no write operation.
All
As appropriate for the block RAM data
(See Figure 4-4)
organization, write data from the DI and DIP
1 1 Ç
input ports to the currently addressed RAM
location.

Table 4-6: RAMB16BWE/R Write Operations (Extended Spartan-3A Family FPGAs Only)

Data Byte-level Write Enables

EN CLK
Organization WE3 WE2 WE1 WE0 Function
All 0 X X X X X Block RAM disabled. No operation.
16Kx1 0 Ç Block RAM enabled but no write operation.
8Kx2
1 Write data from the DI and DIP input ports to the
4Kx4 1 Ç
2Kx9 currently addressed RAM location.

Write 18 bits: Write data from the DI[15:0] and DIP[1:0]

1 1 Ç
input ports to the currently addressed RAM location.
Write lower 9 bits: Write data only from the DI[7:0]
Same Same
0 1 and DIP[0] input ports to the currently addressed
1Kx18 1 as as
RAM location. Other bits in RAM location unaffected.
WE1 WE0
Write upper 9 bits: Write data only from the DI[15:8]
1 0 and DIP[1] input ports to the currently addressed
RAM location. Other bits in RAM location unaffected.

Spartan-3 Generation FPGA User Guide www.xilinx.com 165

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Table 4-6: RAMB16BWE/R Write Operations (Extended Spartan-3A Family FPGAs Only) (Cont’d)

Data Byte-level Write Enables

EN CLK
Organization WE3 WE2 WE1 WE0 Function
Write 36 bits: Write data from the DI[31:0] and DIP[3:0]
1 1 1 1 Ç
input ports to the currently addressed RAM location.
Write lowest 9 bits: Write data only from the DI[7:0]
0 0 0 1 Ç and DIP[0] input ports to the currently addressed
RAM location. Other bits in RAM location unaffected.
Write next 9 bits: Write data only from the DI[15:8] and
0 0 1 0 Ç DIP[1] input ports to the currently addressed RAM
512x36 1
location. Other bits in RAM location unaffected.
Write lower 18 bits: Write data from the DI[15:0] and
0 0 1 1 Ç DIP[1:0] input ports to the currently addressed RAM
location. Other bits in RAM location unaffected.
Write upper 18 bits: Write data from the DI[31:16] and
1 1 0 0 Ç DIP[3:2] input ports to the currently addressed RAM
location. Other bits in RAM location unaffected.

Output Register Enable - REGCE (REGCEA, REGCEB) Spartan-3A DSP

FPGA Only
The Output Register Write enable input, REGCE, controls when data is written to the RAM
Output registers. When both EN and REGCE are asserted at the rising clock edge, the value
on the output of the block RAM is written to the block RAM output register.
The polarity of REGCE is configurable and is active High by default.

Output Latch Synchronous Set/Reset — SSR (SSRA, SSRB)

The synchronous set/reset input, SSR, forces the data output latches to the value specified
by the SRVAL attribute. When SSR and the enable signal, EN, are High, the data output
latches for the DO and DOP outputs are synchronously set to a ‘0’ or ‘1’ according to the
SRVAL parameter.
A Synchronous Set/Reset operation does not affect RAM cells and does not disturb write
operations on the other port.
The polarity of SSR is configurable and is active High by default.
The SSR input is available on the RAMB16 and RAMB16BWE components. The
RAMB16BWER component for the Spartan-3A DSP platform provides the RST input
instead.

Output Latch/Register Synchronous/Asynchronous Set/Reset - RST (RSTA,

RSTB) - Spartan-3A DSP FPGA Only
The Spartan-3A DSP platform block RAM set/reset input is optionally synchronous or
asynchronous and controls both the output latches and the optional output registers. The
control pin for this operation is named RST and is available on the RAMB16BWER
component.

166 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

In synchronous mode, if RST and the enable signal EN are High, the data output latches
and optional output registers for the DO and DOP outputs are synchronously set to a ‘0’ or
‘1’ according to the SRVAL parameter.
In asynchronous mode, if RST and the enable signal EN are High, the data output latches
and optional output registers for the DO and DOP outputs are asynchronously set to a ‘0’
or ‘1’ according to the SRVAL parameter.
The mode is set by setting the RSTTYPE attribute to “SYNC” for synchronous operation or
“ASYNC” for asynchronous operation. The default for RSTTYPE is synchronous. Due to
improved timing and circuit stability, it is recommended to always have this set to "SYNC"
unless an asynchronous reset is absolutely necessary.
A RST operation does not affect block RAM cells and does not disturb write operations on
the other port.
The polarity of RST is configurable and is active High by default.
The RST input is available on the RAMB16BWER component for the Spartan-3A DSP
platform. The RAMB16 and RAMB16BWE components provide the SSR input instead.

Global Set/Reset — GSR

The global set/reset signal, GSR, is asserted automatically and momentarily at the end of
device configuration. By instantiating the STARTUP primitive, the logic application can
also assert GSR to restore the initial FPGA state at any time. The GSR signal initializes the
output latches to the INIT value. A GSR signal has no impact on internal memory contents.
Because GSR is a global signal and automatically connected throughout the device, the
block RAM primitive does not have a GSR input pin.

Inverting Control Pins

For each port, the four control pins—CLK, EN, WE, and SSR/RST—each have an
individual inversion option. Any control signal can be configured as active High or Low,
and the clock can be active on a rising or falling edge without consuming additional logic
resources.

Unused Inputs
Tie any unused data or address inputs to logic ‘1’. Connecting the unused inputs High
saves logic and routing resources compared to connecting the inputs Low.

Attributes
A block RAM has a number of attributes that control its behavior as shown in Table 4-7 for
VHDL and Verilog. The CORE Generator system uses slightly different values, as
described below.

Table 4-7: Block RAM Attributes and VHDL/Verilog Attribute Names

Function VHDL or Verilog Attribute Default Value
Number of Ports Defined by instantiating the N/A
appropriate RAMB16 primitive
Memory Organization Defined by instantiating the N/A
appropriate RAMB16 primitive

Spartan-3 Generation FPGA User Guide www.xilinx.com 167

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Table 4-7: Block RAM Attributes and VHDL/Verilog Attribute Names (Cont’d)
Function VHDL or Verilog Attribute Default Value
Initial Content for Data Memory, INIT_xx Initialized to zero
Loaded during Configuration
Initial Content for Parity Memory, INITP_xx Initialized to zero
Loaded during Configuration
Data Output Latch Initialization INIT (single-port) Initialized to zero
INIT_A, INIT_B (dual-port)
Data Output Latch Synchronous SRVAL (single-port) Reset to zero
Set/Reset Value SRVAL_A, SRVAL_B (dual-port)
Data Output Latch Behavior during WRITE_MODE WRITE_FIRST
Write
Block RAM Location LOC N/A
Reset Type RSTTYPE SYNC
(Spartan-3A DSP FPGA only)

Number of Ports
Although physically dual-port memory, each block RAM performs as either single-port or
dual-port memory. The method to specify the number of ports depends on the design
entry tool.

CORE Generator System

As shown in Figure 4-5, the Xilinx CORE Generator system provides module generators
for various types of memory blocks. Choose single- or dual-port block memories or use the
higher-level functions to create FIFOs, content-addressable memories (CAMs), and so
forth.

X463_05_060606

Figure 4-5: Selecting a Block RAM Function in CORE Generator System

168 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

VHDL or Verilog Instantiation

The Xilinx design libraries contain single- and dual-port memory primitives similar to
those shown in Figure 4-1. Select among the various primitives to choose single- or dual-
port memory, as well as the memory organization or aspect ratio of the memory. See
Table 4-8 and Table 4-9 for single-port and dual-port block RAM primitives, respectively.

Memory Organization/Aspect Ratio

The data organization or aspect ratio of a RAM block is configurable, as shown in
Table 4-8. If the data path is byte-wide or wider, then the block RAM also provides
additional bits to support parity for each byte. Consequently, a 1Kx18 memory
organization is 18 bits wide with 16 bits (two bytes) allocated to data plus two parity bits,
one for each byte. Also, the physical amount of memory accessible from a port depends on
the memory organization. For memories byte-wide and wider, there are 18K memory bits
accessible. For narrower memories, only 16K bits are accessible due to the lack of parity
bits in these organizations. Essentially, 16K bits are allocated to data, 2K bits to parity on
the 18 Kbit block RAM. See Figure 4-4 for details on data mapping for and between each
memory organization.

Table 4-8: Block RAM Data Organizations/Aspect Ratios

Memory Data Parity Single-Port Total RAM
Organization Depth Width Width DI/DO DIP/DOP ADDR Primitive Kbits
512x36 512 32 4 (31:0) (3:0) (8:0) RAMB16_S36 18K
1Kx18 1024 16 2 (15:0) (1:0) (9:0) RAMB16_S18 18K
2Kx9 2048 8 1 (7:0) (0:0) (10:0) RAMB16_S9 18K
4Kx4 4096 4 - (3:0) - (11:0) RAMB16_S4 16K
8Kx2 8192 2 - (1:0) - (12:0) RAMB16_S2 16K
16Kx1 16384 1 - (0:0) - (13:0) RAMB16_S1 16K

CORE Generator System — Memory Size

The CORE Generator system creates a wide variety of memories with very flexible aspect
ratios. Unlike the actual block RAM primitive, the CORE generator system does not
differentiate between data and parity bits and considers all bits data bits. For dual-port
memories, each port can have different organizations or aspect ratios.
Within the CORE Generator system, locate the Memory Size group and enter the desired
memory organization, as shown in Figure 4-6.

Memory Size

Width 16 Valid Range 1..256

Depth 256 Valid Range 2..16384

Figure 4-6: Selecting Memory Width and Depth in CORE Generator System

Spartan-3 Generation FPGA User Guide www.xilinx.com 169

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

VHDL or Verilog Instantiation

The aspect ratio is defined at design time by specifying or instantiating the appropriate
SelectRAM component. Table 4-8 indicates the SelectRAM component for single-port
RAM. For single-port RAM, the proper component name is RAMB16_Sn, where n is the
data path width including both the data bits plus parity bits. For example, a 1Kx18 single-
port RAM uses component RAMB16_S18. In this example, n=18 because there are 16 data
bits plus 2 parity bits.
Selecting a dual-port memory is slightly more complex because the two memory ports can
have different aspect ratios. For dual-port RAM, the proper component name is
RAMB16_Sm_Sn, where m is the data path width for Port A and n is the width for Port B.
For example, using the suffix shown in Table 4-9, if Port A is organized a 2Kx9 and Port B
is organized as 1Kx18, then the proper dual-port RAM component is RAMB16_S9_S18. In
this example, m=9 and n=18.

Table 4-9: Dual-Port RAM Component Suffix Appended to “RAMB16”

Port A
16Kx1 8Kx2 4Kx4 2Kx9 1Kx18 512x36
16Kx1 _S1_S1
8Kx2 _S1_S2 _S2_S2
Port B

4Kx4 _S1_S4 _S2_S4 _S4_S4

2Kx9 _S1_S9 _S2_S9 _S4_S9 _S9_S9
1Kx18 _S1_S18 _S2_S18 _S4_S18 _S9_S18 _S18_S18
512x36 _S1_S36 _S2_S36 _S4_S36 _S9_S36 _S18_S36 _S36_S36

Address and Data Mapping Between Two Ports

In dual-port mode, both ports access the same set of memory cells. However, both ports
can have the same or different memory organization or aspect ratio. Figure 4-4 shows how
the same data set might appear with different aspect ratios.
There are extra bits available to store parity for memory organizations that are byte-wide
or wider. The extra parity bits are designed to be associated with a particular byte and
these parity bits appear as the more-significant bits on the data port. For example, if a x36
data word (32 data, 4 parity) is addressed as two x18 halfwords (16 data, 2 parity), the
parity bits associated with each data byte are mapped within the block RAM to
appropriate parity bits. The same effect happens when the x36 data word is mapped as
four x9 words. The extra parity bits are not available if the data port is configured as x4, x2,
or x1.
The following formulas provide the starting and ending address for data when the two
ports have different memory organizations. Find the starting and ending addresses for
Port X given the address and port width of Port Y and the port width of Port X.

ADDRESS Y • WIDTH Y
START_ADDRESS X = INTEGER ⎛ ------------------------------------------------------------⎞
⎝ WIDTH X ⎠

( ( ADDRESS Y + 1 ) • WIDTH Y ) – 1
END_ADDRESS X = INTEGER ⎛ ----------------------------------------------------------------------------------------⎞
⎝ WIDTH X ⎠

170 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

If, due the memory organization, one port includes parity bits and the other does not, then
the above equations are invalid and the values for width should only include the data bits.
The parity bits are not available on any port that is less than 8 bits wide.

Content Initialization
By default, block RAM is initialized with all zeros during the device configuration
sequence. However, the contents can also be initialized with user-defined data.
Furthermore, the RAM contents are protected against spurious writes during
configuration.

CORE Generator System — Load Init File

To specify the initial RAM contents for a CORE Generator block RAM function, create a
coefficients (.coe) file. A simple example of a coefficients file appears in Figure 4-7. At a
minimum, define the radix for the initialization data—i.e., base 2, 10, or 16—and then
specify the RAM contents starting with the data at location 0, followed by data at
subsequent locations.

memory_initialization_radix=16;
memory_initialization_vector= 80, 0F, 00, 0B, 00, 0C, …, 81;

Figure 4-7: A Simple Coefficients File (.coe) Example

To include the coefficients file, locate the appropriate section in the CORE Generator
wizard and check Load Init File, as shown in Figure 4-8. Then, click Load File and select
the coefficients file.

Initial Contents
Global Init Value:

ff0001 (Hex Value)

Load Init File

Load File ...

C:\MyProject\my_ram_init.coe (.coe File)

Figure 4-8: Specifying Initial RAM Contents in CORE Generator System

VHDL or Verilog Instantiation — INIT_xx, INITP_xx

For VHDL and Verilog instantiation, there are two different types of initialization
attributes. The INIT_xx attributes define the initial contents of the data memory locations.
The INITP_xx attributes define the initial contents of the parity memory locations.
The INIT_xx attributes on the instantiated primitive define the initial memory contents.
There are 64 initialization attributes, named INIT_00 through INIT_3F. Each INIT_xx
attribute is a 64-digit (256-bit) hex-encoded bit vector. The memory contents can be
partially initialized and any unspecified locations are automatically completed with zeros.
The following formula defines the bit positions for each INIT_xx attribute.
Given yy = convert_hex_to_decimal(xx), INIT_xx corresponds to the following memory
cells.
• Starting Location: [(yy + 1) * 256] –1

Spartan-3 Generation FPGA User Guide www.xilinx.com 171

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

• End Location: (yy) * 256

For example, for the attribute INIT_1F, the conversion is as follows:
• yy = convert_hex_to_decimal(0x1F) = 31
• Starting Location: [(31+1) * 256] –1 = 8191
• End Location: 31 * 256 = 7936
Table 4-10: VHDL/Verilog RAM Initialization Attributes for Block RAM
Attribute From To
INIT_00 255 0
INIT_01 511 256
INIT_02 767 512
… … …
INIT_3F 16383 16128

The INITP_xx attributes define the initial contents of the memory cells corresponding to
parity bits, i.e., those bits that connect to the DIP/DOP buses. By default these memory
cells are also initialized to all zeros.
The eight initialization attributes from INITP_00 through INITP_07 represent the
memory contents of parity bits. Each INITP_xx is a 64-digit (256-bit) hex-encoded bit
vector and behaves like an INIT_xx attribute. The same formula calculates the bit
positions initialized by a particular INITP_xx attribute.

Data Output Latch Initialization

The block RAM output latches can be initialized to a user-specified value immediately
after configuration or whenever the global set/reset signal, GSR, is asserted. For dual-port
memories, there is a separate initialization value for each port.
If no value is specified, the output latch is initialized to zero.

CORE Generator System — Global Init Value

Figure 4-9 describes how to specify the initial value for data output latches in the CORE
Generator system. The value, specified in hexadecimal, should include one bit per the
specified data width. For dual-port memories, there is a separate initialization value for
each port.

Initial Contents
Global Init Value:

ff0001 (Hex Value)

Load Init File

Load File ...

C:\MyProject\my_ram_init.coe (.coe File)

Figure 4-9: Specifying Initial Value for Block RAM Data Output Latches

172 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

VHDL or Verilog Instantiation — INIT (INIT_A and INIT_B)

For VHDL or Verilog, the INIT attribute (or INIT_A and INIT_B for dual-port memories)
defines the output latch value after configuration. The INIT (or INIT_A and INIT_B)
attribute specifies the initial value for the data and, if applicable, the parity bits. Figure 4-4
shows the expected bit format for each memory organization with parity bits—if
applicable—as the more significant bits followed by the data bits. For example, the
initialization value for a 2Kx9 memory would be nine bits wide and would include one
parity bit followed by eight data bits. These attributes are hex-encoded bit vectors and the
default value is 0.

Data Output Latch Synchronous Set/Reset Value

When the synchronous set/reset input, SSR (RST for the RAMB16BWER), is asserted, the
data output latches are set or reset according to the set/reset value attribute. For dual-port
memories, there is a separate initialization value for each port.
If no value is specified, the output latch is reset to zero during a valid Synchronous
Set/Reset operation.
For the RAMB16BWER, the optional output register is also set or reset with the output
latch.

CORE Generator System — Init Value (SINIT)

Figure 4-10 describes how to specify the synchronous set/reset value for data output
latches in the CORE Generator system. Check the SINIT pin and then specify the
synchronous set/reset value in hexadecimal, with one bit per the specified data width. For
dual-port memories, there is a separate value for each port.

Output Register Options

Additional Output Pipe Stages 0

SINIT pin (sync. reset of output registers)

Init Value (Hex) a5a50f

Figure 4-10: Specifying the Output Data Latch Set/Reset Value

VHDL or Verilog Instantiation — SRVAL (SRVAL_A and SRVAL_B)

For VHDL or Verilog, the SRVAL attribute (or SRVAL_A and SRVAL_B for dual-port
memories) defines the output latch value after configuration. The SRVAL (or SRVAL_A
and SRVAL_B) attribute specifies the initial value for the data and, if applicable, the parity
bits. Figure 4-4 shows the expected bit format for each memory organization with parity
bits—if applicable—as the more significant bits followed by the data bits. These attributes
are hex-encoded bit vectors, and the default value is 0.

Read Behavior During Simultaneous Write — WRITE_MODE

To maximize data throughput and utilization of the dual-port memory at each clock edge,
block RAM supports one of three write modes for each memory port. These different
modes determine which data is available on the output latches after a valid write clock
edge to the same port. The default mode, WRITE_FIRST, provides backwards
compatibility with the older Virtex, Virtex-E, and Spartan-IIE FPGA architectures and is

Spartan-3 Generation FPGA User Guide www.xilinx.com 173

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

also the default behavior for Virtex-II and Virtex-II Pro devices. However, READ_FIRST
mode is the most useful as it increases the efficiency of block RAM at each clock cycle,
allowing designs to use maximum bandwidth. In READ_FIRST mode, a memory port
supports simultaneous read and write operations to the same address on the same clock
edge, free of any timing complications.
Table 4-11 outlines how the WRITE_MODE setting affects the output data latches on the
same port, and how it affects the output latches on the opposite port during a
simultaneous access to the same address.

Table 4-11: WRITE_MODE Affects Data Output Latches During Write Operations
Effect on Opposite Port
Write Mode Effect on Same Port (Dual-Port Mode Only, Same Address)
WRITE_FIRST Data on DI, DIP inputs written into specified RAM Invalidates data on DO, DOP outputs.
Read After Write location and simultaneously appears on DO, DOP
(Default) outputs.

READ_FIRST Data from specified RAM location appears on DO, Data from specified RAM location appears
Read Before Write DOP outputs. on DO, DOP outputs.
(Recommended) Data on DI, DIP inputs written into specified
location.
NO_CHANGE Data on DO, DOP outputs remains unchanged. Invalidates data on DO, DOP outputs.
No Read on Write Data on DI, DIP inputs written into specified
location.

Mode selection is set by configuration. One of these three modes is set individually for
each port by an attribute. The default mode is WRITE_FIRST.

WRITE_FIRST or Transparent Mode (Default)

The WRITE_FIRST mode is the default operating mode for backward compatibility
reasons. For new designs, READ_FIRST mode is recommended.
In this mode, the input data is written into the addressed RAM location memory and
simultaneously stored in the data output latches, resulting in a transparent write
operation, as shown in Figure 4-11. The WRITE_FIRST mode provides backwards
compatibility with the 4 Kbit block RAMs on Virtex/Virtex-E and Spartan-II/Spartan-IIE
FPGAs and is also the default mode for Virtex-II/Virtex-II Pro FPGA block RAMs.

Data_in Data_out
WE
EN
CLK
Address RAM Location

WRITE_MODE = WRITE_FIRST X463_11_062503

Figure 4-11: Data Flow during a WRITE_FIRST Write Operation

174 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

Figure 4-12 demonstrates that a valid write operation during a valid read operation results
in the write data appearing on the data output.

CLK

WE

Data_in XXXX 1111 2222 XXXX

Address aa bb cc dd

Data_out 0000 MEM(aa) 1111 2222 MEM(dd)

ENABLE

DISABLED READ WRITE WRITE READ

MEM(bb)=1111 MEM(cc)=2222

X463_12_020503

Figure 4-12: WRITE_FIRST Mode Waveforms

READ_FIRST or Read-Before-Write Mode

In READ_FIRST mode, data previously stored at the write address appears on the output
latches, while the new input data is stored in memory, resulting in a read-before-write
operation shown in Figure 4-13. The older RAM data appears on the data output while the
new RAM data is stored in the specified RAM location. READ_FIRST mode is the
recommended operating mode.

Data_in Data_out
WE
EN
CLK
Address RAM Location

WRITE_MODE = READ_FIRST X463_13_062503

Figure 4-13: Data Flow during a READ_FIRST Write Operation

Figure 4-14 demonstrates that the older RAM data always appears on the data output,
regardless of a simultaneous write operation.

Spartan-3 Generation FPGA User Guide www.xilinx.com 175

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

CLK

WE

Data_in XXXX 1111 2222 XXXX

Address aa bb cc dd

Data_out 0000 MEM(aa) old MEM(bb) old MEM(cc) MEM(dd)

ENABLE

DISABLED READ WRITE WRITE READ

MEM(bb)=1111 MEM(cc)=2222

X463_14_020503

Figure 4-14: READ_FIRST Mode Waveforms

This mode is particularly useful for building circular buffers and large, block-RAM-based
shift registers. Similarly, this mode is useful when storing FIR filter taps in digital signal
processing applications. Old data is copied out from RAM while new data is written into
RAM.

NO_CHANGE Mode
In NO_CHANGE mode, the output latches are disabled and remain unchanged during a
simultaneous write operation, as shown in Figure 4-15. This behavior mimics that of
simple synchronous memory where a memory location is either read or written during a
clock cycle, but not both.

Data_in Data_out
WE
EN
CLK
Address RAM Location

WRITE_MODE = NO_CHANGE X463_15_062503

Figure 4-15: Data Flow during a NO_CHANGE Write Operation

The NO_CHANGE mode is useful in a variety of applications, including those where the
block RAM contains waveforms, function tables, coefficients, and so forth. The memory
can be updated without affecting the memory output.
Figure 4-16 shows that the data output retains the last read data if there is a simultaneous
write operation on the same port.

176 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

CLK

WE

Data_in XXXX 1111 2222 XXXX

Address aa bb cc dd

Data_out 0000 MEM(aa) MEM(dd)

ENABLE

DISABLED READ WRITE WRITE READ

MEM(bb)=1111 MEM(cc)=2222

X463_16_020503

Figure 4-16: NO_CHANGE Mode Waveforms

CORE Generator System — Write Mode

To specify the WRITE_MODE in the CORE Generator system, locate the settings for Write
Mode as shown in Figure 4-17. Select between Read After Write (WRITE_FIRST), Read
Before Write (READ_FIRST) or No Read On Write (NO_CHANGE).

Write Mode
Read After Write Read Before Write No Read On Write

Figure 4-17: Selecting the Write Mode in CORE Generator System

VHDL or Verilog Instantiation — WRITE_MODE

When instantiating block RAM, specify the write mode via the WRITE_MODE attribute.
Acceptable values include WRITE_FIRST, READ_FIRST, and NO_CHANGE, as
demonstrated in the examples in the appendices.

Location Constraints (LOC)

In general, it is best to allow the Xilinx ISE® software to assign a block RAM location.
However, block RAMs can be constrained to specific locations on a Spartan-3 device using
an attached LOC property. Block RAM placement locations are device-specific and differ
from the convention used for naming CLB locations, allowing LOC properties to transfer
easily from array to array.
The LOC properties use the following form:
LOC = RAMB16_X#Y#
The RAMB16_X0Y0 is the lower-left block RAM location on the device, as shown in
Figure 4-18. The upper-right block RAM location depends on n, the number of block RAM
columns, and m, the number of block RAM rows, as provided in Table 4-1, page 155. The
Spartan-3A DSP platform has four or five columns of block RAM, similar to the XC3S4000
and XC3S5000 devices.

Spartan-3 Generation FPGA User Guide www.xilinx.com 177

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Upper Upper
Left Right
RAMB16_X0Y(m-1) RAMB16_X(n-1)Y(m-1)

XC3S50/A/AN
XC3S100E

n = total columns
m = total rows

XC3S200/A/AN XC3S4000
XC3S400/A/AN XC3S5000
XC3S700A/AN XC3SD1800A
XC3S1400A/AN XC3SD3400A
XC3S250E
XC3S500E
RAMB16_X0Y0 XC3S1200E
Lower Lower XC3S1600E
Left Right XC3S1000
XC3S1500
XC3S2000
RAMB16_X(n-1)Y0 UG331_c4_13_033007

Figure 4-18: Block RAM LOC Coordinates

Location attributes cannot be specified directly in the CORE Generator system. However,
location constraints can be added to VHDL or Verilog instantiations.

Block RAM Operation

Table 4-12 describes the behavior of block RAM and assumes that all control signals use
their default, active-High behavior. However, the control signals can be inverted in the
design if necessary. The table and following text describe the behavior for a single memory
port. In dual-port mode, both ports perform as independent single-port memories.
All read and write operations to block RAM are synchronous. All inputs have a set-up time
relative to clock and all outputs have a clock-to-output time.

Table 4-12: Block RAM Function Table

Input Signals Output Signals RAM Contents
GSR EN SSR/RST WE CLK ADDR DIP DI DOP DO Parity Data
Immediately After Configuration
Loaded During Configuration X X INITP_xx2 INIT_xx2
Global Set/Reset Immediately after Configuration
1 X X X X X X X INIT3 INIT No Chg No Chg
RAM Disabled
0 0 X X X X X X No Chg No Chg No Chg No Chg
Synchronous Set/Reset
0 1 1 0 Ç X X X SRVAL4 SRVAL No Chg No Chg

178 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Operation

Table 4-12: Block RAM Function Table (Cont’d)

Input Signals Output Signals RAM Contents
GSR EN SSR/RST WE CLK ADDR DIP DI DOP DO Parity Data
Synchronous Set/Reset during Write RAM
0 1 1 1 Ç addr pdata Data SRVAL SRVAL RAM(addr) RAM(addr)
Åpdata Å data
Read RAM, no Write Operation
0 1 0 0 Ç addr X X RAM(pdata) RAM(data) No Chg No Chg
Write RAM, Simultaneous Read Operation
0 1 0 1 Ç addr pdata Data WRITE_MODE = WRITE_FIRST5 (default)
pdata data RAM(addr) RAM(addr)
Åpdata Å data
WRITE_MODE = READ_FIRST6 (recommended)
RAM(data) RAM(data) RAM(addr) RAM(addr)
Åpdata Åpdata
WRITE_MODE = NO_CHANGE7
No Chg No Chg RAM(addr) RAM(addr)
Åpdata Åpdata

Notes:
1. No Chg = No Change, addr = address to RAM, data = RAM data, pdata = RAM parity data.
2. Refer to “Content Initialization,” page 171.
3. Refer to “Data Output Latch Initialization,” page 172.
4. Refer to “Data Output Latch Synchronous Set/Reset Value,” page 173.
5. Refer to “WRITE_FIRST or Transparent Mode (Default),” page 174.
6. Refer to “READ_FIRST or Read-Before-Write Mode,” page 175.
7. Refer to “NO_CHANGE Mode,” page 176.

RAM Contents Initialized During Configuration

The initial RAM contents, if specified, are loaded during the Spartan-3 FPGA configuration
process. If no contents are specified, the RAM cells are loaded with zero. The RAM
contents are protected against spurious writes during configuration.

Global Set/Reset Initializes Data Output Latches Immediately After

Configuration or Global Reset
Immediately following configuration, the Spartan-3 device begins its start-up procedure
and asserts the global set/reset signal, GSR, to initialize the state of all flip-flops and
registers. The initial contents of the block RAM output latches, INIT, are asynchronously
loaded at this time. The GSR signal does not change or re-initialize the RAM contents.

Enable Input Activates or Disables RAM

If the block RAM is disabled—i.e., EN is Low—then the block RAM retains its present
state. The enable input must be High for any other operations to proceed.

Spartan-3 Generation FPGA User Guide www.xilinx.com 179

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Synchronous Set/Reset Initializes Data Output Latches

If the block RAM is enabled (EN is High) and the Synchronous Set/Reset signal is asserted
High, then the data output latches are initialized at the next rising clock edge. The SRVAL
attribute defines the synchronous set/reset state for the data output latches. This operation
is different the operation caused by the global set/reset signal, GSR, immediately after
configuration. The synchronous set/reset input affects the specific RAM block whereas the
GSR signal affects the entire device.

Simultaneous Write and Synchronous Set/Reset Operations

If a simultaneous write operation occurs during the synchronous set/reset operation, then
the data on the DI and DIP inputs is stored at the RAM location specified by the ADDR
input. However, the data output latches are initialized to the SRVAL attribute value as
described immediately above.

Read Operations Occur on Every Clock Edge When Enable is Asserted

Read operations are synchronous and require a clock edge and an asserted clock enable.
The data output behavior depends on whether or not a simultaneous write operation
occurs during the read cycle.
If no simultaneous write cycle occurs during a valid read cycle, then the read address is
registered on the read port and the data stored in RAM at that address is simply loaded
into the output latches after the RAM access interval passes.
However, if there is a simultaneous write cycle during the read cycle, then the output
behavior depends on which of the three write modes is selected, as described immediately
below.

Write Operations Always Have Simultaneous Read Operation, Data

Output Latches Affected
During a Write operation, a simultaneous Read operation occurs. The WRITE_MODE
attribute determines the behavior of the data output latches during the Write operation
(refer to “Read Behavior During Simultaneous Write — WRITE_MODE,” page 173). By
default, WRITE_MODE is WRITE_FIRST and the data output latches and the addressed
RAM locations are updated with the input data during a simultaneous Write operation.
When WRITE_MODE is READ_FIRST, the output latches are updated with the data
previously stored in the addressed RAM location and the new data on the DI and DIP
inputs is stored at the address RAM location. When WRITE_MODE is NO_CHANGE, the
data output latches are unaffected by a simultaneous Write operation and retain their
present state.

General Characteristics
• A write operation requires only one clock edge.
• A read operation requires only one clock edge.
• All inputs are registered with the port clock and have a setup-to-clock timing
specification.
• All outputs have a read-through function or one of three read-during-write functions,
depending on the state of the WE pin. The outputs relative to the port clock are
available after the clock-to-out timing interval.

180 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Dual-Port RAM Conflicts and Resolution

• Block RAM cells are true synchronous RAMs and do not have a combinatorial path
from the address to the output.
• The ports are completely independent of each other without arbitration. Each port has
its own clocking, control, address, read/write functions, initialization, and data
width.
• Output ports are latched with a self-timed circuit, guaranteeing glitch-free read
operations. The state of the output port does not change until the port executes
another read or write operation.

Functional Compatibility with Other Xilinx FPGA Families

The block RAM on Spartan-3 generation FPGAs is functionally identical to block RAM on
the Xilinx Virtex-II/Virtex-II Pro FPGA families. Consequently, design tools that support
Virtex-II and Virtex-II Pro FPGA block RAM also support with Spartan-3 generation
FPGAs.
Extended Spartan-3A family FPGAs, while remaining fully backwards compatible with
Spartan-3/3E FPGAs, also add byte-level write enable controls, similar to those found on
Virtex-4 FPGAs. The Spartan-3A DSP FPGAs also include a block RAM output register
similar to those found in the Virtex-4 FPGAs.

Dual-Port RAM Conflicts and Resolution

As a dual-port RAM, the block RAM allows both ports to simultaneously access the same
memory cell. Potentially, conflicts arise under the following conditions:
1. If the clock inputs to the two ports are asynchronous, then conflicts occur if clock-to-
clock setup time requirements are violated.
2. Both memory ports write different data to the same RAM location during a valid write
cycle.
3. If a port uses WRITE_MODE=NO_CHANGE or WRITE_FIRST, a write to the port
invalidates the read data output latches on the opposite port.
If Port A and Port B different memory organizations and consequently different widths,
only the overlapping bits are invalid when conflicts occur.

Timing Violation Conflicts

When one port writes to a given memory cell, the other port must not address that memory
cell—either for a write or a read operation—within the clock-to-clock setup window,
which is equivalent to the block RAM minimum clock period (TBPWH + TBPWL), specified
in the Spartan-3 generation FPGA family data sheets. Figure 4-19 describes this situation
where both ports operate from asynchronous clock inputs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 181

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

B A B A

READ Port
CLK_A

Address_A aa bb

WE_A
Data_out_A UNKNOWN 4444

CLK_B WRITE Port

Data_in_B 3333 4444 2222

Address_B aa bb cc

WE_B MEM(aa)= MEM(bb)= MEM(cc)=

3333 4444 2222

Clock-to-clock
setup violation X463_19_020503

Figure 4-19: Clock-to-Clock Timing Conflicts

The first rising edge on CLK_A violates the clock-to-clock setup parameter, because it
occurs too soon after the last CLK_B clock edge. The write operation on port B is valid
because Data_in_B, Address_B, and WE_B all had sufficient setup time before the rising
edge on CLK_B. Unfortunately, the read operation on port A is invalid because it depends
on the RAM contents being written to Address_B and the read clock, CLK_A, happened
too soon after the write clock, CLK_B.
On the second rising edge of CLK_B, there is another valid write operation to port B. The
memory location at address (bb) contains 4444. Data on the Data_out_A port is still invalid
because there has not been another rising clock edge on CLK_A. The second rising edge of
CLK_A reads the new data at location (bb), which now contains 4444. This time, the read
operation is valid because there has been sufficient setup time between CLK_B and
CLK_A.

Simultaneous Writes to Both Ports with Different Data Conflicts

If both ports write simultaneously into the same memory cell with different data, then the
data stored in that cell becomes invalid, as outlined in Table 4-13.

Table 4-13: RAM Conflicts During Simultaneous Writes to Same Address

Input Signals
Port A Port B RAM Contents
WEA CLKB DIPA DIA WEB CLKA DIPB DIB Parity Data
1 Ç DIPA DIA 1 · DIPB DIB ? ?

Notes:
1. ADDRA=ADDRB, ENA=1, ENB=1, DIPA ≠ DIPB, DIA ≠ DIB, ?=Unknown or invalid data.

182 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Design Entry

Write Mode Conflicts on Output Latches

Potential conflicts occur when one port writes to memory and the opposite port reads from
memory. Write operations always succeed, and the write port’s output data latches behave
as described by the port’s WRITE_MODE attribute. If the write port is configured with
WRITE_MODE set to NO_CHANGE or WRITE_FIRST, then a write operation to the port
invalidates the data output latches on the opposite port, as shown in Table 4-14.
Using the READ_FIRST mode does not cause conflicts on the opposite port.

Table 4-14: Conflicts to Output Latches Based on WRITE_MODE

Input Signals Output Signals
Port A Port B Port A Port B
WEA CLKA DIPA DIA WEB CLKB DIPB DIB DOPA DOA DOPB DOB
WRITE_MODE_A=NO_CHANGE
1 Ç DIPA DIA 0 Ç DIPB DIB No Chg No Chg ? ?
WRITE_MODE_B=NO_CHANGE
0 Ç DIPA DIA 1 Ç DIPB DIB ? ? No Chg No Chg
WRITE_MODE_A=WRITE_FIRST
1 Ç DIPA DIA 0 Ç DIPB DIB DIPA DIA ? ?
WRITE_MODE_B=WRITE_FIRST
0 Ç DIPA DIA 1 Ç DIPB DIB ? ? DIPB DIB
WRITE_MODE_A=WRITE_FIRST, WRITE_MODE_B=WRITE_FIRST
1 Ç DIPA DIA 1 Ç DIPB DIB ? ? ? ?

Notes:
1. ADDRA=ADDRB, ENA=1, ENB=1, ?=Unknown or invalid data

Conflict Resolution
There is no dedicated monitor to arbitrate the result of identical addresses on both ports.
The application must time the two clocks appropriately. However, conflicting
simultaneous writes to the same location never cause any physical damage.

Block RAM Design Entry

Various tools help create Spartan-3 FPGA block RAM designs, two of which are the Xilinx
CORE Generator system and VHDL or Verilog instantiation of the appropriate Xilinx
library primitives.

Xilinx CORE Generator System

The Xilinx CORE Generator system provides both a Single Port Block Memory and a Dual
Port Block Memory module generator, as shown in Figure 4-5. Both module generators
support RAM, ROM, and Write Only functions, according to the control signals that are
selected. Any size memory that can be created in the architecture is supported.

Spartan-3 Generation FPGA User Guide www.xilinx.com 183

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Both modules are parameterizable as with most CORE Generator modules. To create a
module, specify the component name and choose to include or exclude control inputs, and
choose the active polarity for the control inputs. For the Dual-Port Block Memory, once the
organization or aspect ratio for Port A is selected, only the valid options for Port B are
displayed.
Optionally, specify the initial memory contents. Unless otherwise specified, each memory
location initializes to zero. Enter user-specified initial values via a Memory Initialization
File, consisting of one line of binary data for every memory location. A default file is
generated by the CORE Generator system. Alternatively, create a coefficients file (.coe),
which not only defines the initial contents in a radix of 2, 10, or 16, but also defines all the
other control parameters for the CORE Generator system.
The output from the CORE Generator system includes a report on the options selected and
the device resources required. If a very deep memory is generated, some external
multiplexing might be required, and these resources are reported as the number of logic
slices required. In addition, the software reports the number of bits available in block RAM
that are less than 100% utilized. For simulation purposes, the CORE Generator system
creates VHDL or Verilog behavioral models.
• CORE Generator: Single-Port Block Memory module (RAM or ROM)
• CORE Generator: Dual-Port Block Memory module (RAM or ROM)

VHDL and Verilog Instantiation

VHDL and Verilog synthesis-based designs can either infer or directly instantiate block
RAM, depending on the specific logic synthesis tool used to create the design.

Inferring Block RAM

Some VHDL and Verilog logic synthesis tools, such as the Xilinx Synthesis Tool (XST) and
Synplicity Synplify both infer block RAM based on the hardware described. The Xilinx ISE
Project Navigator includes templates for inferring block RAM in your design. To use the
templates within Project Navigator, select Edit Æ Language Templates from the menu,
and then select VHDL or Verilog, followed by Synthesis Templates Æ RAM from the
selection tree. Finally, select the preferred block RAM template.
It is still possible to directly instantiate block RAM, even if portions of the design infer
block RAM.

Instantiation Templates
For VHDL- and Verilog-based designs, various instantiation templates are available to
speed development. Within the Xilinx ISE Project Navigator, select Edit Æ Language
Templates from the menu, and then select VHDL or Verilog, followed by Component
Instantiation Æ Block RAM from the selection tree.
The appendices include example code showing how to instantiate block RAM in both
VHDL and Verilog.
In VHDL, each template has a component declaration section and an architecture section.
Each part of the template must be inserted within the VHDL design file. The port map of
the architecture section must include the signal names used in the application.
The SelectRAM_Ax templates (with x = 1, 2, 4, 9, 18, or 36) are single-port modules and
instantiate the corresponding RAMB16_Sx module.

184 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

SelectRAM_Ax_By templates (with x = 1, 2, 4, 9, 18, or 36 and y = 1, 2, 4, 9, 18, or 36) are

dual-port modules and instantiate the corresponding RAMB16_Sx_Sy module.

Initialization in VHDL or Verilog Codes

Block RAM structures can be initialized in VHDL or Verilog code for both synthesis and
simulation. For synthesis, the attributes are attached to the block RAM instantiation and
are copied within the EDIF output file compiled by Xilinx tools. The VHDL code
simulation uses a generic parameter to pass the attributes. The Verilog code simulation
uses a defparam parameter to pass the attributes.
The VHDL and Verilog examples in the appendices illustrate these techniques.

Block RAM Applications

Typically, block RAM is used for a variety of local storage applications. However, the
following section describes additional, perhaps less obvious block RAM capabilities,
illustrating some powerful capabilities to spur the imagination.

Creating Larger RAM Structures

Block SelectRAM columns have specialized routing to allow cascading blocks with
minimal routing delays. Wider or deeper RAM structures incur a small delay penalty. For
examples of how to create wider block memories, see application note
XAPP229: Wider Block Memories, which includes a reference design.

Block RAM as Read-Only Memory (ROM)

By tying the write enable input Low, block RAM optionally functions as registered block
ROM. The ROM outputs are synchronous and require a clock input and perform exactly
like a block RAM read operation. The ROM contents are defined by the initial contents at
design time.
After design compilation, the ROM contents can also be updated using the Data2BRAM
utility described below.

FIFOs
First-In, First-Out (FIFO) memories, also known as elastic stores, are perhaps the most
common application of block RAM, other than for random data storage. FIFOs typically
resynchronize data, either between two different clock domains, or between two parts of a
system that have different data rates, even though they operate from a single clock. The
Xilinx CORE Generator system provides two parameterizable FIFO modules, one a
synchronous FIFO where both the read and write clocks are synchronous to one another
and the other an asynchronous FIFO where the read and write clocks are different.
Application note XAPP261 demonstrates that the FIFO read and write ports can be
different data widths, integrating the data width converter into the FIFO.
Application note XAPP291 describes a self-addressing FIFO that is useful for throttling
data in a continuous data stream.
• CORE Generator: Synchronous FIFO module
• CORE Generator: Asynchronous FIFO module
• XAPP258: FIFOs Using Block RAM, includes reference design

Spartan-3 Generation FPGA User Guide www.xilinx.com 185

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

• XAPP261: Data-Width Conversion FIFOs Using Block RAM Memory, includes reference
design
• XAPP291: Self-Addressing FIFO

Storage for Embedded Processors

Block RAM also enables efficient embedded processor applications. RAM performs a
variety functions in an embedded processor such as those listed below.
• Register file for processor register set, although for some processors, distributed RAM
might be a preferred solution.
• Stack or LIFO for stack-based architectures and for call stacks.
• Fast, local code storage. The fast access time to internal block RAM significantly
boosts the performance of embedded processors. However, on-chip storage is limited
by the number of available block RAMs.
• Large dual-ported mailbox memory shared with external processor or DSP device.
• Temporary trace buffers (see “Circular Buffers, Shift Registers, and Delay Lines”) to
ease and enhance application debugging.

Updating Block RAM/ROM Content by Directly Modifying Device

Bitstream
In a typical design flow, the initial contents of block RAM/ROM is defined at design time
and compiled into the device bitstream that is downloaded to and configures a Spartan-3
FPGA.
However, for some applications, the actual memory contents might not be known when
the bitstream is created or might change later. One example is if a processor embedded
with the Spartan-3 FPGA uses block RAM to store program code. To avoid recompiling the
FPGA design just to incorporate a code change, Xilinx provides a utility called Data2MEM
that updates an existing FPGA bitstream with new block RAM/ROM contents.
As shown in Figure 4-20, the inputs to Data2MEM include the new RAM contents—
typically the output from the embedded processor compiler/linker, the present FPGA
bitstream, and a file that describes the mapping between the system address space and the
addressing used on the individual block RAMs and the physical location of each block
RAM.

Present FPGA Bitstream

(.bit file)
New Block RAM Data Mapping Between System
(.elf Compiler/Linker output, Address Space and Block
or .mem file) RAM, Block RAM Location
(.bmm file)

Data2MEM
Utility

New FPGA Bitstream with

Updated Block RAM Contents
(.bit file) UG331_c4_14_111006

Figure 4-20: The Data2BRAM Utility Updates Block RAM Contents in a Bitstream

186 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

Two Independent Single-Port RAMs Using One Block RAM

Some applications might require more single-port RAMs than there are RAM blocks on the
device. However, a simple trick allows a single block RAM to behave as if it were two,
completely independent single-port memories, effectively doubling the number of RAM
blocks on the device. The penalty is that each RAM block is only half the size of the original
block, up to 9K bits total.
Figure 4-21 shows how to create two independent single-port RAMs from one block RAM.
Tie the most-significant address bit of one port High and the most-significant address bit of
the other port Low. Both ports evenly split the available RAM between them.

ADDR[m]
ADDR[m -1:0]

Single-Port A
DIA DOA
WEA
ENA
SSRA
CLKA

CLKB
SSRB
Single-Port B

ENB
WEB
DIB DOB
ADDR[n -1:0]
ADDR[n]
X463_21_062503

Figure 4-21: One Block RAM Becomes Two Independent Single-Port RAMs

Both ports are independent, each with its own memory organization, data inputs and
outputs, clock input, and control signals. For example, Port A could be 256x36 while Port
B is 2Kx4.
Figure 4-21 splits the available memory evenly between the two ports. With additional
logic on the upper address lines, the memory can be split into other ratios.

Spartan-3 Generation FPGA User Guide www.xilinx.com 187

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

A 256x72 Single-Port RAM Using One Block RAM

Figure 4-22 illustrates how to create a 256-deep by 72-bit wide single-port RAM using a
single block RAM. As in the previous example, the memory array is split into halves. One
half contains the lower 36 bits, and the upper half stores the upper 36 bits, effectively
creating a 72-bit wide memory.

ADDR[8]
ADDR[7:0] ADDRA[7:0]
DI[71:36] DIA DOA DO[71:36]
WE WEA
ENA ENA
SSR SSRA
CLK CLKA

CLKB
SSRB
ENB
WEB
DI[35:0] DIB DOB DO[35:0]
ADDRB[7:0]
ADDRB[8]
X463_22_062403

Figure 4-22: A 256x72 Single-Port RAM Using a Single Block RAM

The most-significant address line, ADDR[8] is tied High on one port and Low on the other.
Both ports share the same the address inputs, control inputs, and clock input.

188 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

Circular Buffers, Shift Registers, and Delay Lines

Circular buffers are used in a variety of digital signal processing applications, such as finite
impulse response (FIR) filters, multi-channel filtering, plus correlation and cross-
correlation functions. Circular buffers are also useful simply for delaying data to
resynchronize it with other parts of a data path.
Figure 4-23 conceptually describes how a circular buffer operates. Data is written into the
buffer. After n clock cycles, that same data is clocked out of the buffer while new data is
written to the same location.

n -1 n -2

n …

IN 0 3
OUT
1 2
X463_20_020503

Figure 4-23: Circular Buffer

Figure 4-24 describes the hardware implementation to create a circular buffer using block
RAM. A modulo-n counter drives the address inputs to a single-port block RAM. For
simple data delay lines, the block RAM writes new data on every clock cycle.
The circular buffer also reads the delayed data value on every clock edge. Using block
RAM’s READ_FIRST write mode, both the incoming write data and the outgoing read
data use the same clock input and the same clock edge, both simplifying the design and
improving overall performance. The actual write and read behavior is described in
Figure 4-17.

IN DI DO OUT
ADDR
Counter

WE
EN
SSR
CLK

WRITE_MODE=READ_FIRST
X463_24_020503

Figure 4-24: Circular Buffer Implementation Using Block RAM and Counter

In Figure 4-24, the width of the IN and OUT data ports is identical, although they do not
need be. Using dual-port mode, the ports can be different widths. Figure 4-25 shows an
example where byte-wide data enters the block RAM and a 32-bit word exits the block
RAM. Furthermore, the data can be delayed up to 2,048 byte-clock cycles.

Spartan-3 Generation FPGA User Guide www.xilinx.com 189

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Byte 3
Byte 2
Byte 1
Byte 0
IN

Block RAM
Circular Buffer/ Data delayed up to
2,048 clock cycles
Delay Buffer

OUT

Byte 3 Byte 2 Byte 1 Byte 0

X463_25_020503

Figure 4-25: Merge Circular Buffer and Port-Width Converter into a Single Block
RAM

A single block RAM is configured as dual-port memory. The incoming byte-wide data
feeds Port B, which is configured as a 2Kx9 memory. The outgoing 32-bit data appears on
Port A and consequently, Port A is configured as a 512x36 memory.

512x36
ADDRA[8:0]
÷n
EN DOA[31:0] OUT
WEA
ENA

CLKA

CLKB

ENB
WEB
IN
DIB[7:0]
TC
ADDRB[11:2]
÷4 ADDRB[1:0]
2Kx9
X463_26_062503

Figure 4-26: Incoming Byte-Wide Data is Delayed 4n Clock Cycles, Converted to

32-Bit Data

Manipulating the addresses that feeds both ports creates the 4n-byte clock delay. Every
32-bit output word requires four incoming bytes. Consequently, a divide-by-4 counter
feeds the two lower address bits, ADDRB[1:0]. After four bytes are stored, a terminal
count, TC, from the lower counter enables Port A plus a separate divide-by-n counter. The
enable signal latches the 32-bit output data on Port B and increments the upper counter.
The combination of the divide-by-4 counter and the divide-by-n counter effectively create
a divide-by-4n counter. The output from the divide-by-n counter forms the more-
significant address bits to Port B, ADDRB[11:2] and the entire address to Port A,
ADDRA[9:0].

190 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

Fast Complex State Machines and Microsequencers

Because block RAMs can be configured with any set of initial values, they also make
excellent dual-ported registered ROMs that can be used as state machines. For example, a
128-state, 8-way branch finite state machine with 38 total state outputs, fits in a single block
RAM, as shown in Figure 4-27.

Branch Control 7 bits 1Kx9 7+2 bits

State Machine
or Sequencer

7 State Bits
38 Output Bits

256x36
7 bits State Outputs 36 bits
Control
Output

X463_27_062503

Figure 4-27: 128-State Finite State Machine with 38 Outputs in a Single Block RAM

A dual-port block RAM is divided into two completely independent half-size, single-port
memories by tying the most-significant address bit of one port High and the other one
Low, similar to Figure 4-21. Port A is configured as 2Kx9 but used as a 1Kx9 single-port
ROM. Seven outputs feed back as address inputs, stepping through the 128 states. The
1Kx9 ROM has ten total address lines, seven of which are the current-state inputs and the
remaining three address inputs determine the eight-way branch. Any of the 128 states can
conditionally branch to any set of eight new states, under the control of these three address
inputs.
Port B is configured as 512 x 36 and used as a 256 x 36 single-port ROM. It receives the same
7-bit current-state value from Port A, and drives 36 outputs that can be arbitrarily defined
for each state. However, due to the synchronous nature of block ROM, the 36 outputs from
the 256x36 ROM are delayed by one clock cycle. The eighth address input can invoke an
alternate definition of the 36 outputs. Two additional state bits are available from the 1Kx9
block, but are not delayed by one clock.
This same basic architecture can be modified to form a 256-state finite state machine with
four-way branch, or a 64-state state machine with 16-way branch.
If additional branch-control inputs are needed, they can be combined using an input
multiplexer. The advantages of this design are its low cost (a single block RAM), its high
performance (125+ MHz), the absence of layout or routing issues, and complete design
freedom.

Spartan-3 Generation FPGA User Guide www.xilinx.com 191

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

Fast, Long Counters Using RAM

A counter is an example of a simple state machine, where the next state depends only on
the current state. A binary up counter, for example, simply increments the current state to
create the next state. Figure 4-28 shows a 20-bit binary up counter, with clock enable and
synchronous reset, implemented in a single block RAM.

1Kx18
TC ADDRB[9:0]
DOA[9:0] COUNT[19:10]
CNT[9:0] DOA[10] TERM_COUNT
DOA[17:11]
(unused)
WEA
EN ENA
SSR SSRA
CLK CLKA

TC CLKB
SSRB Terminal Count
CNT[9:0] ENA ENB from Port B enables
WEB Port A every 1024
(unused) clock cycles.
DOB[17:11]
DOB[10]
EN DOB[9:0] COUNT[9:0]
ADDRB[9:0]
1Kx18
UG331_c4_15_120406

Figure 4-28: Two 10-Bit Counters Create a 20-Bit Binary Counter Using a Single Block RAM

A 20-bit binary counter can be constructed from two identical 10-bit binary counters, with
the lower 10-bit counter enabling the upper 10-bit counter every 1024 clock cycles. In this
example, Port B is a 1Kx18 ROM (WEB is Low) that forms the lower 10-bit counter. The 10
less-significant data outputs, representing the current state, connect directly to the 10
address inputs, ADDRB[9:0]. The next state is looked up in the ROM using the current state
applied to the address pins. The 11th data bit, D[10], forms the terminal-count output from
the counter. In this example, the upper seven data bits, DOB[17:11] are unused.
The next-state logic for a binary counter appears in Table 4-15. The counter starts at state
0—or the value specified by the INIT or SRVAL attributes—and counts through to 0x3FF
(1023 decimal) at which time the terminal count, D[10], is active and the counter rolls over
back to 0.

192 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

Table 4-15: Next-State Logic for Binary Up Counter

Current State State Outputs Next State

TC COUNT
ADDR[9:0] D[9:0]
(Hex) D[10] (Hex)
0 0 1
1 0 2
2 0 3
… … …
3FFF 1 0

Port A is configured nearly identically to Port B, except that Port A is enabled by the
terminal count output from Port B. The 10-bit counter in Port A has the identical counting
pattern as Port B, except that it increments at 1/1024th the rate of Port B.
With a simple modification, the 20-bit up counter becomes an 18-bit up/down counter.
Using the most-significant address input as a direction control, the same basic counter
architecture either increments or decrements its count, as shown in Table 4-16. In this
example, the counter increments when the Up/Down control is Low and decrements
when High. The ROM is split between the incrementing and decrementing next-state logic.

Table 4-16: Next-State Logic for Binary Up/Down Counter

Up/Down Control Present State State Outputs Next State
TC COUNT
ADDR[8:0] D[9:0]
ADDR[9] (Hex) D[10] (Hex)
0 0 0 1
(Up)
1 0 2
2 0 3
… … …
1FFF 1 0

1 1FFF 0 1FFE
(Down)
1FFE 0 1FFD
1FFD 0 1FFC
… … …
0 1 1FFF

Various other counter implementations are possible including the following:

• Binary up and up/down counters of various modulos determined by the
combinations of the modulos of the counters implemented in Port A and Port B.

Spartan-3 Generation FPGA User Guide www.xilinx.com 193

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

• Counters with other incrementing and decrementing patterns including fast gray-
code counters.
• A six-digit BCD counter in one block ROM, configured as 512x36, plus one CLB.

Four-Port Memory
Each block RAM is physically a dual-port memory. However, due to the block RAM’s fast
access performance, it is possible to create multi-port memories by time-division
multiplexing the signals in and out of the memory. A block RAM with some additional
logic easily supports up to four ports but at the cost of additional access latency for each
port. The following application note provides additional details and a reference design.
• XAPP228: Quad-Port Memories in Virtex Devices, includes reference design

Content-Addressable Memory (CAM)

Content-Addressable Memory (CAM), sometimes known as associative memory, is used
in a variety of networking and data processing applications. In most memory applications,
content is referenced by an address. In CAM applications, the content is the driving input
and the output indicates whether or not the content exists in memory and, if so, provides
a reference to its location.
An easy way to envision how a CAM operates is to think of an index to a book. Looking up
an item, i.e., the content, first determines whether the item exists in the index, and if it does,
provides a reference to its location, i.e., the page number of where the item can be found.
• CORE Generator: Content-Addressable Memory module
• XAPP260: Using Block RAM for High-Performance Read/Write CAMs
• XAPP201: An Overview of Multiple CAM Designs, written for Virtex/Virtex-E and
Spartan-II/Spartan-IIE architectures but provides a useful overview to the techniques
involved

Implementing Logic Functions Using Block RAM

Inside every Spartan-3 FPGA logic cell, there is a four-input RAM/ROM called a look-up
table or LUT. The LUT performs any possible logic function of its four inputs and forms the
basis of the Spartan-3 logic architecture.
Another possible application for block RAM is as a much larger look-up table. In one of its
organizations, a block RAM—used as ROM in this case—has 14 inputs and a single output.
Consequently, block RAM is capable of implementing any possible arbitrary logic function
of up to 14 inputs, regardless of the complexity and regardless of inversions. There are a
few restrictions, however.
• There cannot be any asynchronous feedback paths in the logic, such as those that
create latches.
• The logic output must be synchronized to a clock input. Block RAM does not support
asynchronous read outputs.
If the logic function meets these requirements, then a single block RAM implements the
following functions.
• Any possible Boolean logic function of up to 14 inputs.
• Nine separate arbitrary Boolean logic functions of 11 inputs, as long as the inputs are
shared.

194 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Block RAM Applications

• Various other combinations are possible, but might have restrictions to the number of
inputs, the number of shared inputs, or the complexity of the logic function.
Due to the flexibility and speed of CLB logic, block RAM might not be faster or more
efficient for simple wide functions like an address decoder, where multiple inputs are
ANDed together. Block RAM is faster and more efficient for complex logic functions, such
as majority decoders, pattern matching, and correlators.

Fuzzy Pattern Matching Circuit Example

For example, Figure 4-29 illustrates a fuzzy pattern matching circuit that detects both exact
matches and those patterns that are close enough. Each incoming bit is matched against the
required MATCH pattern. Then, any “don’t care” bits are masked off, indicating that the
specific bit should always match. Then, the number of matching bits is counted and
compared against an activation threshold. If the number of matching bits is greater than
the activation threshold, then the input data mostly matches the required pattern and the
MATCH output goes High.

Compare bit to MASK any don't

MATCH pattern care bits

1 0
•
•
•
•
•
• Σ
0 1 > MATCH

Threshold
Is number of
matching bits
greater than
threshold?

Number of bits
that must match X463_29_040403

Figure 4-29: A 14-Input Fuzzy Pattern Matching Circuit Implemented in a Single

Block RAM

If the application requires a new matching pattern or different logic function, it could be
loaded via the second memory port.
Implemented in CLB logic, this function would require numerous logic cells and multiple
layers of logic. However, because the MATCH, MASK, and Threshold values are known in
advance, the function can be pre-computed and then stored in block RAM. For each input
condition, i.e., starting at address 0 and incremented through the entire memory, the
output condition can be precomputed. A 14-input fuzzy pattern matching circuit requires
a single block RAM and performs the operation in a single clock cycle.

Mapping Logic into Block RAM Using MAP –bp Option

The Xilinx ISE software does not automatically attempt to map logic functions into block
RAM. However, there is a mapping option to aid the process.
The block RAM mapping option is enabled when using the MAP –bp option. If so enabled,
the Xilinx ISE logic mapping software attempts to place LUTs and attached flip-flops into
an unused single-output, single-port block RAM. The final flip-flop output is required as
block RAMs have a synchronous, registered output. The mapping software packs the flip-

Spartan-3 Generation FPGA User Guide www.xilinx.com 195

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

flop with whatever LUT logic is driving it. No register is packed into block RAM without
LUT logic, and vice versa.
To specify which register outputs are converted to block RAM outputs, create a file
containing a list of the net names connected to the register output(s). Set the environment
variable XIL_MAP_BRAM_FILE to the file name, which instructs the mapping software to
use this file. The MAP program looks for this environment variable whenever the –bp
option is specified. Only those output nets listed in the file are converted into block RAM
outputs.
• PCs:
set XIL_MAP_BRAM_FILE=file_name
• Workstations:
setenv XIL_MAP_BRAM_FILE file_name

Waveform Storage, Function Tables, Direct Digital Synthesis (DDS) Using

Block RAM
Another powerful block RAM application is waveform storage, including function tables
such as trigonometric functions like sine and cosine. Sine and cosine form the backbone of
other functions such as direct digital synthesis (DDS) to generate output waveforms. The
Xilinx CORE Generator system provides parameterizable modules for both:
• CORE Generator: Sine/Cosine Look-Up Table module
• CORE Generator: Direct Digital Synthesizer (DDS) module
Another potential application of waveform storage is in various signal companders
(compressors/expanders) and normalization circuits used to boost important parts of a
signal within the available bandwidth. Examples include converters between linear data,
u-Law encoded data, and A-Law encoded data commonly used in telecommunications.
The dual-port nature of block RAM not only facilitates waveform storage, it also enables an
application to update the waveform, either with a completely new waveform or with
corrected or normalized waveform data. In the example shown in Figure 4-30, Port A
initially contains the currently active waveform. The application can load a new waveform
on Port B.

Port A
Active Waveform

DOA

ADDRA

DIB

ADDRB

Port B
Update Waveform X463_30_062503

Figure 4-30: Dual-Port Block RAM Facilitates Waveform Storage and Updates

196 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

As in real-world engineering, sometimes it is faster to look up an answer than deriving it.

The same is true in digital designs. Block RAM is also useful for storing pre-computed
function tables where the output, y, is a function of the input, x, or y=f(x).
For example, instead of creating the CLB logic that implements the following polynomial
equation, the function can be precomputed and stored in a block RAM.
Y = Ax3 – Bx2 + Cx + D
The values A, B, C, and D are all constants. The output, y, depends only on the input, x. The
output value can be precomputed for each input value of x and stored in memory. There
are obvious limitations as the function might not fit in a single logic block either because of
the range of values for x, or the magnitude of the output, y. For example, a 512x36 block
ROM implements the above equation for input values between 0 and 511. The range of x is
limited by its exponential effect on y. With x at its maximum value for this specific
example, y requires at least 28 output bits.
Some other look-up functions possible in a single block RAM/ROM include the following:
• Various complex arithmetic functions of a single input, including mixtures of
functions such as log(x), square-root(x). Multipliers of two values are possible but are
typically limited by the number of block RAM inputs. The Spartan-3 FPGA embedded
18x18 multipliers are a better solution for pure multiplication functions.
• Two independent 11-bit binary to 4-digit BCD converters with the block ROM
configured as 1Kx18. The least-significant bit (LSB) of each converter bypasses the
ROM as the converted result is the same as the original value, i.e., the LSB indicates
whether the value is odd or even.
• Two independent 3-digit BCD to 10-bit binary converters with the block ROM
configured as 2Kx9 and the LSBs bypass the converters.
• Sine-cosine look-up tables using one port for sine and the other one for cosine with 90
degree-shifted addresses, 18-bit amplitude, and 10-bit angular resolution.
• Two independent 10-bit binary to three-digit, seven-segment LED output converter
with the block ROM configured as 1Kx18. Leading zeros are displayed as blanks.
Because input values are limited to 1023, the LED digits display from “0” to “3FF”.
Consequently, the logic for the most-significant digit requires only four inputs
(segment a=d=g, segment f is always High).

Related Materials and References

• Creative Uses of Block RAM by Peter Alfke, Xilinx, Inc.
• The Myriad Uses of Block RAM by Jan Gray, Gray Research, LLC.
http://www.fpgacpu.org/usenet/bb.html
• Spartan-3A and Spartan-3A DSP FPGA Libraries Guide for HDL Designs, by Xilinx, Inc.
http://www.xilinx.com/support/documentation/dt_ise.htm
This document is also located within Project Navigator by selecting HelpÆSoftware
Manuals. When the Acrobat document appears, click on a Libraries Guide from the
table of contents on the left.

Conclusion
The Spartan-3 generation FPGA’s abundant, fast, and flexible block RAMs provide
invaluable on-chip local storage for scratchpad memories, FIFOs, buffers, look-up tables,

Spartan-3 Generation FPGA User Guide www.xilinx.com 197

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

and much more. Using unique capabilities, block RAM implements such functions as shift
registers, delay lines, counters, and wide, complex logic functions.
Block RAM is supported in applications using the broad spectrum of Xilinx ISE
development software, including the CORE Generator system and can be inferred or
directly instantiated in VHDL or Verilog synthesis designs.

Appendix A: VHDL Instantiation Example

The following VHDL instantiation example XC3S_RAMB_1_PORT uses the
SelectRAM_A36.vhd VHDL template. This and other templates are available for
download from the following Web link. The following example is a VHDL code snippet
and will not compile as is.
• xapp463_vhdl.zip
-- Module: XC3S_RAMB_1_PORT
-- Description: 18Kb Block SelectRAM example
-- Single Port 512 x 36 bits
-- Use template “SelectRAM_A36.vhd"
--
-- Device: Spartan-3 Family
---------------------------------------------------------------------
library IEEE;
use IEEE.std_logic_1164.all;
--
-- pragma translate_off
library UNISIM;
use UNISIM.VCOMPONENTS.ALL;
-- pragma translate_on
--
entity XC3S_RAMB_1_PORT is
port (
DATA_IN : in std_logic_vector (35 downto 0);
ADDRESS : in std_logic_vector (8 downto 0);
ENABLE : in std_logic;
WRITE_EN : in std_logic;
SET_RESET : in std_logic;
CLK : in std_logic;
DATA_OUT : out std_logic_vector (35 downto 0)
);
end XC3S_RAMB_1_PORT;
--
architecture XC3S_RAMB_1_PORT_arch of XC3S_RAMB_1_PORT is
--
-- Components Declarations:
--
component BUFG
port (
I : in std_logic;
O : out std_logic
);
end component;
--
component RAMB16_S36
-- pragma translate_off
generic (
-- "Read during Write" attribute for functional simulation
WRITE_MODE : string := "READ_FIRST" ; -- WRITE_FIRST(default)/ READ_FIRST/
NO_CHANGE
-- Output value after configuration
INIT : bit_vector(35 downto 0) := X"000000000";
-- Output value if SSR active

198 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Appendix A: VHDL Instantiation Example

SRVAL : bit_vector(35 downto 0) := X"012345678";

-- Initialize parity memory content
INITP_00 : bit_vector(255 downto 0) :=
X"000000000000000000000000000000000000000000000000FEDCBA9876543210";
INITP_01 : bit_vector(255 downto 0) :=
X"0000000000000000000000000000000000000000000000000000000000000000";
... (snip)
INITP_07 : bit_vector(255 downto 0) :=
X"0000000000000000000000000000000000000000000000000000000000000000";
-- Initialize data memory content
INIT_00 : bit_vector(255 downto 0) :=
X"000000000000000000000000000000000000000000000000FEDCBA9876543210";
INIT_01 : bit_vector(255 downto 0) :=
X"0000000000000000000000000000000000000000000000000000000000000000";
... (snip)
INIT_3F : bit_vector(255 downto 0) :=
X"0000000000000000000000000000000000000000000000000000000000000000"
);
-- pragma translate_on
port (
DI : in std_logic_vector (31 downto 0);
DIP : in std_logic_vector (3 downto 0);
ADDR : in std_logic_vector (8 downto 0);
EN : in STD_LOGIC;
WE : in STD_LOGIC;
SSR : in STD_LOGIC;
CLK : in STD_LOGIC;
DO : out std_logic_vector (31 downto 0);
DOP : out std_logic_vector (3 downto 0)
);
end component;
--
-- Attribute Declarations:
attribute WRITE_MODE : string;
attribute INIT: string;
attribute SRVAL: string;
-- Parity memory initialization attributes
attribute INITP_00: string;
attribute INITP_01: string;
... (snip)
attribute INITP_07: string;
-- Data memory initialization attributes
attribute INIT_00: string;
attribute INIT_01: string;
... (snip)
attribute INIT_3F: string;
--
-- Attribute "Read during Write mode" = WRITE_FIRST(default)/ READ_FIRST/
NO_CHANGE
attribute WRITE_MODE of U_RAMB16_S36: label is "READ_FIRST";
attribute INIT of U_RAMB16_S36: label is "000000000";
attribute SRVAL of U_RAMB16_S36: label is "012345678";
--
-- RAMB16 memory initialization for Alliance
-- Default value is "0" / Partial initialization strings are padded
-- with zeros to the left
attribute INITP_00 of U_RAMB16_S36: label is
"000000000000000000000000000000000000000000000000FEDCBA9876543210";
attribute INITP_01 of U_RAMB16_S36: label is
"0000000000000000000000000000000000000000000000000000000000000000";
... (snip)
attribute INITP_07 of U_RAMB16_S36: label is
"0000000000000000000000000000000000000000000000000000000000000000";
--
attribute INIT_00 of U_RAMB16_S36: label is

Spartan-3 Generation FPGA User Guide www.xilinx.com 199

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

"000000000000000000000000000000000000000000000000FEDCBA9876543210";
attribute INIT_01 of U_RAMB16_S36: label is
"0000000000000000000000000000000000000000000000000000000000000000";
... (snip)
attribute INIT_3F of U_RAMB16_S36: label is
"0000000000000000000000000000000000000000000000000000000000000000";
--
-- Signal Declarations:
--
-- signal VCC : std_logic;
-- signal GND : std_logic;
signal CLK_BUFG: std_logic;
signal INV_SET_RESET : std_logic;
--
begin
-- VCC <= ’1’;
-- GND <=’0’;
--
-- Instantiate the clock buffer
U_BUFG: BUFG
port map (
I => CLK,
O => CLK_BUFG
);
--
-- Use of the free inverter on SSR pin
INV_SET_RESET <= NOT SET_RESET;
-- Block SelectRAM Instantiation
U_RAMB16_S36: RAMB16_S36
port map (
DI => DATA_IN (31 downto 0), -- insert 32 bits data-in bus (<31 downto 0>)
DIP => DATA_IN (35 downto 32), -- insert 4 bits parity data-in bus (or <35
-- downto 32>)
ADDR => ADDRESS (8 downto 0), -- insert 9 bits address bus
EN => ENABLE, -- insert enable signal
WE => WRITE_EN, -- insert write enable signal
SSR => INV_SET_RESET, -- insert set/reset signal
CLK => CLK_BUFG, -- insert clock signal
DO => DATA_OUT (31 downto 0), -- insert 32 bits data-out bus (<31 downto 0>)
DOP => DATA_OUT (35 downto 32) -- insert 4 bits parity data-out bus (or <35
-- downto 32>)
);
--
end XC3S_RAMB_1_PORT_arch;

200 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Appendix B: Verilog Instantiation Example

Appendix B: Verilog Instantiation Example

The following Verilog instantiation example XC3S_RAMB_1_PORT uses the
SelectRAM_A36.v Verilog template. This and other templates are available for
download from the following Web link. The following example is a Verilog code snippet
and will not compile as is.
• xapp463_verilog.zip
// Module: XC3S_RAMB_1_PORT
// Description: 18Kb Block SelectRAM-II example
// Single Port 512 x 36 bits
// Use template "SelectRAM_A36.v"
//
// Device: Spartan-3 Family
//-------------------------------------------------------------------
module XC3S_RAMB_1_PORT (CLK, SET_RESET, ENABLE, WRITE_EN, ADDRESS, DATA_IN,
DATA_OUT);
input CLK, SET_RESET, ENABLE, WRITE_EN;
input [35:0] DATA_IN;
input [8:0] ADDRESS;
output [35:0] DATA_OUT;
wire CLK_BUFG, INV_SET_RESET;
//Use of the free inverter on SSR pin
assign INV_SET_RESET = ~SET_RESET;
// initialize block ram for simulation
defparam
//”Read during Write” attribute for functional simulation
U_RAMB16_S36.WRITE_MODE = “READ_FIRST”, //WRITE_FIRST(default)/ READ_FIRST/
NO_CHANGE
//Output value after configuration
U_RAMB16_S36.INIT = 36'h000000000,
//Output value if SSR active
U_RAMB16_S36.SRVAL = 36'h012345678,
//Initialize parity memory content
U_RAMB16_S36.INITP_00 =
256'h0123456789ABCDEF000000000000000000000000000000000000000000000000,
U_RAMB16_S36.INITP_01 =
256'h0000000000000000000000000000000000000000000000000000000000000000,
... (snip)
U_RAMB16_S36.INITP_07 =
256'h0000000000000000000000000000000000000000000000000000000000000000,
//Initialize data memory content
U_RAMB16_S36.INIT_00 =
256'h0123456789ABCDEF000000000000000000000000000000000000000000000000,
U_RAMB16_S36.INIT_01 =
256'h0000000000000000000000000000000000000000000000000000000000000000,
... (snip)
U_RAMB16_S36.INIT_3F =
256'h0000000000000000000000000000000000000000000000000000000000000000;
//Instantiate the clock Buffer
BUFG U_BUFG ( .I(CLK), .O(CLK_BUFG));
//Block SelectRAM Instantiation
RAMB16_S36 U_RAMB16_S36 (
.DI(DATA_IN[31:0]),
.DIP(DATA_IN-PARITY[35:32]),
.ADDR(ADDRESS),
.EN(ENABLE),
.WE(WRITE_EN),
.SSR(INV_SET_RESET),
.CLK(CLK_BUFG),
.DO(DATA_OUT[31:0]),
.DOP(DATA_OUT-PARITY[35:32]));
// synthesis attribute declarations

Spartan-3 Generation FPGA User Guide www.xilinx.com 201

UG331 (v1.7) August 19, 2010
 R

Chapter 4: Using Block RAM

/* attribute
WRITE_MODE "READ_FIRST"
INIT "000000000"
SRVAL "012345678"
INITP_00
"0123456789ABCDEF000000000000000000000000000000000000000000000000"
INITP_01
"0000000000000000000000000000000000000000000000000000000000000000"
... (snip)
INITP_07
"0000000000000000000000000000000000000000000000000000000000000000"
INIT_00
"0123456789ABCDEF000000000000000000000000000000000000000000000000"
INIT_01
"0000000000000000000000000000000000000000000000000000000000000000"
... (snip)
INIT_3F
"0000000000000000000000000000000000000000000000000000000000000000"
*/
endmodule

202 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 5

Using Configurable Logic Blocks (CLBs)

CLB Overview
The Configurable Logic Blocks (CLBs) constitute the main logic resource for implementing
synchronous as well as combinatorial circuits. Each CLB contains four slices, and each slice
contains two Look-Up Tables (LUTs) to implement logic and two dedicated storage
elements that can be used as flip-flops or latches. The LUTs can be used as a 16x1 memory
(RAM16) or as a 16-bit shift register (SRL16), and additional multiplexers and carry logic
simplify wide logic and arithmetic functions. Most general-purpose logic in a design is
automatically mapped to the slice resources in the CLBs. The details of the CLB resources
are helpful when estimating the number of resources required for an application or when
optimizing a design to the architecture.

CLB Array
The CLBs are arranged in a regular array of rows and columns as shown in Figure 5-1.
Each density varies by the number of rows and columns of CLBs (see Table 5-1).

X0Y3 X1Y3 X2Y3 X3Y3

X0Y2 X1Y2 X2Y2 X3Y2

X0Y1 X1Y1 X2Y1 X3Y1

Spartan-3E
FPGA
X0Y0 X1Y0 X2Y0 X3Y0

IOBs

Slice
CLB DS312-2_31_021205

Figure 5-1: CLB Locations

Spartan-3 Generation FPGA User Guide www.xilinx.com 203

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

Table 5-1: CLB Resources

CLB CLB CLB LUTs / Equivalent RAM16 / Distributed
Device Slices
Rows Columns Total Flip-Flops Logic Cells SRL16 RAM Bits
Spartan®-3A DSP FPGA CLB Resources
XC3SD1800A 88 48 4,160 16,640 33,280 37,440 16,640 266,240
XC3SD3400A 104 58 5,968 23,872 47,744 53,712 23,872 381,952
Spartan-3A/3AN FPGA CLB Resources
XC3S50A/AN 16 12 176 704 1,408 1,584 704 11,264
XC3S200A/AN 32 16 448 1,792 3,584 4,032 1,792 28,672
XC3S400A/AN 40 24 896 3,584 7,168 8,064 3,584 57,344
XC3S700A/AN 48 32 1,472 5,888 11,776 13,248 5,888 94,208
XC3S1400A/AN 72 40 2,816 11,264 22,528 25,344 11,264 180,224
Spartan-3E FPGA CLB Resources
XC3S100E 22 16 240 960 1,920 2,160 960 15,360
XC3S250E 34 26 612 2,448 4,896 5,508 2,448 39,168
XC3S500E 46 34 1,164 4,656 9,312 10,476 4,656 74,496
XC3S1200E 60 46 2,168 8,672 17,344 19,512 8,672 138,752
XC3S1600E 76 58 3,688 14,752 29,504 33,192 14,752 236,032
Spartan-3 FPGA CLB Resources
XC3S50 16 12 192 768 1,536 1,728 768 12,288
XC3S200 24 20 480 1,920 3,840 4,320 1,920 30,720
XC3S400 32 28 896 3,584 7,168 8,064 3,584 57,344
XC3S1000 48 40 1,920 7,680 15,360 17,280 7,680 122,880
XC3S1500 64 52 3,328 13,312 26,624 29,952 13,312 212,992
XC3S2000 80 64 5,120 20,480 40,960 46,080 20,480 327,680
XC3S4000 96 72 6,912 27,648 55,296 62,208 27,648 442,368
XC3S5000 104 80 8,320 33,280 66,560 74,880 33,280 532,480

CLB Differences between Spartan-3 Generation Families

Each CLB is identical within a family, and the CLBs are identical among all
Spartan-3 generation families. The performance varies slightly between families due to
minor variations in processing and characterization. The only difference between families
is how the number of CLBs relates to the number of rows and columns. In the Spartan-3E
and Extended Spartan-3A family, the number of CLBs is less than the multiple of the
number of rows and columns. This difference is because in the Extended Spartan-3A
family, the DCMs are embedded in the array, and in the Spartan-3E family, both the DCMs
and the block RAM/multiplier blocks are embedded in the array. See Module 1 of the
Spartan-3E, Spartan-3A, Spartan-3AN, and Spartan-3A DSP FPGA data sheets for a figure
showing the array structure.

204 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Slices

Slices
Each CLB comprises four interconnected slices, as shown in Figure 5-3. These slices are
grouped in pairs. Each pair is organized as a column with an independent carry chain. The
left pair supports both logic and memory functions and its slices are called SLICEM. The
right pair supports logic only and its slices are called SLICEL. Therefore half the LUTs
support both logic and memory (including both RAM16 and SRL16 shift registers) while
half support logic only, and the two types alternate throughout the array columns. The
SLICEL reduces the size of the CLB and lowers the cost of the device, and can also provide
a performance advantage over the SLICEM.

Spartan-3 Generation FPGA User Guide www.xilinx.com 205

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

COUT

YBMUX
YB

FXINA
Fi
FXINB GYMUX

G[4:1]

G-LUT DYMUX
D Q YQ

FFY
WS DI CE
CK
SR REV

BY
Top Portion

CE
CLK
SR

Common Logic

WS DI

F[4:1]

F-LUT
WF[4:1]

FFX

CYINIT

BX
Bottom Portion

LEGEND: Logic Functions

Distributed RAM and CIN
Shift Register Functions
UG331_c7_02_110708

Notes:
1. Options to invert signal polarity as well as other options that enable lines for various functions are not shown.
2. The index i can be 6, 7, or 8, depending on the slice. The upper SLICEL has an F8MUX, and the upper SLICEM has an
F7MUX. The lower SLICEL and SLICEM both have an F6MUX.

Figure 5-2: Simplified Diagram of the Left-Hand SLICEM

206 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Slice Overview

Left-Hand SLICEM Right-Hand SLICEL

(Logic or Distributed RAM (Logic Only)
or Shift Register)
COUT

CLB
SLICE
X1Y1

SLICE
X1Y0
COUT
Switch Interconnect
Matrix CIN to Neighbors
SLICE
X0Y1
SHIFTOUT
SHIFTIN
SLICE
X0Y0

CIN DS099-2_05_082104

Figure 5-3: Arrangement of Slices within the CLB

Slice Location Designations

The Xilinx development software designates the location of a slice according to its X and Y
coordinates, starting in the bottom left corner, as shown in Figure 5-1. The letter ‘X’
followed by a number identifies columns of slices, incrementing from the left side of the
die to the right. The letter ‘Y’ followed by a number identifies the position of each slice in
a pair as well as indicating the CLB row, incrementing from the bottom of the die.
Figure 5-3 shows the CLB located in the lower left-hand corner of the die. The SLICEM
always has an even ‘X’ number, and the SLICEL always has an odd ‘X’ number.

Slice Overview
A slice includes two LUT function generators and two storage elements, along with
additional logic, as shown in Figure 5-4.
Both SLICEM and SLICEL have the following elements in common to provide logic,
arithmetic, and ROM functions:
• Two 4-input LUT function generators, F and G
• Two storage elements
• Two wide-function multiplexers, F5MUX and FiMUX
• Carry and arithmetic logic

Spartan-3 Generation FPGA User Guide www.xilinx.com 207

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

SRL16 FiMUX FiMUX

RAM16
LUT4 (G)
Carry LUT4 (G)
Carry
Register Register

F5MUX F5MUX
SRL16
RAM16 Carry Register Carry Register
LUT4 (F) LUT4 (F)

Arithmetic Logic Arithmetic Logic

SLICEM SLICEL DS312-2_13_020905

Figure 5-4: Resources in a Slice

The SLICEM pair supports two additional functions:

• Two 16x1 distributed RAM blocks, RAM16
• Two 16-bit shift registers, SRL16

Logic Cells
The combination of a LUT and a storage element is known as a “Logic Cell”. The
additional features in a slice, such as the wide multiplexers, carry logic, and arithmetic
gates, add to the capacity of a slice, implementing logic that would otherwise require
additional LUTs. Benchmarks have shown that the overall slice is equivalent to 2.25 simple
logic cells. This calculation provides the equivalent logic cell count shown in Table 5-1.

Slice Details
Figure 5-2 is a detailed diagram of the SLICEM. It represents a superset of the elements and
connections to be found in all slices. The dashed and gray lines (blue when viewed in
color) indicate the resources found only in the SLICEM and not in the SLICEL.
Each slice has two halves, which are differentiated as top and bottom to keep them distinct
from the upper and lower slices in a CLB. The control inputs for the clock (CLK), Clock
Enable (CE), Slice Write Enable (SLICEWE1), and Reset/Set (RS) are shared in common
between the two halves.
The LUTs located in the top and bottom portions of the slice are referred to as "G" and "F",
respectively, or the "G-LUT" and the "F-LUT". The storage elements in the top and bottom
portions of the slice are called FFY and FFX, respectively.
Each slice has two multiplexers with F5MUX in the bottom portion of the slice and FiMUX
in the top portion. Depending on the slice, the FiMUX takes on the name F6MUX, F7MUX,
or F8MUX, according to its position in the multiplexer chain. The lower SLICEL and
SLICEM both have an F6MUX. The upper SLICEM has an F7MUX, and the upper SLICEL
has an F8MUX.
The carry chain enters the bottom of the slice as CIN and exits at the top as COUT. Five
multiplexers control the chain: CYINIT, CY0F, and CYMUXF in the bottom portion and

208 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Slice Details

CY0G and CYMUXG in the top portion. The dedicated arithmetic logic includes the
exclusive-OR gates XORF and XORG (bottom and top portions of the slice, respectively) as
well as the AND gates FAND and GAND (bottom and top portions, respectively).
See Table 5-2 for a description of all the slice input and output signals.

Table 5-2: Slice Inputs and Outputs

Name Location Direction Description
F[4:1] SLICEL/M Bottom Input F-LUT and FAND inputs
G[4:1] SLICEL/M Top Input G-LUT and GAND inputs or Write Address (SLICEM)
BX SLICEL/M Bottom Input Bypass to or output (SLICEM) or storage element, or control
input to F5MUX, input to carry logic, or data input to RAM
(SLICEM)
BY SLICEL/M Top Input Bypass to or output (SLICEM) or storage element, or control
input to FiMUX, input to carry logic, or data input to RAM
(SLICEM)
BXOUT SLICEM Bottom Output BX bypass output
BYOUT SLICEM Top Output BY bypass output
ALTDIG SLICEM Top Input Alternate data input to RAM
DIG SLICEM Top Output ALTDIG or SHIFTIN bypass output
SLICEWE1 SLICEM Common Input RAM Write Enable
F5 SLICEL/M Bottom Output Output from F5MUX; direct feedback to FiMUX
FXINA SLICEL/M Top Input Input to FiMUX; direct feedback from F5MUX or another FiMUX
FXINB SLICEL/M Top Input Input to FiMUX; direct feedback from F5MUX or another FiMUX
Fi SLICEL/M Top Output Output from FiMUX; direct feedback to another FiMUX
CE SLICEL/M Common Input FFX/Y Clock Enable
SR SLICEL/M Common Input FFX/Y Set or Reset or RAM Write Enable (SLICEM)
CLK SLICEL/M Common Input FFX/Y Clock or RAM Clock (SLICEM)
SHIFTIN SLICEM Top Input Data input to G-LUT RAM
SHIFTOUT SLICEM Bottom Output Shift data output from F-LUT RAM
CIN SLICEL/M Bottom Input Carry chain input
COUT SLICEL/M Top Output Carry chain output
X SLICEL/M Bottom Output Combinatorial output
Y SLICEL/M Top Output Combinatorial output
XB SLICEL/M Bottom Output Combinatorial output from carry or F-LUT SRL16 (SLICEM)
YB SLICEL/M Top Output Combinatorial output from carry or G-LUT SRL16 (SLICEM)
XQ SLICEL/M Bottom Output FFX output
YQ SLICEL/M Top Output FFY output

Spartan-3 Generation FPGA User Guide www.xilinx.com 209

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

Main Logic Paths

Central to the operation of each slice are two nearly identical data paths at the top and
bottom of the slice. The description that follows uses names associated with the bottom
path. (The top path names appear in parentheses.) The basic path originates at an
interconnect switch matrix outside the CLB. See Chapter 12, “Using Interconnect,” for
more information on the switch matrix and the routing connections.
Four lines, F1 through F4 (or G1 through G4 on the upper path), enter the slice and connect
directly to the LUT. Once inside the slice, the lower 4-bit path passes through a LUT ‘F’ (or
‘G’) that performs logic operations. The LUT Data output, ‘D’, offers five possible paths:
1. Exit the slice via line "X" (or "Y") and return to interconnect.
2. Inside the slice, "X" (or "Y") serves as an input to the DXMUX (or DYMUX) which feeds
the data input, "D", of the FFX (or FFY) storage element. The "Q" output of the storage
element drives the line XQ (or YQ) which exits the slice.
3. Control the CYMUXF (or CYMUXG) multiplexer on the carry chain.
4. With the carry chain, serve as an input to the XORF (or XORG) exclusive-OR gate that
performs arithmetic operations, producing a result on "X" (or "Y").
5. Drive the multiplexer F5MUX to implement logic functions wider than four bits. The
"D" outputs of both the F-LUT and G-LUT serve as data inputs to this multiplexer.
In addition to the main logic paths described above, there are two bypass paths that enter
the slice as BX and BY. Once inside the FPGA, BX in the bottom half of the slice (or BY in the
top half) can take any of several possible branches:
1. Bypass both the LUT and the storage element, and then exit the slice as BXOUT (or
BYOUT) and return to interconnect.
2. Bypass the LUT, and then pass through a storage element via the D input before
exiting as XQ (or YQ).
3. Control the wide function multiplexer F5MUX (or FiMUX).
4. Via multiplexers, serve as an input to the carry chain.
5. Drive the DI input of the LUT.
6. BY can control the REV inputs of both the FFY and FFX storage elements. See “Storage
Element Functions,” page 328.
7. Finally, the DIG_MUX multiplexer can switch BY onto the DIG line, which exits the
slice.
The control inputs CLK, CE, SR, BX, and BY have programmable polarity. The LUT inputs
do not need programmable polarity because their function can be inverted inside the LUT.
The sections that follow provide more detail on individual functions of the slice.

Look-Up Tables
The Look-Up Table or LUT is a RAM-based function generator and is the main resource for
implementing logic functions. Furthermore, the LUTs in each SLICEM pair can be
configured as Distributed RAM or a 16-bit shift register, as described later.
Each of the two LUTs (F and G) in a slice have four logic inputs (A1-A4) and a single output
(D). Any four-variable Boolean logic operation can be implemented in one LUT. Functions
with more inputs can be implemented by cascading LUTs or by using the wide function
multiplexers that are described later.

210 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Wide Multiplexers

The output of the LUT can connect to the wide multiplexer logic, the carry and arithmetic
logic, or directly to a CLB output or to the CLB storage element. See Figure 5-5.

4
G[4:1] A[4:1] D YQ
FFY
G-LUT

4
F[4:1] A[4:1] D XQ
FFX
F-LUT

DS312-2_33_111105

Figure 5-5: LUT Resources in a Slice

Wide Multiplexers
Wide-function multiplexers effectively combine LUTs in order to permit more complex
logic operations. Each slice has two of these multiplexers with F5MUX in the bottom
portion of the slice and FiMUX in the top portion. The F5MUX multiplexes the two LUTs in
a slice. The FiMUX multiplexes two CLB inputs which connect directly to the F5MUX and
FiMUX results from the same slice or from other slices. For more information on the wide
multiplexers, see Chapter 8, “Using Dedicated Multiplexers.”

Carry and Arithmetic Logic

The carry chain, together with various dedicated arithmetic logic gates, support fast and
efficient implementations of math operations. The carry logic is automatically used for
most arithmetic functions in a design. The gates and multiplexers of the carry and
arithmetic logic can also be used for general-purpose logic, including simple wide Boolean
functions. For more information on the carry and arithmetic logic, see Chapter 9, “Using
Carry and Arithmetic Logic.”

Storage Elements
The storage element, which is programmable as either a D-type flip-flop or a level-
sensitive transparent latch, provides a means for synchronizing data to a clock signal,
among other uses. The storage elements in the top and bottom portions of the slice are
called FFY and FFX, respectively. FFY has a fixed multiplexer on the D input selecting
either the combinatorial output Y or the bypass signal BY. FFX selects between the
combinatorial output X or the bypass signal BX.
The functionality of a slice storage element is identical to that described earlier for the I/O
storage elements. All signals have programmable polarity; the default active-High
function is described.

Spartan-3 Generation FPGA User Guide www.xilinx.com 211

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

Table 5-3: Storage Element Signals

Signal Description
D Input. For a flip-flop data on the D input is loaded when R and S (or CLR and PRE) are Low and CE is High
during the Low-to-High clock transition. For a latch, Q reflects the D input while the gate (G) input and gate
enable (GE) are High and R and S (or CLR and PRE) are Low. The data on the D input during the High-to-
Low gate transition is stored in the latch. The data on the Q output of the latch remains unchanged as long as
G or GE remains Low.
Q Output. Toggles after the Low-to-High clock transition for a flip-flop and immediately for a latch.
C Clock for edge-triggered flip-flops.
G Gate for level-sensitive latches.
CE Clock Enable for flip-flops.
GE Gate Enable for latches.
S Synchronous Set (Q = High). When the S input is High and R is Low, the flip-flop is set, output High, during
the Low-to-High clock (C) transition. A latch output is immediately set, output High.
R Synchronous Reset (Q = Low); has precedence over Set.
PRE Asynchronous Preset (Q = High). When the PRE input is High and CLR is Low, the flip-flop is set, output
High, during the Low-to-High clock (C) transition. A latch output is immediately set, output High.
CLR Asynchronous Clear (Q = Low); has precedence over Preset to reset Q output Low
SR CLB input for R, S, CLR, or PRE
REV CLB input for opposite of SR. Must be asynchronous or synchronous to match SR.

The control inputs R, S, CE, and C are all shared between the two flip-flops in a slice.

FDRSE
D Q
CE
C

R
DS312-2_40_021305

Figure 5-6: FD Flip-Flop Component with Synchronous Reset, Set, and Clock
Enable

Table 5-4: FD Flip-Flop Functionality with Synchronous Reset, Set, and Clock
Enable
Inputs Outputs
R S CE D C Q
1 X X X ↑ 0
0 1 X X ↑ 1
0 0 0 X X No Change
0 0 1 1 ↑ 1
0 0 1 0 ↑ 0

212 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Storage Elements

Initialization
The CLB storage elements are initialized at power-up, during configuration, by the global
GSR signal, and by the individual SR or REV inputs to the CLB. The storage elements can
also be re-initialized using the GSR input on the STARTUP primitive. See “Global
Controls,” page 401.
Table 5-5: Slice Storage Element Initialization
Signal Description
SR Set/Reset input. Forces the storage element into the state specified by the attribute
SRHIGH or SRLOW. SRHIGH forces a logic “1” when SR is asserted. SRLOW
forces a logic “0”. For each slice, set and reset can be set to be synchronous or
asynchronous.
REV Reverse of Set/Reset input. A second input (BY) forces the storage element into the
opposite state. The reset condition is predominant over the set condition if both are
active. Same synchronous/asynchronous setting as for SR.
GSR Global Set/Reset. GSR defaults to active High but can be inverted by adding an
inverter in front of the GSR input of the STARTUP element. The initial state after
configuration or GSR is defined by a separate INIT0 and INIT1 attribute. By
default, setting the SRLOW attribute sets INIT0, and setting the SRHIGH attribute
sets INIT1.

Timing Parameters
There are several possible paths through the CLB. For any timing parameter, examine the
source and destination to help define the path. Most timing parameters have names based
on the source and destination. Setup time parameters typically are named according to the
input pin followed by “CK”, such as tCECK for setup from CE to CLK. Hold time
parameters are named with “CK” followed by the input pin, such as tCKCE for hold time
from CLK to CE. Table 5-6 defines the most common CLB timing parameters.

Table 5-6: Slice (LUT and Storage Element) Timing Parameters

Parameter Description
TCKO When reading from the FFX (FFY) flip-flop, the time from the active transition
at the CLK input to data appearing at the XQ (YQ) output.
TAS Time from the setup of data at the F or G input to the active transition at the
CLK input of the CLB; also referred to as TFCK or TGCK.
TAH Time from the active transition at the CLK input to the point where data is last
held at the F or G input; also referred to as TCKF or TCKG.
TDICK Time from the setup of data at the BX or BY input to the active transition at the
CLK input of the CLB.
TCKDI Time from the active transition at the CLK input to the point where data is last
held at the BX or BY input.
TCECK Time from the setup of the CE input to the active transition at the CLK input of
the CLB.
TCKCE Time from the active transition at the CLK input to the point where data is last
held at the CE input.
TCH The High pulse width of the CLK signal.

Spartan-3 Generation FPGA User Guide www.xilinx.com 213

UG331 (v1.7) August 19, 2010
 R

Chapter 5: Using Configurable Logic Blocks (CLBs)

Table 5-6: Slice (LUT and Storage Element) Timing Parameters

Parameter Description
TCL The Low pulse width of the CLK signal
FTOG Toggle frequency (for export control)
TILO The time it takes for data to travel from the CLB’s F (G) input to the X (Y) output
TRPW_CLB The minimum allowable pulse width, High or Low, to the CLB’s SR input

Distributed RAM
The LUTs in the SLICEM can be programmed as distributed RAM. This type of memory
affords moderate amounts of data buffering anywhere along a data path. One SLICEM
LUT stores 16 bits (RAM16). For more information on the distributed RAM, see Chapter 6,
“Using Look-Up Tables as Distributed RAM.”

Shift Registers
It is possible to program each SLICEM LUT as a 16-bit shift register. Used in this way, each
LUT can delay serial data anywhere from 1 to 16 clock cycles without using any of the
dedicated flip-flops. The resulting programmable delays can be used to balance the timing
of data pipelines. For more information on the shift registers, see Chapter 7, “Using Look-
Up Tables as Shift Registers (SRL16).”

Related Materials
The following documents provide supplementary information useful with this chapter:
• WP272: Get Smart About Reset: Think Local, Not Global
Applying a global reset to your FPGA designs is not a very good idea and should be
avoided. This is a controversial issue, so this white paper looks at the reasons why
such a design policy should be considered.
• WP273: Performance + Time = Memory (Cost Saving with 3-D Design)
Operating logic at a higher rate than the processing rate allows operations to be
achieved sequentially. As with a processor, logic is timeshared over multiple clock
cycles. Memory holds values not being used on a given clock cycle. The FPGA can be
considered to be a three-dimensional volume to be filled. "Performance + Time =
Memory" is a strange formula, but when understood, it can often result in
significantly lower cost implementations with Xilinx devices.
• WP275: Get Your Priorities Right - Make Your Design up to 50% Smaller
This white paper describes a rarely noticed design technique that can make a
difference in the size and the performance of your FPGA design. Control signals on
FPGA flip-flops have a built-in priority. If you can learn to write code that is
sympathetic to the priorities, the results will be rewarding. This white paper provides
some simple VHDL and Verilog examples to explain key points.

214 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 6

Using Look-Up Tables as Distributed

RAM
Summary
Each Spartan®-3 generation Configurable Logic Block (CLB) contains up to 64 bits of
single-port RAM or 32 bits of dual-port RAM. This RAM is distributed throughout the
FPGA and is commonly called “distributed RAM” to distinguish it from the 18-Kbit block
RAM. Distributed RAM is also referred to as LUT RAM. Distributed RAM is fast, localized,
and ideal for small data buffers, FIFOs, or register files. This chapter describes the features
and capabilities of distributed RAM and illustrates how to specify the various options
using the Xilinx CORE Generator system or via VHDL or Verilog instantiation.

Introduction
In addition to the embedded 18-Kbit block RAMs,
Spartan-3 generationSpartan-3 generation FPGAs feature distributed RAM within each
Configurable Logic Block (CLB). Each SLICEM function generator or LUT within a CLB
resource optionally implements a 16-deep x 1-bit synchronous RAM. The LUTs within a
SLICEL slice do not have distributed RAM.
Distributed RAM writes synchronously and reads asynchronously. However, if required
by the application, use the register associated with each LUT to implement a synchronous
read function. Each 16 x 1-bit RAM is cascadable for deeper and/or wider memory
applications, with a minimal timing penalty incurred through specialized logic resources.
Spartan-3 generation CLBs support various RAM primitives up to 64-deep by 1-bit-wide.
Two LUTs within a SLICEM slice combine to create a dual-port 16x1 RAM—one LUT with
a read/write port, and a second LUT with a read-only port. One port writes into both 16x1
LUT RAMs simultaneously, but the second port reads independently.
Distributed RAM is crucial to many high-performance applications that require relatively
small embedded RAM blocks, such as FIFOs or small register files. The Xilinx
CORE Generator software automatically generates optimized distributed RAMs for the
Spartan-3 generation architecture. Similarly, the CORE Generator system creates
Asynchronous and Synchronous FIFOs using distributed RAMs.

Single-Port and Dual-Port RAMs

Data Flow
Distributed RAM supports the following memory types:

Spartan-3 Generation FPGA User Guide www.xilinx.com 215

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

• Single-port RAM with synchronous write and asynchronous read. Synchronous reads
are possible using the flip-flop associated with distributed RAM.
• Dual-port RAM with one synchronous write and two asynchronous read ports. As
above, synchronous reads are possible.
As illustrated in Figure 6-1, dual-port distributed RAM has one read/write port and an
independent read port.

Single-Port RAM Dual-Port RAM

D R/W Port D R/W Port

O SPO
Address Address

Write Read Write Read

WCLK WCLK
Read Port
DPO
Address
Read

x464_01_062503

Figure 6-1: Single-Port and Dual-Port Distributed RAM

Any write operation on the D input and any read operation on the SPO output can occur
simultaneously with and independently from a read operation on the second read-only
port, DPO.

Write Operations
The write operation is a single clock-edge operation, controlled by the write-enable input,
WE. By default, WE is active High, although it can be inverted within the distributed RAM.
When the write enable is High, the clock edge latches the write address and writes the data
on the D input into the selected RAM location.
When the write enable is Low, no data is written into the RAM.

Read Operation
A read operation is purely combinatorial. The address port—either for single- or dual-port
modes—is asynchronous with an access time equivalent to a LUT logic delay.

Read During Write

When synchronously writing new data, the output reflects the data as it is written to the
addressed memory cell, which is similar to the WRITE_MODE=WRITE_FIRST
(transparent) mode on the Spartan-3 generation block RAMs. The timing diagram in
Figure 6-2 illustrates a write operation with the previous data read on the output port,
before the clock edge, followed by the new data.

216 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Characteristics

WCLK

DATA_IN d

ADDRESS aa

WRITE_EN
twrite
tread tread

DATA_OUT MEM(aa) d

Previous New
Data Data

x464_02_070303

Figure 6-2: Write Timing Diagram

Characteristics
• A write operation requires only one clock edge.
• A read operation requires only the logic access time.
• Outputs are asynchronous and dependent only on the LUT logic delay.
• Data and address inputs are latched with the write clock and have a setup-to-clock
timing specification. There is no hold time requirement.
• For dual-port RAM, the A[#:0] port is the write and read address, and the DPRA[#:0]
port is an independent read-only address.

Distributed RAM in the CLB

The distributed RAM takes advantage of the resources described in Chapter 5, “Using
Configurable Logic Blocks (CLBs).” One SLICEM LUT stores 16 bits (RAM16). The four
LUT inputs F[4:1] or G[4:1] become the address lines labeled A[4:1] in the device model
and A[3:0] in the design components, providing a 16x1 configuration in one LUT. Multiple
SLICEM LUTs can be combined in various ways to store larger amounts of data, including
16x4, 32x2, or 64x1 configurations in one CLB. The fifth and sixth address lines required for
the 32-deep and 64-deep configurations, respectively, are implemented using the BX and
BY inputs, which connect to the write enable logic for writing and the F5MUX and F6MUX
for reading.
Writing to distributed RAM is always synchronous to the SLICEM clock (WCLK for
distributed RAM) and enabled by the SLICEM SR input which functions as the active-
High write enable (WE). The read operation is asynchronous, and, therefore, during a
write, the output initially reflects the old data at the address being written.
The distributed RAM outputs can be captured using the flip-flops within the SLICEM
element. The WE control for the RAM and the clock-enable (CE) control for the flip-flop are
independent, but the WCLK and CLK clock inputs are shared. Because the RAM read
operation is asynchronous, the output data always reflects the currently addressed RAM
location.
A dual-port option combines two LUTs so that memory access is possible from two
independent data lines. The same data is written to both 16x1 memories but they have
independent read address lines and outputs. The dual-port function is implemented by

Spartan-3 Generation FPGA User Guide www.xilinx.com 217

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

cascading the G-LUT address lines, which are used for both read and write operations, to
the F-LUT write address lines (WF[4:1] in Figure 5-2, page 206), and by cascading the
G-LUT data input DI through the DIF_MUX in Figure 5-2 and to the DI input on the
F-LUT. One CLB provides a 16x1 dual-port memory as shown in Figure 6-5, page 224.
The INIT attribute can be used to preload the memory with data during FPGA
configuration. The default initial contents for RAM is all zeros. If WE is held Low, the
element can be considered a ROM. The ROM function can be implemented in the SLICEL.

Distributed RAM Differences between Spartan-3 Generation

Families
The distributed RAM is identical among all Spartan-3 generation families. There are
different amounts of distributed RAM per device (see Table 6-1). The performance varies
slightly between families due to minor variations in processing and characterization.

Table 6-1: Distributed RAM Resources by FPGA Family and Device

Feature Distributed RAM Blocks Distributed RAM Bits
Extended Spartan-3A Family
XC3SD1800A 16,640 266,240
XC3SD3400A 23,872 381,952
XC3S50A/AN 704 11,264
XC3S200A/AN 1,792 28,672
XC3S400A/AN 3,584 57,344
XC3S700A/AN 5,888 94,208
XC3S1400A/AN 11,264 180,224
Spartan-3E Family
XC3S100E 960 15,360
XC3S250E 2,448 39,168
XC3S500E 4,656 74,496
XC3S1200E 8,672 138,752
XC3S1600E 14,752 236,032
Spartan-3 Family
XC3S50 768 12,288
XC3S200 1,920 30,720
XC3S400 3,584 57,344
XC3S1000 7,680 122,880
XC3S1500 13,312 212,992
XC3S2000 20,480 327,680

218 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Compatibility with Other Xilinx FPGA Families

Table 6-1: Distributed RAM Resources by FPGA Family and Device (Cont’d)
Feature Distributed RAM Blocks Distributed RAM Bits
XC3S4000 27,648 442,368
XC3S5000 33,280 532,480

Compatibility with Other Xilinx FPGA Families

Each Spartan-3 generation distributed RAM operates identically to the distributed RAM
found in Virtex®, Virtex-E, Spartan-II, Spartan-IIE, Virtex-II, and Virtex-II Pro FPGAs.
Table 6-2 shows the basic memory capabilities embedded within the CLBs on various
Xilinx FPGA families. Like Virtex-II/Virtex-II Pro FPGAs, Spartan-3 generation CLBs have
eight LUTs and implement 128 bits of ROM memory. Like the Virtex/Virtex-E and
Spartan-II/Spartan-IIE FPGAs, Spartan-3 generation CLBs have 64 bits of distributed
RAM. Although the Spartan-3 and Virtex-II/Virtex-II Pro FPGA CLBs are identical for
logic functions, the Spartan-3 generation CLBs have half the amount of distributed RAM
within each CLB.

Table 6-2: Distributed Memory Features by FPGA Family

Virtex/Virtex-E, Virtex-II,
Spartan-3 Virtex-4 Virtex-5
Feature Spartan-II/Spartan-IIE Virtex-II Pro
Generation Family Family
Families Families
LUTs per CLB 8 4 8 8 8
ROM bits per CLB 128 64 128 128 256
Single-port RAM bits per CLB 64 64 128 64 256
Dual-port RAM bits per CLB 32 32 64 32 128

Table 6-3 lists the various single- and dual-port distributed RAM primitives supported by
the different Xilinx FPGA families. For each type of RAM, the table indicates how many
instances of a particular primitive fit within a single CLB. For example, two 32x1 single-
port RAM primitives fit in a single Spartan-3 generation CLB. Similarly, two 16x1 dual-
port RAM primitives fit in a Spartan-3 generation CLB but a single 32x1 dual-port RAM
primitive does not.

Table 6-3: Single- and Dual-port RAM Primitives Supported in a CLB by Family
Single-Port RAM Dual-Port RAM
Family
16x1 32x1 64x1 128x1 16x1 32x1 64x1
Spartan-3 Generation FPGAs 4 2 1 - 2 - -
Spartan-II/Spartan-IIE FPGAs 4 2 1 - 2 - -
Virtex/Virtex-E FPGAs
Virtex-II/Virtex-II Pro FPGAs 8 4 2 1 4 2 1
Virtex-4 FPGAs 4 2 1 - 2 - -
Virtex-5 FPGAs 8 6 4 2 4 4 2

Spartan-3 Generation FPGA User Guide www.xilinx.com 219

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

Library Primitives
There are four library primitives that support Spartan-3 generation distributed RAM,
ranging from 16 bits deep to 64 bits deep. All the primitives are one bit wide. Three
primitives are single-port RAMs and one primitive is dual-port RAM, as shown in
Table 6-4.

Table 6-4: Single-Port and Dual-Port Distributed RAMs

RAM Size
Primitive Type Address Inputs
(Depth x Width)
RAM16X1S 16 x 1 Single-port A3, A2, A1, A0
RAM32X1S 32 x 1 Single-port A4, A3, A2, A1, A0
RAM64X1S 64 x 1 Single-port A5, A4, A3, A2, A1, A0
RAM16X1D 16 x 1 Dual-port A3, A2, A1, A0

The input and output data are one bit wide. However, several distributed RAMs,
connected in parallel, easily implement wider memory functions.
Figure 6-3 shows generic single-port and dual-port distributed RAM primitives. The
A[#:0] and DPRA[#:0] signals are address buses.

RAMyX1S RAMyX1D
D D
O SPO
WE WE
WCLK WCLK
R/W Port
A[#:0] A[#:0]

DPO
DPRA[#:0]
Read Port

X464_03_062503

Figure 6-3: Single-Port and Dual-Port Distributed RAM Primitives

Table 6-5: Dual-Port RAM Function

Inputs Outputs
WE (mode) WCLK D SPO DPO
0 (read) X X data_a data_d
1 (read) 0 X data_a data_d
1 (read) 1 X data_a data_d
1 (write) ↑ D D data_d
1 (read) ↓ X data_a data_d

Notes:
1. data_a = word addressed by bits A#-A0.
2. data_d = word addressed by bits DPRA#-DPRA0.

220 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Signal Ports

As shown in Table 6-6, wider library primitives are available for 2-bit and 4-bit RAMs.

Table 6-6: Wider Library Primitives

RAM Size
Primitive Data Inputs Address Inputs Data Outputs
(Depth x Width)
RAM16X2S 16 x 2 D1, D0 A3, A2, A1, A0 O1, O0
RAM32X2S 32 x 2 D1, D0 A4, A3, A2, A1, A0 O1, O0
RAM16X4S 16 x 4 D3, D2, D1, D0 A3, A2, A1, A0 O3, O2, O1, O0

Signal Ports
Each distributed RAM port operates independently of the other while reading the same set
of memory cells.

Clock — WCLK
The clock is used for synchronous writes. The data and the address input pins have setup
times referenced to the WCLK pin. Active on the positive edge by default with built-in
programmable polarity.

Enable — WE
The enable pin affects the write functionality of the port. An inactive Write Enable prevents
any writing to memory cells. An active Write Enable causes the clock edge to write the data
input signal to the memory location pointed to by the address inputs. Active High by
default with built-in programmable polarity.

Address — A0, A1, A2, A3 (A4, A5, A6, A7)

The address inputs select the memory cells for read or write. The width of the port
determines the required address inputs.
Note: The address inputs are not a bus in VHDL or Verilog instantiations.

Dual-Port Read Address — DPRA0, DPRA1, DPRA2, DPRA3

On the RAM16X1D, the dual-port address inputs select the memory cells for reading on
the DPO output. Does not affect the write process.

Data In — D
The data input provides the new data value to be written into the RAM.

Data Out — O, SPO, and DPO

The data output O on single-port RAM or the SPO and DPO outputs on dual-port RAM
reflects the contents of the memory cells referenced by the address inputs. Following an
active write clock edge, the data out (O or SPO) reflects the newly written data. Registered
outputs use the available flip-flop within the SLICEM element.

Spartan-3 Generation FPGA User Guide www.xilinx.com 221

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

Inverting Control Pins

The two control pins, WCLK and WE, each have an individual inversion option. Any
control signal, including the clock, can be active at logic level 0 (negative edge for the clock)
or at logic level 1 (positive edge for the clock) without requiring other logic resources.

Global Set/Reset — GSR

The global set/reset (GSR) signal does not affect distributed RAM modules.

Global Write Enable — GWE

The global write enable signal, GWE, is asserted automatically at the end of device
configuration to enable all writable elements. The GWE signal guarantees that the
initialized distributed-RAM contents are not disturbed during the configuration process.
GWE is also used to ensure that Distributed RAM maintains its value during the
Extended Spartan-3A family FPGA Suspend mode.
Because GWE is a global signal and automatically connected throughout the device, the
distributed RAM primitive does not have a GWE input pin.

Attributes

Content Initialization — INIT

By default, distributed RAM is initialized with all zeros during the device configuration
sequence. To specify [non-zero] initial memory contents after configuration, use the INIT
attributes. Each INIT is a hexadecimal-encoded bit vector, arranged from most-significant
to least-significant bit. In other words, the right-most hexadecimal character represents
RAM locations 3, 2, 1, and 0. Table 6-7 shows the length of the INIT attribute for selected
primitives.

Table 6-7: INIT Attributes Length

Primitive Template INIT Attribute Length
RAM16X1S RAM_16S 4 digits
RAM32X1S RAM_32S 8 digits
RAM64X1S RAM_64S 16 digits
RAM16X1D RAM_16D 4 digits

The INIT attribute is required for any ROM instantiation. The ROM is initialized to the
INIT value at configuration and does not change during operation. For example, on a
ROM16X1, the parameter INIT = 10A7 produces the following datastream:
0001 0000 1010 0111

222 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Attributes

Placement Location — LOC

Each Spartan-3 generation CLB contains four slices, each with its own location coordinate,
as shown in Figure 6-4. Distributed RAM fits only in SLICEM slices. The ‘M’ in SLICEM
indicates that the slice supports memory-related functions and distinguishes SLICEMs
from SLICELs. The ‘L’ indicates that the slice supports logic only although the SLICEL can
also support ROM.

Configurable Logic Block (CLB)

LUT Reg LUT Reg

LUT Reg LUT Reg

X0Y1 X1Y1

LUT Reg LUT Reg

LUT Reg LUT Reg

X0Y0 X1Y0

SLICEM SLICEL
Logic/ROM Logic/ROM only
Distributed RAM
Shift Register x464_04_070803

Figure 6-4: SLICEM Slices within a Spartan-3 Generation CLB

When a LOC property is assigned to a distributed RAM instance, the Xilinx ISE® software
places the instance in the specified location. Figure 6-4 shows the X,Y coordinates for the
slices in a Spartan-3 generation CLB. Again, only SLICEM slices support memory.
Distributed RAM placement locations use the slice location naming convention, allowing
LOC properties to transfer easily from array to array.
For example, the single-port RAM16X1S primitive fits in any LUT within any SLICEM. To
place the instance U_RAM16 in slice X0Y0, use the following LOC assignment:
INST "U_RAM16" LOC = "SLICE_X0Y0";
The 16x1 dual-port RAM16X1D primitive requires both 16x1 LUT RAMs within a single
SLICEM slice, as shown in Figure 6-5. The first 16x1 LUT RAM, with output SPO,
implements the read/write port controlled by address A[3:0] for read and write. The
second LUT RAM implements the independent read-only port controlled by address
DPRA[3:0]. Data is presented simultaneously to both LUT RAMs, again controlled by
address A[3:0], WE, and WCLK.

Spartan-3 Generation FPGA User Guide www.xilinx.com 223

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

SLICEM

D 16x1 SPO
LUT
A[3:0] RAM
(Read/
Write) Optional
WE Register

WCLK

DPO
16x1
LUT
RAM
DPRA[3:0] (Read Optional
Only) Register

x464_05_062603

Figure 6-5: RAM16X1D Placement

A 32x1 single-port RAM32X1S primitive fits in one slice, as shown in Figure 6-6. The 32 bits
of RAM are split between two 16x1 LUT RAMs within the SLICEM slice. The A4 address
line selects the active LUT RAM via the F5MUX multiplexer within the slice.

SLICEM

D Optional
16x1 Register
A[3:0] LUT
A4 RAM
WE
WCLK
F5MUX

16x1
LUT
RAM Optional
Register

x464_06_062603

Figure 6-6: RAM32X1S Placement

224 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Distributed RAM Design Entry

The 64x1 single-port RAM64X1S primitive occupies both SLICEM slices in the CLB. The
read path uses both F5MUX and F6MUX multiplexers within the CLB.
Table 6-8 shows all Distributed RAM design elements and the number of slices required in
the Spartan-3 generation FPGA families.

Table 6-8: Distributed RAM Design Element and Required Slices

Element Slices Element Slices Element Slices
RAM16X1D 1 RAM32X1S_1 1 RAM128X1S_1 4
RAM16X1D_1 1 RAM32X2S 2 ROM16X1 0.5
RAM16X1S 0.5 RAM32X4S 4 ROM32X1 1
RAM16X1S_1 0.5 RAM32X8S 8 ROM64X1 2
RAM16X2S 1 RAM64X1S 2 ROM128X1 4
RAM16X4S 2 RAM64X1S_1 2 ROM256X1 8
RAM16X8S 4 RAM64X2S 4
RAM32X1S 1 RAM128X1S 4

Distributed RAM Design Entry

To specify distributed RAM in an application, use one of the various design entry tools,
including the Xilinx CORE Generator software or VHDL or Verilog.

Xilinx CORE Generator System

The Xilinx CORE Generator system creates distributed memory designs for both single-
port and dual-port RAMs, ROMs, and even SRL16 shift-register functions.
The Distributed Memory module is parameterizable; the depth can range from 16 to 65536
words in multiples of 16, and the width of each word can be anywhere in the range of 1 bit
to 1024 bits. To create a module, specify the component name and choose to include or
exclude control inputs, then choose the active polarity for the control inputs. Options are
available for simple registering of inputs and outputs. Optional asynchronous and
synchronous resets are available for the output registers.
Optionally, specify the initial memory contents. Unless otherwise specified, each memory
location initializes to zero. Enter user-specified initial values via a Memory Initialization
File, consisting of one line of binary data for every memory location. A default file is
generated by the CORE Generator system. Alternatively, create a coefficients file (.coe) as
shown in Figure 6-7, which not only defines the initial contents in a radix of 2, 10, or 16, but
also defines all the other control parameters for the CORE Generator system.

memory_initialization_radix=16;
memory_initialization_vector= 80, 0F, 00, 0B, 00, 0C, …, 81;

Figure 6-7: A Simple Coefficients File (.coe) Example for a Byte-Wide Memory

The output from the CORE Generator system includes a report on the options selected and
the device resources required. If a very deep memory is generated, then some external
multiplexing might be required; these resources are reported as the number of logic slices

Spartan-3 Generation FPGA User Guide www.xilinx.com 225

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

required. For simulation purposes, the CORE Generator system creates VHDL or Verilog
behavioral models.
The CORE Generator FIFO Generator supports both distributed and block RAMs.
• CORE Generator: Distributed Memory Module
http://www.xilinx.com/support/documentation/ip_documentation/dist_mem_gen_ds322.pdf
• CORE Generator: FIFO Generator
http://www.xilinx.com/support/documentation/ip_documentation/fifo_generator_ds317.pdf

VHDL and Verilog

VHDL and Verilog synthesis-based designs can either infer or directly instantiate
distributed RAM, depending on the specific logic synthesis tool used to create the design.

Inferring Distributed RAM

Most VHDL and Verilog logic synthesis tools, such as the Xilinx Synthesis Tool (XST) and
Synplicity Synplify, infer distributed RAM based on the hardware described. The Xilinx
ISE Project Navigator includes templates for inferring distributed RAM in your design. To
use the templates within Project Navigator, select Edit Æ Language Templates from the
menu, and then select VHDL or Verilog, followed by Synthesis Constructs Æ Coding
Examples Æ RAM from the selection tree. Finally, select the preferred distributed RAM
template. Cut and paste the template into the source code for the application and modify it
as appropriate.

VHDL Inference Template Example

process (<clock>)
begin
if (<clock>'event and <clock> = '1') then
if (<write_enable> = '1') then
<ram_name>(conv_integer(<address>)) <= <input_data>;
end if;
end if;
end process;

<ram_output> <= <ram_name>(conv_integer(<address>));

Verilog Inference Template Example

parameter RAM_WIDTH = <ram_width>;
parameter RAM_ADDR_BITS = <ram_addr_bits>;

reg [RAM_WIDTH-1:0] <ram_name> [(2**RAM_ADDR_BITS)-1:0];

wire [RAM_WIDTH-1:0] <output_data>;

<reg_or_wire> [RAM_ADDR_BITS-1:0] <address>;

<reg_or_wire> [RAM_WIDTH-1:0] <input_data>;

always @(posedge <clock>)

if (<write_enable>)
<ram_name>[<address>] <= <input_data>;

assign <output_data> = <ram_name>[<address>];

226 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Distributed RAM Design Entry

It is still possible to directly instantiate distributed RAM, even if portions of the design
infer distributed RAM.

Instantiation Templates
For VHDL- and Verilog-based designs, various instantiation templates are available to
speed development. Within the Xilinx ISE Project Navigator, select Edit Æ Language
Templates from the menu, and then select VHDL or Verilog, followed by Device
Primitive Instantiation Æ FPGA Æ RAM/ROM Æ Distributed RAM from the selection
tree. Cut and paste the template into the source code for the application and modify it as
appropriate.
There are also downloadable VHDL and Verilog templates available for all single-port and
dual-port primitives. The RAM_xS templates (where x = 16, 32, or 64) are single-port
modules and instantiate the corresponding RAMxX1S primitive. The ‘S’ indicates single-
port RAM. The RAM_16D template is a dual-port module and instantiates the
corresponding RAM16X1D primitive. The ‘D’ indicates dual-port RAM.
• VHDL Distributed RAM Templates
xapp464_vhdl.zip
• Verilog Distributed RAM Templates
xapp464_verilog.zip
The following are single-port templates:
• RAM_16S
• RAM_32S
• RAM_64S
The following is a dual-port template:
• RAM_16D
In VHDL, each template has a component declaration section and an architecture section.
Insert both sections of the template within the VHDL design file. The port map of the
architecture section must include the design signal names.
Templates for the RAM_16S module are provided below as examples in both VHDL and
Verilog code.

VHDL Instantiation Template Example

--- RAM16X1S : In order to incorporate this function into the design,
-- VHDL : the following instance declaration needs to be placed
-- instance : in the architecture body of the design code. The
-- declaration : instance name (RAM16X1S_inst) and/or the port
-- code : declarations after the "=>" assignment maybe changed
-- : to properly reference and connect this function to the
-- : design. All inputs and outputs must be connected.

-- Library : In addition to adding the instance declaration, a use

-- declaration : statement for the UNISIM.vcomponents library needs to
-- for : be added before the entity declaration. This library
-- Xilinx : contains the component declarations for all Xilinx
-- primitives : primitives and points to the models that will be used
-- : for simulation.

-- Copy the following two statements and paste them before the
-- Entity declaration, unless they already exist.

Spartan-3 Generation FPGA User Guide www.xilinx.com 227

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

Library UNISIM;
use UNISIM.vcomponents.all;

-- <-----Cut code below this line and paste into architecture body---->

-- RAM16X1S: 16 x 1 posedge write distributed => Distributed RAM

-- Xilinx HDL Language Template

RAM16X1S_inst : RAM16X1S
generic map (
INIT => X"0000")
port map (
O => O, -- RAM output
A0 => A0, -- RAM address[0] input
A1 => A1, -- RAM address[1] input
A2 => A2, -- RAM address[2] input
A3 => A3, -- RAM address[3] input
D => D, -- RAM data input
WCLK => WCLK, -- Write clock input
WE => WE -- Write enable input
);

-- End of RAM16X1S_inst instantiation

Verilog Instantiation Template Example

// RAM16X1S : In order to incorporate this function into the design,
// Verilog : the following instance declaration needs to be placed
// instance : in the body of the design code. The instance name
// declaration : (RAM16X1S_inst) and/or the port declarations within
// code : the parenthesis may be changed to properly reference and
// : connect this function to the design. All inputs
// : and outputs must be connected.

// <-----Cut code below this line---->

// RAM16X1S: 16 x 1 posedge write distributed (LUT) RAM

// Xilinx HDL Language Template

RAM16X1S #(
.INIT(16'h0000) // Initial contents of RAM
) RAM16X1S_inst (
.O(O), // RAM output
.A0(A0), // RAM address[0] input
.A1(A1), // RAM address[1] input
.A2(A2), // RAM address[2] input
.A3(A3), // RAM address[3] input
.D(D), // RAM data input
.WCLK(WCLK), // Write clock input
.WE(WE) // Write enable input
);

// End of RAM16X1S_inst instantiation

228 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Conclusion

Wider Distributed RAM Modules

Table 6-9 shows the VHDL and Verilog distributed RAM examples that implement n-bit-
wide memories.
Table 6-9: VHDL and Verilog Submodules
Submodules Primitive Size Type
XC3S_RAM16XN_S_SUBM RAM16X1S 16 words x n-bit Single-port
XC3S_RAM32XN_S_SUBM RAM32X1S 32 words x n-bit Single-port
XC3S_RAM64XN_S_SUBM RAM64X1S 64 words x n-bit Single-port
XC3S_RAM16XN_D_SUBM RAM16X1D 16 words x n-bit Dual-port

Initialization in VHDL or Verilog Codes

Distributed RAM structures can be initialized in VHDL or Verilog code for both synthesis
and simulation. For synthesis, the attributes are attached to the distributed RAM
instantiation and are copied in the EDIF output file to be compiled by Xilinx ISE Series
tools. The VHDL code simulation uses a generic parameter to pass the attributes. The
Verilog code simulation uses a defparam parameter to pass the attributes.

Conclusion
Frequently FPGA designs require multiple small, fast, and flexible memories for system
configuration, control, and status functions. These memories are usually distributed
throughout the design. The distributed RAM in the Spartan-3 generation FPGAs is ideal
for such applications, and allows the CLBs to be changed from logic to memory "on
demand". These memories can then be linked together for various data width or depth
requirements. The Xilinx tools automatically use distributed RAM for small arrays or they
can be instantiated in a design.

Related Materials and References

The following list provides additional information:
• Chapter 4, “Using Block RAM”
• Chapter 5, “Using Configurable Logic Blocks (CLBs)”
• Chapter 7, “Using Look-Up Tables as Shift Registers (SRL16)”
• RAM and ROM Application Notes
www.xilinx.com/support/documentation/ram_and_rom.htm
• Distributed Memory Generator Xilinx IP Core
www.xilinx.com/xlnx/xebiz/designResources/ip_product_details.jsp?key=DIST_MEM_GEN
• Xilinx ISE Software Manuals
www.xilinx.com/support/documentation/dt_ise.htm

Spartan-3 Generation FPGA User Guide www.xilinx.com 229

UG331 (v1.7) August 19, 2010
 R

Chapter 6: Using Look-Up Tables as Distributed RAM

230 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 7

Using Look-Up Tables as Shift Registers

(SRL16)
Summary
The SRL16 is an alternative mode for the look-up tables where they are used as 16-bit shift
registers. Using this Shift Register LUT (SRL) mode can improve performance and rapidly
lead to cost savings of an order of magnitude. Although the SRL16 can be automatically
inferred by the software tools, considering their effective use can lead to more cost-
effective designs.

Shift Register Differences between Spartan-3 Generation Families

This chapter applies to all Spartan®-3 generation FPGA families. Each SRL16 shift register
is identical within a family, and the SRL16 function is identical among all
Spartan-3 generation families. The performance varies slightly between families due to
minor variations in processing and characterization. The number of SRL16 shift registers is
the same as the number of distributed RAM blocks, as shown in Table 6-1, page 218.

Introduction
Spartan-3 generation FPGAs can configure the look-up table (LUT) in a SLICEM slice as a
16-bit shift register without using the flip-flops available in each slice. Shift-in operations
are synchronous with the clock, and output length is dynamically selectable. A separate
dedicated output allows the cascading of any number of 16-bit shift registers to create
whatever size shift register is needed. Each CLB resource can be configured using four of
the eight LUTs as a 64-bit shift register.
This document provides generic VHDL and Verilog submodules and reference code
examples for implementing from 16-bit up to 64-bit shift registers. These submodules are
built from 16-bit shift-register primitives and from dedicated MUXF5, MUXF6, and
MUXF7 multiplexers.
These shift registers enable the development of efficient designs for applications that
require delay or latency compensation. Shift registers are also useful in synchronous FIFO
and Content-Addressable Memory (CAM) designs. To quickly generate a Spartan-3 shift
register without using flip-flops (i.e., using the SRL16 element(s)), use the CORE Generator
RAM-based Shift Register module.

Spartan-3 Generation FPGA User Guide www.xilinx.com 231

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

Shift Register Architecture

The structure of the SRL16 is described from the bottom up, starting with the shift register
and then building up to the surrounding FPGA structure.

LUT Structure
The LUT can be described as a 16:1 multiplexer with the four inputs serving as binary
select lines, and the values programmed into the LUT serving as the data being selected
(see Figure 7-1).

1 0 1 1 1 0 0 0 1 1 1 0 1 0 0 1

A[3:0] 0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111

D x465_01_070603

Figure 7-1: LUT Modeled as a 16:1 Multiplexer

With the SRL16 configuration, the fixed LUT values are configured instead as an
addressable shift register (see Figure 7-2). The shift register inputs are the same as those for
the synchronous RAM configuration of the LUT: a data input, clock, and clock enable (not
shown). A special output for the shift register is provided from the last flip-flop, called Q15
on the library primitives or MC15 in the FPGA Editor. The LUT inputs asynchronously (or
dynamically) select one of the 16 storage elements in the shift register.

Q15 or
MC15
DIN DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ DQ
CLK

A[3:0] 0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111

D x465_02_040203

Figure 7-2: LUT Configured as an Addressable Shift Register

Dynamic Length Adjustment

The address can be thought of as dynamically changing the length of the shift register. If D
is used as the shift register output instead of Q15, setting the address to 7 (0111) selects Q7
as the output, emulating an 8-bit shift register. Note that since the address lines control the
mux, they provide an asynchronous path to the output.

Logic Cell Structure

The F-LUT and the G-LUT in the SLICEM are used as the basis of the SRL16 (see the details
of the CLB structure in Figure 5-2, page 206). The SLICEM LUTs cascade from the G-LUT
MC15 output to the F-LUT DI input through the DIFMUX. The SHIFTIN and SHIFTOUT
lines cascade a SLICEM to the SLICEM below through the DIGMUX to form larger shift
registers.

232 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Architecture

Each shift register provides a shift output MC15 for the last bit in each LUT, in addition to
providing addressable access to any bit in the shift register through the normal D output
(Figure 7-3). The address inputs A[3:0] are the same as the distributed RAM address lines,
which come from the LUT inputs F[4:1] or G[4:1].

SHIFTIN SRLC16

SHIFT-REG
4 D Output
A[3:0] A[3:0]
MC15 Registered
D Q
WS DI Output

DI (BY)
WSG (optional)

CE (SR) WE
CLK CK

SHIFTOUT
or YB X465_03_040203

Figure 7-3: Logic Cell SRL Structure

Registered Output
Each SRL16 LUT has an associated flip-flop that makes up the overall logic cell. The
addressable bit of the shift register can be stored in the flip-flop for a synchronous output
or can be fed directly to a combinatorial output of the CLB. When using the register, it is
best to have fixed address lines selecting a static shift register length to avoid timing
hazards. The CLB flip-flop can be used to provide one more shift delay for the addressable
bit. Since the clock-to-output delay of the flip-flop is faster than the shift register,
performance can be improved by addressing the second-to-last bit and then using the flip-
flop as the last stage of the shift register. Using the flip-flop also allows for asynchronous or
synchronous set or reset of the output.
The shift register input can come from a dedicated SHIFTIN signal, and the Q15/MC15
signal from the last stage of the shift register can drive a SHIFTOUT output. The
addressable D output is available in all SRL primitives, while the Q15/MC15 signal that
can drive SHIFTOUT is only available in the cascadable SRLC16 primitive.
The SRL16 can shift from either LSB to MSB or MSB to LSB according to the application.
Although the device arbitrarily names the output MC15, it can be the LSB of the user
function.

Slice Structure
The two logic cells within a slice are connected for cascading a shift register up to 32 bits
(see Figure 7-4). These connect the Q15/MC15 of the first shift register to the DI (or Q0 flip-
flop) of the second shift register.

Spartan-3 Generation FPGA User Guide www.xilinx.com 233

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

SLICEM

LC
SRL16
MC15

DI

SRL16
LC

UG331_c9_04_072406

Figure 7-4: Shift Register Connections Between Logic Cells in a Slice

If dynamic addressing (or "dynamic length adjustment") is desired, the two separate data
outputs from each SRL16 must be multiplexed together. One of the two SRL16 bits can be
selected by using the F5MUX to make the selection (see Figure 7-5).

SRL16
LC F5MUX

A[3:0]

SRL16
LC

A4
X465_05_070703

Figure 7-5: Using F5MUX for Addressing Multiple SRL16 Components

CLB Structure
The Spartan-3 generation CLB contains four slices, each with two LUTs, but only two allow
LUTs to be used as SRL16 components or distributed RAM. The two left-hand SLICEM
components allow their two LUTs to be configured as a 16-bit shift register. SHIFTOUT to
SHIFTIN connections are available to cascade the two SLICEM components. The four left-
hand LUTs of a single CLB can be combined to produce delays up to 64 clock cycles (see
Figure 7-6). It is also possible to combine shift registers across more than one CLB.

234 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Architecture

1 Shift Chain
in CLB
IN DI D FF
SRLC16
MC15

DI D FF
SRLC16
MC15
SLICEM S1

SHIFTOUT

SHIFTIN
DI D FF
SRLC16
MC15

DI D FF
SRLC16
MC15
SLICEM S0

OUT
CASCADABLE OUT
X465_06_040503

Figure 7-6: Cascading Shift Register LUTs in a CLB

The multiplexers can be used to address multiple SLICEMs similar to the description for
combining the two LUTs within a SLICEM. The F6MUX can be used to select from three or
four SRL16 components in a CLB, providing up to 64 bits of addressable shift register (see
Figure 7-7).

Spartan-3 Generation FPGA User Guide www.xilinx.com 235

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

D
LUT

F5

LUT
SLICEM S1

F6
LUT

F5

LUT
SLICEM S0 CLB
X465_07_040203

Figure 7-7: Using F6MUX to Address a 64-Bit Shift Register

Library Primitives
The shift register element is known as the SRL16 (Shift Register LUT 16-bit), with a C
added to signify a cascade ability (Q15 output) and E to indicate a clock enable. See
Figure 7-8 for an example of the SRLC16E component.

D SRLC16E

CE Q
CLK Q15
A0
A1
A2
A3

X465_19_040503

Figure 7-8: SRLC16E Primitive

Eight library primitives are available that offer optional clock enable (CE), inverted clock
(CLK), and cascadable output (Q15) combinations.

236 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Library Primitives

Table 7-1 lists all of the available primitives for synthesis and simulation.
Table 7-1: Shift Register Primitives
Primitive Length Control Address Inputs Output
SRL16 16 bits CLK A3, A2, A1, A0 Q
SRL16E 16 bits CLK, CE A3, A2, A1, A0 Q
SRL16_1 16 bits CLK A3, A2, A1, A0 Q
SRL16E_1 16 bits CLK, CE A3, A2, A1, A0 Q
SRLC16 16 bits CLK A3, A2, A1, A0 Q, Q15
SRLC16E 16 bits CLK, CE A3, A2, A1, A0 Q, Q15
SRLC16_1 16 bits CLK A3, A2, A1, A0 Q, Q15
SRLC16E_1 16 bits CLK, CE A3, A2, A1, A0 Q, Q15

Initialization in VHDL and Verilog Code

A shift register can be initialized in VHDL or Verilog code for both synthesis and
simulation. For synthesis, the INIT attribute is attached to the 16-bit shift register
instantiation and is copied in the EDIF output file to be compiled by Xilinx tools. The
VHDL code simulation uses a generic parameter to pass the attributes. The Verilog code
simulation uses a defparam parameter to pass the attributes.
The S3_SRL16E shift register instantiation code examples (in VHDL and Verilog) illustrate
these techniques (see “VHDL and Verilog Templates,” page 245). S3_SRL16E.vhd and
S3_SRL16E.v files are not a part of the documentation.

Port Signals
Clock — CLK
Either the rising edge or the falling edge of the clock is used for the synchronous shift-in.
The data and clock enable input pins have set-up times referenced to the chosen edge of
CLK.

Data In — D
The data input provides new data (one bit) to be shifted into the shift register.

Clock Enable — CE (optional)

The clock enable pin affects shift functionality. An inactive clock enable pin does not shift
data into the shift register and does not write new data. Activating the clock enable allows
the data in (D) to be written to the first location and all data to be shifted by one location.
When available, new data appears on output pins (Q) and the cascadable output pin (Q15).

Address — A3, A2, A1, A0

Address inputs select the bit (range 0 to 15) to be read. The nth bit is available on the output
pin (Q). Address inputs have no effect on the cascadable output pin (Q15), which is always
the last bit of the shift register (bit 15).

Spartan-3 Generation FPGA User Guide www.xilinx.com 237

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

Data Out — Q
The data output Q provides the data value (1 bit) selected by the address inputs.

Data Out — Q15 (optional)

The data output Q15 provides the last bit value of the 16-bit shift register. New data
becomes available after each shift-in operation.

Inverting Control Pins

The two control pins (CLK, CE) have an individual inversion option. The default is the
rising clock edge and active High clock enable.

GSR
The global set/reset (GSR) signal has no impact on shift registers.

Attributes
Content Initialization — INIT
The INIT attribute defines the initial shift register contents. The INIT attribute is a hex-
encoded bit vector with four digits (0000). The left-most hexadecimal digit is the most
significant bit. By default the shift register is initialized with all zeros during the device
configuration sequence, but any other configuration value can be specified.

Location Constraints
Figure 7-9 shows how the slices are arranged within a CLB. Each CLB has four slices, but
only the two at the bottom-left of the CLB can be used as shift registers. These are both
designated SLICEM in CLB positions S0 and S1. The relative position coordinates are X0Y0
and X0Y1. To constrain placement, these coordinates can be used in a LOC property
attached to the SRL primitive. Note that the dedicated CLB shift chain runs from the top to
the bottom, but the start and end of the shift register can be in any of the four SLICEM
LUTs.

238 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Library Primitives

Left-Hand SLICEM Right-Hand SLICEL

(Logic or Distributed RAM (Logic Only)
or Shift Register)
COUT

CLB
SLICE
X1Y1

SLICE
X1Y0
COUT
Switch Interconnect
Matrix CIN to Neighbors
SLICE
X0Y1
SHIFTOUT
SHIFTIN
SLICE
X0Y0

CIN X465_08_040203

Figure 7-9: Arrangement of Slices within the CLB

Shift Register Operations

The functionality of the shift register is shown in Table 7-2. The SRL16 shifts on the rising
edge of the clock input when the Clock Enable control is High. This shift register cannot be
initialized either during configuration or during operation except by shifting data into it.
The clock enable and clock inputs are shared between the two LUTs in a SLICEM. The
clock enable input is automatically kept active if unused.

Table 7-2: SRL16 Shift Register Function

Inputs Outputs
Am CLK CE D Q Q15
Am X 0 X Q[Am] Q[15]
Am ↑ 1 D Q[Am-1] Q[15]

Notes:
1. m = 0, 1, 2, 3.

Data Flow
Each shift register (SRL16 primitive) supports:
• Synchronous shift-in
• Asynchronous 1-bit output when the address is changed dynamically
• Synchronous shift-out when the address is fixed

Spartan-3 Generation FPGA User Guide www.xilinx.com 239

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

In addition, cascadable shift registers (SRLC16) support synchronous shift-out output of

the last (16th) bit. This output has a dedicated connection to the input of the next SRLC16
inside the CLB resource. Two primitives are illustrated in Figure 7-10.

SRLC16E

D Q
Address

SRL16E
CE
Q15
CLK
D Q
Address

CE SRLC16E

CLK D Q
Address

CE

CLK Q15

X465_09_070703

Figure 7-10: Shift Register and Cascadable Shift Register

Shift Operation
The shift operation is a single clock-edge operation with an active-High clock enable
feature. When enable is High, the input (D) is loaded into the first bit of the shift register,
and each bit is shifted to the next highest bit position. In a cascadable shift register
configuration (such as SRLC16), the last bit is shifted out on the Q15 output.
The bit selected by the 4-bit address appears on the Q output.

Dynamic Read Operation

The Q output is determined by the 4-bit address. Each time a new address is applied to the
4-input address pins, the new bit position value is available on the Q output after the time
delay to access the LUT. This operation is asynchronous and independent of the clock and
clock enable signals.

240 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Inference

Figure 7-11 illustrates the shift and dynamic read operations.

CLK

CE

D
tshift

Q15

Shift Timing Diagram

Address 7 10

Q Position (7) Position (10)

taccess taccess

Dynamic Length Timing Diagram

X465_10_040203

Figure 7-11: Shift- and Dynamic-Length Timing Diagrams

Static Read Operation

If the 4-bit address is fixed, the Q output always uses the same bit position. This mode
implements any shift register length up 1 to 16 bits in one LUT. Shift register length is
(N+1) where N is the input address.
The Q output changes synchronously with each shift operation. The previous bit is shifted
to the next position and appears on the Q output.

Characteristics
• A shift operation requires one clock edge.
• Dynamic-length read operations are asynchronous (Q output).
• Static-length read operations are synchronous (Q output).
• The data input has a setup-to-clock timing specification.
• In a cascadable configuration, the Q15 output always contains the last bit value.
• The Q15 output changes synchronously after each shift operation.

Shift Register Inference

When a shift register is described in generic HDL code, synthesis tools infer the use of the
SRL16 component. Since the SRL16 does not have either synchronous or asynchronous set

Spartan-3 Generation FPGA User Guide www.xilinx.com 241

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

or reset inputs, and does not have access to all bits at the same time, using such capabilities
precludes the use of the SRL16, and the function is implemented in flip-flops. The
cascadable shift register (SRLC16) might be inferred if the shift register is larger than 16 bits
or if only the Q15 is used.
In fact, adding a reset is one way to force a synthesis tool to use flip-flops instead of the
SRL16 when flip-flops are preferred for performance or other reasons. If a reset is not
needed, simply connect a dummy signal and use an appropriate KEEP attribute to prevent
the synthesis tool from optimizing it out of the design.
Although the SRL16 shift register does not have a parallel load capability, an equivalent
function can be implemented simply by anticipating the load requirement and shifting in
the proper data. This requires predictable timing for the load command.

VHDL Inference Code

The following code infers an SRL16 in VHDL.
architecture Behavioral of srl16 is

signal Q_INT: std_logic_vector(15 downto 0);

begin

process(C)
begin
if (C’event and C=’1’) then
Q_INT <= Q_INT(14 downto 0) & D;
end if;
end process;

Q <= Q_INT(15);

end Behavioral;
An inverted clock (SRL16_1) is inferred by replacing C='1' with C='0'. A clock enable
(SRL16E) is inferred by inserting if (CE='1') then after the first if-then statement.

Verilog Inference Code

The following code infers an SRL16 in Verilog.
always @ (posedge C)
begin
Q_INT <= {Q_INT[14:0],D};
end

always @(Q_INT)
begin
Q <= Q_INT[15];
end
An inverted clock (SRL16_1) is inferred by replacing (posedge C) with (negedge C). A
clock enable (SRL16E) is inferred by inserting if(CE) after the begin statement.

242 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Submodules

Shift Register Submodules

In addition to the 16-bit primitives, two submodules that implement 32-bit and 64-bit
cascadable shift registers are provided in VHDL and Verilog code. Table 7-3 lists available
submodules.
Table 7-3: Shift Register Submodules
Submodule Length Control Address Inputs Output
SRLC32E_SUBM 32 bits CLK, CE A4, A3, A2, A1, A0 Q, Q31
SRLC64E_SUBM 64 bits CLK, CE A5, A4, A3, A2, A1, A0 Q, Q63

The submodules are based on SRLC16E primitives, which are associated with dedicated
multiplexers (MUXF5, MUXF6, and so forth). This implementation allows a fast static- and
dynamic-length mode, even for very large shift registers.
Figure 7-12 represents the cascadable shift registers (32-bit and 64-bit) implemented by the
submodules in Table 7-3.

A4 A5, A4 A5
5 6
Add. Add.
A3, A2, A1, A0 A3, A2, A1, A0 A4

D D Q D D Q
4 4
A[3:0] A[3:0]

CE CE

Q15 Q15
SRLC16E SRLC16E

MUXF5 MUXF5
D Q D Q
4 4
A[3:0] A[3:0]

CE CE

Q15 Q31 Q15

SRLC16E SRLC16E

Q
32-bit Shift Register
MUXF6
D Q
4
A[3:0]

CE

Q15
SRLC16E

MUXF5
D Q
4
A[3:0]

CE

Q15 Q63
SRLC16E

64-bit Shift Register X465_11_051505

Figure 7-12: Shift-Register Submodules (32-bit, 64-bit)

Spartan-3 Generation FPGA User Guide www.xilinx.com 243

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

All clock enable (CE) and clock (CLK) inputs are connected to one global clock enable and
one clock signal per submodule. If a global static- or dynamic-length mode is not required,
the SRLC16E primitive can be cascaded without multiplexers.

Fully Synchronous Shift Registers

All shift-register primitives and submodules do not use the register(s) available in the
same slice(s). To implement a fully synchronous read and write shift register, output pin Q
must be connected to a flip-flop. Both the shift register and the flip-flop share the same
clock, as shown in Figure 7-13.

FF

D Q Synchronous
D Q
SRLC16E Output
Address

CE (Write Enable)
CLK Q15

X465_12_040203

Figure 7-13: Fully Synchronous Shift Register

This configuration provides a better timing solution and simplifies the design. Because the
flip-flop must be considered to be the last register in the shift-register chain, the static or
dynamic address should point to the desired length minus one. If needed, the cascadable
output can also be registered in a flip-flop. The delay from the SRL16 to the flip-flop is a
fixed CLB setup time delay and is not controlled by a PERIOD constraint.

Static-Length Shift Registers

The cascadable 16-bit shift register implements any static length mode shift register
without the dedicated multiplexers (MUXF5, MUXF6, and so on). Figure 7-14 illustrates a
40-bit shift register. Only the last SRLC16E primitive needs to have its address inputs tied
to “0111”. Alternatively, shift register length can be limited to 39 bits (address tied to
“0110”) and a flip-flop can be used as the last register. (In an SRLC16E primitive, the shift
register length is the address input + 1.)

244 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Submodules

D D D D
LUT LUT
Q15 Q15
SRLC16 SRLC16

D D
LUT LUT
Q15 Q15
SRLC16 SRLC16

FF
D Q OUT D Q D Q OUT
"0111" 4 A[3:0] (40-bit SRL) "0110" A[3:0] (40-bit SRL)
LUT LUT
Q15 Q15
SRLC16 SRLC16
X465_13_051505

Figure 7-14: 40-bit Static-Length Shift Register

VHDL and Verilog Instantiation

VHDL and Verilog instantiation templates are available for all primitives and submodules:
• xapp465_vhdl.zip
• xapp465_verilog.zip
In VHDL, each template has a component declaration section and an architecture section.
Each part of the template should be inserted within the VHDL design file. The port map of
the architecture section should include the design signal names.
The ShiftRegister_C_x (with x = 16, 32, or 64) templates are cascadable modules and
instantiate the corresponding SRLCxE primitive (16) or submodule (32 or 64).
The ShiftRegister_16 template can be used to instantiate an SRL16 primitive.

VHDL and Verilog Templates

In template names, the number indicates the number of bits (for example,
SHIFT_SELECT_16 is the template for the 16-bit shift register) and the “C” extension
means the template is cascadable.
The following are templates for primitives:
• SHIFT_REGISTER_16
• SHIFT_REGISTER_16_C
The following are templates for submodules:
• SHIFT_REGISTER_32_C (submodule: SRLC32E_SUBM)
• SHIFT_REGISTER_64_C (submodule: SRLC64E_SUBM)
The corresponding submodules have to be synthesized with the design.
Templates for the SHIFT_REGISTER_16_C module are provided in VHDL and Verilog
code as an example.

Spartan-3 Generation FPGA User Guide www.xilinx.com 245

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

VHDL Template:
-- Module: SHIFT_REGISTER_C_16
-- Description: VHDL instantiation template
-- CASCADABLE 16-bit shift register with enable (SRLC16E)
-- Device: Spartan-3 Generation Family
---------------------------------------------------------------------
-- Components Declarations:
--
component SRLC16E
-- pragma translate_off
generic (
-- Shift Register initialization ("0" by default) for functional
simulation:
INIT : bit_vector := X"0000"
);
-- pragma translate_on
port (
D : in std_logic;
CE : in std_logic;
CLK : in std_logic;
A0 : in std_logic;
A1 : in std_logic;
A2 : in std_logic;
A3 : in std_logic;
Q : out std_logic;
Q15 : out std_logic
);
end component;
-- Architecture Section:
--
-- Attributes for Shift Register initialization (“0” by default):
attribute INIT: string;
--
attribute INIT of U_SRLC16E: label is “0000”;
--
-- ShiftRegister Instantiation
U_SRLC16E: SRLC16E
port map (
D => , -- insert input signal
CE => , -- insert Clock Enable signal (optional)
CLK => , -- insert Clock signal
A0 => , -- insert Address 0 signal
A1 => , -- insert Address 1 signal
A2 => , -- insert Address 2 signal
A3 => , -- insert Address 3 signal
Q => , -- insert output signal
Q15 => -- insert cascadable output signal
);

Verilog Template:
// Module: SHIFT_REGISTER_16
// Description: Verilog instantiation template
// Cascadable 16-bit Shift Register with Clock Enable (SRLC16E)
// Device: Spartan-3 Generation Family
//-------------------------------------------------------------------

defparam

246 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Shift Register Submodules

//Shift Register initialization ("0" by default) for functional

simulation:
U_SRLC16E.INIT = 16'h0000;

//SelectShiftRegister-II Instantiation
SRLC16E U_SRLC16E ( .D(),
.A0(),
.A1(),
.A2(),
.A3(),
.CLK(),
.CE(),
.Q(),
.Q15()
);

// synthesis attribute declarations

/* attribute
INIT "0000"
*/

CORE Generator System

The Xilinx CORE Generator system generates fast, compact, FIFO-style shift registers,
delay lines, or time-skew buffers using the SRL16. The RAM-based Shift Register module
shown in Figure 7-15 provides a very efficient multibit wide shift for widths up to 256 and
depths to 1024. Fixed-length shift registers and variable-length shift registers can be
created. An option is also provided to register the outputs of the module. If output
registering is selected, there are additional options for Clock Enable, Asynchronous Set,
Clear, and Init, and Synchronous Set, Clear, and Init of the output register. The module can
optionally be generated as a relationally placed macro (RPM) or as unplaced logic.

ASET SSET
D[N:0] Q[N:0]
A[M:0]
CE

CLK

ACLR SCLR AINIT SINIT

x465_14_040203

Figure 7-15: CORE Generator RAM-Based Shift Register Module

Spartan-3 Generation FPGA User Guide www.xilinx.com 247

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

Applications

Delay Lines
The register-rich nature of the Xilinx FPGA architecture allows for the addition of pipeline
stages to increase throughput. Data paths must be balanced to keep the desired
functionality. The SRL16 can be used when additional clock cycles of delay are needed
anywhere in the design (see Figure 7-16).

12 Cycles

Operation A Operation B

4 Cycles 8 Cycles

Operation C

3 Cycles

9-cycle imbalance
3 Cycles

12 Cycles

Operation A Operation B

4 Cycles 8 Cycles

Operation C Pipeline

9 Cycles
3 Cycles using SRL16

Paths statically
balanced
12 Cycles X465_20_040603

Figure 7-16: Using SRL16 as a Delay Line

Linear Feedback Shift Registers

Linear Feedback Shift Registers (LFSRs) sequence through 2n-1 states, where n is the
number of flip-flops. The sequence is created by feeding specific bits back through an XOR
or XNOR gate. LFSRs can replace conventional binary counters in performance critical
applications where the count sequence is not important (e.g., FIFOs). LFSRs are also used
as pseudo-random number generators. They are important building blocks in encryption
and decryption algorithms.
Maximal-length LFSRs need taps taken from specific positions within the shift register.
There are multiple ways these taps can be made available in the SRL16 configuration. One
is by addressing the necessary bit in a given SRL16 while allowing the Q15 to cascade to
the next SRL16. Another is to use flip-flops to "extend" the SRL16 where necessary to access
the tap points. For example, Figure 7-17 shows how a 52-bit LFSR can be implemented

248 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Applications

with the feedback coming from bits 49 and 52. A third method is to duplicate the LFSR in
multiple SRLs and address different bits from each one. Yet another method is to generate
multiple addresses in one SRL clock cycle to capture multiple bit positions. The XNOR gate
required for any LFSR can be conveniently located in the SLICEL part of the CLB. More
detail is available in XAPP210.

Bit 52

Bit 1
D Q
XNOR
Bit 49

Bit 17
D Q Output
D Q
SRL16
Bit 52

Address = 15

Bit 33
D Q D Q
SRL16 Bit 51

Address = 15

Bit 49
D Q D Q
SRL16 Bit 50

Address = 15
x465_15_040203

Figure 7-17: 52-bit LFSR

Gold Code Generator

Gold code generators are used in CDMA systems to generate code sequences with good
correlation properties (see Figure 7-18). R. Gold suggested that sets of small correlation
codes could be generated by modulo 2 addition of the results of two LFSRs, primed with
factor codes. The result is a set of codes ideally suited to distinguish one code from another
in a spectrum full of coded signals. Figure 7-18 shows an implementation of a Gold code
generator. The logic required to initially fill the LFSR and provide the feedback can be
located in the SLICEL parts of the CLB. See XAPP217 for more details.

LFSR 1

Gold Code Out

LFSR 2
x465_16_051505

Figure 7-18: Gold Code Generator

Spartan-3 Generation FPGA User Guide www.xilinx.com 249

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

FIFOs
Synchronous FIFOs can be built out of the SRL16 components. These are useful when other
resources become scarce, providing up to 64 bits per CLB. For larger FIFOs, the block RAM
is the most efficient resource to use. See XAPP256 for more detail.

CLK
SRL16
SINIT Based DATA_OUT
FIFO
DATA_IN Address
Counter
RD_EN
FIFO FIFO_COUNT
WR_EN Count

Status FULL
Flag
Generation EMPTY

x465_17_051505

Figure 7-19: Synchronous FIFO Using SRLC16 Shift Registers

Counters
Any desired repeated sequence of 16 states can be achieved by feeding each output with an
SRL16. Cascading the SRL16 allows even longer arbitrary count sequences. A terminal
count can be generated by using the standard carry chain (see Figure 7-20).

TC

Q3
SRL 0 1

Q2
SRL 0 1

Q1
SRL 0 1

Q0
SRL VCC
0 1

x465_18_051505

Figure 7-20: SRL-Based Counter with Terminal Count

250 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

Related Materials and References

The following documents provide supplementary information useful with this chapter:
• XAPP210: Linear Feedback Shift Registers in Virtex® Devices
Linear Feedback Shift Registers are very efficient counters in the FPGA architecture.
Using the SRL16 as the basis of the shift register, a 15-bit counter can fit in one slice
and a 52-bit counter in two slices.
• XAPP211: PN Generators Using the SRL Macro
Pseudo-random Noise sequences are used to code and spread signals across a wide
band of transmission frequencies for spread spectrum modulation. PN generators are
based upon LFSRs, which can be effectively built from the SRL16 components.
• XAPP217: Gold Code Generators in Virtex Devices
A special type of PN sequence is a Gold code generator, which can be created from
SRL16-based LFSRs.
• XAPP256: FIFOs Using Virtex-II Shift Registers
The SRL16 is ideal for building smaller synchronous FIFOs. FIFOs can be built in any
width while producing a 1-bit resolution. With cascaded SRL16 shift registers, a
flexible depth in multiples of 16 is available. These techniques are useful for even
larger FIFOs when block RAM resources are not available.
• WP271: “Saving Costs with the SRL16E"
Describes the SRL16 function and its application in pipeline compensation, pseudo
random noise generators, serial frame synchronizers, running averages, pulse
generation and clock division, pattern generation, state machines, dynamically
addressable shift registers, FIFOs, and an RS232 receiver.
• DS228: RAM-Based Shift Register LogiCORE Module
Generates fast, compact, FIFO-style shift registers, delay lines or time-skew buffers
using the SRL16.
• SRL16 Primitives in Libraries Guide
Describes the usage and functionality of the SRL16 primitive and its variations.

Conclusion
The SRL16 configuration of the Spartan-3 generation LUT provides a space-efficient shift
register that otherwise require 16 flip-flops. This feature is automatically used when a
small shift register is described in HDL code. However, creative consideration of the uses
of the SRL16 as described here can provide even more significant advantages in many
applications.

Spartan-3 Generation FPGA User Guide www.xilinx.com 251

UG331 (v1.7) August 19, 2010
 R

Chapter 7: Using Look-Up Tables as Shift Registers (SRL16)

252 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 8

Using Dedicated Multiplexers

Summary
The Spartan®-3 generation architecture includes dedicated multiplexers within the
Configurable Logic Blocks (CLBs). These specialized multiplexers improve the
performance and density of not just wide multiplexers but almost any wide-input
function. Using these resources, a 32:1 multiplexer fits in just one level of logic, as do some
Boolean logic functions of up to 79 inputs. The dedicated multiplexers are identical in all
Spartan-3 generation FPGAs: Spartan-3, Spartan-3E, and Extended Spartan-3A families.

Introduction
A multiplexer, or mux, is a common building block of almost every logic design, selecting
one of several possible input signals. Spartan-3 generation FPGAs are very efficient at
implementing multiplexers: small ones in the look-up tables and larger ones using
dedicated multiplexer resources. Any Spartan-3 generation device easily implements:
• a 4:1 mux in one slice
• a 16:1 mux in one CLB
• a 32:1 mux in two CLBs
The same logic resources also can be used for wide, general-purpose logic functions. For
applications like comparators, encoder-decoders, or case statements, these resources
provide an optimal solution. These resources are used automatically by the Xilinx
development system, especially when a CASE statement is used, and then optimized for
the timing requirements of a given design. This chapter explains how to further optimize
the use of dedicated multiplexers and how to analyze their use in a design.
This chapter describes the dedicated multiplexer resources in the Spartan-3 generation
architecture. The signals and parameters associated with the multiplexers are defined. The
many methods to include multiplexers in a design are described along with
recommendations and guidelines for their use.

Dedicated Multiplexer Differences between Spartan-3 Generation

Families
Each CLB multiplexer structure is identical within a family, and the CLBs are identical
among all Spartan-3 generation families. The performance varies slightly between families
due to minor variations in processing and characterization.

Spartan-3 Generation FPGA User Guide www.xilinx.com 253

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Advantages of Dedicated Multiplexers

Spartan-3 generation FPGAs are based on four-input look-up tables (LUTs) that can
provide any possible function of the four inputs. The largest mux that a single LUT
supports is a 2:1 mux, with the fourth input available as a possible enable. One method to
construct larger muxes would be to cascade multiple LUTs. For example, a 4:1 mux could
be built by combining the outputs of two LUTs into a third LUT. However, this method
adds two full levels of logic delays plus an additional routing delay between the LUTs.
Without special resources, an 8:1 mux would consume s LUTs as well as add three levels of
logic delays plus two levels of routing delays, as shown in Figure 8-1.

LUT
net
net LUT

LUT
net
net LUT

LUT
net
net LUT

LUT
X466_01_040303

Figure 8-1: 8:1 Mux, 7 LUTs, 3 Levels of Logic

To increase multiplexer speed and density, Spartan-3 generation FPGAs provide a

dedicated 2:1 mux following every LUT, which replaces additional levels of LUT-based
logic. One of these, called the F5MUX, combines adjacent LUTs to create a 4:1 mux. The
other mux, following every pair of LUTs, combines muxes into wider functions with
different capabilities depending on its location in the CLB. This mux is called the FiMUX,
where the index "i" equals 6, 7, or 8. For example, the F6MUX combines the results of two
F5MUX elements to create an 8:1 mux as shown in Figure 8-2. The connections from the
LUTs to the muxes and between the muxes are dedicated and have zero connection delay.
The combination of LUTs and dedicated multiplexers allows very efficient implementation
of large multiplexers.

254 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

CLB Multiplexer Resources

LUT F5MUX

LUT
F6MUX

LUT
F5MUX

LUT
X466_02_050505

Figure 8-2: 8:1 Mux, 4 LUTs, 1 Level of Logic

CLB Multiplexer Resources

The Spartan-3 generation architecture consists of an array of identical Configurable Logic
Blocks, or CLBs. Each CLB is made up of four slices: two SLICEMs with memory capability
and two SLICELs with logic-only capability. Each slice is identical with respect to logic and
mux resources. Each slice has two LUTs, an F5MUX, and a second expansion mux (see
Figure 8-3).

FiMUX

4
LUT Reg

F5MUX
4
LUT Reg

Any Slice X466_03_050505

Figure 8-3: LUTs and F5MUX in a Slice

F5MUX
The F5MUX always combines the two LUTs in a slice. If those two LUTs contain 2:1 muxes
with the same control input, then the overall result is a 4:1 mux (see Figure 8-4).

Spartan-3 Generation FPGA User Guide www.xilinx.com 255

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

LUT

4:1 MUX
F5MUX
LUT

X466_04_050505

Figure 8-4: 4:1 Mux Implemented Using F5MUX

The F5MUX is so named because it generates any possible Boolean logic function of five
inputs (see Figure 8-5). If the two LUTs contain independent functions of the same four
inputs, the mux select line becomes the fifth input. The F5MUX becomes a function
expander that is just as efficient as another 3-input LUT for implementing any 5-input
function. This is a significant advantage over other FPGA architectures.

4
LUT

Any 5-input Function

F5MUX

LUT

X466_05_050505

Figure 8-5: Any 5-input Function Can Be Implemented Using F5MUX

As shown in Figure 8-6, the F5MUX also produces some functions of up to nine inputs, if
they can be partitioned into two 4-input LUTs and a mux.

4
LUT

Some 9-input Functions

F5MUX
4
LUT

X466_06_050605

Figure 8-6: Some 9-Input Functions Can Be Implemented Using a F5MUX

Consequently, the F5MUX generates any 5-input function, the 4:1 mux 6-input function, or
some 9-input functions.

256 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

CLB Multiplexer Resources

FiMUX
The second mux, called the FiMUX, functions as either an F6MUX, F7MUX, or F8MUX,
depending on its location and connections to the other muxes.

F8MUX
SLICEL S3

F6MUX
SLICEL S2

F7MUX
SLICEM S1

F6MUX
SLICEM S0
CLB
X466_07_040303

Figure 8-7: FiMUX Positions in a CLB

Each FiMUX receives inputs from muxes of the next lower number; for example, the two
F6MUX results drive the F7MUX. Like the F5MUX, the FiMUX has the flexibility to
implement other types of functions besides just multiplexers. The F6MUX is so named
because it creates any function of six inputs. Similarly, the F7MUX generates any function
of seven inputs, and the F8MUX generates any function of eight inputs.
Table 8-1: Mux Capabilities
Total Number of Inputs per Function
Mux Usage Input Source For Any For Limited
For Mux
Function Functions
F5MUX F5MUX LUTs 5 6 (4:1 mux) 9
FiMUX F6MUX F5MUX 6 11 (8:1 mux) 19
F7MUX F6MUX 7 20 (16:1 mux) 39
F8MUX F7MUX 8 37 (32:1 mux) 79

Naming Conventions
In this document and in the Spartan-3 generation data sheets, the mux that serves as either
F6MUX, F7MUX, or F8MUX generically is called an FiMUX (i = 6, 7, or 8). This name
avoids confusion with the static CLB mux that generates the X output, which the FPGA
Editor refers to as the "FXMUX". The FiMUX is always referred to as the "F6MUX" in the
FPGA Editor. The timing analyzer also refers to the path through the FiMUX to the CLB
pin as "TIF6Y", although it can be used as an F7MUX or F8MUX.
The library components are called MUXF5, MUXF6, MUXF7, and MUXF8. MUXF6,
MUXF7, and MUXF8 use the FiMUX and restrict the placement to a specific relative
location in the CLB.

Spartan-3 Generation FPGA User Guide www.xilinx.com 257

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Dedicated Local Routing

A significant benefit of the dedicated multiplexers is the dedicated routing that connects
between levels. Although each mux is implemented as one pass through the CLB, the
outputs connect back to the CLB inputs through local interconnect with zero routing delay.
The result is the same as if the muxes were in series within the CLB.
The F5MUX feeds the F5 CLB output pin, which only connects back to an FiMUX input on
the same CLB (called FXINA and FXINB). The FiMUX feeds the FX CLB output pin, which
also feeds back to an FiMUX input on the same CLB, or in the case of the F7MUX, also to
the CLB below. If the mux result is needed elsewhere, it connects to a general-purpose CLB
output (X for the F5MUX, Y for the FiMUX).

FXINB
F8 X
FXINA

F5 F5

FXINB FX
F6
FXINA

F5
F5

FXINB
FX
F7
FXINA

F5 F5

FXINB
F6 FX
FXINA

F5 F5

X466_08_050505

Figure 8-8: Muxes and Dedicated Feedback in a Spartan-3 Generation CLB

258 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Implementation Examples

Mux Select Inputs

The select inputs for the multiplexers come from general-purpose routing. The select input
for the F5MUX is the BX input on the CLB, and the select input for the FiMUX is the BY
input on the CLB.

FiMUX
FXINA 1
FX (Local Feedback to FXIN)
FXINB 0
Y (General Interconnect)
BY
D Q YQ

F5MUX
F[4:1] LUT 1
F5 (Local Feedback to FXIN)
G[4:1] LUT 0
X (General Interconnect)
BX
D Q XQ

x466_13_050505

Figure 8-9: Dedicated Multiplexers in a Spartan-3 Generation CLB

Implementation Examples

Wide-Input Multiplexers
Each LUT optionally implements a 2:1 multiplexer. In each slice, the F5MUX and two LUTs
can implement a 4:1 multiplexer. As shown in Figure 8-10, the F6MUX and two slices
implement an 8:1 multiplexer. The F7MUX and the four slices of any CLB implement a 16:1
multiplexer, and the F8MUX and two CLBs implement a 32:1 multiplexer.

Spartan-3 Generation FPGA User Guide www.xilinx.com 259

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

DATA[0]
LUT
DATA[1]

8:1
DATA[7:0]
F5 (S2 & S3)

DATA[2]
LUT
DATA[3]

16:1 output
S1 F7

DATA[4] 8:1 Output

LUT F6
DATA[5]

F5
8:1
DATA[15:8]
(S0 & S1)

DATA[6]
LUT
DATA[7]

S0 CLB
SELECT[0]
SELECT[1] SELECT[2:0]
SELECT[2] SELECT[3]

8:1 MUX 16:1 MUX

X466_09_030603

Figure 8-10: 8:1 and 16:1 Multiplexers

260 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Implementation Examples

Wide-Input Functions
Slices S0 and S2 have an F6MUX, designed to combine the outputs of two F5MUX
resources. Figure 8-11 illustrates a combinatorial function up to 19 inputs in the slices S0
and S1, or in the slices S2 and S3.

SLICEL OR SLICEM

FiMUX

4
LUT Reg

F5MUX
4
LUT Reg

S_F5

S_F6

OUT_F6

F6MUX
4
LUT Reg

F5MUX
4
LUT Reg

S_F5

SLICEL OR SLICEM X466_10_030603

Figure 8-11: 19-input Function Using F6MUX in Two Slices

Spartan-3 Generation FPGA User Guide www.xilinx.com 261

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

The slice S1 has an F7MUX, designed to combine the outputs of two F6MUXs. Figure 8-12
illustrates a combinatorial function up to 39 inputs in a Spartan-3 generation CLB.

SLICEL S3
FiMUX

4
LUT Reg

F5MUX
4
LUT Reg

S_F5

S_F6 SLICEL S2

F6MUX
4
LUT Reg

F5MUX
4
LUT Reg

S_F5

S_F7 SLICEM S1

F7MUX OUT_F
4
LUT Reg

F5MUX
4
LUT Reg

S_F5

S_F6
SLICEM S0
F6MUX
4
LUT Reg

F6MUX
4
LUT Reg

S_F5

X466_11_030603

Figure 8-12: 39-input Function Using F7MUX in One CLB

262 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Implementation Examples

The slice S3 of each CLB has an F8MUX. Combinatorial functions of up to 79 inputs fit in
two CLBs as shown in Figure 8-13. The outputs of two F7MUXs are combined through
dedicated routing resources between two adjacent CLBs in a column.

FiMUX

Slice S3

F6MUX

Slice S2

F7MUX

Slice S1

F6MUX

Slice S0
CLB

F8MUX OUT_F8

Slice S3

F6MUX

Slice S2

F7MUX

Slice S1

F6MUX

Slice S0
CLB
X466_12_060606

Figure 8-13: 79-input Function Using F8MUX in Two Adjacent CLBs

Spartan-3 Generation FPGA User Guide www.xilinx.com 263

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Timing Parameters
There are several possible paths through the CLB multiplexers. The two types of
multiplexers are considered separately (F5MUX and FiMUX). Each multiplexer type has
two types of inputs: data inputs and select lines. The output of the mux drives the local
interconnect through the F5 and FX CLB pins, the general interconnect through the X and
Y CLB pins, or the D input on the flip-flop. See Figure 8-9, page 259 for a block diagram
showing dedicated multiplexers in a Spartan-3 generation CLB. Note that although the
mux functionality is identical between the slices with memory and those without, the
timing values are independent and can vary slightly.
Although the multiplexers are connected in series inside the CLB, each mux actually feeds
a CLB output pin, which feeds back to an input pin through zero-delay local interconnect.
Thus each reported block delay element will have only one mux from input to output. The
Spartan-3 generation architecture improves on the Virtex®-II architecture by providing a
direct path from the F5MUX or FiMUX to the flip-flop in the CLB.

Table 8-2: Multiplexer Timing Paths

Symbol CLB Input Through CLB Output
tIF5 F/G LUT Inputs LUT and F5MUX Inputs F5
tIF5X F/G LUT Inputs LUT and F5MUX Inputs X
tIF5CK F/G LUT Inputs LUT and F5MUX Inputs D input on flip-flop
tBXF5 BX F5MUX Select F5
tBXX BX F5MUX Select X
tINAFX FXINA FiMUX Inputs FX
tINBFX FXINB FiMUX Inputs FX
tIF6Y FXINA or FXINB FiMUX Inputs Y
tBYFX BY FiMUX Select FX
tBYY BY FiMUX Select Y

Programmable Polarity
As with most resources in the Spartan-3 generation FPGA, inverters are free in large
multiplexers. The functions in the LUT can have inverters added to inputs or outputs with
no effect on performance or utilization. The control inputs to the F5MUX (BX) and FiMUX
(BY) have programmable polarity inside the CLB.

Floorplanning Multiplexers
The wide multiplexers force a particular placement on the LUTs being combined. The
LUTs must always be in the same slice for the F5MUX and in adjacent vertical slices for the
wider muxes. This vertical orientation aligns nicely with the arithmetic logic.
The wide multiplexers cannot be used in conjunction with the arithmetic logic because the
arithmetic XOR gate is multiplexed with the F5MUX result. Also, the 32x1 configuration of
the distributed RAM uses the F5MUX for the fifth address input.

264 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Uses of Multiplexers

Related Uses of Multiplexers

Multiplexers and Three-State Buffers

The LUT and mux resources multiplex one of several input signals onto an internal routing
resource, using the routing like an internal bus. This is equivalent to the BUFT-based
multiplexers found in other FPGA architectures. In most modern FPGA families, these
three-state buffers actually are implemented as dedicated logic gates to avoid possible
contention when more than one is enabled at a time. The Spartan-3 generation families
reduce die size and cost by eliminating the overhead of these internal three-state buffer
gates. Instead, internal functions defined as a three-state buffer in the Spartan-3 generation
families must be implemented in the LUTs and dedicated muxes.
The CLB multiplexers provide binary encoding of the select lines, requiring fewer signals
than the one-hot encoding of the BUFT-based multiplexers. CLB-based multiplexers have
no limit on width as BUFT-based multiplexers did, nor any special placement
considerations.
The BUFT component, representing a three-state buffer, is not available in the
Spartan-3 generation libraries, except for the output function in the IOBs. The CORE
Generator functions of the BUFT-based Multiplexer (and the equivalent BUFE-based
Multiplexer) will be implemented as multiplexers in the CLBs.

Using Memory in Place of Multiplexers

To optimize designs, consider replacing multiplexers with memories. A 4:1 mux requires
two LUTs and an F5MUX. If the inputs are static, the same function can be thought of as a
4-bit memory and can fit in less than one LUT. In fact, the LUT can be considered to be a
16:1 mux with the LUT inputs serving as the select lines. In any situation where the mux
inputs are static, a memory-based implementation saves resources by using the built-in
address decode as the mux logic. The 32x1 distributed RAM uses the F5MUX for the fifth
address input. For more information, see Chapter 6, “Using Look-Up Tables as Distributed
RAM.”
A 4:1 mux with changeable inputs still can be built in one level of logic using the LUT RAM
by reprogramming the RAM as the method of selecting one of the four inputs. An easy
way of doing this is to use the SRL16 mode to write data into the RAM in 16 clock cycles.
For more information, see Chapter 7, “Using Look-Up Tables as Shift Registers (SRL16).”
Creative design concepts such as these can save significant resources. More information is
found in the WP273 “Performance + Time = Memory (Cost-saving with 3-D Design)”.

Other Multiplexers
The CLB also contains other multiplexers for routing signals through the logic resources.
The CYMUX for propagating carry signals is the only other dynamic mux. Several other
muxes are used for selecting one of multiple paths. One is called the FXMUX in the FPGA
Editor, since it routes the F LUT signal to the X CLB output. Do not confuse this static mux
with the FXMUX name that is sometimes used for the FiMUX described here.
When multiplexing clock signals, remember to use the BUFGMUX, which helps eliminate
glitches on the resulting clock. Another special multiplexer is found in the I/O to support
DDR interfaces. The DDR mux combines two signals onto one output by automatically
muxing back and forth between them as they are clocked into the IOB. See the Spartan-3
generation data sheets for more information on these other multiplexing features.

Spartan-3 Generation FPGA User Guide www.xilinx.com 265

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Designing with Multiplexers

There are several ways multiplexers can be used in a design. The most common is to
simply have them inferred by synthesis tools when appropriate for a design. Library
primitives can be used to instantiate specific multiplexers. This document provides HDL
submodules that combine the library primitives into larger muxes. The CORE Generator
system includes the Bus Multiplexer and Bit Multiplexer functions, and many other CORE
solutions take advantage of the dedicated multiplexers.

Inference
Multiplexers are typically inferred by a conditional statement, most commonly the CASE
or IF-THEN-ELSE statement. The IF statement generally produces priority-encoded logic.
The CASE statement is more likely to generate an optimized multiplexer.
Synthesis options can determine whether multiplexers are inferred and how they are
implemented. For XST, the MUX_EXTRACT constraint specifies whether multiplexers are
inferred, and the MUX_STYLE constraint specifies whether they are implemented in the
dedicated logic multiplexers or the carry multiplexers (CY_MUX). The default is to infer
automatically the best resource.
CASE statements should be full (all branches defined) to avoid creating a latch. Undefined
branches assume the current value needs to be maintained, implying memory. They also
should be parallel (branch conditions all mutually exclusive) to avoid a priority encoder.
Some synthesis tools, such as XST, have options to assume full and parallel CASE
statements even if not written that way. It is good practice to include a “When Others”
(VHDL) or “Default” (Verilog) branch to make sure even undefined inputs do not generate
a latch.
An IF statement can contain a set of different expressions while a CASE statement is
evaluated against a common controlling expression. In general, use the CASE statement
for complex decoding and use the IF statement for speed critical paths.
Most current synthesis tools can determine if the IF-ELSIF conditions are mutually
exclusive, and will not create extra logic to build the priority tree. The following are points
to consider when writing IF statements:
• Make sure that all outputs are defined in all branches of an IF statement. If not, they
can create latches or long equations on the CE signal. A good way to prevent this is to
have default values for all outputs before the IF statements.
• Limit the number of input signals into an IF statement to reduce the number of logic
levels. If there are a large number of input signals, see if some of them can be
predecoded and registered before the IF statement.
• Avoid bringing the dataflow into a complex IF statement. Only control signals should
be generated in complex IF-ELSE statements.
Make sure you do not write the code such that your synthesis tool will infer BUFT-based
multiplexers. A BUFT-based multiplexer usually requires a statement with a "Z" value.
Some synthesis tools might automatically or optionally convert BUFT logic to
multiplexers.
A decoder is a special case of a multiplexer where the inputs are fixed as one-hot values.
Decoders of up to 4:16 in size are easily implemented in individual LUTs for each output
and do not need to use the dedicated multiplexers, or they can even use the Carry muxes
for high performance.

266 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with Multiplexers

The following subsections provide examples of 2:1 muxes described using the CASE
statement in Verilog and VHDL code.

Verilog Inference
module MUX_2_1 (DATA_I, SELECT_I, DATA_O);

input [1:0]DATA_I;
input SELECT_I;

output DATA_O;
reg DATA_O;

always @ (DATA_I or SELECT_I)

case (SELECT_I)
1'b0 : DATA_O <= DATA_I[0];
1'b1 : DATA_O <= DATA_I[1];
default : DATA_O <= 1'bx;
endcase

endmodule

VHDL Inference
entity MUX_2_1 is
port (
DATA_I: in std_logic_vector (1 downto 0);
SELECT_I: in std_logic;
DATA_O: out std_logic
);
end MUX_2_1;

architecture MUX_2_1_arch of MUX_2_1 is

--
begin
--
SELECT_PROCESS: process (SELECT_I, DATA_I)
begin
case SELECT_I is
when '0' => DATA_O <= DATA_I (0);
when '1' => DATA_O <= DATA_I (1);
when others => DATA_O <= 'X';
end case;
end process SELECT_PROCESS;
--
end MUX_2_1_arch;

Library Primitives
Four library primitives are available that offer access to the dedicated multiplexers in each
slice: MUXF5, MUXF6, MUXF7, and MUXF8. These use the F5MUX and FiMUX CLB
resources (see “Naming Conventions,” page 257). Each of the multiplexer primitives looks
identical (see Figure 8-14). The actual selection simply determines where in the CLB the
multiplexer can be located, as shown in Table 8-5.

Spartan-3 Generation FPGA User Guide www.xilinx.com 267

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

I0
O
I1
S
x466_14_040303

Figure 8-14: MUXF5 Primitive

Table 8-3: MUX Inputs and Outputs

Signal Function
I0 Input selected when S is Low
I1 Input selected when S is High
S Select input
LO Local Output that connects to the F5 or FX CLB pins, which use local feedback to
the FXIN inputs to the FiMUX for cascading (see “Modeling Local Output
Timing,” page 269)
O General Output that connects to the general-purpose combinatorial or registered
outputs of the CLB

Table 8-4: MUX Function

Inputs Outputs
S I0 I1 O LO
0 1 X 1 1
0 0 X 0 0
1 X 1 1 1
1 X 0 0 0

Table 8-5: Multiplexer Resources

Physical
Primitive Slice Control Input Output
Location
MUXF5 S0, S1, S2, S3 F5MUX S I0, I1 O
MUXF6 S0, S2 FiMUX S I0, I1 O
MUXF7 S1 FiMUX S I0, I1 O
MUXF8 S3 FiMUX S I0, I1 O

The generic multiplexer components also can take advantage of the dedicated
multiplexers. The M2_1 schematic library component is implemented in a LUT, while the
larger multiplexers in the library use the F5MUX and FiMUX components.

Enable Signals in Multiplexers

An enable signal on a multiplexer can be used to keep the multiplexer output Low when
disabled. Although the dedicated multiplexers do not have enable signals, the enable can
be implemented on the preceding 2:1 mux that will be implemented in a LUT. The M4_1E
and M8_1E schematic library components are built this way, using the F5MUX and F6MUX

268 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with Multiplexers

for the final result, respectively, while the M16_1E schematic library component keeps the
enable on the final mux, forcing it into a LUT instead of the F7MUX. Figure 8-15 shows the
M4_1E schematic library component logic.

M2_1E
D0 D0
O
D1 D1
S0 MUXF5
E M01 M01 I0
O O
M23 I1
M2_1E 0
D2 D0 S
O
D3 D1
S0 S0
E E M23
S1
x466_15_040303

Figure 8-15: M4_1E Library Component Logic

Modeling Local Output Timing

There are also two alternative versions of each library component that are functionally
identical but can be used for more accurate timing estimation before implementation. As
mentioned previously, the multiplexers can drive one or both CLB outputs. The first
output is the special CLB output that feeds directly back through local interconnect to the
next multiplexer in series, known as the local output. The second output is the general-
purpose CLB output, which can be routed to any other logic. For better pre-
implementation timing estimation, the user can substitute special primitives that specify
whether to use the local output timing or the general-purpose output timing. The
MUXF5_L primitive models the local output, while the MUXF5_D primitive models both
output paths (see Figure 8-16). The functionality is identical to that for the MUXF5
primitive.

I0 I0 LO
LO
I1 I1 O
S S
x466_16_040303

Figure 8-16: MUXF5_L and MUXF5_D Primitives to Model Local Output Timing

Submodules
In addition to the primitives, five submodules that implement multiplexers from 2:1 to 32:1
are provided in VHDL and Verilog code. Synthesis tools can automatically infer the above
primitives (MUXF5, MUXF6, MUXF7, and MUXF8); however, the submodules described
in this section use instantiation of the multiplexers to guarantee an optimized result.
Table 8-6 lists available submodules.
• xapp466_vhdl.zip
• xapp466_vhdl.zip

Spartan-3 Generation FPGA User Guide www.xilinx.com 269

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Table 8-6: Available Submodules

Submodule Multiplexer Control Input Output
MUX_2_1_SUBM 2:1 SELECT_I DATA_I[1:0] DATA_O
MUX_4_1_SUBM 4:1 SELECT_I[1:0] DATA_I[3:0] DATA_O
MUX_8_1_SUBM 8:1 SELECT_I[2:0] DATA_I[7:0] DATA_O
MUX_16_1_SUBM 16:1 SELECT_I[3:0] DATA_I[15:0] DATA_O
MUX_32_1_SUBM 32:1 SELECT_I[4:0] DATA_I[31:0] DATA_O

Port Signals
Data In — DATA_I
The data input provides the data to be selected by the SELECT_I signal(s).

Control In — SELECT_I
The select input signal or bus determines the DATA_I signal to be connected to the output
DATA_O. For example, the MUX_4_1_SUBM multiplexer has a 2-bit SELECT_I bus and a
4-bit DATA_I bus. Table 8-7 shows the DATA_I selected for each SELECT_I value.
Table 8-7: Selected Inputs
SELECT_I[1:0] DATA_O
00 DATA_I[0]
01 DATA_I[1]
10 DATA_I[2]
11 DATA_I[3]

Data Out — DATA_O

The data output O provides the data value (1 bit) selected by the control inputs.

Applications
Multiplexers are used in various applications. These are often inferred by synthesis tools
when a “case” statement is used (see the example below). Comparators, encoder-decoders,
and wide-input combinatorial functions are optimized when they are based on one level of
LUTs and dedicated multiplexer resources of the Spartan-3 generation CLBs.

VHDL and Verilog Instantiation

The primitives (MUXF5, MUXF6, and so forth) can be instantiated in VHDL or Verilog
code, to design wide-input functions.
The submodules (MUX_2_1_SUBM, MUX_4_1_SUBM, and so forth) can be instantiated in
VHDL or Verilog code to implement multiplexers. However, the corresponding
submodule must be added to the design directory as a hierarchical submodule. For
example, if a module is using the MUX_16_1_SUBM, the MUX_16_1_SUBM.vhd file
(VHDL code) or MUX_16_1_SUBM.v file (Verilog code) must be compiled with the design

270 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with Multiplexers

source code. The submodule code can also be “cut and pasted” into the designer source
code.

VHDL and Verilog Submodules

VHDL and Verilog submodules are available to implement multiplexers up to 32:1. They
illustrate how to design with the MUX resources. When synthesis infers the corresponding
MUX resource(s), the VHDL or Verilog code is behavioral code (“case” statement).
Otherwise, the equivalent “case” statement is provided in comments and the correct MUX
resources are instantiated. However, most synthesis tools support the inference of all of the
MUXs. The following examples can be used as guidelines for designing other wide-input
functions.
The following submodules are available:
• MUX_2_1_SUBM (behavioral code)
• MUX_4_1_SUBM
• MUX_8_1_SUBM
• MUX_16_1_SUBM
• MUX_32_1_SUBM
The corresponding submodules have to be synthesized with the design.
The submodule MUX_16_1_SUBM is provided in VHDL and Verilog as an example:

VHDL Template
-- Module: MUX_16_1_SUBM
-- Description: Multiplexer 16:1
--
-- Device: Spartan-3 Family
---------------------------------------------------------------------
library IEEE;
use IEEE.std_logic_1164.all;

library UNISIM;
use UNISIM.VCOMPONENTS.ALL;

entity MUX_16_1_SUBM is
port (
DATA_I: in std_logic_vector (15 downto 0);
SELECT_I: in std_logic_vector (3 downto 0);
DATA_O: out std_logic
);
end MUX_16_1_SUBM;

architecture MUX_16_1_SUBM_arch of MUX_16_1_SUBM is

-- Component Declarations:
component MUXF7
port (
I0: in std_logic;
I1: in std_logic;
S: in std_logic;
O: out std_logic
);
end component;
--
-- Signal Declarations:

Spartan-3 Generation FPGA User Guide www.xilinx.com 271

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

signal DATA_MSB : std_logic;

signal DATA_LSB : std_logic;
--
begin
--
-- If synthesis tools support MUXF7 :
--SELECT_PROCESS: process (SELECT_I, DATA_I)
--begin
--case SELECT_I is
-- when "0000" => DATA_O <= DATA_I (0);
-- when "0001" => DATA_O <= DATA_I (1);
-- when "0010" => DATA_O <= DATA_I (2);
-- when "0011" => DATA_O <= DATA_I (3);
-- when "0100" => DATA_O <= DATA_I (4);
-- when "0101" => DATA_O <= DATA_I (5);
-- when "0110" => DATA_O <= DATA_I (6);
-- when "0111" => DATA_O <= DATA_I (7);
-- when "1000" => DATA_O <= DATA_I (8);
-- when "1001" => DATA_O <= DATA_I (9);
-- when "1010" => DATA_O <= DATA_I (10);
-- when "1011" => DATA_O <= DATA_I (11);
-- when "1100" => DATA_O <= DATA_I (12);
-- when "1101" => DATA_O <= DATA_I (13);
-- when "1110" => DATA_O <= DATA_I (14);
-- when "1111" => DATA_O <= DATA_I (15);
-- when others => DATA_O <= 'X';
--end case;
--end process SELECT_PROCESS;
--
-- If synthesis tools DO NOT support MUXF7 :
SELECT_PROCESS_LSB: process (SELECT_I, DATA_I)
begin
case SELECT_I (2 downto 0) is
when "000" => DATA_LSB <= DATA_I (0);
when "001" => DATA_LSB <= DATA_I (1);
when "010" => DATA_LSB <= DATA_I (2);
when "011" => DATA_LSB <= DATA_I (3);
when "100" => DATA_LSB <= DATA_I (4);
when "101" => DATA_LSB <= DATA_I (5);
when "110" => DATA_LSB <= DATA_I (6);
when "111" => DATA_LSB <= DATA_I (7);
when others => DATA_LSB <= 'X';
end case;
end process SELECT_PROCESS_LSB;
--
SELECT_PROCESS_MSB: process (SELECT_I, DATA_I)
begin
case SELECT_I (2 downto 0) is
when "000" => DATA_MSB <= DATA_I (8);
when "001" => DATA_MSB <= DATA_I (9);
when "010" => DATA_MSB <= DATA_I (10);
when "011" => DATA_MSB <= DATA_I (11);
when "100" => DATA_MSB <= DATA_I (12);
when "101" => DATA_MSB <= DATA_I (13);
when "110" => DATA_MSB <= DATA_I (14);
when "111" => DATA_MSB <= DATA_I (15);
when others => DATA_MSB <= 'X';
end case;
end process SELECT_PROCESS_MSB;

272 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with Multiplexers

--
-- MUXF7 instantiation
U_MUXF7: MUXF7
port map (
I0 => DATA_LSB,
I1 => DATA_MSB,
S => SELECT_I (3),
O => DATA_O
);
--
end MUX_16_1_SUBM_arch;
--

Verilog Template
// Module: MUX_16_1_SUBM
//
// Description: Multiplexer 16:1
// Device: Spartan-3 Family
//-------------------------------------------------------------------
//
module MUX_16_1_SUBM (DATA_I, SELECT_I, DATA_O);

input [15:0]DATA_I;
input [3:0]SELECT_I;

output DATA_O;

wire [2:0]SELECT;

reg DATA_LSB;
reg DATA_MSB;

assign SELECT[2:0] = SELECT_I[2:0];

/*
//If synthesis tools support MUXF7 :
always @ (DATA_I or SELECT_I)

case (SELECT_I)
4'b0000 : DATA_O <= DATA_I[0];
4'b0001 : DATA_O <= DATA_I[1];
4'b0010 : DATA_O <= DATA_I[2];
4'b0011 : DATA_O <= DATA_I[3];
4'b0100 : DATA_O <= DATA_I[4];
4'b0101 : DATA_O <= DATA_I[5];
4'b0110 : DATA_O <= DATA_I[6];
4'b0111 : DATA_O <= DATA_I[7];
4'b1000 : DATA_O <= DATA_I[8];
4'b1001 : DATA_O <= DATA_I[9];
4'b1010 : DATA_O <= DATA_I[10];
4'b1011 : DATA_O <= DATA_I[11];
4'b1100 : DATA_O <= DATA_I[12];
4'b1101 : DATA_O <= DATA_I[13];
4'b1110 : DATA_O <= DATA_I[14];
4'b1111 : DATA_O <= DATA_I[15];
default : DATA_O <= 1'bx;
endcase
*/

Spartan-3 Generation FPGA User Guide www.xilinx.com 273

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

//If synthesis tools do not support MUXF7 :

always @ (SELECT or DATA_I)

case (SELECT)
3'b000 : DATA_LSB <= DATA_I[0];
3'b001 : DATA_LSB <= DATA_I[1];
3'b010 : DATA_LSB <= DATA_I[2];
3'b011 : DATA_LSB <= DATA_I[3];
3'b100 : DATA_LSB <= DATA_I[4];
3'b101 : DATA_LSB <= DATA_I[5];
3'b110 : DATA_LSB <= DATA_I[6];
3'b111 : DATA_LSB <= DATA_I[7];
default : DATA_LSB <= 1'bx;
endcase

always @ (SELECT or DATA_I)

case (SELECT)
3'b000 : DATA_MSB <= DATA_I[8];
3'b001 : DATA_MSB <= DATA_I[9];
3'b010 : DATA_MSB <= DATA_I[10];
3'b011 : DATA_MSB <= DATA_I[11];
3'b100 : DATA_MSB <= DATA_I[12];
3'b101 : DATA_MSB <= DATA_I[13];
3'b110 : DATA_MSB <= DATA_I[14];
3'b111 : DATA_MSB <= DATA_I[15];
default : DATA_MSB <= 1'bx;
endcase

// MUXF7 instantiation

MUXF7 U_MUXF7 (.I0(DATA_LSB),

.I1(DATA_MSB),
.S(SELECT_I[3]),
.O(DATA_O)
);
endmodule

CORE Generator System

The CORE Generator system offers the basic logic functions of the Bit Multiplexer and the
Bus Multiplexer. The Bit Multiplexer, shown in Figure 8-17, supports sizes up to 256
inputs. The Bus Multiplexer, shown in Figure 8-18, supports muxes of up to 32 inputs for
buses of up to 256 bits each. These core solutions have a parameter Mux Type to select a
BUFT or LUT based multiplexer. Select the appropriate radio button in the CORE
Generator system for the construction of the multiplexer. The default setting is LUT based,
which is required for Spartan-3 generation multiplexers. The CORE Generator system also
offers options for registering the output of the multiplexer.

274 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with Multiplexers

ASET SSET

M[N:0] O • D Q

CE
S[M:0]
CLK

ACLR SCLR

x465_17_041003

Figure 8-17: Bit Multiplexer CORE Symbol

MCH[N:0]
ASET SSET
•
•
• O[N:0] D[N:0] Q[N:0]

MA[N:0] CE
S[M:0]
CLK

ACLR SCLR AINIT SINIT

X466_18_060506

Figure 8-18: Bus Multiplexer CORE Symbol

The CORE Generator system also offers the specific functions of the BUFT-based
Multiplexer (and the equivalent BUFE-based Multiplexer). As with the generic Bit and Bus
Multiplexers, they are implemented in LUTs and/or muxes.

Spartan-3 Generation FPGA User Guide www.xilinx.com 275

UG331 (v1.7) August 19, 2010
 R

Chapter 8: Using Dedicated Multiplexers

Related Materials
The following document provides supplementary information useful with this chapter:
WP274: Multiplexer Selection
This white paper considers a variety of ways in which multiplexers can be implemented
within Xilinx FPGA devices, including some alternative techniques that can lead to more
efficient and lower cost implementations.

Summary
The dedicated multiplexers in the Spartan-3 generation architecture enable wider
functions than possible in the four-input LUTs. These multiplexers are automatically used
by the software tools but careful coding can help optimize their use to minimize resource
requirements and improve performance of designs.

276 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 9

Using Carry and Arithmetic Logic

Summary
Dedicated carry and arithmetic logic improves the performance of adders, counters,
comparators, multipliers, wide logic gates, and related functions in the
Spartan®-3 generation FPGA family. Carry logic consists of dedicated gates, multiplexers,
and routing that are independent of the general-purpose logic resources and provide both
higher density and higher performance. Carry logic can be explicitly called out in the
design using primitives, implemented via library or user-defined macros, or inferred by
the synthesis tools. Most arithmetic components automatically use the carry logic. This
chapter describes the carry logic resources and how they can be used efficiently in
Spartan-3 generation FPGA designs.

Introduction
The basic building block of the FPGA is the look-up table, or LUT. Although arithmetic
functions can be implemented in the LUTs, they require the generation of a sum and a
carry for every input and could quickly use up LUT and routing resources. Arithmetic
functions are common enough to warrant dedicating their own special resources for
implementation. The arithmetic logic allows the generation of a sum outside the LUT and
the carry logic provides dedicated routing resources for cascading a carry signal between
slices of a CLB and between CLBs. The carry chain cascades from the bottom to the top of
each column of CLB slices.
The arithmetic logic consists of a discrete XOR component for single level sum completion,
an AND gate for multiplication, and multiplexers for controlling signal flow. These gates
work in conjunction with the LUTs to implement efficient arithmetic functions, including
counters and multipliers, typically at two bits per slice. Each CLB provides two separate
carry chains of four bits each. The resources can be used to improve the performance of
arithmetic functions and can also be used to cascade LUTs for wide-input logic functions.

Carry and Arithmetic Logic Differences between

Spartan-3 Generation Families
The carry and arithmetic logic is identical among all Spartan-3 generation families. The
performance varies slightly between families due to minor variations in processing and
characterization. In the Spartan-3E and Extended Spartan-3A families, most of the DCMs
are embedded in the CLB array, and therefore limit the maximum length of the carry chain
for those CLB columns containing DCMs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 277

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

Look-Ahead Carry Addition

To understand the basic resources of the Spartan-3 generation carry logic, it is important to
understand the basics of look-ahead carry addition. Normal addition of two digits requires
simply an XOR gate to generate the Sum and an AND gate to generate a Carry, as shown
in Table 9-1.

Table 9-1: Binary Addition

Sum Carry Out
A B
(A XOR B) (A AND B)
0 0 0 0
0 1 1 0
1 0 1 0
1 1 0 1

This logic is known as a half adder because it does not include a Carry input. Accounting
for a Carry input can be done simply by repeating the half adder to add the first Sum and
the Carry input and then generating a final Carry output if either half-adder generated a
Carry (using an OR gate). A full adder is created as shown in Figure 9-1.

Half Adder Half Adder

B
A Sum

Carry Out
Carry In
UG331_c11_01_072906

Figure 9-1: Full Adder

This logic can easily be implemented in two LUTs with three inputs each to generate Sum
and Carry. The problem with this implementation is that it requires two LUTs for every
input bit, and the Carry propagates through the full LUT delay for each bit.
A better implementation is to "look ahead" and determine if the input Carry signal needs to
be propagated (the inputs are different) or generated (both inputs are High). See Table 9-2.

Table 9-2: Look-Ahead Carry

A B Propagate Generate
0 0 0 0
0 1 1 0
1 0 1 0
1 1 0 1

This case is similar to the half Sum and Carry values described earlier. The Propagate
signal, which is the same as the half Sum in the first half adder, can be implemented using
the same XOR gate.
If Propagate is not True, then A = B and either signal can be used directly as the Generate
signal. Thus the Carry output can be defined by a multiplexer controlled by Propagate that

278 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Resource Details

allows the Carry input through when Propagate = 1 and allows A (or B) through when
Propagate = 0.
The full Sum is still generated by a second XOR gate, resulting in the logic shown in
Figure 9-2.

COUT

Propagate/Half Sum
0 1

Sum

Generate
CIN UG331_c11_02_011508

Figure 9-2: Look-Ahead Carry Implementation

The logic has been split into three functions that cannot all be combined into one or two
LUTs. To optimize the implementation of this logic, the Spartan-3 generation CLB provides
a dedicated XOR gate outside the LUT to generate the Sum, called XORCY, and a dedicated
mux to provide the Carry, called MUXCY, as shown in Figure 9-3.

COUT
Function
Generator MUXCY

B
A 0 1

Sum

CIN
XORCY
UG331_c11_03_073006

Figure 9-3: Carry Logic in Spartan-3 Generation FPGAs

An advantage of this structure is that it provides a very fast carry propagation, since it only
requires the delay of a 2:1 mux. Also, this structure uses only one LUT, or one half of a slice,
allowing two bits per slice and therefore an efficient, high-density implementation. With
dedicated connections at each cascade point, from COUT to the CIN in the other half of a
slice, to the CIN in the other slice in a CLB, and to the CIN of the next CLB, the carry chain
can propagate up a column of CLBs with very high performance.

Resource Details
The Spartan-3 generation carry and arithmetic logic consists of dedicated CLB resources
and inter-CLB routing. The logic is almost identical within the two logic cells in each of the
slice and is identical in both the logic-only SLICEL and the SLICEM that adds distributed
RAM capability. A simplified view of one logic cell is shown in Figure 9-4.

Spartan-3 Generation FPGA User Guide www.xilinx.com 279

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

COUT

1
MUXCY
0 1

B LUT
A
Sum
XORCY
D Q

MULT_AND

1
0
Bypass CIN UG331_c11_04_073006

Figure 9-4: Simplified View of Spartan-3 FPGA Carry and Arithmetic Logic in One
Logic Cell

This implementation adds flexibility to the MUXCY beyond the standard functionality
described earlier. The A/B "Generate" input to the MUXCY can come from either signal, or
even an AND of the two signals. The carry chain can be initialized with a 1 or 0 or fed by
an independent bypass input to MUXCY. The MUXCY control input can be fixed to 1 to
always propagate the carry. The output sum, the XORCY, can be optionally registered.
Figure 9-5 shows the entire carry logic and connections for one slice. The dashed lines
indicate an additional fixed multiplexer that is only found in the SLICEM half of the CLB.

COUT

YB
CYSELG
1 CYMUXG Y
G4 0 1
G3 D Q YQ
G2 G-LUT
G1 XORG

CY0G

GAND 1
0
BY

XB
CYSELF
1 CYMUXF X
F4 0 1
F3 D Q XQ
F2 F-LUT
F1 XORF

CY0F

FAND 1 CYINT
0 1
0

CIN
BX UG331_c11_05_073006

Figure 9-5: Simplified View of Carry Logic in One Slice

280 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Resource Details

MUXCY
The dynamic mux generically referred to as MUXCY is available at both the bottom (called
CYMUXF) and top (CYMUXG) of each slice.
The "0" input to the MUXCY typically comes from one of the LUT inputs. It can be fed by
two of the four LUT inputs (F1 or F2 on the bottom and G1 or G2 on the top). In addition,
the "0" input can come from a dedicated AND gate (MULT_AND) of those two inputs (for
multiplier functions, as discussed later). It can also be fed directly by a 0 or 1 to use it as a
simple wide gate (to be discussed later). A sixth input comes from the CLB bypass input
(BX or BY) as an alternative to using a LUT input, allowing the carry chain to be initialized
or continued from anywhere in the device. This fixed 6:1 mux driving the 0 input on the
MUXCY is called CY0F in the bottom half of the slice and CY0G in the top half.
The "1" input to the MUXCY is the carry input CIN, which also feeds the XORCY input.
The select input to the MUXCY is the LUT output, where the LUT is typically configured as
an XOR gate for the Propagate selection. This is the same LUT output that provides the
other XORCY input.

Carry Chain Bypass and Initialization

To bypass the carry chain logic and always propagate the carry in signal, the MUXCY can
be set to always select the "1" input. This is done via a Carry Select Mux called CYSELF or
CYSELG at the bottom and top of the slice, respectively.
The "1" input to the MUXCY also supports initialization. Another mux allows the BX input
to drive the MUXCY "1" input instead of CIN. This mux, CYINIT, is only available on the
bottom LUT within a CLB slice. The BX signal comes from outside the CLB and can be
sourced from the dedicated VCC points in the interconnect or from any LUT forced to a 0,
or even from internal logic, allowed initialization to 1, 0, or a variable.
The MUXCY can also be initialized via the "0" input. The "0" input can be permanently
selected by forcing the LUT to a constant 0 and selecting the LUT through the Carry Select
Mux CYSELF or CYSELG. The "0" input comes from the CY0F/G signal. The CY0F/G
mux, in turn, can select a fixed "0" or "1" directly for carry initialization.

XORCY
The XOR gate generically referred to as XORCY is available at both the bottom (called
XORF) and top (XORG) of each slice. The inputs are sourced by the carry signal input CIN
and the LUT output. The XORCY output goes to the primary output of the logic cell (to
both the combinatorial output and the flip-flop). This path goes through a fixed mux that
chooses between the LUT, the wide multiplexers, or the XORCY, which is called FXMUX at
the bottom and GYMUX at the top of the slice.

Carry Logic Connections

The carry path is very fast because it has dedicated connections within and between CLBs.
These dedicated connections have zero delays.

Connections within a Slice

The carry output of the bottom half of a slice connects directly to the carry input on the top
half of the slice (CYMUXF drives directly into CYMUXG), as shown in Figure 9-5,
page 280.

Spartan-3 Generation FPGA User Guide www.xilinx.com 281

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

Connections between Slices and CLBs

The carry output (COUT) of the bottom slice of one side of a CLB then connects directly to
the carry in (CIN) of the top slice. This appears as a net in the design but has zero delay.
SLICEM at X0Y0 connects to SLICEM X0Y1 on the left side, while SLICEL X1Y0 connects to
SLICEL X1Y1 on the right side.
In addition, the COUT of the top slice on one side of a CLB, connects directly to the CIN of
the bottom slice of the CLB above. This net also has a zero delay. SLICEM at X0Y1 connects
to SLICEM at X0Y0 in the CLB above, and SLICEL at X1Y1 connects to SLICEL at X1Y0 in
the CLB above. See Figure 9-6, page 283.

282 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Resource Details

COUT COUT
to S0 of the next CLB to CIN of S2 of the next CLB

0 1 MUXCY
FF
LUT
(First Carry Chain)
SLICEL X1Y1
0 1 MUXCY
FF
LUT

CIN

COUT

0 1 MUXCY
FF
LUT

SLICEL X1Y0
0 1 MUXCY
FF
0 1 MUXCY LUT
FF
LUT

SLICEM X0Y1
0 1 MUXCY
FF
LUT

CIN
COUT (Second Carry Chain)
0 1 MUXCY
FF
LUT

SLICEM X0Y0
0 1 MUXCY
FF
LUT

CIN CIN CLB

UG331_06_073006

Figure 9-6: Carry Logic Connections within a CLB

As a result of the dedicated routing structure, the carry chain runs vertically up the
columns of CLBs, with four bits per CLB in the two slices on one side. The other side of the
CLB has a completely independent carry chain, so there are two chains per column.
The total number of carry chains is twice the number of CLB columns, as shown in
Table 9-3. The number of bits per column is limited by the number of logic cells per

Spartan-3 Generation FPGA User Guide www.xilinx.com 283

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

column. In the Spartan-3E and Extended Spartan-3A family architectures, some of the CLB
columns are interrupted by the DCMs and provide fewer bits per column.

Table 9-3: Number of Carry Chains per Device

Device Number of Carry Chains Bits per Column
XC3S50 24 64
XC3S200 40 96
XC3S400 56 128
XC3S1000 80 192
XC3S1500 104 256
XC3S2000 128 320
XC3S4000 144 384
XC3S5000 160 416

XC3S100E 24 88
XC3S250E 36 136
XC3S500E 52 184
XC3S1200E 76 240
XC3S1600E 100 304

XC3S50A/AN 24 64
XC3S200A/AN 32 128
XC3S400A/AN 48 160
XC3S700A/AN 64 192
XC3S1400A/AN 80 288

XC3SD1800A 96 352
XC3SD3400A 116 416

Carry chains can be split or cascaded to provide even more flexibility. Splitting the carry
chain means connecting the COUT of one MUXCY to the CIN signal of multiple MUXCYs,
continuing the carry into two chains without having to duplicate the logic. Cascading the
carry chain means connecting COUT through normal logic to a CIN other than one directly
above. This can be used to continue a carry chain into a second column.
Splitting and cascading can be done since the COUT from each MUXCY not only feeds the
next MUXCY up the column, but is also available at a CLB bypass output (XB on the
bottom, YB on the top). These CLB outputs can only be driven by the MUXCY or by the
SRL16 shiftout. Also, the CIN can come from the CLB bypass inputs BX/BY or from a LUT
input in either one of the two MUXCY components.

284 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Multiplication Resources

Multiplication Resources
Special resources are also available for multiplication. One-bit multiplication is logically
very simple, requiring only sets of AND gates.

Table 9-4: Binary Multiplication

Product
A B
(A AND B)
0 0 0
0 1 0
1 0 0
1 1 1

Multiplication of larger values is performed by generating partial products by multiplying

each value by one bit of the other factor. These AND gates either allow the input value to
be passed, or force the partial product completely to zero. The partial products are then
added to generate the final product, as shown in Figure 9-7. The carry logic is very effective
for the adder, so a common function preceding it will be the partial product multiplication.

+ P4
A3

+ P3
A2

+ P2
A1

+ P1
A0

+ P0
‘0’
B1 B0
UG331_c11_07_080306

Figure 9-7: Partial Product Multiplication

While the latter stages of the addition tree are pure add functions, look at the way in which
the first two partial products are formed and then applied to the first stage adder in
Figure 9-7. In the majority of cases, the two adder inputs are each driven by a 2-input AND
gate. As these AND gates would each occupy a LUT, a multiplier suddenly becomes very
large in an FPGA. In the case of a 12-bit by 8-bit multiplier it would require 12 x 8 = 96
LUTs (48 slices) just to implement the AND gates.
However, an optimization is quickly visible. The AND gate associated with one of the
adder inputs can be absorbed into the LUT forming the half sum for addition. This in itself
reduces the size of the 12-bit by 8-bit multiplier by 48 LUTs (24 slices).

Spartan-3 Generation FPGA User Guide www.xilinx.com 285

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

COUT

Am
Bn+1
Am+1 0 1
Bn Pm+1

CIN UG331_c11_08_073106

Figure 9-8: Implementing Partial Product Multiplication in a CLB

An ideal situation would be to absorb the other AND gate into the LUT, but the signal it
produces is also required by the MUXCY part of the addition function. So this circuit
appears to be the optimal that can be achieved.
However, Spartan-3 generation FPGAs allow this second AND gate to be absorbed. Next
to each LUT is yet another component called the MULT_AND. It has the effect of recreating
the same input to the MUXCY, even though the desired signal is now buried within the
LUT.

COUT

Am
Bn+1
Am+1 0 1
Bn Pm+1

CIN

MULT_AND UG331_c11_09_073106

Figure 9-9: MULT_AND Optimizes Partial Product Multiplication

The generic MULT_AND gate is called FAND at the bottom of the slice, combining the F1
and F2 LUT inputs into the MUXCY at the bottom, CYMUXF. GAND at the top of the slice
combines G1 and G2 into the MUXCY at the top, CYMUXG (see Figure 9-5).
This dedicated AND gate can be used for any other function besides a multiplier, but it can
only connect to a MUXCY, which in turn can feed a CLB output to any logic.

Component and Pin Names

Table 9-5 summarizes all the names used for the elements of the carry and arithmetic logic.
Italicized names are Slice pins.

Table 9-5: Carry and Arithmetic Logic Names

FPGA Editor
Name Loc in Slice Inputs Outputs
Names
XORCY Bottom XORF CYINIT XOR F X (XQ)
XORCY Top XORG CYMUXF XOR G Y (YQ)
CYMUXF (and
MUXCY Bottom CYMUXF CYINIT or CY0F
XB)

286 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Component and Pin Names

Table 9-5: Carry and Arithmetic Logic Names

FPGA Editor
Name Loc in Slice Inputs Outputs
Names
MUXCY Top CYMUXG CYMUXF or CY0G COUT (and YB)
CYINIT Bottom CYINIT CIN or BX CYINIT
MULT_AND Bottom FAND F1 AND F2 CY0F
MULT_AND Top GAND G1 AND G2 CY0G

COUT and YB are always the same signal in the SLICEL components (SLICEM allows
driving YB from the SRL16 output). In the same way the CYMUXF output always drives
the XB output in the SLICEL.
Table 9-6 summarizes the carry logic functions. For a detailed picture of the CLB slice, see
Chapter 5, “Using Configurable Logic Blocks (CLBs).”

Table 9-6: Carry Logic Functions

Function Description
CYINIT Initializes carry chain for a slice. Fixed selection of:
• CIN carry input from the slice below
• BX input
CY0F Carry generation for bottom half of slice. Fixed selection of:
• F1 or F2 inputs to the LUT (both equal 1 when a carry is to be generated)
• FAND gate for multiplication
• BX input for carry initialization
• Fixed "1" or "0" input for use as a simple Boolean function
CY0G Carry generation for top half of slice. Fixed selection of:
• G1 or G2 inputs to the LUT (both equal 1 when a carry is to be generated)
• GAND gate for multiplication
• BY input for carry initialization
• Fixed "1" or "0" input for use as a simple Boolean function
CYMUXF Carry generation or propagation mux for bottom half of slice. Dynamic selection via CYSELF of:
• CYINIT carry propagation (CYSELF = 1)
• CY0F carry generation (CYSELF = 0)
CYMUXG Carry generation or propagation mux for top half of slice. Dynamic selection via CYSELF of:
• CYMUXF carry propagation (CYSELG = 1)
• CY0G carry generation (CYSELG = 0)
CYSELF Carry generation or propagation select for bottom half of slice. Fixed selection of:
• F-LUT output (typically XOR result)
• Fixed "1" to always propagate
CYSELG Carry generation or propagation select for top half of slice. Fixed selection of:
• G-LUT output (typically XOR result)
• Fixed "1" to always propagate

Spartan-3 Generation FPGA User Guide www.xilinx.com 287

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

Table 9-6: Carry Logic Functions

Function Description
XORF Sum generation for bottom half of slice. Inputs from:
• F-LUT
• CYINIT carry signal from previous stage
Result is sent to either the combinatorial or registered output for the top of the slice.
XORG Sum generation for top half of slice. Inputs from:
• G-LUT
• CYMUXF carry signal from previous stage
Result is sent to either the combinatorial or registered output for the top of the slice.
FAND Multiplier partial product for bottom half of slice. Inputs:
• F-LUT F1 input
• F-LUT F2 input
Result is sent through CY0F to become the carry generate signal into CYMUXF
GAND Multiplier partial product for top half of slice. Inputs:
• G-LUT G1 input
• G-LUT G2 input
Result is sent through CY0G to become the carry generate signal into CYMUXG

Performance
Performance of carry logic based functions is determined by three components: the delay
to get into the carry chain, the delay for each bit of the function in the carry chain, and the
delay to generate the last result (see Figure 9-10). The delay to get into the carry chain is
from the F inputs to the COUT output, tOPCYF, is approximately 0.9 ns (see Figure 9-11).
The delay for each slice is the zero delay routing plus the delay from CIN to COUT, which
is tBYP, approximately 0.2 ns (see Figure 9-12). Some functions can fit four bits per slice
while most fit two bits per slice. The delay to generate the final result is typically the CIN
delay to the YQ output or tCINCK, which is approximately 1.3 ns, or tCINY for a
combinatorial result, which is approximately 1.2 ns (see Figure 9-13). Thus the total delay
is 2.1 ns for four bits (2.2 ns registered) plus 0.2 ns for every additional two bits.

288 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Performance

CLB
0 1 MUXCY
FF
LUT Q7

TCINCK SLICEL X1Y1

0 1 MUXCY
FF
LUT Q6

CIN
TNET = 0
COUT

0 1 MUXCY
FF
LUT Q5

TBYP
SLICEL X1Y0
0 1 MUXCY
FF
LUT Q4

TNET = 0

CLB
0 1 MUXCY
FF
LUT Q3

TBYP SLICEL X1Y1

0 1 MUXCY
FF
LUT Q2

CIN
TNET = 0
COUT

0 1 MUXCY
FF
LUT Q1

TOPCYF
SLICEL X1Y0
0 1 MUXCY
FF
LUT Q0

UG331_10_073106

Figure 9-10: Carry Delay Path

Spartan-3 Generation FPGA User Guide www.xilinx.com 289

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

COUT

YB
CYSELG
1 CYMUXG Y
G4 0 1
G3 D Q YQ
G2 G-LUT
G1 XORG

CY0G

GAND 1
0
BY

XB
CYSELF
1 CYMUXF X
F4 0 1
F3 D Q XQ
F2 F-LUT
F1 XORF

CY0F

FAND 1 CYINT TOPCYF

0 1
0

CIN
BX UG331_c11_11_073106

Figure 9-11: Bottom Operand Input to Carry Out, TOPCYF

290 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Performance

COUT

YB
CYSELG
1 CYMUXG Y
G4 0 1
G3 D Q YQ
G2 G-LUT
G1 XORG

CY0G

GAND 1
0
BY

XB
CYSELF
1 CYMUXF X
F4 0 1
F3 D Q XQ
F2 F-LUT
F1 XORF

CY0F

FAND 1 CYINT
0 1 TBYP
0

CIN
BX UG331_c11_12_073106

Figure 9-12: Carry Propagation, tBYP

Spartan-3 Generation FPGA User Guide www.xilinx.com 291

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

COUT

YB
CYSELG
1 CYMUXG Y
G4 0 1
G3 D Q YQ
G2 G-LUT
G1 XORG

CY0G

GAND 1
0
BY

XB
CYSELF
1 CYMUXF X
F4 0 1
F3 D Q XQ
F2 F-LUT
F1 XORF

CY0F

FAND 1 CYINT TCINY

0 1
0

CIN
BX UG331_c11_13_073106

Figure 9-13: Carry Input to Top Sum Combinatorial Output, tCINY

Optimize the delays to get to and from the carry-based function. Bring the source inputs
close to the carry function; use registered signals in a column to the left of the carry
function if possible to provide near-zero routing delay to the carry function. The output of
the carry function can be registered directly in the same CLB to provide high performance
through pipelining.
The SLICEL timing is often faster than the SLICEM timing due to the simpler structure,
and therefore the SLICEL should be favored for the highest performance.
Use the following estimates for Spartan-3 generation carry-based adders, counters, and
accumulators:
• 8 bits: 3.0 ns or 333 MHz
• 16 bits: 3.8 ns or 263 MHz
• 32 bits: 5.4 ns or 185 MHz
• 64 bits: 8.6 ns or 116 MHz

Specifications
The carry and arithmetic logic is defined by multiple timing specifications to cover each of
the possible signal paths, as described in Table 9-7.

292 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Specifications

Table 9-7: Other Specifications

Specification Description Path
CIN input through CYINIT through CYMUXF
TCINYB Carry split or cascade to top slice
through CYMUXG to YB output
CIN input through CYINIT through CYMUXF to
TCINXB Carry split or cascade to bottom slice
XB output
CIN input through CYINIT through XORF to X
TCINX Carry input to bottom sum combinatorial output
output
CIN input through CYINIT through XORF to FFX
Carry input to top or bottom sum registered
TCINCK setup; or to top logic cell through CYINIT
output setup
through CYMUXF through XORG to FFY setup
Hold time for TCINCK, from CIN through XOR to
TCKCIN Carry input hold time
flip-flops
Bottom bypass input to carry output for BX input through CYINIT (or CY0F) through
TBXCY
initialization CYMUXF through CYMUXG to COUT output
BX input through CYINIT (or CY0F) through
TBXYB Carry split
CYMUXF through CYMUXG to YB output
BX input through CYINIT (or CY0F) through
TBXXB Carry split
CYMUXF to XB output
BY input through CY0G through CYMUXG to
TBYCY Top bypass input to carry output for initialization
COUT output
BY input through CY0G through CYMUXG to YB
TBYYB Carry split
output
BX input through CYINIT through CYMUXF
TBXY Bottom bypass input to top sum for initialization
through XORG to Y output
Bottom bypass input to bottom sum for BX input through CYINIT through XORF to X
TBXX
initialization output
BX input through CYINIT through XORF to FFX
TDICK Bottom bypass input to sum setup time setup (or through CYMUXF through XORG to
FFY setup)
F1-F2 inputs through CY0F through CYMUXF (or
TOPFYB Bottom operand input to carry split out F1-F4 inputs through F-LUT through CYSELF to
CYMUXF select) through CYMUXG to YB
F1-F4 inputs through F- LUT through CYSELF
TOPXB Bottom operand input to carry split out
through CYMUXF select to XB output
F1-F4 inputs through F-LUT through CYSELF
TOPY Bottom operand input to top sum through CYMUXF select through XORG to Y
output
F1-F4 inputs through F-LUT through XORF to X
TOPX Bottom operand input to bottom sum
output
G1-G2 inputs through CY0G through CYMUXG
TOPCYG Top operand input to carry out (or G1-G4 through G-LUT through CYSELG to
CYMUXG select) to COUT output

Spartan-3 Generation FPGA User Guide www.xilinx.com 293

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

Table 9-7: Other Specifications

Specification Description Path
G1-G2 inputs through CY0G through CYMUXG
TOPGYB Top operand input to carry split out (or G1-G4 through G-LUT through CYSELG to
CYMUXG select) to YB output
G1-G4 inputs through G-LUT through XORG to Y
TOPGY Top operand input to sum
output
F1/F2 inputs through FAND through CY0F
TFANDCY Bottom factor input to carry out
through CYMUXF through CYMUXG to COUT
F1/F2 inputs through FAND through CY0F
TFANDYB Bottom factor input to carry split out through CYMUXF through CYMUXG to YB
output
F1/F2 inputs through FAND through CY0F
TFANDXB Bottom factor input to bottom carry split out
through CYMUXF to XB
G1/G2 inputs through GAND through CY0G
TGANDCY Top factor input to carry out
through CYMUXG to COUT output
G1/G2 inputs through GAND through CY0G
TGANDYB Top factor input to carry split out
through CYMUXG to YB output

Designing with the Carry and Arithmetic Logic

It is important to understand whether a design uses the carry logic and make sure it is
being used effectively. Carry logic can be instantiated via primitives or macros, selected for
components created by the CORE Generator system, or used automatically by high-level
synthesis tools.

Library Elements Using Carry

The most direct way to use the carry logic is to instantiate the library components that
already have it built in. Library components are designed to use the carry and arithmetic
logic when they will be efficient in most designs. The adders, adder/subtractors, and
accumulators use the carry, along with the 8-bit and 16-bit counters that start with "CC".
The widest logic gates of 16 inputs use the carry multiplexers. Any custom function
beyond these can be built from the carry primitives.

294 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with the Carry and Arithmetic Logic

Primitives
MUXCY

MUXCY
S
0 1

DI CI
UG331_c11_14_073106

Figure 9-14: MUXCY Primitive

The MUXCY primitive is used to implement a 1-bit high-speed carry propagate function.
DI is mapped to a CLB Direct Input while CI is the Carry Input. The select input S comes
from the LUT; when Low, S selects DI; when High, S selects CI. The O output can cascade
to the CI of the next MUXCY above or be fed to a CLB output.
The MUXCY primitive gets mapped to the CYMUXF or CYMUXG components at the
bottom and top of the Slice, respectively. The S select input is normally driven by an XOR
gate in a LUT, but the LUT can be fixed to zero to always select the DI input. A fixed 1 on
the S input always selects the CI carry input, and can be implemented inside the mux itself,
saving the LUT for other functions.
The MUXCY is also available as two additional primitives with "local" outputs. Local
outputs reflect the dedicated connections between logic elements, in this case the direct
connections from COUT to CIN. The local output on the primitive does not control the
routing but allows the design tools to better estimate the timing before implementation. An
O pin on a MUXCY connected to a CI pin on another MUXCY almost always uses zero-
delay connections, reflected by the local output (LO). The general-purpose output reflects
the longer block and routing delays for splitting the carry chain by feeding it to the bypass
outputs XB or YB. MUXCY_L will model the zero delay of the COUT path while
MUXCY_D has both local and general-purpose outputs. Both paths are always available
and can be used at the same time. If the O pin connects to a CI of another MUXCY, use the
LO output of the MUXCY_L or MUXCY_D. If O connects to anything else, use a generic
MUXCY or a the O output of a MUXCY_D.

LO LO O

MUXCY_L MUXCY_D
S S
0 1 0 1

DI CI DI CI
UG331_c11_15_073106

Figure 9-15: MUXCY_D and MUXCY_L Primitives

Spartan-3 Generation FPGA User Guide www.xilinx.com 295

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

VHDL Instantiation
-- Component Declaration for MUXCY should be placed
-- after architecture statement but before begin keyword
component MUXCY
port (O : out STD_ULOGIC;
CI : in STD_ULOGIC;
DI : in STD_ULOGIC;
S : in STD_ULOGIC);
end component;
-- Component Attribute specification for MUXCY
-- should be placed after architecture declaration but
-- before the begin keyword
-- Attributes should be placed here
-- Component Instantiation for MUXCY should be placed
-- in architecture after the begin keyword
MUXCY_INSTANCE_NAME : MUXCY
port map (O => user_O,
CI => user_CI,
DI => user_DI,
S => user_S);

Verilog Instantiation
MUXCY MUXCY_instance_name (.O (user_O),
.CI (user_CI),
.DI (user_DI),
.S (user_S));

XORCY
The XORCY primitive is a special dedicated XOR with general output used for generating
faster and smaller arithmetic functions. The XORCY primitive gets mapped to the XORF or
XORG component in the bottom or top of the slice, respectively. The Logic Input (LI) is
driven by the LUT output, typically the same as the S input on the MUXCY. The Carry
Input (CI) is driven by the output of a MUXCY or initialized by another signal. The O
output drives the combinatorial or registered output of the slice.

LI
O
CI
UG331_c11_16_073106

Figure 9-16: XORCY Primitive

XORCY is also available as two additional primitives with Local Outputs or LO pins.
Although there is no special routing for the output of an XORCY as there is for the
MUXCY, the fast delay modeled by the Local Output can be used when direct connections
are to be used to the adjacent CLB or back to the same CLB, or when the XORCY directly
feeds a flip-flop.
As mentioned earlier the XORCY is used to complete the sum initiated by an XOR in the
LUT. The XOR in the LUT is represented by a general-purpose XOR2 component or similar
function.

VHDL Instantiation
-- Component Declaration for XORCY should be placed
-- after architecture statement but before begin keyword

296 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Designing with the Carry and Arithmetic Logic

component XORCY
port (O : out STD_ULOGIC;
CI : in STD_ULOGIC;
LI : in STD_ULOGIC;
end component;
-- Component Attribute specification for XORCY
-- should be placed after architecture declaration but
-- before the begin keyword
-- Attributes should be placed here
-- Component Instantiation for XORCY should be placed
-- in architecture after the begin keyword
XORCY_INSTANCE_NAME : XORCY
port map (O => user_O,
CI => user_CI,
LI => user_LI);

Verilog Instantiation
XORCY XORCY_instance_name (.O (user_O),
.CI (user_CI),
.LI (user_LI));

MULT_AND
The MULT_AND primitive is an AND component used almost exclusively for building
faster and smaller multipliers. The MULT_AND primitive maps into the FAND or GAND
gate in the Spartan-3 FPGA Slices. The inputs come from two specific LUT inputs, F1 and
F2 or G1 and G2. The output can only connect to the DI input on a MUXCY, so the
primitive is only available with an "LO" Local Output to reflect the lack of any routing
delays in pre-implementation timing analysis.

I1
LO
I0
UG331_c11_17_073106

Figure 9-17: MULT_AND Primitive

Even if a generic AND2 gate is used, if it feeds into the MUXCY data input, it will likely be
placed in the MULT_AND for efficiency. In the same way, a MULT_AND feeding into a
generic M2_1 mux typically forces the mux into the MUXCY. The MULT_AND is used to
"duplicate" one of two AND gates in the LUT in a typical multiplier. Those AND gates are
designated by general-purpose AND2 components.

VHDL Instantiation
-- Component Declaration for MULT_AND should be placed
-- after architecture statement but before begin keyword
component MULT_AND
port (LO : out STD_ULOGIC;
I0 : in STD_ULOGIC;
I1 : in STD_ULOGIC);
end component;
-- Component Attribute specification for MULT_AND
-- should be placed after architecture declaration but
-- before the begin keyword
-- Attributes should be placed here
-- Component Instantiation for MULT_AND should be placed

Spartan-3 Generation FPGA User Guide www.xilinx.com 297

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

-- in architecture after the begin keyword

MULT_AND_INSTANCE_NAME : MULT_AND
port map (LO => user_LO,
I0 => user_I0,
I1 => user_I1);

Verilog Instantiation
MULT_AND MULT_AND_instance_name (.LO (user_LO),
.I0 (user_I0),
.I1 (user_I1));

Emulating Virtex-II FPGA ORCY Components

The ORCY primitive found in the Virtex®-II architecture is not available in the
Spartan-3 generation FPGAs. Although labeled similarly to the carry primitives, ORCY is
not typically used for arithmetic functions. It is used to create a Sum-Of-Products solution
in the Virtex-II family, creating an OR of the wide AND gates that can be created using the
MUXCY resources. In Spartan-3 generation FPGAs, similar logic can be implemented in
the LUTs with more placement flexibility. If an ORCY component is found in a
Spartan-3 generation design, it is mapped to a LUT.

Macros
Table 9-8 shows the library macros that use the carry logic. All adders, adder/subtractors,
and accumulators use the carry logic. Only counters that begin with "CC" use the carry
logic, along with Magnitude Comparators with "MC". Wide gates of 16 bits use the carry
logic while 12-bit and smaller gates do not. None of the standard library macros use the
MULT_AND gate.

Table 9-8: Library Macros Using Carry Logic

Carry Resources
Macro Name Macro Description Slices
Used
Acct. 4-bit Accumulator 6 MUXCY, XORCY
Acct. 8-bit Accumulator 10 MUXCY, XORCY
Acct. 16-bit Accumulator 18 MUXCY, XORCY
Add 4-bit Adder 3 MUXCY, XORCY
Add 4-bit Adder 5 MUXCY, XORCY
Add 4-bit Adder 9 MUXCY, XORCY
Addis 4-bit Adder/Subtractor 3 MUXCY, XORCY
ADSU8 4-bit Adder/Subtractor 5 MUXCY, XORCY
ADSU16 4-bit Adder/Subtractor 9 MUXCY, XORCY
AND16 16-bit AND gate 2 MUXCY
8-bit binary counter with clear and
CC8CE 5 MUXCY, XORCY
clock enable
16-bit binary counter with clear and
CC16CE 9 MUXCY, XORCY
clock enable

298 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Using the CORE Generator System

Table 9-8: Library Macros Using Carry Logic

Carry Resources
Macro Name Macro Description Slices
Used
8/16-bit binary counter with clear,
CC8/16CLE 9/17 MUXCY, XORCY
load, and clock enable
8-bit binary counter with clear, load,
CC8/16CLED 17/33 MUXCY, XORCY
bidirectional, and clock enable
8/16-bit binary counter with reset and
CC8/16RE 9/17 MUXCY, XORCY
clock enable
COMPMC8 8-bit Magnitude Comparator 8 MUXCY
COMPMC16 16-bit Magnitude Comparator 16 MUXCY
NAND16 16-bit NAND gate 2 MUXCY
NOR16 16-bit NAND gate 2 MUXCY
OR16 16-bit NAND gate 2 MUXCY

Using the CORE Generator System

The Xilinx CORE Generator system can be used to generate more complex or customized
functions. The CORE Generator software should be used for multipliers, since it can use
the MUXCY and MULT_AND components or trade off resources with the dedicated
multipliers.

Adder
The adder and adder/subtractor components in the CORE Generator software
automatically use the carry logic in a similar fashion to the library macros. The CORE
Generator version allows for more flexibility in terms of data widths and registered
outputs, among other functions.

Spartan-3 Generation FPGA User Guide www.xilinx.com 299

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

ASET SSET

D_OVFL Q_OVFL
ADD D_C_OUT Q_C_OUT
OVFL
A[N:0] C_OUT

A_SIGNED
S[P:0] D[P:0] Q[P:0]
B_SIGNED
B[M:0] BYPASS
BYPASS CE
C_IN
CLK

ACLR SCLR AINIT SINIT

UG331_c11_18_073106

Figure 9-18: CORE Generator Adder/Subtractor

Accumulator
The CORE Generator system's Accumulator is similar to the Adder/Subtractor but with
the registered output Q feeding back in the A input of the adder. Functions built using the
Accumulator use the Spartan-3 generation carry logic.

Comparator
The CORE Generator system's Comparator function also takes advantage of the carry logic
in implementation.

Multiplier
The CORE Generator system also includes a multiplier that can be targeted to either the
carry logic or the dedicated 18 x 18 multipliers. The graphical interface for the Multiplier
core includes the Multiplier Construction option. If Use LUTs is selected, the multiplier is
built using the carry and MULT_AND logic. If Use 18 x 18 Multiplier Blocks is selected, the
multiplier is built using the dedicated 18 x 18 multipliers.

300 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Carry and Synthesis Constraints

UG331_c11_19_073106

Figure 9-19: Multiplier Interface in CORE Generator System

Logic Gates
The 16-input gates in the library are implemented using carry logic. However, the logic
gates in the CORE Generator system basic elements get implemented in LUTs, not the
carry logic. When Use RPMs is selected, the function is placed in a column similar to the
carry-based functions. The arithmetic functions always use the carry logic even if Use
RPMs is not selected.

Carry and Synthesis Constraints

Most synthesis tools recognize arithmetic operators as opportunities to use the carry logic.
Synplicity Synplify and Mentor Precision tools automatically infer usage of the MUXCY
for adders and related arithmetic functions and the MULT_AND for multipliers that do
not use the dedicated 18 x 18 resources. Check with your synthesis tool vendor for specific
information on when it will use these resources.
The Xilinx Synthesis Tool (XST) also automatically infers the use of the carry and
arithmetic logic. A synthesis constraint USE_CARRY_CHAIN can be applied locally or
globally to force the tool to use the carry logic. After trying automatic synthesis, use this
option in designs containing arithmetic logic to see if it provides a more efficient
implementation. The parameter has two settings: Yes or No.

MUX_STYLE Constraint
The MUX_STYLE constraint guides the Xilinx XST synthesis tool to the type of multiplexer
implementation desired. This constraint controls the way the macrogenerator implements
the multiplexer functions. Allowed values are:

Spartan-3 Generation FPGA User Guide www.xilinx.com 301

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

• Auto: let synthesis tools decide which implementation is best (default)

• MUXF: use dedicated MUXF5, MUXF6, MUXF7, or MUXF8 multiplexers
• MUXCY: use dedicated MUXCY carry multiplexers.
The MUXF is used specifically to combine the results of LUTs or other MUXF functions,
under control of a dedicated input, for the purpose of expanding a general-use function to
more inputs. The MUXCY is restricted to combining a single LUT input with a dedicated
carry input, under control of a LUT, and drives the carry output, for the purpose of
propagating an arithmetic carry. MUXF and MUXCY have very different functions, and
typically it will be clear whether to use one or the other. See Chapter 8, “Using Dedicated
Multiplexers,” for more information on MUXF.

MULT_STYLE Constraint
The MULT_STYLE constraint guides the XST synthesis tool to the type of multiplier
implementation desired. This constraint controls the way the macrogenerator implements
the multiplier macros. Allowed values are:
• Auto: let synthesis tools decide which implementation is best (default in Project
Navigator)
• Block: use dedicated MULT18X18 multipliers
• LUT (default at command line): implement using carry and MULT_AND logic
• Pipe_block: use dedicated MULT18X18S pipeline multipliers (to be supported in a
future release)
• KCM: Constant Coefficient Multiplier (command line only)
• Pipe_LUT: pipeline multiplier using carry and MULT_AND logic with registered
inputs and outputs
Try selecting specific types of multiplier implementations to determine if the performance
or density improves for your own designs.

Carry and Relative Location Constraints

Carry logic is most efficient when it is implemented in a single column of slices, since the
carry chain has direct connections running up each column. This placement can be found
automatically by the place & route tools, or defined by the user via one of several
alternative methods.
Xilinx library macros and CORE Generator System macros using the carry logic include
mapping, placement and routing information using two types of constraints, FMAPs and
RLOCs. The FMAP symbol is used to map logic to the function generator of a slice. FMAP
constraints define the inputs and outputs of each LUT. Note that for carry-based functions,
only the I1 or I2 inputs on the LUT can connect to the MUXCY or the MULT_AND, so the
FMAP symbol needs to have the arithmetic inputs on one of those two pins. Dummy
signals can be placed on the other two inputs if unused to prevent the arithmetic inputs
from being moved there. The FMAP symbol is primarily being used for relative placement,
however, which requires the addition of an RLOC property.
Relative Location (RLOC) constraints define the placement of the carry logic components
and LUTs relative to each other. These macros are known as Relationally Placed Macros or
RPMs. RPMs provide order and structure to related design elements without requiring
specification of their absolute placement on the FPGA die. This gives the implementation
tools more flexibility to meet timing whereas floorplanning requires absolute placement of
the logic. The advantage of relative placement is that it allows the function to move as a

302 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Carry and Floorplanning

complete whole anywhere in the device. For example, an adder can be defined as requiring
adjacent slices in one column, but that stack of slices can be placed by the tools wherever it
is most efficient inside the device.
RLOC constraints can be applied to any of the carry primitives - MUXCY, XORCY, and
MULT_AND, along with flip-flops. An RLOC constraint cannot be applied directly to a
gate primitive but it can be applied to a LUT defined via the FMAP component. In
Spartan-3 generation designs, the RLOC constraint is specified using the slice-based XY
coordinate system (RLOC = X0Y0, etc.). Slices are numbered on an XY grid beginning in
the lower left corner of the chip. X ascends in value horizontally from left to right. Y
ascends in value vertically from bottom to top. A CLB actually encompasses two rows and
two columns of the coordinate system; slices S2 (X1Y0) and S3 (X1Y1) are considered
horizontally adjacent to slices S0 (X0Y0) and S1 (X0Y1) in the CLB. Each part of the
hierarchy can have its own independent set of relative constraints, and the user can even
define multiple sets per hierarchical block.

Left-Hand SLICEM Right-Hand SLICEL

(Logic or Distributed RAM (Logic Only)
or Shift Register)
COUT

CLB
SLICE
X1Y1

SLICE
X1Y0
COUT
Switch Interconnect
Matrix CIN to Neighbors
SLICE
X0Y1
SHIFTOUT
SHIFTIN
SLICE
X0Y0

CIN DS099-2_05_082104

Figure 9-20: Arrangement of Slices within the CLB

Carry and Floorplanning

RLOCs can be specified directly in HDL code or a schematic, defined in the PlanAhead™
software or written into a constraints file. The RPMs are always placed as a complete
group.
Because carry-based functions are in a column with the MSB at the top, related functions
should be placed in a similar fashion. The data inputs should come from a column of CLBs,
or IOBs on the left or right edges, also with MSB at the top.
RLOC constraints allow the user to place logic blocks relative to each other to increase
speed and use die resources efficiently. When defining relative location constraints, it is
important to remember how "tall" the target device is and stay within one column.
However, if an area constraint is too tall, the carry path is automatically split to fit across
multiple columns. With two LUTs per slice, the bottom two FMAPs should be assigned

Spartan-3 Generation FPGA User Guide www.xilinx.com 303

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

RLOC=X0Y0, the next two RLOC=X0Y1, and so on. See the example for the ADD4 library
component in Figure 9-21.

FMAP
B3 I3 RLOC=X0Y1 dummy
A3 dummy I3
B3
A3
RLOC=X0Y1
FMAP
B2 I2 dummy
A2 RLOC=X0Y1
dummy I2
B2
A2
RLOC=X0Y1
FMAP
B1 I1 RLOC=X0Y0 dummy
A1 dummy I1
B1
A1
RLOC=X0Y0
FMAP
B0 I0 RLOC=X0Y0 dummy
A0 dummy I0
B0
A[3:0] A0
B[3:0] RLOC=X0Y0

UG331_c11_21_080606

Figure 9-21: ADD4 Schematic Implementation

For more information on relative location constraints, see the Libraries and Constraints
guides. If relative locations are to be applied to both slices and other resources such as
block RAM, consider using the RPM GRID system described in XAPP416.

Applications
Although the carry logic is most directly applicable to arithmetic functions, it can also be
used for other types of logic.

Wide Gates
The MUXCY is useful for general-purpose logic because it is controlled by the LUT and can
have a fixed 0 or 1 input without using resources. An AND gate can be implemented in a
mux by selecting the input "A" when B is High, as shown in Figure 9-22. The 0 data on the
0 side of the mux is available as a fixed input within the slice and does not require any
resources.

304 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Applications

A*B

B 0 1

0 A
UG331_c11_22_080106

Figure 9-22: Using a MUXCY Multiplexer as an AND Gate

With two LUTs and two MUXCYs per slice, two four-input functions can be combined into
one result in each slice, as shown in Figure 9-23.

F*G

G
LUT MUXCY
0 1

0
F
LUT MUXCY
0 1

0 1
UG331_c11_23_080106

Figure 9-23: Using MUXCY in Slice to AND Two 4-Input Functions

The initial 1 on the 1 input of the MUXCY can be sourced by one of the global 1 signals
available within the FPGA structure. Alternatively, it can be connected to a ninth input to
create a 9-input AND gate. With appropriate inversions, the AND function can be turned
into a NAND, OR, or NOR gate, with the same efficiency. A 1 is also available on the
MUXCY data input, while an initial zero can be generated in any unused LUT.
This implementation of wide logic functions provides higher performance and more
efficient utilization of the FPGA resources. The carry chain eliminates multiple levels of
logic and provides a fast path to the final result. The only limit on the width of the gate is
the number of LUTs in a column, allowing over 400 inputs in one function. Common
applications include wide input decoding, comparators, and counters.
The 16-input gates in the Xilinx library use the MUXCY logic (see Figure 9-24). 12-input
gates and smaller use multiple levels of LUTs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 305

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

OUT

4 MUXCY
0 1
LUT
0
Slice
4 MUXCY
0 1
LUT
0
16
AND OUT

4 MUXCY
0 1
LUT
0
Slice
4 MUXCY
0 1
LUT
VCC
UG331_c11_24_080106

Figure 9-24: Library AND16 Implementation Using Carry Logic

Remember that another alternative for wide logic functions is to use the F5MUX and
FiMUX (F6MUX, F7MUX, F8MUX). These multiplexers are more efficient for registered
functions since they feed directly into the flip-flop in the same CLB, and can create any
function of up to 8 inputs in one level of logic, and some functions of up to 79 inputs. See
Chapter 8, “Using Dedicated Multiplexers” for more details.

Sum of Products
Generic logic descriptions will always be optimized into the four-input LUTs of the
Spartan-3 architecture, minimizing the number of resources required. The Xilinx software
is very efficient at optimizing logic to fit into the LUT structure, where the only limit is the
number of inputs, not the type of function.
Some logic architectures, including CPLDs, incorporate a sum-of-products structure, using
wide AND gates followed by OR gates. The wide AND gates can be implemented in the
Spartan-3 FPGA using the carry logic. The OR gates would be implemented in LUTs, with
up to four wide AND gates able to be combined in one fast LUT. Since carry-based AND
gates will be vertical with the result at the top, the OR LUT should be placed in a CLB
above the column-based AND gates.
Note that the ORCY function, used in the Virtex-II and Virtex-II Pro families to OR together
carry-based AND gates, is not available in the Spartan-3 family. Using the LUT instead
provides for a smaller CLB (and therefore lower cost), and offers more placement
flexibility.

Comparators
The AND function in the MUXCY can be extended to implement an equality comparator
of two four-bit values per slice.

306 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Applications

A=B

A3 A3=B3
B3
A(2:3)=B(2:3)
0 1 MUXCY
A2
B2 A2=B2
0

A1 A1=B1
B1
A(0:1)=B(0:1)
0 1 MUXCY
A0
B0 A0=B0
0 1
UG331_c11_25_080106

Figure 9-25: Equality Comparator in One Slice

A magnitude comparator can also be implemented using the carry logic, at two bits per
slice.

A1=B1
B1
A1 0 1 MUXCY

A0 < B0

A0=B0
A0
B0 0 1 MUXCY

UG331_c11_26_080106

Figure 9-26: Magnitude Comparator in One Slice

As with the logic gates, inverters can be used to generate inequality or other comparisons.

Adders
The adder is the fundamental function of the carry logic, as described earlier. Two bits per
Logic Cell can be added.

Spartan-3 Generation FPGA User Guide www.xilinx.com 307

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

COUT
Function
Generator MUXCY

B
A 0 1
Sum
D Q

CIN
XORCY
UG331_c11_27_080106

Figure 9-27: Basic Adder Cell

Counters
A binary counter can be implemented by toggling each flip-flop when all the lower-order
flip-flops are High. Figure 9-28 shows a typical binary up counter using toggle flip-flops.
Each bit toggles if all the lower-order bits are high. The AND gates are required for each bit
and get wider as the counter gets larger.

T Q

T Q

T Q

T Q

1 T Q

UG331_c11_28_080106

Figure 9-28: Binary Counter

These wide AND gates are also a candidate for the carry logic, especially since it avoids
having to duplicate the gate at different widths. Figure 9-29 shows the same binary counter
using the MUXCY in place of the AND gates. Each MUXCY expands the width of the AND
gate by the additional bit needed for each stage of the counter, and there is no redundant
logic. The performance limit is only the propagation of the carry chain instead of wide
AND gates.

308 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Applications

MUXCY
0 1

T Q
0

MUXCY 0 1

T Q
0

MUXCY 0 1

T Q
0

MUXCY 0 1

T Q
0

1 T Q

UG331_c11_29_080106

Figure 9-29: Binary Counter Using MUXCY for AND Gates

Figure 9-30 shows the implementation of the carry-based binary counter using the D flip-
flops available in the slices. The inverter and XOR gates are implemented in the LUT
preceding each flip-flop.

MUXCY
0 1

D Q
0

MUXCY 0 1

D Q
0

MUXCY 0 1

D Q
0

MUXCY 0 1

D Q
0

MUXCY 0 1

D Q
0 1

UG331_c11_30_080606

Figure 9-30: Binary Counter Using D Flip-Flops

Spartan-3 Generation FPGA User Guide www.xilinx.com 309

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

Multipliers
Multiplication is typically done by generating partial products and then adding the results.
The carry logic optimizes both aspects of the design.
One-bit multiplication is logically very simple, requiring only sets of AND gates. These
gates either allow the input value to be passed or force the partial product completely to
zero.

A3 A3

A2 A2
First Second
Partial Partial
A1 Product A1 Product

A0 A0

B0 B1 UG331_c11_31_080206

Figure 9-31: Partial Product Generation

Then, all the partial products need to be added together with the appropriate bit
weighting. If there is sufficient time (enough clock cycles), the classical serial "shift and
add" technique can be adopted based on an accumulator; however, for a maximum
performance parallel multiplier, an addition tree is needed. This tree is effectively
implemented using carry-based adders, with pipelining registers available if desired for
better performance.

COUT
+ P4
A3

+ P3
A2

+ P2
A1

A0 + P1

+ P0
‘0’ CIN
B1 B0
UG331_c11_32_011508

Figure 9-32: Partial Product Multiplication

These AND gates can be implemented in the MULT_AND and the LUTs available in the
CLBs.

310 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Applications

LUT

CY_MUX

CY_XOR
MULT_AND

Dedicated
AxB AND Gate

B
UG331_c11_33_080206

Figure 9-33: Multiplier Implementation

Optimizing Carry-Based Multipliers

Figure 9-9, page 286 shows that two bits can be multiplied at a time using the carry and
arithmetic logic. Therefore one of the inputs must be divided into 2-bit pieces and then the
partial products combined in an adder tree. Fewer levels of logic will be required in the
adder tree if the smaller value is the one divided into 2-bit sections. Also consider dividing
the input whose number of bits is a power of 2, since that will provide more symmetry.
For example, a multiplier of an 8-bit value by a 12-bit value has the two possible
implementations shown in Figure 9-34 and Figure 9-35.

12
A 14
+
16
+
14
+
20
+
14
+
16
+
14
8 +
B
3 Levels of Logic (3-Stage Pipeline)
54 Slices Fully Pipelined
UG331_c11_34_080606

Figure 9-34: 12 x 8 Multiplier Dividing 8-Bit Input into 2-Bit Sections

Spartan-3 Generation FPGA User Guide www.xilinx.com 311

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

8
A 10
+
12
+
10
+
16
+
10
+
12 20
+ +
10
+
Compensation
10 Delay
+
12
+
10
12 +
B

4 Levels of Logic
72 Slices Fully Pipelined UG331_c11_35_080606

Figure 9-35: 8 x 12 Multiplier Dividing 12-Bit Input into 2-Bit Sections

The fact that 8 is a power of 2 means that the "12 x 8" breaks down nicely into 4 multiplier
adders in the first stage; hence, it leads to a symmetrical addition tree of 3 levels. In
contrast, the "8 x 12" is less elegant: the 6 multiplier adders of the first stage do not sum
easily, leading to more adders and 4 levels of logic. For a fully pipelined multiplier, there is
even the requirement for delay compensation.
The multiplier adders and pure adders of the "8 x 12" are generally a smaller number of
bits than in the "12 x 8"; but with the efficient carry-based adders in the Spartan-3
architecture, this has a very minimal impact on performance. In any case, both multipliers
have the same largest-size adder at the final stage. Combinatorial multiplier performance
will be set by the number of logic levels, and in this case, the "12 x 8" will definitely win.
It is difficult to know for certain how each design entry tool handles the implementation of
complex functions, so experiment with alternative implementations. Even simply
switching the order of the inputs could have a significant affect on the performance and
resource requirements.

MULT_AND vs. MULT18X18

The combination of the MULT_AND with the carry logic provides an efficient
implementation of small multipliers. The multipliers can be placed in any column of the
device. Larger multipliers should use the dedicated MULT18X18 resource, which provides
high-speed multiplication of two 18-bit signed or two 17-bit unsigned values, without
using any CLB resources. The different Spartan-3 FPGA densities have from 4 to 104 of the
dedicated multipliers, and the Spartan-3A DSP platform has 84 to 126 DSP48A blocks. The
dedicated resources should be used even for small functions if CLB resources are at a
premium. The dedicated resources are faster for larger functions, especially if the inputs
and outputs are pipelined with CLB flip-flops, allowing over 150 MHz multiplication. See
Chapter 11, “Using Embedded Multipliers,” for more information.

MULT_AND vs. CLB Logic

For smaller and simpler multipliers even the MULT_AND logic might be unnecessary.
Multiplying by 2n simply requires shifting the value n places and adding is eliminated, so
the SRL16 or other resources can be used. Small multipliers can be implemented in LUTs,

312 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Conclusion

possibly using Muxes to expand to more inputs, and removes the column-based
requirement of the MULT_AND based multipliers.

Other Types of Multipliers

There are many alternative variations on multipliers that might be more efficient in the
LUT-based multiplier logic. For example, using Canonic Signed Digits instead of a binary
representation provides the fewest number of non-zero bits by using subtraction to
compress the representation and therefore can provide more efficiency. For example,
multiplying by 119 would normally be broken down into (26 + 25 + 24 + 22 + 21 + 20) * x. A
Canonic Signed Digits version would be implemented as (27 - 23 - 20)*x

X 26 X 25 X 24 X 22 X 21

+ + + + + 119*x
UG331_c11_36_080306

Figure 9-36: Binary Multiplication by 119 Uses 5 Shifters and 5 Adders

X 27 X 23

- - 119*x
UG331_c11_37_011508

Figure 9-37: CSD Multiplication by 119 Uses 2 Shifters and 2 Adder/Subtractors

Conclusion
Dedicated carry logic provides fast arithmetic addition and subtraction. The
Spartan-3 generation CLB has two separate carry chains with two bits per slice. The
dedicated carry path and carry multiplexer can also be used to cascade function generators
for implementing wide logic functions. The arithmetic logic includes an XOR gate that
allows a two-bit full adder to be implemented within a slice. In addition, a dedicated AND
improves the efficiency of multiplier implementations. These resources are used
automatically by synthesis tools or can be explicitly called out by the user.

Related Materials and References

Information on the carry primitives and carry-based macros can be found in the Libraries
Guide at:
http://www.xilinx.com/support/software_manuals.htm
CORE Generator System component information can be found at:
www.xilinx.com/products/design_tools/logic_design/design_entry/coregenerator.htm

Spartan-3 Generation FPGA User Guide www.xilinx.com 313

UG331 (v1.7) August 19, 2010
 R

Chapter 9: Using Carry and Arithmetic Logic

314 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 10

Using I/O Resources

All signals entering and exiting a Spartan®-3 generation FPGA must pass through the I/O
resources, known as I/O blocks or IOBs. Because FPGAs are used in more advanced
applications, they must support an increasing variety of I/O features. The revolutionary
SelectIO input/output capabilities of Spartan-3 generation FPGAs have met this need by
providing a highly configurable, high-performance resource suitable for applications such
as high-speed memory and programmable backplane interfaces.
The Spartan-3 generation FPGAs simplify high-performance design by offering selectable
I/O standards for inputs and outputs. Over 20 different standards are supported in each
family, with different specifications for current, voltage, I/O buffering, and termination
techniques. As a result, the Spartan-3 generation FPGA can be used to integrate discrete
translators and directly drive the most advance backplanes, buses, and memories. Directly
providing the necessary interface standard not only eliminates the cost of external
translators, but also significantly improves the chip-to-chip speed and reduces power
consumption.
This chapter describes how to take full advantage of the flexibility of the I/O capabilities
and the design considerations to improve and simplify system level design. The following
I/O topics are covered:
• “IOB Overview”
• “I/O Differences between Spartan-3 Generation Families”
• “Design Entry”
• “Architectural Details”
• “SelectIO Signal Standards”
• “Supply Voltages for the IOBs”

IOB Overview
The Input/Output Block (IOB) provides a programmable, unidirectional or bidirectional
interface between a package pin and the FPGA’s internal logic, supporting a wide variety
of standard interfaces. The robust feature set includes programmable control of output
strength and slew rate, registered or combinatorial inputs and outputs with dedicated
double data rate (DDR) registers, programmable input delays, on-chip termination, and
hot-swap capability.
Figure 10-1, page 317 is a simplified diagram of the IOB’s internal structure. There are
three main signal paths within the IOB: the output path, input path, and 3-state path. Each
path has its own pair of storage elements that can act as either registers or latches. For more
information, see “Storage Element Functions,” page 328. The three main signal paths are as
follows:

Spartan-3 Generation FPGA User Guide www.xilinx.com 315

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

• The input path carries data from the pad, which is bonded to a package pin, through
an optional programmable delay element directly to the I line. After the delay
element, there are alternate routes through a pair of storage elements to the IQ1 and
IQ2 lines. The IOB outputs I, IQ1, and IQ2 lead to the FPGA’s internal logic. The delay
element can be set to ensure a hold time of zero (see “Input Delay Functions”).
• The output path, starting with the O1 and O2 lines, carries data from the FPGA’s
internal logic through a multiplexer and then a three-state driver to the IOB pad. In
addition to this direct path, the multiplexer provides the option to insert a pair of
storage elements.
• The 3-state path determines when the output driver is high impedance. The T1 and T2
lines carry data from the FPGA’s internal logic through a multiplexer to the output
driver. In addition to this direct path, the multiplexer provides the option to insert a
pair of storage elements.
All signal paths entering the IOB, including those associated with the storage elements,
have an inverter option. Any inverter placed on these paths is automatically absorbed into
the IOB.

316 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

IOB Overview

T
TFF1
T1 D Q
CE
CK
SR REV
DDR
MUX
TCE

T2 D Q
TFF2
CE
CK
SR REV

Three-state Path

VCCO
OFF1
O1 D Q
CE
OTCLK1 CK Pull-Up ESD (1)
SR REV
DDR I/O
MUX Pin
OCE
Program- Pull-
O2 Q mable ESD
D Down
OFF2 Output
CE Driver
OTCLK2 CK
SR REV Keeper
Latch
Output Path

Programmable
I LVCMOS, LVTTL, PCI
Delay

IQ1 Programmable
Delay Single-ended Standards
using VREF
IDDRIN1 D Q
IDDRIN2 IFF1 VREF
CE Pin
ICLK1 CK
SR REV Differential Standards
ICE
I/O Pin
from
IQ2 Adjacent
D Q IOB
IFF2
CE
ICLK2 CK
SR REV
SR
REV Input Path DS312-2_19_080210

Figure 10-1: Simplified IOB Diagram

Note relevant to Figure 10-1:

1. The clamp diode is disabled by default in the Extended Spartan-3A family. (See
“Clamp Diodes,” page 336.)

Spartan-3 Generation FPGA User Guide www.xilinx.com 317

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

I/O Differences between Spartan-3 Generation Families

The Spartan-3 generation families have the same basic I/O capabilities, but there are many
differences in the details between each family.

Number of Resources per Device

The details showing the number of I/O resources in each part/package combination are
found in Chapter 1, “Overview.” Table 10-1 summarizes the maximum number of I/Os for
each device.

Table 10-1: Maximum Number of I/Os per Spartan-3 Generation Device

Spartan-3A DSP Spartan-3AN Spartan-3A Spartan-3E Spartan-3
I/O I/O I/O I/O I/O
FPGA FPGA FPGA FPGA FPGA
XC3SD1800A 519 XC3S50AN 108 XC3S50A 144 XC3S100E 108 XC3S50 124
XC3SD3400A 469 XC3S200AN 195 XC3S200A 248 XC3S250E 172 XC3S200 173
XC3S400AN 311 XC3S400A 311 XC3S500E 232 XC3S400 264
XC3S700AN 372 XC3S700A 372 XC3S1200E 304 XC3S1000 391
XC3S1400AN 502 XC3S1400A 502 XC3S1600E 376 XC3S1500 487
XC3S2000 565
XC3S4000 633
XC3S5000 633

The devices offer complementary solutions for different applications. The Spartan-3A DSP
platform is optimized for digital signal processing and similar logic-intensive applications.
The Spartan-3AN platform offers a non-volatile FPGA solution. The Spartan-3A platform
has the highest number of I/Os per gate, and is most cost-effective for applications that are
I/O intensive. The Spartan-3E family offers a higher number of gates per I/O, making it
cost-effective for applications requiring more logic than I/O. The I/O ratios differ
primarily because the Extended Spartan-3A family has a dual, staggered I/O ring around
the device, while the Spartan-3E family has a single in-line I/O ring. The Spartan-3 family
offers even higher density solutions for both gates and I/Os, and also has a staggered I/O
ring.

Input-Only Pins
To optimize the I/O ring and reduce cost, some I/O blocks in the Extended Spartan-3A
and Spartan-3E families are input-only pins. Dedicated Inputs are IOBs usable only as
inputs. Pin names designate a Dedicated Input if the name starts with IP, for example, IP_x
or IP_Lxxx_x. Dedicated inputs retain the full functionality of the IOB for input functions
with a single exception for differential inputs (IP_Lxxx_x). For the differential Dedicated
Inputs, the on-chip differential termination is not available. To use the on-chip differential
termination, either choose a differential pair that supports outputs (IO_Lxxx_x) or use an
external 100Ω termination resistor on the board.
The unidirectional, input-only block has a subset of the full IOB capabilities. Thus there are
no connections or logic for an output path. The following paragraphs assume that any
reference to output functionality does not apply to the input-only blocks. The number of
input-only blocks varies with device size but is never more than 25% of the total IOB count.

318 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

I/O Differences between Spartan-3 Generation Families

For details on the number of input-only pins in each part/package combination, see
Chapter 1, “Overview.”

Package Footprint Compatibility

Sometimes, applications outgrow the logic capacity of a specific FPGA, or it is possible to
optimize down to a lower density solution. Fortunately, each family is designed so that
multiple part types are available in pin-compatible package footprints, as described in
Module 4 of each family’s data sheet. In some cases, there are subtle differences between
devices available in the same footprint. These differences are outlined for each package,
such as pins that are unconnected on one device but connected on another in the same
package, or pins that are input-only pins on one package but full I/O on another. When
designing the printed circuit board (PCB), plan for potential future upgrades and package
migration. For details on the package pinout compatibility within a family, see Module 4 of
the data sheet for each family.
The Spartan-3A and Spartan-3AN platforms are pin compatible, and the XC3SD1800A in
the Spartan-3A DSP platform offers a straightforward upward migration. There is no other
pin compatibility between families; the Extended Spartan-3A family FPGA pinouts,
Spartan-3E FPGA pinouts, and Spartan-3 FPGA pinouts are not compatible. There are
significant differences between the other families in terms of the dedicated pins, number of
I/Os per bank, and package options. Each family has been optimized to maximize the
efficiency of the pinout for the features found in that family. Although, for example, the
Spartan-3A XC3S200A and XC3S400A devices are completely pin-compatible in the FT256
package, they are not compatible with the Spartan-3E or Spartan-3 devices in the same
FT256 package.
Also note that the XC3S200A and XC3S400A pinouts are different than the XC3S700A and
XC3S1400A pinouts for the FT256 package. The larger two devices require additional
power and ground pins due to their higher density. Therefore some of the I/O pins on the
smaller parts become power and ground pins in the larger parts. Pinout compatibility
could be maintained by using those I/O pins as virtual power and ground pins in the
smaller parts (see “Optionally Place Virtual Ground Pins Around DCM Input and Output
Connections” in Chapter 3. If the design uses differential I/O or HSTL/SSTL, the
differences in differential pairing and VREF pins will also need to be accounted for.
The I/O pins are separated into independent banks, typically four per device. Each bank
has a common output voltage supply (VCCO) and a common reference voltage for HSTL
and SSTL standards (VREF). The banks are numbered clockwise from the top of the device
(see Figure 10-35, page 356).

Summary of Differences
Table 10-2 highlights the major differences between the I/O resources of the different
Spartan-3 generation FPGA families. Some of these differences are described in more detail
later in this chapter.

Spartan-3 Generation FPGA User Guide www.xilinx.com 319

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-2: Differences Between Architectures

Extended Spartan-3A
Features Spartan-3E FPGA Spartan-3 FPGA
Family FPGA
Input-Only Pins Yes Yes No
I/O Structure Staggered In-Line Staggered
Programmable Input Delay - Dynamic, Programmable, Programmable,
Combinatorial 16 Values 12 Values 1 Value
Programmable Input Delay - Programmable, Programmable, Programmable,
Registered 8 Values 6 Values 1 Value
IOSTANDARDs Vary – see details later Vary – see details later Vary – see details later
Banks 4 4 8
Differential Termination Yes, ~100Ω Yes, ~ 120Ω N/A
Single-Ended Termination N/A N/A Digitally Controlled
Impedance (DCI)
Slew Rates 3 2 2
Hot Swap Fully Supported Requires Sequencing Requires Sequencing
VIN Absolute Maximum 4.6V 4.4V VCCO + 0.5V or 4.05V
IDDR2 Register Cascade Yes Yes N/A
ODDR2 Register Cascade Yes N/A N/A

Design Entry
In many cases the I/O resources are automatically selected by the implementation tools.
Users might want to specify particular components for special purposes, such as using
dedicated clock inputs. The components listed below can be instantiated in HDL code or in
a schematic.

Library Components
The Xilinx library includes an extensive list of components designed to provide support for
the variety of I/O features (Table 10-3). Most of these components represent variations of
the five generic I/O elements:
• IBUF (input buffer)
• IBUFG (global clock input buffer)
• OBUF (output buffer)
• OBUFT (3-state output buffer)
• IOBUF (input/output buffer)

320 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Entry

Table 10-3: Spartan-3 Generation I/O Components

Extended Spartan-3A Spartan-3E Spartan-3 Three-
Input Output Differential
Family FPGA FPGA FPGA State
IBUF Y Y Y Y N N N
IBUFG Y Y Y Y N N N
IBUFDS Y Y Y Y N N Y
IBUFGDS Y Y Y Y N N Y
IBUF_DLY_ADJ Y N N Y N N N
IBUFDS_DLY_ADJ Y N N Y N N Y
OBUF Y Y Y N Y N N
OBUFDS Y Y Y N Y N Y
OBUFT Y Y Y N Y Y N
OBUFTDS Y Y Y N Y Y Y
IOBUF Y Y Y Y Y Y N
IOBUFDS(1) Y Y Y Y Y Y Y
IFDDRCPE
IDDR2 Y Y and Y N N optional(2)
IFDDRRSE
OFDDRCPE
ODDR2 Y Y and N Y N optional(2)
OFDDRRSE
PULLUP Y Y Y Y Y - -
PULLDOWN Y Y Y Y Y - -
KEEPER Y Y Y Y Y - -

Notes:
1. Must use a bidirectional differential IOSTANDARD such as BLVDS.
2. Must be differential if the DDR_ALIGNMENT = C0/C1 feature is used.

Earlier families had additional I/O components, but these are not recommended for use in
new designs. These components included:
• Bus I/O (Example: IBUF4)
These are still available for schematic entry only at 4, 8, and 16 bits wide, but
individual components allow more control over constraints.
• Registered I/O (Example: IFD)
These are also available for schematic entry and include both registered and latched
I/Os. However, it is recommended that the software be allowed to optimize to either
the IOB or the CLB, whichever is more efficient.
• I/O Standard Suffix (Example: IBUF_LVCMOS18)
These components included the IOSTANDARD as part of the component name. It is
recommended to apply an IOSTANDARD constraint to a generic component instead.

Spartan-3 Generation FPGA User Guide www.xilinx.com 321

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Registered I/O
The Spartan-3 generation IOB includes an optional flip-flop or latch on the input path,
output path, and 3-state control input. However, there are no special library components
for the I/O registers. To simplify design, especially synthesis, the standard register
primitives are automatically absorbed into the IOB when possible. This feature is selected
by the user by turning on the Map Property "Pack I/O Registers/Latches into IOBs", which
can be set to Off (default), For Inputs Only, For Outputs Only, or For Inputs and Outputs.
Alternatively, the IOB = TRUE property can be placed on a register to force the mapper to
place the register in an IOB.
An optional delay element is associated with the input path in each logic input primitive
(IBUF or IOBUF). When the buffer drives an input register within the IOB, the delay
element activates by default to ensure a zero hold time requirement. The delay element is
not used for non-registered inputs, to provide higher performance. The user can override
the defaults; see “Input Delay Functions,” page 325 for more details.

Differential I/O
The Spartan-3 generation IOBs include differential I/O standards such as LVDS, BLVDS,
and RSDS. Differential I/O requires two pins for every signal, which toggle in opposite
directions. To support differential signaling, most I/O components have differential
versions with DS in the name and two I/O pins on the component.
On the inputs, if only the P side of the differential pair is called out, the N side is
automatically configured as the other half of the differential pair. If the N input is called
out in a design for simulation and system-level integration, it is trimmed during the
mapping process, although physically it is still used in conjunction with the P input, and
the software does not allow it to be used for any other purpose.
On the outputs, both the P and N sides of the differential pair must be defined. The IOB
must have the same net source the control pins: clock, set/reset, three-state, three-state
clock enable, and output clock enable. In addition, the output pins must be inverted with
respect to each other, and, if output registers are used, the D inputs must be inverted to
each other and the INIT states must be opposite values (one High and one Low). Three-
state registers must have the same inputs and have the same INIT states. INIT states must
be set correctly for the power-up state even if the INIT function is not used in the design
(INIT is connected to ground).
The pins that can be used as differential pairs are specified in the Module 4 pinout tables,
including the special pairs that can be used for clock inputs.

IBUF
Signals used as inputs to the device must source an input buffer (IBUF) via an external
input port. Figure 10-2 shows the generic IBUF symbol.

IBUF
I O

x133_01_111699

Figure 10-2: Input Buffer (IBUF) Symbol

322 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Entry

IBUFG
IBUFG is a special global clock input buffer that can connect directly to the BUFG (global
clock buffer) and DCM components. A standard input driving a clock signal is put onto an
IBUFG by the Xilinx tools, or the user can instantiate the IBUFG directly. See Chapter 2,
“Using Global Clock Resources,” for more details. Figure 10-3 shows the generic IBUFG
symbol.

IBUFG
I O

x133_03_111699

Figure 10-3: Global Clock Input Buffer (IBUFG) Symbol

IBUFDS
IBUFDS is an input buffer that supports differential signaling. In IBUFDS, a design level
interface signal is represented as two ports (I and IB), one deemed the "master" and the
other the "slave." The master and the slave are opposite phases of the same logical signal
(for example, MYNET and MYNETB). Figure 10-4 shows the generic IBUFDS symbol.

I
O
IB

UG331_c10_03_111106

Figure 10-4: Differential Input Buffer (IBUFDS) Symbol

OBUF
An OBUF must drive outputs through an external output port. Figure 10-5 shows the
generic output buffer (OBUF) symbol.

OBUF
I O

x133_04_111699

Figure 10-5: Output Buffer (OBUF) Symbol

OBUFT
The generic 3-state output buffer OBUFT, shown in Figure 10-7, typically implements
3-state outputs. Unused I/Os are configured with a disabled OBUFT.

OBUFT
T

I O

x133_05_111699

Figure 10-6: 3-State Output Buffer (OBUFT) Symbol

Spartan-3 Generation FPGA User Guide www.xilinx.com 323

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

IOBUF
Use the IOBUF symbol for bidirectional signals that require both an input buffer and a
3-state output buffer with an active high 3-state pin. This symbol combines the
functionality of the OBUFT and IBUF symbols. Figure 10-7 shows the generic
input/output buffer IOBUF.

IOBUF
T

I IO

x133_06_111699

Figure 10-7: Input/Output Buffer (IOBUF) Symbol

Pull-Up, Pull-Down, and Keeper Circuits

Three-state output buffers and bidirectional buffers can have a pull-up resistor, a pull-
down resistor, or a keeper circuit. Control this feature by adding the appropriate symbol to
the output net of the OBUFT or IOBUF (PULLUP, PULLDOWN, or KEEPER).
VREF is typically needed only for inputs that use an IOSTANDARD requiring VREF, such as
HSTL and SSTL inputs. However, an IOB configured using an OBUFT with a weak keeper
circuit requires the input buffer to sample the I/O signal. Therefore, using an OBUFT
requires the use of the VREF pins in the bank if the OBUFT is configured with KEEPER and
a standard that requires VREF. In most applications, the VREF pins in the bank are needed
anyway because the OBUFT is usually combined with an input IBUF component.

DDR and Adjustable Delay I/O Components

The DDR components (IDDR2 and ODDR2) are discussed in “Double-Data-Rate
Transmission,” page 330. The adjustable delay (IBUF_DLY_ADJ) is discussed in “Input
Delay Functions,” page 325.

HDL Entry
I/O components can be easily instantiated in VHDL or Verilog code. The Xilinx
development system includes language templates for any of the standard I/O
components.
Following is an example of the template for the IOBUF input/output buffer component.
Registers can automatically be merged into the I/O block, simplifying the generation of
the HDL code.
-- INOUT_PORT : inout STD_LOGIC;
--**Insert the following between the
-- 'architecture' and 'begin' keywords**
signal IN_SIG, OUT_SIG, T_ENABLE: std_logic;
component IOBUF
port (I, T: in std_logic;
O: out std_logic;
IO: inout std_logic);

324 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

end component;
--**Insert the following after the 'begin' keyword**
U1: IOBUF port map (I => OUT_SIG, T => T_ENABLE,
O => IN_SIG, IO => INOUT_PORT);

Architectural Details

Input Delay Functions

An optional delay element is associated with each input path. When the buffer drives an
input register within the IOB, the delay element activates by default to ensure a zero hold
time requirement. This is desirable because the clock signal has a longer path to the IOB
through the global clock buffer and global clock routing, as shown in Figure 10-8. The
delay element slows down the data input so that when data and clock pins change at the
same time, the clock arrives first and clocks in the data set up on the previous clock edge.
The delayed data signal then arrives at the flip-flop, ready for the next clock edge.

Delay
Pad D Q

IOB

Pad BUFG

UG331_c10_08_111106

Figure 10-8: Simplified View of Data and Clock Routing to Input Flip-Flop

There are actually two flip-flops on the input path to support double data rate signaling
called IFF1 and IFF2. They generate IOB signals IQ1 and IQ2, respectively, as shown in
Figure 10-1, page 317. The delay element choice affects both flip-flops.
The delay element is not used for non-registered (combinatorial) inputs in order to provide
higher performance. An IOB can supply both a registered and a non-registered version of
the same input pad if required in the application. When both paths are used, the delay
element choice is independent for the two paths, for example, allowing the registered path
to be delayed while the combinational path is not.
The user can override the defaults, either adding the delay to a combinatorial input or
removing it from a registered input. Extra delay might be required on some clock or data
inputs, for example, in interfaces to various types of RAM. If the design uses a DCM in the
clock path, then the delay element can be removed from registered inputs, still without a
hold time requirement.

Programmable Delay
In the Spartan-3E and Extended Spartan-3A families, the delay block itself has
programmable delay values.

Spartan-3 Generation FPGA User Guide www.xilinx.com 325

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Each IOB has a programmable delay block that can delay the input signal by a
programmable amount. In Figure 10-9, the signal path has a coarse delay element that can
be bypassed. The input signal then feeds a 6-tap delay line in the Spartan-3E family (an
8-tap delay line in the Extended Spartan-3A family). All six taps are available via a
multiplexer for use as an asynchronous input directly into the FPGA fabric. Three of the six
taps are also available via a multiplexer to the D inputs of the synchronous storage
elements. The coarse delay element is common to both asynchronous and synchronous
paths, and must be either used or not used for both paths.

IFD_DELAY_VALUE
Synchronous input (IQ1)
D Q

Synchronous input (IQ2)

D Q

Coarse Delay
PAD

Asynchronous input (I)

IBUF_DELAY_VALUE
UG331_c10_09_011508

Figure 10-9: Programmable Fixed Input Delay Elements

These delay values are set up in the silicon once at configuration time through the
IBUF_DELAY_VALUE and the IFD_DELAY_VALUE parameters. The default
IBUF_DELAY_VALUE is 0, bypassing the delay elements for the asynchronous input. The
user can set this parameter to 0-12 in the Spartan-3E family. The default
IFD_DELAY_VALUE is AUTO; the Xilinx software chooses the default value automatically
because the value depends on device size. The default values are shown in the data sheet
timing specifications, and are indicated in the Map report generated by the
implementation tools. The user can select a specific IFD_DELAY_VALUE from 0-6 in the
Spartan-3E family, and the resulting timing is reported by the Timing Analyzer tool.
IBUF_DELAY_VALUE and IFD_DELAY_VALUE are independent for each input. If the
same input pin uses both registered and non-registered input paths, both parameters can
be used, but they must both be in the same half of the total delay (both either bypassing or
using the initial delay element).

Dynamic Delay in the Extended Spartan-3A Family

The Extended Spartan-3A family has the same input delay structure as described for
Spartan-3E devices, but add more taps (8 on the registered path and 16 on the
combinatorial path) and dynamic adjustment on the combinatorial path. The delay on the
combinatorial input can be dynamically adjusted during operation without having to
reconfigure the device, allowing the device to be fine-tuned to the specific operating
conditions of the system. Three control inputs to the delay element taps allow immediate

326 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

changes to the delay amount. The choice of the coarse delay element is still fixed as part of
the device configuration.

IFD_DELAY_VALUE
Synchronous input (IQ1)
D Q

Synchronous input (IQ2)

D Q

Coarse Delay
PAD

S[2:0]

Asynchronous input (I)

IBUF_DELAY_VALUE
UG331_c10_10_011508

Figure 10-10: Extended Spartan-3A Family FPGA Programmable Dynamic Input

Delay Elements

The delay values at configuration are still controlled by the IFD_DELAY_VALUE and
IBUF_DELAY_VALUE parameters. To use the dynamic adjustment delay for
combinatorial inputs, replace the IBUF component with the IBUF_DLY_ADJ component
(see Figure 10-11) and connect the three select inputs. The IBUF_DLY_ADJ component is
only used for the combinatorial (non-registered) path, and has no affect on the IFD
(registered) path.

IBUF_DLY_ADJ
I O

S[2:0]
UG331_c10_11_111206

Figure 10-11: Input Buffer Symbol with Dynamic Delay Adjustment

The IBUF_DLY_ADJ only allows moving up or down half of the total delay amount. The
DELAY_OFFSET parameter specifies whether it is the first half or the second half of the
delay amounts. DELAY_OFFSET = ON feeds the coarse delay element into the dynamic
mux, while DELAY_OFFSET = OFF bypasses the coarse delay element. Table 10-4 shows
how the IBUF_DELAY_VALUE corresponds to the Select lines. The binary equivalents of
the Select lines, 0 to 7, correspond to the IBUF_DELAY_VALUE options of 1-8 or 9-16. An
IBUF_DELAY_VALUE of 0 corresponds to completely bypassing the delay functions, and
is available with the IBUF component only, not IBUF_DLY_ADJ.

Spartan-3 Generation FPGA User Guide www.xilinx.com 327

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-4: Fixed and Dynamic Delay Values

S[2:0] DELAY_OFFSET Equivalent IBUF_DELAY_VALUE

0 1
1 2
2 3
3 4
OFF
4 5
5 6
6 7
7 8
0 9
1 10
2 11
3 12
ON
4 13
5 14
6 15
7 16

Figure 10-12 shows how the two types of delay specifications control the muxes.

IBUF_DELAY_VALUE > 8 (fixed or

DELAY_OFFSET (dynamic)
1/9 2/10 3/11 4/12 5/13 6/14 7/15 8/16
0
1

0 1 2 3 4 5 6 7
3
(dynamic)

IBUF_DELAY_VALUE 0 >0
UG331_c10_12_111206

Figure 10-12: Fixed and Dynamic Delay Controls

Storage Element Functions

There are three pairs of storage elements in each IOB, one pair for each of the three paths.
It is possible to configure each of these storage elements as an edge-triggered D-type flip-
flop (FD) or a level-sensitive latch (LD).
The storage-element pair on either the Output path or the Three-State path can be used
together with a special multiplexer to produce Double-Data-Rate (DDR) transmission.

328 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

This is accomplished by taking data synchronized to the clock signal’s rising edge and
converting it to bits synchronized on both the rising and the falling edge. The combination
of two registers and a multiplexer is referred to as a Double-Data-Rate D-type flip-flop
(ODDR2).
Table 10-5 describes the signal paths associated with the storage element.

Table 10-5: Storage Element Signal Description

Storage
Element Description Function
Signal
D Data input Data at this input is stored on the active edge of CK and enabled by CE. For latch
operation when the input is enabled, data passes directly to the output Q.
Q Data output The data on this output reflects the state of the storage element. For operation as a latch
in transparent mode, Q mirrors the data at D.
CK Clock input Data is loaded into the storage element on this input’s active edge with CE asserted.
CE Clock Enable input When asserted, this input enables CK. If not connected, CE defaults to the asserted state.
SR Set/Reset input This input forces the storage element into the state specified by the SRHIGH/SRLOW
attributes. The SYNC/ASYNC attribute setting determines if the SR input is
synchronized to the clock or not. If both SR and REV are active at the same time, the
storage element gets a value of 0.
REV Reverse input This input is used together with SR. It forces the storage element into the state opposite
from what SR does. The SYNC/ASYNC attribute setting determines whether the REV
input is synchronized to the clock or not. If both SR and REV are active at the same time,
the storage element gets a value of 0.

As shown in Figure 10-1, page 317, the upper registers in both the output and three-state
paths share a common clock. The OTCLK1 clock signal drives the CK clock inputs of the
upper registers on the output and three-state paths. Similarly, OTCLK2 drives the CK
inputs for the lower registers on the output and three-state paths. The upper and lower
registers on the input path have independent clock lines: ICLK1 and ICLK2.
Clock routing resources are often shared between adjacent IOBs, including differential
pairs. In these situations, the two OTCLK1, OTCLK2, ICLK1, and ICLK2 signals must be
identical when both IOBs used them. The software can swap between the upper and lower
registers if necessary, unless both are used in a DDR configuration.
The OCE enable line controls the CE inputs of the upper and lower registers on the output
path. Similarly, TCE controls the CE inputs for the register pair on the three-state path and
ICE does the same for the register pair on the input path.
The Set/Reset (SR) line entering the IOB controls all six registers, as is the Reverse (REV)
line.
In addition to the signal polarity controls, each storage element additionally supports the
controls described in Table 10-6.

Spartan-3 Generation FPGA User Guide www.xilinx.com 329

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-6: Storage Element Options

Option Switch Function Specificity
FF/Latch Chooses between an edge-triggered flip-flop or a Independent for each storage element
level-sensitive latch
SYNC/ASYNC Determines whether the SR set/reset control is Independent for each storage element
synchronous or asynchronous
SRHIGH/SRLOW Determines whether SR acts as a Set, which forces Independent for each storage element, except
the storage element to a logic "1" (SRHIGH) or a when using ODDR2. In the latter case, the
Reset, which forces a logic "0" (SRLOW) selection for the upper element will apply to
both elements.
INIT1/INIT0 When Global Set/Reset (GSR) is asserted or after Independent for each storage element, except
configuration this option specifies the initial state when using ODDR2, which uses two IOBs. In
of the storage element, either set (INIT1) or reset the ODDR2 case, selecting INIT0 for one IOBs
(INIT0). By default, choosing SRLOW also selects applies to both elements within the IOB,
INIT0; choosing SRHIGH also selects INIT1. although INIT1 could be selected for the
elements in the other IOB.

Double-Data-Rate Transmission
Double-Data-Rate (DDR) transmission describes the technique of synchronizing signals to
both the rising and falling edges of the clock signal. Register pairs are available in all three
IOB paths to perform DDR operations.
The pair of storage elements on the IOB’s Output path (OFF1 and OFF2), used as registers,
combine with a special multiplexer to form a DDR D-type flip-flop (ODDR2). This
primitive permits DDR transmission where output data bits are synchronized to both the
rising and falling edges of a clock. DDR operation requires two clock signals (usually 50%
duty cycle), one the inverted form of the other. These signals trigger the two registers in
alternating fashion, as shown in Figure 10-13. The Digital Clock Manager (DCM) generates
the two clock signals by mirroring an incoming signal, and then shifting it 180 degrees.
This approach ensures minimal skew between the two signals. Alternatively, the inverter
inside the IOB can be used to invert the clock signal, thus only using one clock line and
both rising and falling edges of that clock line as the two clocks for the DDR flip-flops.
The storage-element pair on the Three-State path (TFF1 and TFF2) also can be combined
with a local multiplexer to form a DDR primitive. This permits synchronizing the output
enable to both the rising and falling edges of a clock. This DDR operation is realized in the
same way as for the output path.
The storage-element pair on the input path (IFF1 and IFF2) allows an I/O to receive a DDR
signal. An incoming DDR clock signal triggers one register, and the inverted clock signal
triggers the other register. The registers take turns capturing bits of the incoming DDR data
signal. The primitive to allow this functionality is called IDDR2.
Note that the ODDR2 and IDDR2 primitives must be used to access the special DDR logic
in the IOBs.
Aside from high bandwidth data transfers, DDR outputs also can be used to reproduce, or
mirror, a clock signal on the output. This approach is used to transmit clock and data
signals together (source synchronously). A similar approach is used to reproduce a clock
signal at multiple outputs. The advantage for both approaches is that skew across the
outputs is minimal.

330 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

DCM DCM
180˚ 0˚ 0˚

FDDR FDDR
D1 D1
Q1 Q1

CLK1 CLK1

DDR MUX Q DDR MUX Q

D2 D2
Q2 Q2

CLK2 CLK2

DS312-2_20_021105

Figure 10-13: Two Methods for Clocking the DDR Register

Register Cascade Feature

In the Spartan-3E and Extended Spartan-3A families, one of the IOBs in a differential pair
can cascade its storage elements with those in the other IOB of the differential pair. This is
intended to make DDR operation at high speed much simpler to implement. The new
IDDRIN1/2 connections that are available are shown in Figure 10-1 (dashed lines), and are
only available for routing between IOBs and are not accessible to the FPGA fabric and are
only used within the IDDR2/ODDR2 components. Note that this feature is only available
when using differential I/O and is not available in the Spartan-3 family and is supported
on inputs only (IDDR2) in the Spartan-3E family. The supported differential standards
include the true differential signaling standards, such as LVDS, MINI_LVDS, and RSDS,
but do not include the pseudo-differential standards, such as DIFF_HSTL, DIFF_SSTL, and
LVPECL.
Note that the register cascade feature is accessed using the same IDDR2 and ODDR2
primitives as for the standard DDR interface, but with the DDR_ALIGNMENT attribute
set to either clock C0 or clock C1. The cascaded register does not need to be instantiated,
but the adjacent I/O must be unused (as the other half of a differential pair). The default is
DDR_ALIGNMENT=NONE, which does not use the cascade feature.

IDDR2
As a DDR input pair, the master IOB registers incoming data on the rising edge of ICLK1
(= D1) and the rising edge of ICLK2 (= D2), which is typically the same as the falling edge
of ICLK1. This data is then transferred into the FPGA fabric. At some point, both signals
must be brought into the same clock domain, typically ICLK1. This can be difficult at high
frequencies because the available time is only one half of a clock cycle assuming a 50%
duty cycle. See Figure 10-14 for a graphical illustration of this function.

Spartan-3 Generation FPGA User Guide www.xilinx.com 331

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

D Q D1

PAD
To Fabric

D Q D2

ICLK2
ICLK1

ICLK1

ICLK2

PAD d d+1 d+2 d+3 d+4 d+5 d+6 d+7 d+8

D1 d d+2 d+4 d+6 d+8
D2 d-1 d+1 d+3 d+5 d+7
DS312-2_21_021105

Figure 10-14: Input DDR (without Cascade Feature)

IDDR2
D
C0 Q0
C1
CE Q1
R
S

UG331_c10_15_111206

Figure 10-15: IDDR2 Component

When using the cascade feature (DDR_ALIGNMENT=C0/C1), the signal D2 can be

cascaded into the storage element of the adjacent slave IOB. There it is re-registered to
ICLK1, and only then fed to the FPGA fabric where it is now already in the same time
domain as D1. Here, the FPGA fabric uses only the clock ICLK1 to process the received
data. See Figure 10-16 for a graphical illustration of this function. DDR_ALIGNMENT is
only available on differential pairs (see “Register Cascade Feature,” page 331).

332 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

D Q D1

PAD
To Fabric

IQ2 IDDRIN2
D Q D Q D2

ICLK1

ICLK2

ICLK1

ICLK2

PAD d d+1 d+2 d+3 d+4 d+5 d+6 d+7 d+8

D1 d d+2 d+4 d+6 d+8

D2 d-1 d+1 d+3 d+5 d+7

DS312-2_22_030105

Figure 10-16: Input DDR Using Cascade Feature (IDDR2 with

DDR_ALIGNMENT=C0/C1)

ODDR2
As a DDR output pair, the master IOB registers data coming from the FPGA fabric on the
rising edge of OCLK1 (= D1) and the rising edge of OCLK2 (= D2), which is typically the
same as the falling edge of OCLK1. These two bits of data are multiplexed by the DDR mux
and forwarded to the output pin. The D2 data signal must be resynchronized from the
OCLK1 clock domain to the OCLK2 domain using FPGA slice flip-flops. Placement is
critical at high frequencies, because the time available is only one half a clock cycle. See
Figure 10-17 for a graphical illustration of this function.
In the ODDR2 component for the Extended Spartan-3A family, the DDR_ALIGNMENT
attribute allows both data bits to be captured on C0 or C1 (DDR_ALIGNMENT=C0 or
DDR_ALIGNMENT=C1). DDR_ALIGNMENT is only available on differential pairs (see
“Register Cascade Feature,” page 331).
Note: The Spartan-3E family does not support using the C0 or C1 alignment feature of the ODDR2
flip-flop. The ODDR2 flip-flop without the alignment feature is fully supported, as is the IDDR2 flip-flop
with alignment. Without the alignment feature, the ODDR2 component behaves equivalently to the
ODDR flip-flop components on previous Xilinx FPGA families. The Spartan-3A/3AN production
devices and Spartan-3A DSP devices fully support this feature.
Clock routing resources are often shared between adjacent IOBs, including differential
pairs. In these situations, the two OTCLK1, OTCLK2, ICLK1, and ICLK2 signals must be
identical when both IOBs used them. The software can swap between the upper and lower
registers if necessary, unless both are used in a DDR configuration.

Spartan-3 Generation FPGA User Guide www.xilinx.com 333

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

D1 D Q
PAD
From
Fabric

D2 D Q

OCLK1
OCLK2

OCLK1

OCLK2

D1 d d+2 d+4 d+6 d+8 d+10

D2 d+1 d+3 d+5 d+7 d+9

PAD d d+1 d+2 d+3 d+4 d+5 d+6 d+7 d+8

DS312-2_23_030105

Figure 10-17: Output DDR (without Cascade Feature)

ODDR2
D0
D1
C0
C1 Q
CE
R
S

UG331_c10_17_111206

Figure 10-18: ODDR2 Component

Pull-Up and Pull-Down Resistors

Pull-up and pull-down resistors inside each IOB optionally force a floating I/O or Input-
only pin to a determined state. Pull-up and pull-down resistors are commonly applied to
unused I/Os, inputs, and three-state outputs, but can be used on any I/O or Input-only
pin. The pull-up resistor connects an IOB to VCCO through a resistor. The resistance value
depends on the VCCO voltage (see Module 3 for the specifications). The pull-down resistor
similarly connects an IOB to ground with a resistor. The pull-down resistor is powered by
VCCAUX in the Extended Spartan-3A family and by VCCO in the Spartan-3/3E families.
The PULLUP and PULLDOWN attributes and library primitives turn on these optional
resistors. By default, PULLDOWN resistors terminate all unused I/O and Input-only pins.
Unused I/O and Input-only pins can alternatively be set to PULLUP or FLOAT. To change
the unused I/O Pad setting, set the Bitstream Generator (BitGen) option UnusedPin to
PULLUP, PULLDOWN, or FLOAT. The UnusedPin option is accessed through the
Properties for Generate Programming File in the ISE software.

334 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Architectural Details

During configuration, a Low logic level on the PUDC_B pin (HSWAP in the
Spartan-3E family and HSWAP_EN in the Spartan-3 family) activates pull-up resistors on
all I/O and Input-only pins not actively used in the selected configuration mode.

FPGA Pull-Up Resistor Values

The value of the dedicated and optional pull-up resistors is specified as a current, symbol
IPU in the respective Spartan-3 generation data sheet. The equivalent resistor values
provided in Table 10-7 are for reference. Note that these resistor values are stronger than a
typical weak pull-up resistor, especially in the earlier Spartan-3 and Spartan-3E families,
and require proper external resistor values to overcome them. See the data sheets for more
exact values:
• Spartan-3 FPGA: DS099
• Spartan-3E FPGA: DS312
• Spartan-3A FPGA: DS529
• Spartan-3AN FPGA: DS557
• Spartan-3A DSP FPGA: DS610
Note: The pull-up resistors in Spartan-3 generation FPGAs are strong, especially at higher supply
voltages.
Table 10-7: Pull-Up Resistor Ranges by Spartan-3 Generation Family
Spartan-3 Spartan-3E Extended Spartan-3A
Voltage Range Units
FPGA FPGA Family FPGA

VCCAUX or VCCO = 3.0 to 3.6V

5.1 to 23.9
VCCO = 3.0 to 3.45V 1.27 to 4.11 2.4 to 10.8
kΩ
VCCAUX or VCCO = 2.3 to 2.7V 1.15 to 3.25 2.7 to 11.8 6.2 to 33.1(1)

VCCO = 1.7 to 1.9V 2.45 to 9.10 4.3 to 20.2 8.4 to 52.6

Notes:
3. Spartan-3AN FPGAs require VCCAUX = 3.0 to 3.6V

Keeper Circuit
Each I/O has an optional keeper circuit (see Figure 10-19) that keeps bus lines from
floating when not being actively driven. The KEEPER circuit retains the last logic level on
a line after all drivers have been turned off. Apply the KEEPER attribute or use the
KEEPER library primitive to use the KEEPER circuitry. Pull-up and pull-down resistors
override the KEEPER settings.

Spartan-3 Generation FPGA User Guide www.xilinx.com 335

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Pull-up

Output Path
Input Path

Keeper
Pull-down

UG331_c10_19_111206

Figure 10-19: Keeper Circuit

JTAG Boundary-Scan Capability

All IOBs support Boundary-Scan testing compatible with IEEE 1149.1/1532 standards. For
more information, see Chapter 21, “Using Boundary-Scan and BSDL Files.”

SelectIO Signal Standards

The Input/Output Blocks (IOBs) feature inputs and outputs that support a wide range of
single-ended I/O signaling standards. The majority of the I/Os also can be used to form
differential pairs to support any of the differential signaling standards. This flexibility
allows the user to select the best I/O standard on each pin that meets the interface and
signal integrity requirements of the application.
The I/O pins are separated into independent banks, typically four per device. Each bank
has a common output voltage supply (VCCO) and a common reference voltage for HSTL
and SSTL standards (VREF). The banks are numbered clockwise from the top of the device
(see Figure 10-35, page 356).

Overview of I/O Standards

Modern bus applications, pioneered by the largest and most influential companies in the
digital electronics industry, are commonly introduced with a new I/O standard tailored
specifically to the needs of that application. The bus I/O standards provide specifications
to other vendors who create products designed to interface with these applications. Each
standard often has its own specifications for current, voltage, I/O buffering, and
termination techniques.
The ability to provide the flexibility and time-to-market advantages of programmable logic
is increasingly dependent on the capability of the programmable logic device to support an
ever increasing variety of I/O standards. The SelectIO resources feature highly
configurable input and output buffers which provide support for a wide variety of I/O
standards.

Clamp Diodes
In the Spartan-3/3E FPGAs, internal clamp diodes protect all device pins against excess
voltage transients. Each I/O pin has two clamp diodes that are always connected to the
pin, regardless of the signal standard selected. One diode extends P-to-N from the pin to
VCCO, and the second diode extends N-to-P from the pin to GND. During normal
operation, these diodes are reverse biased. The VIN absolute maximum rating (see

336 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

DS099, Spartan-3 FPGA Family Data Sheet and DS312, Spartan-3E FPGA Family Data Sheet)
specifies the voltage range that each I/O pin can tolerate. When interfacing to a signal that
exceeds the VIN absolute maximum rating, use external components to limit the applied
voltage to the device I/O pin. For example, inserting a resistor between the signal and the
I/O pin will form a voltage divider with the internal clamp diodes when they become
forward biased. Other methods to limit excess voltage transients can be used including
using external level translators, external signal clamps, or external resistor voltage
dividers.
In the Extended Spartan-3A family FPGAs, the I/O uses a floating-well technique to
provide superior hot-swap capability. A clamp diode between the I/O pin and VCCO is
provided for PCI bus applications, but this diode is typically disabled because it would
defeat the purpose of the floating-well technique. As in Spartan-3/3E FPGAs, when
interfacing with signals that exceed the VIN absolute maximum rating, external
components must be used to limit the applied voltage to the device I/O pin. Unlike
Spartan-3/3E FPGAs, it is not possible to use a series resistor, because no clamp diodes are
present. However, all other methods listed are applicable.
See XAPP459, Eliminating I/O Coupling Effects when Interfacing Large-Swing Single-Ended
Signals to User I/O Pins on Spartan-3 Generation FPGAs for more information.
Table 10-8 provides a brief overview of the I/O standards supported by the
Spartan-3 generation FPGAs, including the sponsors and common uses for the standard.
The standard numbers are indicated where appropriate.

Table 10-8: I/O Signaling Standards

Standard Description Spec Use/Sponsor Input Buffer Output Buffer
Single-Ended Standards
LVTTL Low Voltage TTL JESD8C General purpose LVTTL Push-Pull
3.3V
LVCMOS Low Voltage CMOS JESD8C General purpose CMOS Push-Pull
PCI Peripheral PCI SIG PCI bus LVTTL Push-Pull
Component
Interconnect
GTL Gunning Transceiver JESD8-3 High-speed bus, VREF-based Open Drain
Logic backplane; Xerox
GTL+ GTL Plus Intel® Pentium® Pro
HSTL High-Speed JESD8-6 Hitachi SRAM; IBM; VREF-based Push-Pull
Transceiver Logic three of four classes
supported
SSTL3 Stub Series JESD8-8 SRAM/ SDRAM VREF-based Push-Pull
Terminated Logic for bus; Hitachi and
3.3V IBM; two classes
SSTL2 SSTL for 2.5V JESD8-9
SSTL18 SSTL for 1.8V JC42.3
Differential Standards
LVDS Low Voltage ANSI/TIA/EIA- High-speed Differential Pair Differential Pair
Differential Signaling 644-A interface, backplane,
video; National, TI

Spartan-3 Generation FPGA User Guide www.xilinx.com 337

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-8: I/O Signaling Standards

Standard Description Spec Use/Sponsor Input Buffer Output Buffer
BLVDS Bus LVDS ANSI/TIA/EIA- Multipoint LVDS Differential Pair Differential Pair
644-A
LVPECL Low Voltage Positive Freescale High-speed clocks Differential Pair Differential Pair
ECL Semiconductor
(formerly
Motorola)
LDT Lightning Data HyperTransport Bidirectional Differential Pair Differential Pair
Transport Spec v3.0 serial/parallel high-
(HyperTransport™) bandwidth, low-
latency computer
bus; HyperTransport
Consortium
MINI_LVDS mini-LVDS TI Flat panel displays Differential Pair Differential Pair
LVDSEXT LVDS Extended Extension of Higher drive Differential Pair Differential Pair
LVDS requirements
RSDS Reduced Swing National Flat panel displays Differential Pair Differential Pair
Differential Signaling Semiconductor
TMDS Transition Minimized Digital Display Silicon Image; Differential Pair Differential Pair
Differential Signaling Working Group DVI/HDMI
PPDS Point-to-Point National LCDs Differential Pair Differential Pair
Differential Signaling Semiconductor

LVTTL — Low-Voltage TTL

The Low-Voltage TTL (LVTTL) standard is a general-purpose EIA/JESD standard for 3.3V
applications that uses an LVTTL input buffer and a Push-Pull output buffer. This standard
requires a 3.3V output source voltage (VCCO), but does not require the use of a reference
voltage (VREF) or a termination voltage (VTT).

LVCMOS — Low-Voltage CMOS

The Low-Voltage CMOS standard is used for general-purpose applications at voltages
from 1.2V to 3.3V. This standard does not require the use of a reference voltage (VREF) or a
board termination voltage (VTT).

PCI — Peripheral Component Interface

The Peripheral Component Interface (PCI) standard specifies support for both 33 MHz
and 66 MHz PCI bus applications. It uses an LVTTL input buffer and a Push-Pull output
buffer. This standard does not require the use of a reference voltage (VREF) or a board
termination voltage (VTT); however, it does require a 3.3V output source voltage (VCCO).

GTL — Gunning Transceiver Logic Terminated

The Gunning Transceiver Logic (GTL) standard is a high-speed bus standard invented by
Xerox. Xilinx has implemented the terminated variation for this standard. This standard
requires a VREF-based input buffer and an Open-Drain output buffer.

338 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

GTL+ — Gunning Transceiver Logic Plus

The Gunning Transceiver Logic Plus (GTL+) standard is a high-speed bus standard
(JESD8.3) first used by the Intel Pentium Pro processor.

HSTL — High-Speed Transceiver Logic

The High-Speed Transceiver Logic (HSTL) standard is a general-purpose, high-speed 1.5V
or 1.8V bus standard sponsored by IBM. This standard has four variations or classes: Class
I, II, III, and IV. This standard requires a VREF-based input buffer and a Push-Pull output
buffer.

SSTL3 — Stub Series Terminated Logic for 3.3V

The Stub Series Terminated Logic standard is a general-purpose memory bus standard
sponsored by Hitachi and IBM (JESD8-8). This standard has multiple voltages from 1.8V to
3.3V, and two classes, I and II. This standard requires a VREF-based input buffer and a
Push-Pull output buffer.

SSTL2 — Stub Series Terminated Logic for 2.5V

The Stub Series Terminated Logic standard is a general-purpose memory bus standard
sponsored by Hitachi and IBM (JESD8-8). This standard has multiple voltages from 1.8V to
3.3V, and two classes, I and II. This standard requires a VREF-based input buffer and a
Push-Pull output buffer.

SSTL18 — Stub Series Terminated Logic for 1.8V

The SSTL18 standard, specified by JEDEC Standard JC42.3, is a general-purpose 1.8V
memory bus standard. This voltage-referenced standard has two variations or classes, both
of which require a reference voltage of 0.90 V, an input/output source voltage of 1.8 V, and
a termination voltage of 0.90 V. This standard requires a VREF-based input buffer and a
Push-Pull output buffer. SSTL18 is used for high-speed SDRAM interfaces.

LVDS — Low Voltage Differential Signal

LVDS is a differential I/O standard. As with all differential signaling standards, LVDS
requires that one data bit is carried through two signal lines, and it has an inherent noise
immunity over single-ended I/O standards. The voltage swing between two signal lines is
approximately 350 mV. The use of a reference voltage (VREF) or a board termination
voltage (VTT) is not required. LVDS requires the use of two pins per input or output. LVDS
requires resistor termination.

BLVDS — Bus LVDS

Allows for bidirectional LVDS communication between two or more devices. The Bus
LVDS standard requires external resistor termination.

LVPECL — Low Voltage Positive Emitter Coupled Logic

Differential I/O standard with voltage swing between two signal lines of approximately
850 mV. The use of a reference voltage (VREF) or a board termination voltage (VTT) is not
required. The LVPECL standard requires external resistor termination.

Spartan-3 Generation FPGA User Guide www.xilinx.com 339

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

LDT — HyperTransport (formerly known as Lightning Data Transport)

A differential high-speed, high-performance I/O interface standard. It is a point-to-point
standard requiring a 2.5V VCCIO, in which each HyperTransport technology bus consists
of two point-to-point unidirectional links. Each link is 2 to 32 bits. The HyperTransport
technology standard does not require an input reference voltage. However, it does require
a 100Ω termination resistor between the two signals at the input buffer.

mini-LVDS
A serial, intra-flat panel solution that serves as an interface between the timing control
function and an LCD source driver.

LVDS Extended — Extended Mode LVDS

Provides a higher drive capability and voltage swing (350 - 750 mV), which makes it ideal
for long-distance or cable LVDS links. This LVDS Extended Mode driver is intended for
situations that require higher drive capabilities in order to produce an LVDS signal that is
within EIA/TIA specification at the receiver.

RSDS — Reduced Swing Differential Signaling

A signaling standard that defines the output characteristics of a transmitter and inputs of
a receiver along with the protocol for a chip-to-chip interface between Flat Panel timing
Controllers and Column Drivers.

TMDS — Transition Minimized Differential Signaling

Technology for transmitting high-speed serial data used by the DVI and HDMI video
interfaces. The TMDS standard requires external 50Ω resistor pull-ups to 3.3V on inputs.

PPDS — Point-to-Point Differential Signaling

Differential next-generation LCD standard for interface to row and column drivers.

I/O Standard Differences between Spartan-3 Generation Families

The Spartan-3 generation FPGA families all support the common single-ended I/O
standards of LVCMOS at 1.2V to 3.3V, LVTTL, PCI, SSTL at 1.8V and 2.5V, and HSTL at
1.8V. All families also support the common differential I/O standards of LVDS, LVPECL,
BLVDS, and RSDS. The primary differences between the families result from the
optimization of the drive capability of the I/O transistors to reduce die size and cost. The
Spartan-3E family is optimized with smaller transistors and lower drive, while the
Extended Spartan-3A family has high drive on banks 1 and 3 and smaller transistors
(lower output drive) on banks 0 and 2. The Extended Spartan-3A family also offers the
most flexibility for differential standards, but limits differential outputs to banks 0 and 2.

Table 10-9: I/O Standard Differences between Spartan-3 Generation Families

Extended Spartan-3A Spartan-3E Spartan-3
Feature
Family FPGAs FPGAs FPGAs
LVCMOS Drive, Max 24 mA, Banks 1/3 16 mA 24 mA
Differential Outputs Banks 0/2 All Banks All Banks
Differential VCCO 3.3V or 2.5V 2.5V 2.5V

340 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

Table 10-9: I/O Standard Differences between Spartan-3 Generation Families

Extended Spartan-3A Spartan-3E Spartan-3
Feature
Family FPGAs FPGAs FPGAs
LVDS/RSDS Solution Excellent Very Good Good
Newest Differential Standard TMDS/PPDS mini-LVDS RSDS
PCI 66 MHz 66 MHz 33 MHz
Number of Banks 4 4 4-8
Digitally Controlled Impedance No No Yes

Table 10-10 and Table 10-11 show the available I/O standards for the Spartan-3 generation
families.

Table 10-10: Available Single-Ended I/O Standards

Extended Spartan-3A Spartan-3E Spartan-3
Standard VCCO Drive/Class
Family FPGAs FPGAs FPGAs
2 mA √ √ √
1.2V
up to 6 mA Banks 1/3 √
up to 6 mA √ √ √
1.5V
up to 12 mA Banks 1/3 √
up to 8 mA √ √ √
LVCMOS 1.8V
up to 16 mA Banks 1/3 √
up to 12 mA √ √ √
2.5V
up to 24 mA Banks 1/3 √
up to 16 mA √ √ √
3.3V
up to 24 mA Banks 1/3 √
up to 16 mA √ √ √
LVTTL 3.3V
up to 24 mA √ √
3.0V - √ √ √
PCI33
3.3V - √ √ √
PCI66 3.3V - √ √
I √ √ √
1.8V
II √(1) √
I √ √ √
SSTL 2.5V
II √(1) √
I √
3.3V
II √

Spartan-3 Generation FPGA User Guide www.xilinx.com 341

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-10: Available Single-Ended I/O Standards (Cont’d)

Extended Spartan-3A Spartan-3E Spartan-3
Standard VCCO Drive/Class
Family FPGAs FPGAs FPGAs
I √(1) √
1.5V
III √(1) √
HSTL I √ √ √
1.8V II √(1) √
III √ √ √
- - √
GTL
- Plus √
DCI option - - √

Notes:
1. Outputs are restricted to banks 1 and 3. Inputs are unrestricted.

Table 10-11: Available Differential I/O Standards

Extended Spartan-3A Spartan-3E Spartan-3
Standard VCCO
Family FPGAs FPGAs FPGAs
2.5V √(1) √ √
LVDS
3.3V √(1)
BLVDS 2.5V √ √ √
2.5V √(1) √
MINI_LVDS
3.3V √(1)
2.5V √ √ √
LVPECL
3.3V √
2.5V √(1) √ √
RSDS
3.3V √(1)
2.5V √
TMDS
3.3V √(1)
2.5V √(1)
PPDS
3.3V √(1)
LDT 2.5V √
LVDSEXT 2.5V √
DIFF_SSTL18_I 1.8V √ √
DIFF_SSTL18_II 1.8V √(2)
DIFF_SSTL2_I 2.5V √ √
DIFF_SSTL2_II 2.5V √(2) √

342 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

Table 10-11: Available Differential I/O Standards (Cont’d)

Extended Spartan-3A Spartan-3E Spartan-3
Standard VCCO
Family FPGAs FPGAs FPGAs
DIFF_SSTL3_I 3.3V √
DIFF_SSTL3_II 3.3V √
DIFF_HSTL_I_18 1.8V √ √
DIFF_HSTL_II_18 1.8V √(2) √
DIFF_HSTL_III_18 1.8V √ √
DIFF_TERM - ~100Ω ~120Ω
DCI Option - √

Notes:
1. These differential outputs are restricted to banks 0 and 2. Inputs are unrestricted.
2. These high-drive outputs are restricted to banks 1 and 3. Inputs are unrestricted.

Specifying an I/O Standard with the IOSTANDARD Attribute

To define the I/O signaling standard in a design, set the IOSTANDARD attribute to the
appropriate setting. The IOSTANDARD attribute can be applied to any I/O primitive. For
each I/O primitive, there is a version supporting single-ended IOSTANDARD attributes
and a version supporting differential IOSTANDARD attributes.

Table 10-12: Components for Single-Ended and Differential Standards

Component Single-Ended IOSTANDARD Differential IOSTANDARD
Input IBUF IBUFDS
Clock Input IBUFG IBUFGDS
Adjustable Delay
(Extended Spartan-3A family IBUF_DLY_ADJ IBUFDS_DLY_ADJ
FPGAs only)
Output OBUF OBUFDS
Output Three-State OBUFT OBUFTDS
Input DDR IDDR2
Output DDR ODDR2

Earlier libraries had I/O components with IOSTANDARD already specified as part of the
component, such as IBUF_LVTTL. These are not recommended for use in new designs. The
preferred method is to use the I/O component IBUF and assign IOSTANDARD = LVTTL.
IOSTANDARD can be attached to a net or signal when the net or signal is connected to a
pad. In this case, IOSTANDARD is treated as attached to the IOB primitive. When attached
to a design element, IOSTANDARD propagates to all applicable elements in the hierarchy
within the design element.
In VHDL, before using IOSTANDARD, it must be declared with the following syntax:
attribute iostandard: string;
After IOSTANDARD has been declared, specify the VHDL constraint as follows:

Spartan-3 Generation FPGA User Guide www.xilinx.com 343

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

attribute iostandard of {component_name|label_name}: {component|label}

is “iostandard_name”;
In Verilog, specify IOSTANDARD as follows:
// synthesis attribute iostandard [of] {module_name|instance_name} [is]
iostandard_name;
IOSTANDARD can also be specified in a constraints file, which can be created directly or
by using the PlanAhead tool.
The resulting IOSTANDARD syntax in the user constraints file (UCF) is the following:
NET “pad_net_name” IOSTANDARD=iostandard_name;
IOSTANDARD can also be specified on the I/O component:
INST “instance_name” IOSTANDARD=iostandard_name;
For more details on all constraints and entry methods, see the Constraints Guide in the
software documentation, especially the section on "Entry Strategies for Xilinx Constraints".

Timing Analysis
The choice of IOSTANDARD affects the timing for the I/O pin. The data sheet provides
example timing for the LVCMOS25 I/O standard with Fast slew rate and 12 mA drive.
This delay requires adjustment whenever a signal standard other than LVCMOS25 is
assigned to an Input or a standard other than LVCMOS25 with 12 mA drive and Fast slew
rate is assigned to an Output. The adjustments are automatically included in the Timing
Analyzer reports generated by the Xilinx development tools.
When measuring timing parameters at the programmable I/Os, different signal standards
call for different test conditions. The data sheets list the conditions to use for each standard.
The method for measuring Input timing is as follows: a signal that swings between a Low
logic level of VL and a High logic level of VH is applied to the Input under test. Some
standards also require the application of a bias voltage to the VREF pins of a given bank to
properly set the input-switching threshold. The measurement point of the Input signal
(VM) is commonly located halfway between VL and VH.
For the Output test setup, one end of the termination resistor RT is connected to a
termination voltage VT and the other end is connected to the Output. For each standard, RT
and VT generally take on the standard values recommended for minimizing signal
reflections. If the standard does not ordinarily use terminations (for example, LVCMOS,
LVTTL), then RT is set to 1MΩ to indicate an open connection, and VT is set to zero. The
same measurement point (VM) that was used at the Input is also used at the Output.

VT (VREF)

FPGA Output
RT (RREF)

VM (VMEAS)

CL (CREF)

ds312-3_04_090105

Figure 10-20: Output Test Setup

344 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

The capacitive load (CL) is connected between the output and GND. The Output timing for
all standards, as published in the speed files and the data sheet, is always based on a CL
value of zero. High-impedance probes (less than 1 pF) are used for all measurements. Any
delay that the test fixture might contribute to test measurements is subtracted from those
measurements to produce the final timing numbers as published in the speed files and
data sheet.

Using IBIS Models to Simulate Load Conditions in Application

IBIS models permit the most accurate prediction of timing delays for a given application.
The parameters found in the IBIS model (VREF, RREF, and VMEAS) correspond directly with
the parameters found in the data sheet (VT, RT, and VM). Do not confuse VREF (the
termination voltage) from the IBIS model with VREF (the input-switching threshold) from
the table. A fourth parameter, CREF, is always zero. The four parameters describe all
relevant output test conditions. IBIS models are found in the Xilinx development software
as well as at the following link:
http://www.xilinx.com/support/download/index.htm
Delays for a given application are simulated according to its specific load conditions as
follows:
1. Simulate the desired signal standard with the output driver connected to the test setup
shown in the data sheet. Use parameter values VT, RT, and VM from the data sheet;
CREF is zero.
2. Record the time to VM.
3. Simulate the same signal standard with the output driver connected to the PCB trace
with load. Use the appropriate IBIS model (including VREF, RREF, CREF, and VMEAS
values) or capacitive value to represent the load.
4. Record the time to VMEAS.
5. Compare the results of steps 2 and 4. Add (or subtract) the increase (or decrease) in
delay to (or from) the appropriate Output standard adjustment to yield the worst-case
delay of the PCB trace.

LVCMOS/LVTTL Slew Rate Control and Drive Strength

Each IOB has a slew-rate control that sets the output switching edge rate for LVCMOS and
LVTTL outputs. The SLEW attribute controls the slew rate and can be set to SLOW
(default), FAST, or QUIETIO (Extended Spartan-3A family devices only; slowest slew
rate). The slowest slew rate setting provides the lowest noise and power consumption,
while the faster slew rate settings improve timing.
Each LVCMOS and LVTTL output additionally supports up to seven different drive
current strengths as shown in Table 10-13 and Table 10-14. To adjust the drive strength for
each output, the DRIVE attribute is set to the desired drive strength: 2, 4, 6, 8, 12, 16, and 24.
Unless otherwise specified in the FPGA application, the software default IOSTANDARD is
LVCMOS25, SLOW slew rate, and 12 mA output drive.
Each method of specifying IOSTANDARD (schematic, HDL, constraints file, or the
PlanAhead tool) also supports specification of the LVCMOS/LVTTL DRIVE and SLEW
options.

Spartan-3 Generation FPGA User Guide www.xilinx.com 345

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-13: Extended Spartan-3A Family FPGA Programmable Output Drive

Current
Output Drive Current (mA)
IOSTANDARD
2 4 6 8 12 16 24
LVTTL 9 9 9 9 9 9 9
LVCMOS33 9 9 9 9 9 9 Banks
1,3
LVCMOS25 9 9 9 9 9 Banks Banks
1,3 1,3
LVCMOS18 9 9 9 9 Banks Banks
1,3 1,3
LVCMOS15 9 9 9 Banks Banks -
1,3 1,3
LVCMOS12 9 Banks Banks - - -
1,3 1,3

Table 10-14: Spartan-3E FPGA Programmable Output Drive Current

Output Drive Current (mA)
IOSTANDARD
2 4 6 8 12 16
LVTTL 9 9 9 9 9 9
LVCMOS33 9 9 9 9 9 9
LVCMOS25 9 9 9 9 9 -
LVCMOS18 9 9 9 9 - -
LVCMOS15 9 9 9 - - -
LVCMOS12 9 - - - - -

Table 10-15: Spartan-3 FPGA Programmable Output Drive Current

Output Drive Current (mA)
IOSTANDARD
2 4 6 8 12 16 24
LVTTL 9 9 9 9 9 9 9
LVCMOS33 9 9 9 9 9 9 9
LVCMOS25 9 9 9 9 9 9 9
LVCMOS18 9 9 9 9 9 9
LVCMOS15 9 9 9 9 9 -
LVCMOS12 9 9 9 - - -

High output current drive strength and FAST output slew rates generally result in fastest
I/O performance. However, these same settings can also result in transmission line effects
on the PCB for all but the shortest board traces. Each IOB has independent slew rate and

346 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

drive strength controls. Use the slowest slew rate and lowest output drive current that
meets the performance requirements for the end application. Note that in the
Extended Spartan-3A family, the 16 mA drive setting is faster than the 24 mA drive setting
for the slow slew rate. If 24 mA drive and the highest performance is needed, use the fast
slew rate instead.
LVCMOS25/33 and LVTTL standards have about 100 mV of hysteresis on inputs.

Simultaneously Switching Outputs

Likewise, due to lead inductance, a given package supports a limited number of
simultaneous switching outputs (SSOs) when using fast, high-drive outputs. Only use fast,
high-drive outputs when required by the application.
Module 3 of each family's data sheet provides guidelines for the recommended maximum
allowable number of SSOs. These guidelines describe the maximum number of user I/O
pins of a given output signal standard that should simultaneously switch in the same
direction, while maintaining a safe level of switching noise. Meeting these guidelines for
the stated test conditions ensures that the FPGA operates free from the adverse effects of
ground and power bounce.
Ground or power bounce occurs when a large number of outputs simultaneously switch in
the same direction. The output drive transistors all conduct current to a common voltage
rail. Low-to-High transitions conduct to the VCCO rail; High-to-Low transitions conduct to
the GND rail. The resulting cumulative current transient induces a voltage difference
across the inductance that exists between the die pad and the power supply or ground
return. The inductance is associated with bonding wires, the package lead frame, and any
other signal routing inside the package. Other variables contribute to SSO noise levels,
including stray inductance on the PCB as well as capacitive loading at receivers. Any SSO-
induced voltage consequently affects internal switching noise margins and ultimately
signal quality.
For each device/package combination, the data sheet provides the number of equivalent
VCCO/GND pairs. For each output signal standard and drive strength, the data sheet
recommends the maximum number of SSOs, switching in the same direction, allowed per
VCCO/GND pair within an I/O bank. The guidelines are categorized by package style,
slew rate, and output drive current. Furthermore, the number of SSOs is specified by I/O
bank. Multiply the appropriate numbers from each table to calculate the maximum
number of SSOs allowed within an I/O bank. Exceeding these SSO guidelines might result
in increased power or ground bounce, degraded signal integrity, or increased system jitter.
The recommended maximum SSO values assumes that the FPGA is soldered on the
printed circuit board and that the board uses sound design practices. The SSO values do
not apply for FPGAs mounted in sockets, due to the lead inductance introduced by the
socket.
The number of SSOs allowed for quad-flat packages (TQ) is lower than for ball grid array
packages (FG) due to the larger lead inductance of the quad-flat packages. Ball grid array
packages are recommended for applications with a large number of simultaneously
switching outputs.
SSO effects can be minimized by adding virtual ground pins. A virtual ground is created
by defining an output pin driven by a logic 0. See “Optionally Place Virtual Ground Pins
Around DCM Input and Output Connections” in Chapter 3

Spartan-3 Generation FPGA User Guide www.xilinx.com 347

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

HSTL/SSTL VREF Reference Voltage

HSTL and SSTL inputs use the reference voltage (VREF) to bias the input-switching
threshold, as shown in Figure 10-21. Each input also has an associated board termination
voltage, called VTT.

SSTL2 Class I
V = 1.25V
TT
V = 2.5V
CCO

50Ω
25Ω
Z = 50

VREF = 1.25V

x133_15_011900

Figure 10-21: Example Terminated SSTL2 Class I

Once a configuration data file is loaded into the FPGA that calls for the inputs of a given
bank to use HSTL/SSTL, a few specifically reserved I/O pins on the same bank
automatically convert to VREF inputs. All the VREF inputs on a bank need to be connected,
and all need to connect to the same voltage. As a result, HSTL and SSTL inputs can only be
combined in a bank if they use the same VREF voltage (for example, the 1.8V versions of the
SSTL and HSTL standards, where VREF = 0.9V.) For banks that do not contain HSTL or
SSTL, VREF pins remain available for user I/Os or input pins.
VREF is also required for inputs using the GTL and GTLP I/O standards, which are
supported only in the Spartan-3 family. LVTTL and LVCMOS standards do not require
VREF . The only differential standards that require VREF are the differential forms of the
HSTL and SSTL I/O standards (DIFF_HSTL and DIFF_SSTL).

Table 10-16: VREF Values for IOSTANDARD Settings

IOSTANDARD VREF VTT
HSTL_I_18 0.9 0.9
HSTL_II_18 0.9 0.9
HSTL_III_18 1.1 1.8
HSTL_I 0.75 0.75
HSTL_III 0.9 1.5
SSTL18_I 0.9 0.9
SSTL18_II 0.9 0.9
SSTL2_I 1.25 1.25
SSTL2_II 1.25 1.25
SSTL3_I 1.5 1.5
SSTL3_II 1.5 1.5
GTL 0.8 1.2
GTLP 1.0 1.5

348 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

Single-Ended I/O Termination Techniques

The delay of an electrical signal along a wire is dominated by the rise and fall times when
the signal travels a short distance. Transmission line delays vary with inductance and
capacitance, but a well-designed board can experience delays of approximately 180 ps per
inch.
Transmission line effects, or reflections, typically start at 1.5 inches for fast (1.5 ns) rise and
fall times. Poor (or non-existent) termination or changes in the transmission line
impedance cause these reflections and can cause additional delay in longer traces. As
system speeds continue to increase, the effect of I/O delays can become a limiting factor,
and therefore transmission line termination becomes increasingly more important.
A variety of termination techniques reduce the impact of transmission line effects. Output
termination techniques include the following:
• None
• Series
• Parallel (Shunt)
• Series and Parallel (Series-Shunt)
Input termination techniques include the following:
• None
• Parallel (Shunt)
These termination techniques can be applied in any combination. A generic example of
each combination of termination methods appears in Figure 10-22, page 349.

Unterminated Double Parallel Terminated

VTT VTT
Z=50

Z=50

VREF

Unterminated Output Driving Series Terminated Output Driving

a Parallel Terminated Input a Parallel Terminated Input
VTT VTT

Z=50 Z=50

VREF VREF

Series-Parallel Terminated Output

Driving a Parallel Terminated Input
VTT VTT
Series Terminated Output

Z=50
Z=50
VREF
VREF

x133_07_112499

Figure 10-22: Overview of Standard Input and Output Termination Methods

Sample circuits illustrating valid termination techniques for several HSTL and SSTL
standards appear in Figure 10-23 through Figure 10-29. LVTTL, LVCMOS, and PCI
standards require no termination. For GTL or DCI termination in the Spartan-3 family, see
the Spartan-3 FPGA data sheet at DS099.

Spartan-3 Generation FPGA User Guide www.xilinx.com 349

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

HSTL Class I
VTT = 0.75V
VCCO = 1.5V
50Ω
Z = 50

VREF = 0.75V

x133_10_112499

Figure 10-23: Terminated HSTL Class I

HSTL Class III

VTT = 1.5V
VCCO = 1.5V
50Ω
Z = 50

VREF = 0.9V

x133_11_112499

Figure 10-24: Terminated HSTL Class III

HSTL Class IV
VTT = 1.5V VTT = 1.5V
VCCO = 1.5V
50Ω 50Ω
Z = 50

VREF = 0.9V

x133_12_112499

Figure 10-25: Terminated HSTL Class IV

SSTL3 Class I
VTT = 1.5V
VCCO = 3.3V
50Ω
25Ω
Z = 50

VREF = 1.5V
x133_13_112499

Figure 10-26: Terminated SSTL3 Class I

SSTL3 Class II
VTT = 1.5V VTT = 1.5V
VCCO = 3.3V
50Ω 50Ω
25Ω
Z = 50

VREF = 1.5V
x133_14_112499

Figure 10-27: Terminated SSTL3 Class II

350 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

SSTL2 Class I
V = 1.25V
TT
V = 2.5V
CCO

50Ω
25Ω
Z = 50

VREF = 1.25V

x133_15_011900

Figure 10-28: Terminated SSTL2 Class I

SSTL2 Class II
VTT= 1.25V VTT= 1.25V
VCCO = 2.5V
50Ω 50Ω
25Ω
Z = 50

VREF = 1.25V
x133_16_011900

Figure 10-29: Terminated SSTL2 Class II

Differential I/O Standards

Differential standards employ a pair of signals, one the opposite polarity of the other. The
noise canceling properties (for example, Common-Mode Rejection) of these standards
permit exceptionally high data transfer rates. This subsection introduces the differential
signaling capabilities of Spartan-3 generation devices.
Each device-package combination designates specific I/O pairs specially optimized to
support differential standards. A unique L-number, part of the pin name, identifies the line
pairs associated with each bank. For each pair, the letters P and N designate the true and
inverted lines, respectively. For example, the pin names IO_L43P_3 and IO_L43N_3
indicate the true and inverted lines comprising the line pair L43 on Bank 3.
Each family offers a different combination of differential I/O standards and specifications.
The Extended Spartan-3A family offers the largest number of differential standards and
also offers the best differential signaling characteristics.

On-Chip Differential Termination

Extended Spartan-3A family and Spartan-3E devices provide an on-chip differential
termination across the input differential receiver terminals. The on-chip input differential
termination potentially eliminates the external 100Ω termination resistor commonly found
in differential receiver circuits (see Figure 10-30). Differential termination is used for LVDS,
mini-LVDS, and RSDS as applications permit. On-chip differential termination is not
supported on input-only pins.

Spartan-3 Generation FPGA User Guide www.xilinx.com 351

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Spartan-3A Z0 = 50Ω Spartan-3E

Differential Differential Input
Output

100Ω
Z0 = 50Ω

Spartan-3E
Differential Input
Spartan-3A with On-Chip
Z0 = 50Ω Differential
Differential
Terminator

~100Ω
Output

Z0 = 50Ω
DS312-2_24_061606

Figure 10-30: Differential Inputs and Outputs

The on-chip differential termination is powered by VCCO. Therefore, the VCCO level in a
bank must match the voltage standard for any input using differential termination. In the
Extended Spartan-3A family, on-chip differential termination is specified at 100Ω nominal
in banks with VCCO = 3.3V. The on-chip differentiation termination can be used in banks
powered by VCCO = 2.5V, but a wider resistance range is specified. See Module 3 of DS529,
Spartan-3A FPGA Family Data Sheet for specific values. Figure 10-31 shows the details of
using the differential termination in the Extended Spartan-3A family.

Bank 0 and 2 Any Bank

Bank 0 Bank 0

Bank 1
Bank 3

1/4th of Bourns No VCCO Restrictions

Part Number LVDS_33, LVDS_25,
Bank 2 Bank 2 MINI_LVDS_33,
Z0 = 50Ω CAT16-PT4F4
MINI_LVDS_25,
VCCO = 3.3V VCCO = 2.5V RSDS_33, RSDS_25,
PPDS_33, PPDS_25
LVDS_33, LVDS_25,
MINI_LVDS_33, MINI_LVDS_25, Z0 = 50Ω 100Ω
RSDS_33, RSDS_25,
PPDS_33 PPDS_25
DIFF_TERM=No
a) Input-only differential pairs or pairs not using DIFF_TERM=Yes constraint

VCCO = 3.3V VCCO = 2.5V

Z0 = 50Ω
LVDS_33, LVDS_25,
VCCO = 3.3V VCCO = 2.5V MINI_LVDS_33, MINI_LVDS_25,
RSDS_33, RSDS_25,
LVDS_33, LVDS_25, PPDS_33 PPDS_25
MINI_LVDS_33, MINI_LVDS_25, Z0 = 50Ω RDT
RSDS_33, RSDS_25,
PPDS_33 PPDS_25

DIFF_TERM=Yes
b) Differential pairs using DIFF_TERM=Yes constraint DS529-3_09_022507

Figure 10-31: External Input Termination Resistors for Extended Spartan-3A Family FPGA LVDS, RSDS,
MINI_LVDS, and PPDS I/O Standards

In the Spartan-3E family, on-chip differential termination is only supported on banks with
VCCO = 2.5V, and is specified at 120Ω nominal (see Module 3 of DS312, Spartan-3E FPGA
Family Data Sheet).
The DIFF_TERM attribute is set to TRUE to enable differential termination on a differential
I/O pin pair. This attribute uses the following syntax in the UCF:
INST <I/O_BUFFER_INSTANTIATION_NAME> DIFF_TERM = “<TRUE/FALSE>”;

352 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

SelectIO Signal Standards

TMDS_33 Termination
The Extended Spartan-3A family TMDS_33 standard requires pull-up resistors as shown
in Figure 10-32.

Bank 0 and 2 Any Bank

Bank 0 Bank 0

Bank 1
Bank 3
3.3V
Bank 2 Bank 2
50Ω 50Ω
VCCO = 3.3V VCCAUX = 3.3V

TMDS_33 TMDS_33

DVI/HDMI cable DS529-3_08_020107

Figure 10-32: External Input Resistors Required for TMDS_33 I/O Standard

BLVDS Output Termination

BLVDS outputs require external termination as shown in Figure 10-33.

1/4th of Bourns 1/4th of Bourns

Part Number Part Number
CAT16-LV4F12 CAT16-PT4F4

VCCO = 2.5V Z0 = 50Ω VCCO = 2.5V

165Ω

140Ω 100Ω
Z0 = 50Ω

165Ω
ds312-3_07_102105

Figure 10-33: External Output and Input Termination Resistors for BLVDS I/Os

For a bidirectional BLVDS connection, the design must be simulated using the IBIS models
to verify resistor values and their effect on rise and fall times. Refer to XAPP243, Bus LVDS
with Virtex-E Devices for information on an example termination method.
In the Extended Spartan-3A family, the BLVDS outputs are allowed on any bank, and there
is no VCCO restriction on inputs, as shown in Figure 10-34.

Any Bank Any Bank

Bank 0 Bank 0
1/4 th of Bourns
Bank 1

Bank 1
Bank 3

1/4 th of Bourns
Bank 3

Part Number Part Number

Bank 2 CAT16-LV4F12 CAT16-PT4F4 Bank 2
VCCO = 2.5V 165Ω Z0 = 50Ω No VCCO Requirement

BLVDS_25 140Ω Z0 = 50Ω 100Ω BLVDS_25

165Ω
DS529-3_07_020107

Figure 10-34: Extended Spartan-3A Family FPGA External Output and Input
Termination Resistors for BLVDS I/Os

Spartan-3 Generation FPGA User Guide www.xilinx.com 353

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

DCI Digitally Controlled Impedance

In the Spartan-3 family, many standards support the Digitally Controlled Impedance (DCI)
feature, which uses integrated terminations to eliminate unwanted signal reflections. This
feature is also known as XCITE technology. DCI provides two kinds of on-chip
terminations. Parallel terminations make use of an integrated resistor network. Series
terminations result from controlling the impedance of output drivers.
DCI is available only by selecting certain IOSTANDARD options, as listed in the
Spartan-3 FPGA data sheet. The DCI feature operates independently for each of the
device’s eight banks. Each bank has an ‘N’ reference pin (VRN) and a ‘P’ reference pin
(VRP) to calibrate driver and termination resistance. Only when using certain DCI
standards on a given bank do these two pins function as VRN and VRP. When not using a
DCI standard that requires them, the two pins function as user I/Os. Add an external
reference resistor to pull the VRN pin up to VCCO and another reference resistor to pull the
VRP pin down to GND.
For more details on DCI, see DS099, Spartan-3 FPGA Family Data Sheet.

354 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Supply Voltages for the IOBs

The IOBs are powered by three supplies:
1. The VCCO supplies, one for each of the FPGA’s I/O banks, power the output drivers.
The voltage on the VCCO pins determines the voltage swing of the output signal. All
VCCO pins should be connected. If a bank is unused, VCCO pins should be connected to
an available VCCAUX or VCCO rail.
2. VCCINT is the main power supply for the FPGA’s internal logic.
3. VCCAUX is an auxiliary source of power, primarily to optimize the performance of
various FPGA functions, such as the DCM, differential signals including LVPECL_33
and TMDS_33 inputs, and some single-ended input signals, such as LVCMOS25 and
LVCMOS33 inputs.

Spartan-3A and Spartan-3A DSP FPGA Dual-Range VCCAUX

The Spartan-3A and Spartan-3A DSP FPGAs in the Extended Spartan-3A family allow
VCCAUX to be either 2.5V or 3.3V. The option provides greater flexibility to the user, and
allows VCCAUX to be set to the same level as an existing VCCO rail to minimize the number
of power rails. The user must set the CONFIG VCCAUX constraint to either 2.5 or 3.3
according to the voltage being provided to the VCCAUX rails. The CONFIG VCCAUX
constraint is used by the ISE software to determine if LVCMOS25 inputs can be powered
by VCCAUX. If CONFIG VCCAUX = 2.5, VCCAUX is used to power LVCMOS25 inputs. If
CONFIG VCCAUX = 3.3, VCCO must be 2.5V for any banks with LVCMOS25 inputs. The
Spartan-3AN platform requires that VCCAUX be set to 3.3V for the In-System Flash
memory.

ESD Protection
Protection circuitry on all Spartan-3 generation I/Os protects all device pads against
damage from electro-static discharge (ESD) as well as excessive voltage transients. ESD
protection specifications are typically ±2000V for the Human Body Model. Details are
provided in Module 3 of each family’s data sheet.
In the Extended Spartan-3A family, this protection circuitry does not limit I/O voltage
range.
In the Spartan-3E and Spartan-3 families, clamp diodes protect all device pads against
damage from both ESD as well as excessive voltage transients. Each I/O has two clamp
diodes: one diode extends P-to-N from the pad to VCCO, and a second diode extends
N-to-P from the pad to GND. During operation, these diodes are normally biased in the off
state. These clamp diodes are always connected to the pad, regardless of the signal
standard selected. The presence of diodes limits the ability of Spartan-3/3E FPGA I/Os to
tolerate high signal voltages. The VIN absolute maximum rating in Module 3 of each data
sheet specifies the voltage range that I/Os can tolerate. Input voltages outside the VIN max
voltage range are permissible provided that the IIK input diode clamp diode rating is met
and no more than 100 pins exceed the range simultaneously.
The Extended Spartan-3A family of FPGAs has clamp diodes to VCCO only after
configuring the I/Os to the PCI33 or PCI66 I/O standards.

IOBs Organized into Banks

Spartan-3 generation FPGAs allow multiple I/O standards to be combined in the same
device. Although the outputs are always powered by VCCO, multiple standards are

Spartan-3 Generation FPGA User Guide www.xilinx.com 355

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

available under one of the five possible VCCO values. In addition, inputs often do not need
to match the voltage applied to VCCO.
Further flexibility is achieved by offering multiple VCCO levels in a single device. The
VCCO power rails are provided independently each bank of I/Os, or side of the device.
Most Spartan-3 generation devices organize IOBs into four I/O banks as shown in
Figure 10-35. Each bank maintains separate VCCO and VREF supplies. The separate
supplies allow each bank to independently set VCCO (which provides current to the
outputs and additionally powers the on-chip differential termination) and VREF (which
supplies the reference voltage for HSTL and SSTL). Refer to Table 10-17 through
Table 10-19 for VCCO and VREF requirements. Most members of the Spartan-3 family have
eight I/O banks—see DS099, Spartan-3 FPGA Family Data Sheet for more details.

Bank 0

Bank 3

Bank 1
Bank 2

DS312-2_26_021205

Figure 10-35: Spartan-3 Generation I/O Banks (top view)

The design implementation tools automatically assign pins to separate banks when
necessary to meet VCCO or VREF requirements. The user can also assign pins using any of
the floorplanning tools available. In the pinout, bank numbers are specified for each I/O
pin on the device. For example, IO_L18P_0 is a part of differential pair L18 on bank 0.

Single-Ended I/O Standard Bank Compatibility

For a particular VCCO voltage, Table 10-17 through Table 10-19 list all of the single-ended
IOSTANDARDs that can be combined, and if IOSTANDARD is supported only for inputs
or can be used for both inputs and outputs.

Table 10-17: Extended Spartan-3A Family FPGA Single-Ended IOSTANDARD Bank Compatibility
VCCO Supply/Compatibility Input Requirements
Single-Ended Board
IOSTANDARD 1.2V 1.5V 1.8V 2.5V 3.3V VREF Termination
Voltage (VTT)
Input/
LVTTL Input Input Input Input N/R(1) N/R
Output
Input/
LVCMOS33 Input Input Input Input N/R N/R
Output
Input/
LVCMOS25 Input(2) Input(2) Input(2) Input(2) N/R N/R
Output
Input/
LVCMOS18 Input Input Input Input N/R N/R
Output

356 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Table 10-17: Extended Spartan-3A Family FPGA Single-Ended IOSTANDARD Bank Compatibility (Cont’d)
VCCO Supply/Compatibility Input Requirements
Single-Ended Board
IOSTANDARD 1.2V 1.5V 1.8V 2.5V 3.3V VREF Termination
Voltage (VTT)
Input/
LVCMOS15 Input Input Input Input N/R N/R
Output
Input/
LVCMOS12 Input Input Input Input N/R N/R
Output
Input/
PCI33_3 Input Input Input Input N/R N/R
Output
Input/
PCI66_3 Input Input Input Input N/R N/R
Output
Input/
HSTL_I_18 Input Input Input Input 0.9 0.9
Output
Input/
HSTL_II_18 Input Input Input Input 0.9 0.9
Output
Input/
HSTL_III_18 Input Input Input Input 1.1 1.8
Output
Input/
HSTL_I Input Input Input Input 0.75 0.75
Output
Input/
HSTL_III Input Input Input Input 0.9 1.5
Output
Input/
SSTL18_I Input Input Input Input 0.9 0.9
Output
Input/
SSTL18_II Input Input Input Input 0.9 0.9
Output
Input/
SSTL2_I Input Input Input Input 1.25 1.25
Output
Input/
SSTL2_II Input Input Input Input 1.25 1.25
Output
Input/
SSTL3_I Input Input Input Input 1.5 1.5
Output
Input/
SSTL3_II Input Input Input Input 1.5 1.5
Output

Notes:
1. N/R - Not required for input operation.
2. To use LVCMOS25 inputs when VCCO is not 2.5V, VCCAUX must be set to 2.5V.

Spartan-3 Generation FPGA User Guide www.xilinx.com 357

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-18: Spartan-3E FPGA Single-Ended IOSTANDARD Bank Compatibility

VCCO Supply/Compatibility Input Requirements
Single-Ended Board
IOSTANDARD 1.2V 1.5V 1.8V 2.5V 3.3V VREF Termination
Voltage (VTT)
Input/
LVTTL - - - - N/R(1) N/R
Output
Input/
LVCMOS33 - - - - N/R N/R
Output
Input/
LVCMOS25 - - - Input N/R N/R
Output
Input/
LVCMOS18 - - Input Input N/R N/R
Output
Input/
LVCMOS15 - Input Input Input N/R N/R
Output
Input/
LVCMOS12 Input Input Input Input N/R N/R
Output
Input/
PCI33_3 - - - - N/R N/R
Output
Input/
PCI66_3 - - - - N/R N/R
Output
Input/
HSTL_I_18 - - Input Input 0.9 0.9
Output
Input/
HSTL_III_18 - - Input Input 1.1 1.8
Output
Input/
SSTL18_I - - Input Input 0.9 0.9
Output
Input/
SSTL2_I - - - Input 1.25 1.25
Output

Notes:
1. N/R - Not required for input operation.

358 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Table 10-19: Spartan-3 FPGA Single-Ended IOSTANDARD Bank Compatibility

VCCO Supply/Compatibility Input Requirements
Single-Ended Board
IOSTANDARD 1.2V 1.5V 1.8V 2.5V 3.3V VREF(1) Termination
Voltage (VTT)
Input/
LVTTL - - - - N/R(2) N/R
Output
Input/
LVCMOS33 - - - - N/R N/R
Output
Input/
LVCMOS25 - - - - N/R N/R
Output
Input/
LVCMOS18 - - - - N/R N/R
Output
Input/
LVCMOS15 - - - - N/R N/R
Output
Input/
LVCMOS12 - - - - N/R N/R
Output
Input/
PCI33_3 - - - - N/R N/R
Output
Input/
HSTL_I_18 - - Input Input 0.9 0.9
Output
Input/
HSTL_II_18 - - Input Input 0.9 0.9
Output
Input/
HSTL_III_18 - - Input Input 1.1 1.8
Output
Input/
HSTL_I - Input Input Input 0.75 0.75
Output
Input/
HSTL_III - Input Input Input 0.9 1.5
Output
Input/
SSTL18_I - - Input Input 0.9 0.9
Output
Input/
SSTL18_II - - Input Input 0.9 0.9
Output
Input/
SSTL2_I - - - Input 1.25 1.25
Output
Input/
SSTL2_II - - - Input 1.25 1.25
Output
GTL Note 2 0.8 1.2
GTLP - Note 2 1.0 1.5

Notes:
1. Banks 4 and 5 of any Spartan-3 device in a VQ100 package do not support signal standards using VREF.
2. The VCCO level used for the GTL and GTLP standards must be no lower than the termination voltage (VTT), nor can it be lower than
the voltage at the I/O pad.

Spartan-3 Generation FPGA User Guide www.xilinx.com 359

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Differential I/O Standard Bank Compatibility

Most of the differential I/O standards are compatible and can be combined within any
given bank. In the Extended Spartan-3A family, banks 0 and 2 can each support any two of
the following 2.5V differential standards:
• LVDS_25 outputs
• MINI_LVDS_25 outputs
• RSDS_25 outputs
• PPDS_25 outputs
Extended Spartan-3A family banks 0 and 2 alternatively support any two of the following
3.3V differential outputs:
• LVDS_33 outputs
• MINI_LVDS_33 outputs
• RSDS_33 outputs
• PPDS_33 outputs
• TMDS_33 outputs

Maximizing Availability of Differential Pins in the Extended Spartan-3A Family

In the Extended Spartan-3A family, differential outputs are limited to two of the four
banks. If a large number of differential signals are needed, put differential inputs on banks
1 and 3, freeing up banks 0 and 2 for differential outputs and I/O. A larger device or larger
package will further increase the available differential I/O. Note that the pseudo-
differential standards (BLVDS_25 and Differential SSTL/HSTL) do not have this bank
restriction. BLVDS_25 could be considered as an alternative to LVDS_25. The advantage is
that BLVDS_25 allows outputs in banks 1 and 3, but the disadvantage is that it requires
external resistors.
In the Spartan-3E family, each bank can support any two of the following differential
standards:
• LVDS_25 outputs
• MINI_LVDS_25 outputs
• RSDS_25 outputs
As an example, LVDS_25 outputs, RSDS_25 outputs, and any other differential inputs
while using on-chip differential termination are a valid combination. A combination not
allowed is a single bank with LVDS_25 outputs, RSDS_25 outputs, and MINI_LVDS_25
outputs.

Table 10-20: Extended Spartan-3A Family FPGA Differential IOSTANDARD Bank Compatibility

Differential Differential Pad Type Differential Bank

I/O Type VCCAUX(3) VCCO
IOSTANDARD IP IO Restriction(1)

Input 2.5/3.3 Any Yes Yes No

BLVDS_25 Input with DIFF_TERM Not available No
Output 2.5/3.3 2.5 No Yes No

360 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Table 10-20: Extended Spartan-3A Family FPGA Differential IOSTANDARD Bank Compatibility (Cont’d)

Differential Differential Pad Type Differential Bank

I/O Type VCCAUX(3) VCCO
IOSTANDARD IP IO Restriction(1)

Input 2.5/3.3 Any Yes Yes No

LVDS_25 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 2.5 No Bank 0/2 Yes
Input 2.5/3.3 Any Yes Yes No
LVDS_33 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 3.3 No Bank 0/2 Yes
Input 2.5/3.3 Any Yes Yes No
MINI_LVDS_25 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 2.5 No Bank 0/2 Yes
Input 2.5/3.3 2.5 - 3.3 Yes Yes No
MINI_LVDS_33 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 3.3 No Bank 0/2 Yes
Input 2.5/3.3 Any Yes Yes No
RSDS_25 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 2.5 No Bank 0/2 Yes
Input 2.5/3.3 2.5 - 3.3 Yes Yes No
RSDS_33 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 3.3 No Bank 0/2 Yes
Input 2.5/3.3 Any Yes Yes No
PPDS_25 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 2.5 No Bank 0/2 Yes
Input 2.5/3.3 2.5 - 3.3 Yes Yes No
PPDS_33 Input with DIFF_TERM 2.5/3.3 2.5 - 3.3 No Yes No
Output 2.5/3.3 3.3 No Bank 0/2 Yes
Input 2.5/3.3 Any Yes Yes No
LVPECL_25 Input with DIFF_TERM Not available No
Output Not available No
Input 3.3(4) 2.5 - 3.3 Yes Yes No
LVPECL_33 Input with DIFF_TERM Not available No
Output Not available No
Input 3.3(4) 2.5 - 3.3 Yes Yes No
TMDS_33 Input with DIFF_TERM 3.3(4) 3.3 No Yes No
Output 2.5/3.3 3.3 No Bank 0/2 Yes
DIFF_HSTL 2.5/3.3 Same as Single-Ended Standards

Spartan-3 Generation FPGA User Guide www.xilinx.com 361

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-20: Extended Spartan-3A Family FPGA Differential IOSTANDARD Bank Compatibility (Cont’d)

Differential Differential Pad Type Differential Bank

I/O Type VCCAUX(3) VCCO
IOSTANDARD IP IO Restriction(1)

DIFF_SSTL 2.5/3.3 Same as Single-Ended Standards

Notes:
1. Banks 0 and 2 can each support any two of the following 2.5V differential standards: LVDS_25 outputs, MINI_LVDS_25 outputs,
RSDS_25 outputs, PPDS_25 outputs, or any two of the following 3.3V differential standards: LVDS_33 outputs, MINI_LVDS_33
outputs, RSDS_33 outputs, PPDS_33 outputs, TMDS_33 outputs. Other I/O bank restrictions might apply.
2. VREF is not used for the differential I/O standards.
3. Spartan-3AN FPGAs require VCCAUX = 3.3V
4. Power VCCAUX rails at 3.3V and set CONFIG VCCAUX = 3.3.

Table 10-21: Spartan-3E FPGA Differential IOSTANDARD Bank Compatibility

Differential VCCO Supply Differential Bank

IOSTANDARD 1.8V 2.5V 3.3V Restriction(1)

Input,
Applies to
LVDS_25 Input On-chip Differential Termination(2), Input
Outputs Only
Output
Input,
Applies to
RSDS_25 Input On-chip Differential Termination(2), Input
Outputs Only
Output
Input,
Applies to
MINI_LVDS_25 Input On-chip Differential Termination(2), Input
Outputs Only
Output
LVPECL_25 Input Input Input
BLVDS_25 Input Input, Output Input
No Differential
DIFF_HSTL_I_18 Input, Output Input Input Bank Restriction
(other I/O bank
DIFF_HSTL_III_18 Input, Output Input Input restrictions might
apply)
DIFF_SSTL18_I Input, Output Input Input
DIFF_SSTL2_I Input Input, Output Input

Notes:
1. Each bank can support one of the following: LVDS_25 outputs, MINI_LVDS_25 outputs, RSDS_25 outputs.
2. On-chip differential termination is not supported on input-only pins (differential pad type IP).
3. VREF is not used for the differential I/O standards.

Table 10-22: Spartan-3 FPGA Differential IOSTANDARD Bank Compatibility

Differential VCCO Supply Differential Bank

IOSTANDARD 1.8V 2.5V 3.3V Restriction(1)

Applies to Outputs
LVDS_25 Input Input, Output Input
Only
Applies to Outputs
RSDS_25 Input Input, Output Input
Only

362 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Table 10-22: Spartan-3 FPGA Differential IOSTANDARD Bank Compatibility (Cont’d)

Differential VCCO Supply Differential Bank

IOSTANDARD 1.8V 2.5V 3.3V Restriction(1)

Applies to Outputs
LDT_25 (ULVDS_25) Input Input, Output Input
Only
Applies to Outputs
LVDSEXT_25 Input Input, Output Input
Only
LVPECL_25 Input Input, Output Input No Differential
BLVDS_25 Input Input, Output Input Bank Restriction
(other I/O bank
DIFF_HSTL_II_18 Input, Output Input Input restrictions might
DIFF_SSTL2_II Input Input, Output Input apply)

Notes:
1. Each bank can support any two of the following: LVDS_25 outputs, RSDS_25 outputs, LDT_25 (ULVDS_25) outputs, LVDSEXT_25
outputs.
2. VREF is not used for the differential I/O standards.

I/O Banking Rules

When assigning I/Os to banks, these VCCO rules must be followed:
1. All VCCO pins on the FPGA must be connected even if a bank is unused.
2. All VCCO lines associated within a bank must be set to the same voltage level.
3. The VCCO levels used by all standards assigned to the I/Os of any given bank must
agree. The Xilinx development software checks for this. Table 10-17 through
Table 10-22 describe how different standards use the VCCO supply.
4. If a bank does not have any VCCO requirements, connect VCCO to an available voltage,
such as 2.5V or 3.3V. Some configuration modes might place additional VCCO
requirements.
If any of the standards assigned to the inputs of the bank use VREF , then the following
additional rules must be observed:
1. All VREF pins must be connected within a bank.
2. All VREF lines associated with the bank must be set to the same voltage level.
3. The VREF levels used by all standards assigned to the inputs of the bank must agree.
The Xilinx development software checks for this. Table 10-17 through Table 10-19
describe how different standards use the VREF supply.
If VREF is not required to bias the input switching thresholds, all associated VREF pins
within the bank can be used as user I/Os or input pins.

Using Large-Swing Signals

Independent of the I/O standard compatibility with VCCO , care must be taken to ensure
that VIN input voltages do not exceed the maximum specifications, which are sometimes
specified in relation to VCCO. For example, the Spartan-3 family specifies a VIN
recommended maximum of VCCO + 0.3V to avoid turning on the input protection diodes
(see Module 3 of DS099, Spartan-3 FPGA Family Data Sheet for the specifications). The
Spartan-3E family specifies a VIN recommended maximum of VCCO + 0.5V to avoid

Spartan-3 Generation FPGA User Guide www.xilinx.com 363

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

turning on the input protection diodes (see Module 3 of DS312, Spartan-3E FPGA Family
Data Sheet for the specifications). The Extended Spartan-3A family FPGA maximum VIN
values are independent of VCCO, except for the PCI standards —see Module 3 of
DS529, Spartan-3A FPGA Family Data Sheet for the specifications.
In some applications it might be desirable to receive signals with a greater voltage swing
than the I/Os ordinarily permit. The most common case is receiving 5V signals on pins set
to a 3.3V I/O standard. These large-swing signals might be by design or can be a result of
severe overshoot.
A similar situation might exist on the outputs, where the Spartan-3 generation FPGA
needs to drive external devices supporting standards with larger swing. The
Spartan-3 generation outputs at 3.3V can directly drive most 5V devices, although with
less margin. Similarly, the LVCMOS25 dedicated configuration outputs can directly drive
most 3.3V external devices.
For the specific case of interfacing to the PCI bus, see XAPP457, Powering and Configuring
Spartan-3 Generation FPGAs in Compliant PCI Applications.

Voltage Translators
Xilinx recommends the use of voltage level translators as the preferred solution when
interfacing with large-swing signals. A voltage level translator can be as simple as a two-
resistor voltage divider, or as complex as a PCI-to-PCI bridge.

Open-Drain Interfacing
According to another approach, the outputs of certain external devices can be configured
as open-drain outputs. Such outputs with pull-up resistors tied to a low voltage rail can be
used to limit the signal swing so that the FPGA’s Power Diode does not turn on. See
Figure 10-36.

VCCAUX = 2.5V VCCAUX = 2.5V

OFF
Input
ROH

IOH VOH
Open-Drain OFF
ROL
FPGA

X484_10_121205

Figure 10-36: A 3.3V Open-Drain Output Connected to an FPGA’s Dedicated Input

The open-drain output is comparatively slow with reduced noise margin; it is most
suitable for cases where timing is not too critical. Fortunately, in most cases, Dedicated
Inputs (PROG_B, TDI, TMS, and TCK) do not usually need to switch very fast.

364 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

Voltage Clamps Using Internal Diodes

Input voltages outside the recommended range are permissible provided that the IIK input
clamp diode rating is met and I/O coupling effects are accounted for. To meet the IIK input
clamp diode rating in the Spartan-3 and Spartan-3E families, a series resistor can be added
to the input. This solution can be used to apply 3.3V signals to configuration pins
dedicated to LVCMOS25 standards. For more information on the configuration interface,
see UG332, Spartan-3 Generation Configuration User Guide, and XAPP453, The 3.3V
Configuration of Spartan-3 FPGAs.

In this type of solution, the internal Power Diode between the FPGA’s output and its
associated power rail is allowed to turn on, and a resistor in series with the output is used
to limit the current. The current, which flows back into the regulator, is known as reverse
current. Another resistor can be put across the power supply’s output to help ensure proper
regulation.
In this solution, parasitic leakage current between user I/Os in differential pin pairs can
occur, even though the I/Os are configured with single-ended standards. This parasitic
leakage can occur when the VIN is beyond the recommended operating conditions, either
positive or negative, even if the input current is limited. This parasitic current can cause
unexpected device behavior, but does not damage the device. If you use this technique for
large-swing signals, you should either leave the other pin of the differential pair unused, or
manage the potential effects of the increased leakage. See XAPP459, “Eliminating I/O
Coupling Effects when Interfacing Large-Swing Single-Ended Signals to User I/O Pins on
Spartan-3 Generation FPGAs.”

Parasitic Leakage
Parasitic leakage current between user I/Os in differential pin pairs can occur even though
the I/Os are configured with single-ended standards. To provide flexibility, differential
I/O pairs can be used either as a differential I/O or as two single-ended I/Os using
transistors to disable the DIFF_TERM resistor when it is not needed, as shown in
Figure 10-37. This affects Spartan-3, Spartan-3E, and Spartan-3A family differential I/O
pairs.
Parasitic leakage occurs in all Spartan-3 generation families. Although the Spartan-3
family does not support DIFF_TERM resistors, parasitic leakage still exists and must be
accounted for. The Spartan-3 family general recommended operating conditions is limited
to –0.3V.

Spartan-3 Generation FPGA User Guide www.xilinx.com 365

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Victim IBUF

Parasitic Leakage
Aggressor IBUF

–0.2 < VAGGRESSOR < –0.5V

DIFF_TERM IBUFDS
Disabled

DIFF_TERM
Disabled

UG331_c10_20_080410

Figure 10-37: Differential Pairs Used as Single-Ended I/Os

If either of the inputs is negatively driven for any extended periods of time, parasitic
leakage might exist. Pins not affected include these cases:
• The input is located on an input-only pin: IP_L<XXY>_<Z>
• XX: Differential pair numbering
• Y: P/N
• Z: Bank number
• The input is located on a pin that only supports single-ended I/O standards: IO_<Z>
• Z: Bank number
• The aggressor undershoot is 0.0V to –0.2V
• The victim is an input with a pull-up resistor less than 1 kΩ with the aggressor pulse
less than 1 ns for the following IOSTANDARDs:
• LVCMOS33
• LVCMOS25
• LVCMOS18
• LVCMOS15
• The victim is an input pin actively driven High with at least 7.0 mA
• The victim is an output pin with any of the following IOSTANDARDs:
• LVCMOS33_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16, 24]
• LVCMOS25_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16, 24]
• LVCMOS18_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16]
• LVCMOS15_[FAST, SLOW, QUIETIO]_[4, 6, 8, 12]
• LVCMOS12_[FAST, SLOW, QUIETIO]_[6]
• PCI33_3, PCI66_3
• HSTL_I, HSTL_III, HSTL_I_18, HSTL_II_18, HSTL_III_18
• SSTL18_I, SSTL_18_II, SSTL2_I, SSTL2_II, SSTL3_I, SSTL3_II
Two typical situations can occur that might cause parasitic leakage to affect a design. In the
first situation, as is more common, an undershoot on input voltages can cause fast-falling
edges to drop below 0V for short periods of time (see Figure 10-38). Undershoot is only a

366 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

problem if the aggressor's input voltage drops below –0.2 V long enough for the
differential termination resistors to be enabled.

VCCO

0V

–0.2V

–0.5V

1 ns < TUNDERSHOOT < 10 ns

UG331_c10_21_080910

Figure 10-38: Timing Considerations for Input Undershoot

To analyze the effect of the parasitic leakage associated with undershoot, testing is
performed using a weak pull-up termination resistor, as shown in Figure 10-39. A negative
pulse is injected into the aggressor. The leakage current is then measured on the victim.
Leakage current for undershoot pulses of 1 ns is shown in Table 10-23.
In the case of LVCMOS12, the victim is pulled down to 0.62V when the aggressor voltage
is –0.5V. For LVCMOS12 inputs, this is below the recommended input voltage
(VIH = 0.7V) required to ensure that a high value is correctly detected.

VTT

IVICTIM (mA) 1000Ω

IBUF
Victim
DIFF_TERM
Disabled

DIFF_TERM IBUF
Disabled
Aggressor

UG331_c10_22_080910

Figure 10-39: Undershoot Test Setup

Spartan-3 Generation FPGA User Guide www.xilinx.com 367

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

Table 10-23: Victim Voltage with Parasitic Leakage at 125°C

VIH [V]
VVICTIM [V] VVICTIM [V]
Recommended
(VAGGRESSOR = –0.3V) (VAGGRESSOR = –0.5V)
Input Voltage
LVCMOS33, VTT = 3.0V 2.93 2.44 2.0
LVCMOS25, VTT = 2.3V 2.23 1.77 1.7
LVCMOS18, VTT = 1.65V 1.59 1.14 0.8
LVCMOS15, VTT = 1.4V 1.34 0.90 0.8
LVCMOS12, VTT = 1.1V 1.04 0.62 0.7

Notes:
1. 1 kΩ pull-up resistor terminated to VTT, TUNDERSHOOT = 1 ns

In the second situation where the undershoot lasts long enough, understanding how the
parasitic leakage behaves in a steady state is important. For example, during hot-swap,
input signals coming over a backplane might cause an input voltage to switch below 0.0V
long enough to cause parasitic leakage. Figure 10-40 shows the leakage current for the
different voltages of the victim pin. Because the leakage curves change with different
aggressor voltages, four curves are shown in the figure with the aggressor voltage set at –
0.2V, –0.3V,
–0.4V, and –0.5V.

6
VAGGRESSOR = –0.5V

4
IVICTIM (mA)

VAGGRESSOR = –0.4V
2

1
VAGGRESSOR = –0.3V
VAGGRESSOR = –0.2V
0
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
VVICTIM (V)
UG331_c10_23_080410

Figure 10-40: Figure Parasitic Leakage Across Different VAGGRESSOR at 85°C

Leakage current increases at higher temperatures, as shown in Figure 10-41 and

Figure 10-42.

368 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

6.00
VAGGRESSOR = –0.5V
5.00

4.00

IVICTIM (mA)
3.00
VAGGRESSOR = –0.4V
2.00

1.00
VAGGRESSOR = –0.3V
VAGGRESSOR = –0.2V
0.00
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
VVICTIM (V) UG331_c10_24_080610

Figure 10-41: Parasitic Leakage Across Different VAGGRESSOR at 100°C

7.00

VAGGRESSOR = –0.5V
6.00

5.00
IVICTIM (mA)

4.00

3.00
VAGGRESSOR = –0.4V
2.00

VAGGRESSOR = –0.3V
1.00

VAGGRESSOR = –0.2V
0.00
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
VVICTIM (V) UG331_c10_25_080610

Figure 10-42: Parasitic Leakage Across Different VAGGRESSOR at 125°C

Given the parasitic leakage, it is possible to understand how different termination

solutions are affected by the parasitic leakage. Figure 10-43 shows how three resistor
values can be plotted against the leakage curves to approximate the parasitic leakage effect
on a signal being pulled up by a resistor for LVCMOS33.
For LVCMOS33, VIH = 2.0V. Using the parasitic curve for VAGGRESSOR = –0.5V (85°C),
IVICTIM = 5.1 mA when VVICTIM = 2.0V. Because the pull-up resistor is terminated to 3.3V,
the pullup resistor must be 254Ω or lower.
RPULLUP = (VCCO – VIH)/IVICTIM = 254Ω

Spartan-3 Generation FPGA User Guide www.xilinx.com 369

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

6.00
RPULLUP = 691Ω
VCCO = 3.3V
5.00
RPULLUP = 254Ω
VAGGRESSOR = –0.5V
VCCO = 3.3V
4.00

IVICTIM (mA)
VIH = 2.0V
3.00 RPULLUP = 2955Ω
VCCO = 3.3V
VAGGRESSOR = –0.4V
2.00

1.00
VAGGRESSOR = –0.3V
VAGGRESSOR = –0.2V
0.00
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
VVICTIM (V)
UG331_c10_26_080610

Figure 10-43: Resistor Sizing Using Parasitic Leakage for LVCMOS33 (85°C)

Figure 10-44 and Figure 10-45 show three resistor values plotted against the leakage curves
for LVCMOS25 and LVCMOS12, respectively.

6.00 RPULLUP = 158Ω

VCCO = 2.5V
5.00
VAGGRESSOR = –0.5V
4.00
IVICTIM (mA)

RPULLUP = 438Ω
VIH = 1.7V
VCCO = 2.5V
3.00

2.00
VAGGRESSOR = –0.4V

RPULLUP = 1889Ω
1.00
VCCO = 2.5V
VAGGRESSOR = –0.3V
VAGGRESSOR = –0.2V
0.00
-0.5 0 0.5 1 1.5 2 2.5
VVICTIM (V) UG331_c10_28_080610

Figure 10-44: Resistor Sizing Using Parasitic Leakage Curve for LVCMOS25 (85°C)

370 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Supply Voltages for the IOBs

6.00
RPULLUP = 107Ω
RPULLUP = 302Ω VCCO= 1.2V
VAGGRESSOR = –0.5V
5.00 VCCO= 1.2V

4.00

IVICTIM (mA)
VIH = 0.7V
3.00

VAGGRESSOR = –0.4V
2.00

RPULLUP = 1388Ω
1.00 VCCO= 1.2V
VAGGRESSOR = –0.3V
VAGGRESSOR = –0.2V
0.00
-0.5 0 0.5 1 1.5
VVICTIM (V)
UG331_c10_28_080610

Figure 10-45: Resistor Sizing Using Parasitic Leakage Curve for LVCMOS12 (85°C)

Similarly, it is possible to determine if an output driver is strong enough to keep the output
at the logical High level, VOH. Using the Spartan-3A FPGA IBIS models, the output drivers
can be compared against the parasitic leakage curves. As shown in Figure 10-46,
LVCMOS33 (drive = 2 or 4 mA) crosses the VAGGRESSOR = –0.5V curve close to the VOH
limit of 2.9V. Using LVCMOS33 with an output drive of 6 mA or more provides additional
margin.

25
LVCMOS33 24 mA

LVCMOS33 16 mA
20
LVCMOS33 12 mA

LVCMOS33 6,8 mA
IVICTIM (mA)

15 LVCMOS33 2,4 mA

10

VAGGRESSOR = –0.5V
5
VOH = 2.9V
VAGGRESSOR = –0.4V
VAGGRESSOR = –0.3V
0 VAGGRESSOR = –0.2V
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
VVICTIM (V) UG331_c10_29_080610

Figure 10-46: Parasitic Leakage for LVCMOS33 Outputs

If the design encounters an issue, these design suggestions are provided:

• Separate aggressor and victim pins
• Aggressor adjustments:
• Minimize undershoot using capacitors or series resistors on potential aggressors
• Use a slower SLEW rate, if available

Spartan-3 Generation FPGA User Guide www.xilinx.com 371

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

• Reduce drive strength, if available

• Victim adjustments:
• For inputs using pull-up resistors, decrease the pull-up resistor value to
compensate for the parasitic leakage currents
• For outputs on a victim pin, increase drive strength to ensure VOH levels are met:
- LVCMOS33_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16, 24]
- LVCMOS25_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16, 24]
- LVCMOS18_[FAST, SLOW, QUIETIO]_[6, 8, 12, 16]
- LVCMOS15_[FAST, SLOW, QUIETIO]_[4, 6, 8, 12]
- LVCMOS12_[FAST, SLOW, QUIETIO]_[6]

I/O and Input-Only Pin Behavior During Power-On, Configuration, and

User Mode
In this section, all behavior described for I/O pins also applies to input-only pins and dual-
purpose I/O pins that are not actively involved in the currently selected configuration
mode.
The VCCINT (1.2V), VCCAUX, and VCCO supplies can be applied in any order. Before the
FPGA can start its configuration process, VCCINT , VCCO Bank 2 (Bank 4 in the Spartan-3
family) and VCCAUX must have reached their respective minimum recommended
operating levels indicated in Module 3 of each family’s data sheet. At this time, all output
drivers are in a high-impedance state. VCCO Bank 2, VCCINT, and VCCAUX serve as inputs
to the internal power-on reset (POR) circuit.
A Low level applied to the Pull Up During Configuration (PUDC_B) input enables the
pull-up resistors on user-I/O and input-only pins from power-on throughout
configuration. A High level on PUDC_B disables the pull-up resistors, allowing the I/Os to
float. PUDC_B contains a weak pull-up and defaults to High if left floating. In the
Spartan-3E family, this pin is called HSWAP, and in the Spartan-3 family, it is named
HSWAP_EN.
As soon as power is applied, the FPGA begins initializing its configuration memory. At the
same time, the FPGA internally asserts the Global Set-Reset (GSR), which asynchronously
resets all IOB storage elements to a default Low state.
Upon the completion of initialization and the beginning of configuration, INIT_B goes
High, sampling the M0, M1, and M2 inputs to determine the configuration mode.
Configuration data is then loaded into the FPGA. The I/O drivers remain in a high-
impedance state (with or without pull-up resistors, as determined by the PUDC_B input)
throughout configuration.
At the end of configuration, the GSR net is released, placing the IOB registers in a Low state
by default, unless the loaded design reverses the polarity of their respective SR inputs.
The Global Three State (GTS) net is released during start-up, marking the end of
configuration and the beginning of design operation in the User mode. After the GTS net is
released, all user I/Os go active while all unused I/Os are weakly pulled down
(PULLDOWN). The designer can control how the unused I/Os are terminated after GTS is
released by setting the UnusedPin Bitstream Generator (BitGen) option to PULLUP,
PULLDOWN, or FLOAT.
One clock cycle later (default), the Global Write Enable (GWE) net is released allowing the
RAM and registers to change states. Once in User mode, any pull-up resistors enabled by

372 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

PUDC_B revert to the user settings and PUDC_B is available as a general-purpose I/O. For
more information on PULLUP and PULLDOWN, see “Pull-Up and Pull-Down Resistors,”
page 334.
For more information on the power-up and configuration processes, see UG332, Spartan-3
Generation Configuration User Guide.

Behavior of Unused I/O Pins After Configuration

By default, the Xilinx ISE development software automatically configures all unused I/O
pins as input pins with individual internal pull-down resistors to GND.
This default behavior is controlled by the UnusedPin BitGen option.

Related Materials and References

• Spartan-3 Generation Data Sheets
I/O specifications and pin-outs.
http://www.xilinx.com//support/documentation/index.htm
• UG332: Spartan-3 Generation Configuration User Guide
I/O pin function during configuration.
http://www.xilinx.com/support/documentation/user_guides/ug332.pdf
• I/O Block Application Notes
http://www.xilinx.com/support/documentation/topicfpgafeaturedesign_ioblock.htm
• Signal Integrity Central
Documents and links designed to give you everything you need to achieve reliable
PCB designs on the first try.
http://www.xilinx.com/signalintegrity

Spartan-3 Generation FPGA User Guide www.xilinx.com 373

UG331 (v1.7) August 19, 2010
 R

Chapter 10: Using I/O Resources

374 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 11

Using Embedded Multipliers

Summary
Dedicated 18x18 multipliers speed up DSP logic in the Spartan®-3 generation families. The
multipliers are fast and efficient at implementing signed or unsigned multiplication of up
to 18 bits. In addition to basic multiplication functions, the embedded multiplier block can
be used as a shifter or to generate magnitude or two’s-complement return of a value. The
multipliers can be cascaded with each other or CLB logic for larger or more complex
functions.
The Spartan-3A DSP platform devices includes high-performance DSP48A blocks that are
compatible with the Virtex®-4 FPGA DSP48 architecture. These blocks support multiply
accumulate operations at over 250 MHz.

Introduction
Spartan-3 generation FPGAs have a number of features to fortify the chip’s arithmetic
capabilities. Carry logic and dedicated carry routing continues to be provided as in past
generations. Dedicated AND gates in the CLBs accelerate array multiplication operations.
The most significant addition is the dedicated 18x18 two’s-complement multiplier block.
With 3 to 104 of these dedicated multipliers in each device, fast arithmetic functions can be
implemented with minimal use of the general-purpose resources. In addition to the
performance advantage, dedicated multipliers require less power than CLB-based
multipliers.
The embedded multipliers offer fast, efficient means to create 18-bit signed by 18-bit
signed multiplication products. The multiplier blocks share routing resources with the
Block SelectRAM™ memory, allowing for increased efficiency for many applications.
Applications such as signed-signed, signed-unsigned, and unsigned-unsigned
multiplication, logical, arithmetic, and barrel shifters, two’s-complement and magnitude
return are easily implemented.
High-level synthesis tools usually automatically infer the dedicated multiplier for generic
multiplication operations in VHDL or Verilog. To allow more user control or to use special
features of the multiplier, it can be instantiated in a design or defined using the
CORE Generator™ system.

Spartan-3 Generation FPGA User Guide www.xilinx.com 375

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

Embedded Multiplier Resource Differences between

Spartan-3 Generation Families
The Spartan-3E and Spartan-3A/3AN FPGA families have similar embedded multiplier
resources, with the Spartan-3A/3AN families offering additional routing flexibility.
Although the resources are similar, there are timing differences between the families. The
families share the same multiplier primitive, MULT18X18SIO, with optional input and
output registers and cascade signals. The original Spartan-3 family offers a simpler
multiplier without pipelining or cascade capabilities and uses the MULT18X18
(combinatorial) or MULT18X18S (registered) primitives. This chapter focuses on the
architecture of the Spartan-3E and Spartan-3A/3AN families. For details on the Spartan-3
multiplier, see DS099, Spartan-3 FPGA Family Data Sheet and XAPP467, Using Embedded
Multipliers in Spartan-3 FPGAs.
The Spartan-3A DSP platform has a new high-performance DSP48A block that is
compatible with the Virtex-4 FPGA DSP48 and supports multiply accumulate operations
at over 250 MHz. This block is backwards compatible with the Spartan-3E and
Spartan-3A/3AN FPGA multipliers. For more information, see UG431, XtremeDSP Slice for
Spartan-3A DSP FPGAs User Guide.

Two’s-Complement Signed Multiplier

The multiplier primitive MULT18X18SIO is shown in Figure 11-1. The multiplier blocks
primarily perform two’s complement numerical multiplication but can also perform some
less obvious applications, such as simple data storage and barrel shifting. Each multiplier
performs the principle operation P = A × B, where A and B are 18-bit words in two’s
complement form, and P is the 36-bit full-precision product, also in two’s complement
form. The 18-bit inputs represent values ranging from -131,07210 to +131,07110 and the
resulting product ranges from -17,179,738,11210 to +17,179,869,18410.

MULT18X18SIO
A[17:0] P[35:0]
B[17:0]
CEA
CEB
CEP
CLK
RSTA
RSTB
RSTP
BCIN[17:0] BCOUT[17:0]

DS312-2_28_021205

Figure 11-1: MULT18X18SIO Primitive

Optional registers are available on both the A and B inputs and the P output. The
registered paths share a common clock CLK and have independent active-High clock
enables and synchronous resets. The CLK, CE, and RST inputs all have programmable
polarity.

376 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Two’s-Complement Signed Multiplier

Table 11-1: Number of Multipliers per Spartan-3 Generation Device

Device Multiplier Columns Multipliers
Spartan-3A DSP FPGAs
XC3SD1800A 4 84 DSP48A
XC3SD3400A 5 126 DSP48A
Spartan-3A/3AN FPGAs
XC3S50A/AN 1 3
XC3S200A/AN 2 16
XC3S400A/AN 2 20
XC3S700A/AN 2 20
XC3S1400A/AN 2 32
Spartan-3E FPGAs
XC3S100E 1 4
XC3S250E 2 12
XC3S500E 2 20
XC3S1200E 2 28
XC3S1600E 2 36
Spartan-3 FPGAs
XC3S50 1 4
XC3S200 2 12
XC3S400 2 16
XC3S1000 2 24
XC3S1500 2 32
XC3S2000 2 40
XC3S4000 4 96
XC3S5000 4 104

The 18-bit width of the Spartan-3 generation multiplier is unusual but matches with the 18-
bit width of the block RAM, which includes parity bits. Standard 8-bit or 16-bit multipliers
can be created by using part of the multiplier block, or a 32-bit multiplier can be created via
cascading. The Xilinx architecture allows any non-standard bit width to be implemented,
exactly matching the needs of the application. Unused multiplier inputs are connected
automatically to zero via connections to unused LUTs that are set to zero.

Location Constraints
MULT18X18SIO embedded multiplier instances can have LOC properties attached to them
to constrain placement. MULT18X18SIO placement locations differ from the convention
used for naming CLB locations, allowing LOC properties to transfer easily from array to
array.

Spartan-3 Generation FPGA User Guide www.xilinx.com 377

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

The LOC properties use the following form:

LOC = MULT18X18_X#Y#
For example, MULT18X18_X0Y0 is the bottom-left MULT18X18SIO location on the device.

Multiplier/Block RAM Routing Interaction

Each multiplier is located adjacent to an 18 Kbit block RAM and shares some interconnect
resources. In the Spartan-3 and Spartan-3E families, configuring an 18 Kbit block RAM for
32/36-bit wide data (512 x 36 mode) prevents use of the associated dedicated multiplier
because the lower 16 bits of the A multiplicand input are shared with the upper 16 bits of
the block RAM’s Port A Data input. Similarly, the lower 16 bits of the B multiplicand input
are shared with Port B’s Data input.
The Spartan-3A/3AN platforms offer additional routing between the block RAM and
multiplier. The A port inputs are independent, so the multiplier can always be used even if
the block RAM outputs the full 36-bit width On port A. Because Port B is still shared, it is
recommended to define a mixed-width block RAM with the 36-bit data on port A and the
narrower data (up to x18) on port B.
The Spartan-3A/3AN platforms also offer direct routing from the block RAM into the
multiplier. This path helps improve routability and performance when the multiplier
coefficients are stored in the adjacent block RAM. See Figure 11-2 for details of the
Spartan-3A/3AN platform block RAM and multiplier connections.

DIA[#:0] DOA[#:0]
Port A DOA[17:0]

A[17:0] A
RAMB16BWE
MULT18X18SIO
DIB[#:0] DOB[#:0]
DOB[17:0]
Port B B[17:16]
DIB[31:16] B[15:0] B[17:0] B

UG331_c13_03_081106

Figure 11-2: Spartan-3A/3AN Multiplier/Block RAM Connectivity

The Spartan-3A DSP platform offers enhanced routing and features that avoid conflicts
between block RAM and multiplier routing.

Optional Pipeline Registers

As shown in Figure 11-3, each multiplier block has optional registers on each of the
multiplier inputs and the output. The registers, named AREG, BREG, and PREG, and can
be used in any combination. The clock input is common to all the registers within a block,
but each register has an independent clock enable and synchronous reset controls making
it ideal for storing data samples and coefficients.

378 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Two’s-Complement Signed Multiplier

AREG
(Optional)

CEA CE
A[17:0] D Q

PREG
(Optional)
RST
CEP CE
RSTA X D Q P[35:0]

BREG
(Optional)
RST
CEB CE
B[17:0] D Q RSTP

RST

RSTB
DS312-2_27_021205
CLK

Figure 11-3: Principle Ports and Functions of Dedicated Multiplier Blocks

The pipeline registers can be inferred or instantiated. A single-stage multiplier typically

infers usage of the output PREG register. A two-stage multiplier typically infers the usage
of both the AREG/BREG input registers and the PREG output register.
To instantiate the pipeline registers on the MULT18X18SIO primitive, the individual
AREG, BREG, and PREG attributes are set to 1 to insert the associated register or to 0 to
remove it and make the signal path combinatorial. The default is 1 or fully pipelined. The
attribute can be modified via the generic map (VHDL) or named parameter value
assignment (Verilog) as a part of the instantiated component.

Timing Specification
Multiplier performance can be enhanced by limiting the number of bits or putting critical
signals on the LSBs, or by pipelining. When pipelining, the registers boost the multiplier
clock rate, beneficial for higher performance applications.
The result is generated faster for the LSBs than the MSBs, since the MSBs require more
levels of addition, so timing specifications are different for each of the 36 multiplier
outputs. Designs should use only as many output bits as are necessary. For example, if two
unsigned numbers will never have a product of 235 or higher, the P[35] output is always
zero. For any pair of signed numbers of n bits, if you will never have -2n-1 x -2n-1, then the
MSB is always identical to the next lower-order bit (P[2n-1] = P[2n-2]). Also consider that if
some outputs must have longer routing delays, they should be put on the output LSBs to
balance with the MSB delays.
For the same reason, the data input setup time for the registered output multiplier is
shorter for the MSBs than the LSBs, but the timing parameters do not differentiate between
pins for setup time. For additional safety margin in a design, slower inputs should be put
on the MSBs.

Spartan-3 Generation FPGA User Guide www.xilinx.com 379

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

Expanding Multipliers
Multiplication using inputs with more than 18 bits is possible by decomposing the
multiplication process into smaller subprocesses. The binary representation of either input
can be split at any point, provided the proper weighting and sign of the MSBs is taken into
account. Splitting off the 18 MSBs of the input makes the best use of the 18-bit signed
multipliers.

Cascading Multipliers
The Spartan-3E/3A/3AN FPGA MULT18X18SIO primitive has two additional ports
called BCIN and BCOUT to cascade or share the multiplier’s B input among several
multiplier blocks. The 18-bit BCIN “cascade” input port offers an alternate input source
from the more typical B input. The B_INPUT attribute specifies whether the specific
implementation uses the BCIN or B input path. Setting B_INPUT to DIRECT chooses the B
input. Setting B_INPUT to CASCADE selects the alternate BCIN input. The BREG register
then optionally holds the selected input value, if required.
BCOUT is an 18-bit output port that always reflects the value applied to the multiplier’s
second input. This value is the B input, the cascaded value from the BCIN input, or the
output of the BREG, if it is inserted.
Figure 11-4 illustrates the four possible configurations using different settings for the
B_INPUT attribute and the BREG attribute.

BCOUT[17:0] BCOUT[17:0]

BREG
CEB CE
X X

D Q

CLK BREG = 0
RST B_INPUT = CASCADE

BREG = 1
RSTB B_INPUT = CASCADE

BCIN[17:0] BCIN[17:0]

BCOUT[17:0]
BCOUT[17:0]

BREG
X
CEB CE
X
B[17:0]
B[17:0] D Q

BREG = 0
CLK B_INPUT = DIRECT
RST

BREG = 1
RSTB B_INPUT = DIRECT DS312-2_29_021505

Figure 11-4: Four Configurations of the B Input

The BCIN and BCOUT ports have associated dedicated routing that connects adjacent
multipliers within the same column. Via the cascade connection, the BCOUT port of one
multiplier block drives the BCIN port of the multiplier block directly above it. There is no
connection to the BCIN port of the bottom-most multiplier block in a column or a

380 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Expanding Multipliers

connection from the BCOUT port of the top-most block in a column. As an example,
Figure 11-5 shows the multiplier cascade capability for a column of multipliers four blocks
tall. For clarity, the figure omits the register control inputs.

BCOUT
A
P
B

B_INPUT = CASCADE BCIN

BCOUT
A
P
B

B_INPUT = CASCADE BCIN

BCOUT
A
P
B

B_INPUT = CASCADE BCIN

BCOUT
A
P
B

B_INPUT = DIRECT BCIN

DS312-2_30_021505

Figure 11-5: Multiplier Cascade Connection

When using the BREG register, the cascade connection forms a shift register structure
typically used in DSP algorithms such as direct-form FIR filters. When the BREG register is
omitted, the cascade structure essentially feeds the same input value to more than one
multiplier. This parallel connection serves to create wide-input multipliers and implement
transpose FIR filters. It is used in any application requiring several multipliers to have the
same input value.

Examples
For example, Figure 11-6 shows how a 22x16 multiplier could be implemented. The 22-bit
value is decomposed into an 18-bit signed value and a 4-bit unsigned value from the LSBs.
Two partial products are formed. The first is a 20-bit signed product, which is the result of
multiplying the 16-bit signed value by the 4-bit unsigned section. The second is a 34-bit
signed product, formed by multiplying the 16-bit signed value by the 18-bit signed section.
The addition process restores the weighting of the products (note the least significant bits
of the first product bypass the addition) and forms the final 38-bit product. Since the first

Spartan-3 Generation FPGA User Guide www.xilinx.com 381

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

product is signed, the 20-bit value needs to be sign-extended before addition. The adder
itself only needs to be 34 bits, requiring 17 slices.

18
22 34
A
16 34
MULT18X18SIO +
16
B 16
16 38
P
20
Unsigned 4 4
UG331_c13_07_081106

Figure 11-6: 22x16 Multiplier Implementation

The implementation can vary depending on the performance needs and available
resources. The second multiplier can be implemented in the MULT18X18SIO resource or in
CLBs if it is small. Pipelining can be added to improve performance, using the built-in
capabilities of the dedicated multipliers. If both inputs are greater than 18 bits, then four
partial products are formed, but the purely unsigned result from the LSBs simply can be
concatenated with the 36-bit signed product of the MSBs and added to the other two
results.
Figure 11-7 represents the cascaded scheme used to implement a 35-bit by 35-bit signed
multiplier utilizing four embedded multipliers and two adders.
The fixed adder is 53 bits wide (17 LSBs are always 0 on one input).
The 34-bit by 34-bit unsigned submodule is constructed in a similar manner with the most
significant bit on each operand being tied to logic Low.

382 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Two Multipliers in a Single Primitive

A[34:17] A
36 36 36
P [69:34]
B[34:17] B
MULT18X18SIO
[33:0]

0, A[16:0] A
34 34 34
P
0, B[16:0] B

MULT18X18SIO 70 70
+

A[34:17] A
36 36 MSB 69
P
0, B[16:0] B • •
• •
MULT18X18SIO MSB 53
36 36
+ [52:17]
0 16
0, A[16:0] A • •
36 36 • •
P 0 0
B[34:17] B

MULT18X18SIO
UG331_c13_08_081106

Figure 11-7: 35x35 Signed Multiplier

Two Multipliers in a Single Primitive

The dedicated multiplier can be used to multiply two smaller numbers at the same time.
By putting one value on the LSBs and one on the MSBs, two independent results can be
obtained as long as the results do not overlap with each other on the outputs. Shifting one
of the values n positions to the MSBs is the same as multiplying it by 2n. If the value shifted
to the MSBs is X, then the new value is X * 2n. If the value on the LSBs is Y, then the
complete multiplier input is X * 2n + Y.
For simplified illustration purposes, an assumption of two squares being implemented in
the same MULT18X18SIO primitive is used. The following equation shows the form of the
multiplication.
Two Multipliers per Primitive:
(X * 2n + Y)(X * 2n + Y) = (X2 * 22n) + (XY * 2n+1) + (Y2)
For values 0 on X or Y, the equation becomes:
X2 * 22n {Y=0} (X2 on the output MSBs)
Y2 {X=0} (Y2 on the output LSBs)
0 {X=0, Y=0}
With both X and Y at non-zero values, care must be taken to avoid overlap between the
results on the MSBs and LSBs and the middle term (XY * 2n+1). Two multipliers can coexist
in one MULT18X18SIO primitive, if the conditions in the following inequalities are met
when neither X nor Y are 0.
Inequality Conditions for Two Multipliers per Primitive:

Spartan-3 Generation FPGA User Guide www.xilinx.com 383

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

(X2 * 22n)min > (XY * 2n+1)max, (XY * 2n+1)min > (Y2)max

Table 11-2 shows values for X and Y where these conditions are met.
Table 11-2: Two Multipliers per MULT18X18SIO Allowable Sizes
X*X Y*Y
Signed Size Unsigned Size Unsigned Size
6X6 5X5 5X5
5X5 4X4 6X6
4X4 3X3 7X7

Figure 11-8 represents the MULT18X18SIO connections for calculating the square of both a
6-bit signed number and a 5-bit unsigned number.

A_6S [17:12]
x00 [11:5] A
A_5U [4:0]
[35:24] P_6S
P [23:10] NC
[9:0] P_5U
B_6S [17:12]
x00 [11:5] B
B_5U [4:0]
MULT18X18SIO
UG331_c13_11_081106

Figure 11-8: Two Multipliers in One Primitive

Design Entry
There are many options for including the Spartan-3 generation multiplier in a design. The
library primitive MULT18X18SIO described earlier can be instantiated in the schematic or
HDL code. Synthesis tools can infer a multiplier block from the multiply operator,
including Xilinx XST, Synplicity Synplify, and Mentor Precision. They will infer the register
when the operation is controlled by a clock for a synchronous multiplier.
Mentor synthesis features a pipeline multiplier that involves putting levels of registers in
the logic to introduce parallelism and, as a result, use CLB resources instead of the
dedicated multipliers. A certain construct in the input RTL source code description is
required to allow the pipelined multiplier feature to take effect. See the Synthesis and
Simulation Design Guide for more information.
The following VHDL example will infer the MULT18X18SIO with the PREG output
register:
library ieee;
use ieee.std_logic_1164.all;
use ieee.std_logic_arith.all;
use ieee.std_logic_unsigned.all;
entity mult18x18sio is
port ( a : in std_logic_vector(7 downto 0);
b : in std_logic_vector(7 downto 0);

384 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Entry

clk : in std_logic;
prod : out std_logic_vector(15 downto 0));
end mult18x18sio;
architecture arch_mult18x18sio of
mult18x18sio is
begin
process(clk) is begin
if clk’event and clk = ’1’ then
prod <= a*b;
end if;
end process;
end arch_mult18x18sio;
The following is a Synchronous Multiplier VHDL example coded for Mentor:
library ieee;
use ieee.std_logic_1164.all;
use ieee.std_logic_arith.all;
use ieee.std_logic_unsigned.all;
entity mult18x18sio is
port( clk: in std_logic;
a: in std_logic_vector(7 downto 0);
b: in std_logic_vector(7 downto 0);
prod: out std_logic_vector(15 downto 0));
end mult18x18sio;
architecture arch_mult18x18sio of
mult18x18sio is
signal reg_prod : std_logic_vector(15 downto 0);
begin
process(clk)
begin
if(rising_edge(clk))then
reg_prod <= a * b;
prod <= reg_prod;
end if;
end process;
end arch_mult18x18sio;
The following is a Synchronous Multiplier Verilog example coded for Synplify and XST:
module mult18x18sio(a,b,clk,prod);
input [7:0] a;
input [7:0] b;
input clk;
output [15:0] prod;
reg [15:0] prod;
always @(posedge clk) prod <= a*b;
endmodule
The following is a Synchronous Multiplier Verilog example coded for Mentor:
module mult18x18sio (a,b,clk,prod);
input [7:0] a;
input [7:0] b;
input clk;
output [15:0] prod;
reg [15:0] reg_prod, prod;
always @(posedge clk) begin
reg_prod <= a*b;
prod <= reg_prod;
endmodule

Spartan-3 Generation FPGA User Guide www.xilinx.com 385

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

MULT_STYLE Constraint
The MULT_STYLE constraint controls the implementation of the MULT18X18SIO
primitives. In the Project Navigator, the default is that the Xilinx Synthesis Tool (XST) will
select the best type of implementation. To ensure that the embedded multipliers are used,
set MULT_STYLE = Block or select "Block" for the "Multiplier Style" property in the Project
Navigator. The MULT_STYLE constraint can also be applied globally at the XST command
line. For the MULT18X18SIO, the MULT_STYLE constraint is attached to the component,
not the output bus. See the Constraints Guide for more information.

386 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Using the CORE Generator System

Using the CORE Generator System

Multipliers that make use of the embedded Spartan-3 generation 18-bit x 18-bit two’s-
complement multipliers can be easily generated using the CORE Generator Multiplier
module. This core is available with the CORE Generator system. Features of the Multiplier
Generator include:
• Easy-to-use graphical interface generates instantiation templates for VHDL or Verilog
• Generates parallel multipliers using the dedicated multiplier blocks
Also can use other resources for parallel multipliers or generate sequential/serial-
sequential, and fixed/reloadable constant coefficient multipliers
• Supports two’s-complement signed/unsigned modes
• Supports inputs ranging from 1 to 64 bits wide
• Supports outputs ranging from 1 to 129 bits wide
• Generates purely combinatorial and fully pipelined implementations
• Provides optional registered input or output with optional clock enable and
asynchronous and synchronous clears
• Provides optional handshaking signals
Figure 11-9 shows the logic symbol for the Core Multiplier Generator. The RFD (Ready For
Data) output goes High to indicate the multiplier is ready to accept data. The ND (New
Data) input can be asserted to indicate new data is available on the multiplier inputs. The
RDY (Ready) signal indicates that the output is the current product. LOADB and SWAPB
are used in constant coefficient multipliers.

A O
A_SIGNED Q
B

LOADB LOAD_DONE
SWAPB
RFD
ND RDY

CE
CLK

ACLR
SCLR

X467_06_032403

Figure 11-9: Core Multiplier Generator Symbol

The CORE Generator system uses the embedded multiplier for the default Parallel
multiplier type. The Multiplier Construction XCO parameter option or the c_mult_type
Generic option gives the user the choice to implement the function in look-up tables
instead.

Spartan-3 Generation FPGA User Guide www.xilinx.com 387

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

Figure 11-10 shows the timing diagram for the Multiplier Generator.

CLK

SCLR

RFD RFD active unless ACLR or SCLR active

ND
new multiplier inputs A(n) & B(n)
A&B XXX XXX XXX XXX A0 A1 An An A
n+1
Input
interval depends on multiplier latency

RDY
new multiplier output
DOUT = 0 (SCLR was 1)

DOUT D0 Dn Dn Dn+1

multiplier output still valid

but RDY low (ND was 0) X467_07_040303

Figure 11-10: Multiplier Generator Timing Diagram

System Generator
The Multiplier Generator is used by the System Generator for DSP when the MULT block
is used. System Generator presents a high level and abstract view of the design, but also
exposes key features in the underlying silicon, making it possible to build extremely high-
performance FPGA implementations. The System Generator also provides blocks for
compiling MATLAB M-code into synthesizable HDL code. The System Generator uses the
embedded multiplier when a parallel multiplier is selected.

MAC Cores
The CORE Generator system and the System Generator can also implement more complex
functions using the multiplier as a building block. The Multiply Accumulator (MAC) core
supports up to 32-bit inputs and optional user-defined pipelining. The options of an
Embedded or LUT based implementation control whether the dedicated multipliers or
CLB resources are used for the function. The MAC implementation uses relatively few CLB
resources beyond the dedicated multipliers and provides flexibility that is key to matching
a design to the lowest density and lowest cost solution possible.
The MAC and MAC-based FIR filters include an automatic pipeline control which is based
on required system clock performance. Levels of pipeline will automatically be inserted
based on the design requirement for a perfect speed/area trade-off.

388 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Family Library Primitives

Spartan-3 Family Library Primitives

The original Spartan-3 family uses the MULT18X18 (combinatorial) and MULT18X18S
(registered) library primitives for the embedded multipliers. Table 11-3 defines each port of
the MULT18X18SIO primitive.

Table 11-3: MULT18X18SIO Embedded Multiplier Primitives Description

Signal Name Direction Function
A[17:0] Input The primary 18-bit two’s complement value for multiplication. The block multiplies by this
value asynchronously if the optional AREG and PREG registers are omitted. When AREG
and/or PREG are used, the value provided on this port is qualified by the rising edge of
CLK, subject to the appropriate register controls.
B[17:0] Input The second 18-bit two’s complement value for multiplication if the B_INPUT attribute is set
to DIRECT. The block multiplies by this value asynchronously if the optional BREG and
PREG registers are omitted. When BREG and/or PREG are used, the value provided on this
port is qualified by the rising edge of CLK, subject to the appropriate register controls.
BCIN[17:0] Input The second 18-bit two’s complement value for multiplication if the B_INPUT attribute is set
to CASCADE. The block multiplies by this value asynchronously if the optional BREG and
PREG registers are omitted. When BREG and/or PREG are used, the value provided on this
port is qualified by the rising edge of CLK, subject to the appropriate register controls.
P[35:0] Output The 36-bit two’s complement product resulting from the multiplication of the two input
values applied to the multiplier. If the optional AREG, BREG and PREG registers are
omitted, the output operates asynchronously. Use of PREG causes this output to respond to
the rising edge of CLK with the value qualified by CEP and RSTP. If PREG is omitted, but
AREG and BREG are used, this output responds to the rising edge of CLK with the value
qualified by CEA, RSTA, CEB, and RSTB. If PREG is omitted and only one of AREG or BREG
is used, this output responds to both asynchronous and synchronous events.
BCOUT[17:0] Output The value being applied to the second input of the multiplier. When the optional BREG
register is omitted, this output responds asynchronously in response to changes at the
B[17:0] or BCIN[17:0] ports according to the setting of the B_INPUT attribute. If BREG is
used, this output responds to the rising edge of CLK with the value qualified by CEB and
RSTB.
CEA Input Clock enable qualifier for the optional AREG register. The value provided on the A[17:0]
port is captured by AREG in response to a rising edge of CLK when this signal is High,
provided that RSTA is Low.
RSTA Input Synchronous reset for the optional AREG register. AREG content is forced to the value zero
in response to a rising edge of CLK when this signal is High.
CEB Input Clock enable qualifier for the optional BREG register. The value provided on the B[17:0] or
BCIN[17:0] port is captured by BREG in response to a rising edge of CLK when this signal is
High, provided that RSTB is Low.
RSTB Input Synchronous reset for the optional BREG register. BREG content is forced to the value zero
in response to a rising edge of CLK when this signal is High.
CEP Input Clock enable qualifier for the optional PREG register. The value provided on the output of
the multiplier port is captured by PREG in response to a rising edge of CLK when this signal
is High, provided that RSTP is Low.
RSTP Input Synchronous reset for the optional PREG register. PREG content is forced to the value zero
in response to a rising edge of CLK when this signal is High.

Notes:
1. The control signals CLK, CEA, RSTA, CEB, RSTB, CEP, and RSTP have the option of inverted polarity.

Spartan-3 Generation FPGA User Guide www.xilinx.com 389

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

Binary multiplication is similar to regular multiplication with the multiplicand multiplied

by each bit of the multiplier to generate partial products, and then the partial products
added together to create the result. The Xilinx multiplier block uses the modified Booth
algorithm, in effect using multiplexers to create the partial products.

Data Flow
Each embedded multiplier block (MULT18X18SIO primitive) supports two independent
dynamic data input ports: 18-bit signed or 17-bit unsigned. The two inputs are referred to
as the multiplicand and the multiplier, or the factors, while the output is the product.
Multipliers with inputs less than 18 bits are implemented by sign-extending the inputs
(i.e., replicating the most-significant bit). Wider multiplication operations are performed
by combining the dedicated multipliers and slice-based logic in any viable combination or
by time-sharing a single multiplier.
Unsigned multiplication is performed by restricting the inputs to the positive range. The
most-significant bit is tied Low and the unsigned value is represented in the remaining 17
less-significant bits.

Multipliers in the Spartan-3 Generation Architecture

The multipliers are located adjacent to the block RAM, making it convenient to store inputs
or results in the block memory (see Figure 11-11). There are two columns of multipliers in
most devices. The smallest devices have one column, while the largest devices have four to
five columns (see Table 11-1). Where there are two columns, they have two columns of
CLBs between them and the edge, allowing the multiplier to be easily driven by CLB or
IOB logic. There are four CLBs, or 16 slices and 32 LUTs, on either side of a given multiplier
block, allowing 32 input and output signals to be connected immediately adjacent to the
multiplier block. One possible high-speed layout is to put A[15:0] on one side, B[15:0] on
the other side, and intersperse the P[31:0] outputs on both sides. For a full-size 18x18
multiplier, the extra inputs and outputs can connect to the next CLB column.

390 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Alternative Applications to Multiplication

IOBs

CLB

Block RAM

Multiplier
DCM

IOBs
OBs
DCM
Block RAM / Multiplier

CLBs
IOBs

IOBs

DCM

IOBs
DS312-1_01_032606

Notes:
1. The XC3S700A/AN and XC3S1400A/AN have two additional DCMs on both the left and right sides as indicated by
the dashed lines. The XC3S50A/AN has only two DCMs at the top and only one Block RAM/Multiplier column.

Figure 11-11: Location of Multipliers in Spartan-3 Generation Architecture

Alternative Applications to Multiplication

Since binary multiplication by 2n is the same as shifting the value n places, a multiplier can
be used as a shifter or other general-purpose resource. These can be considered in
applications that otherwise would not need the large number of available multipliers.

Shifter
A multiplier can be used as a shifter. One operand is routed to the output, shifted by n
positions, if the other operand is a power of two (2n). Since the sign-bit (MSB) cannot be
used to control the shift, the 18x18 two’s-complement multiplier can shift by 0 to 16
positions.
Of the 36 output lines, those less significant than the shifted data lines are automatically
filled with zeros; those more significant than the shifted data are filled with zeros or ones,
depending on the state of the MSB input. This is the natural result of the two’s-complement
multiplication.

Spartan-3 Generation FPGA User Guide www.xilinx.com 391

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

The user can either perform a logic shift of 17 input bits by holding the MSB input Low, or
perform an arithmetic shift of an 18-bit two’s-complement number, effectively sign-
extending the MSB.
A conventional CLB-based shifter would use an array of n multiplexers, each with n
inputs, and require a large amount of routing resources. Multiplier-based shifters larger
than 18 bits, and barrel shifters of any length, require external OR gating of the outputs, but
use far fewer CLB resources.

Magnitude Return
To generate the absolute value of a number by using multiplication, multiply by 1 if it is
positive (MSB is zero), and multiply by -1 if it is negative (MSB is one). In two’s-
complement notation, 1 is all zeros ending in a one as the LSB, and -1 is all ones, including
the LSB. Therefore, a magnitude return or absolute value generator can be implemented by
multiplying by a value with a one as the LSB and the MSB of the input value in all the other
bit positions. Figure 11-12 shows a magnitude return generator.

• 17
16
•
• A
•
1
0
16
•
P •
•
0
• 17
• 16
• B
•
•
1
1 0

X467_12_032503

Figure 11-12: Magnitude Return

Two’s-Complement Return
Generating the two’s complement of a number typically requires only one LUT per bit
with the carry logic used for larger numbers. However, if LUTs are heavily used, the
multiplier can be used to return the two’s complement of the input. Multiplying an input
number by an equivalent length number of all ones generates the two’s complement of the
number over the same length of the output bits. Any extraneous higher-order bits are
ignored. Figure 11-13 shows a two’s complement return generator.

392 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Alternative Applications to Multiplication

10

•
• A 35
•
•
0
P •
• NC
11

10
1 10 •
P •
•
• B 0
•
•

1 0

X467_13_032503

Figure 11-13: Two’s-Complement Return

Complex Multiplication
Complex multiplication is multiplication of complex numbers, which contain real and
imaginary components with the imaginary unit i equal to the square root of -1. Complex
multiplication can be carried out using only three real multiplications: ac, bd, and
(a + b)(c + d). The real part of (a + ib)(c + id) is ac - bd, and the imaginary part is
(a + b)(c + d) - ac - bd. The large number of multipliers in the Spartan-3 generation
architecture makes it convenient to do even complex multiplication.

Time Sharing in Matrix Multiplication

Many pipelined functions in the computer graphics and video fields are expressed in
matrix mathematics. A 3 x 3 matrix multiplication would require 27 multiplies and 18 adds
to generate the 3 x 3 matrix result. Color conversion can be described as a 3 x 3 matrix
multiplication by a constant, which requires nine multiplies and six adds to generate the
three results.
The high-speed capability of a Spartan-3 generation device allows the user to "time share"
the multipliers. Instead of nine multipliers, the design feeds nine sets of inputs resulting in
nine sets of results at nine times the clock rate of the system, reducing the multiplier count
to one. The adder logic is implemented in CLB resources, and at every third clock, the
adder output is stored in output registers to capture the three results. See XAPP284 for
more information.

Floating-Point Multiplication
Floating-point values add an exponent to the number and sign bit used in binary
multiplication. A 32-bit floating-point multiplier can be implemented using four of the
dedicated multiplier blocks and CLB resources. Such multipliers are available from Xilinx
AllianceCORE partners.

Spartan-3 Generation FPGA User Guide www.xilinx.com 393

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

Related Materials and References

• Spartan-3 generation Data Sheets
Architectural description and timing parameters.
http://www.xilinx.com/support/documentation/index.htm
• XtremeDSP Technology Solutions
http://www.xilinx.com/products/design_resources/dsp_central/grouping/
Information that will enable you to achieve the maximum benefit from our DSP
solutions.
• IP Center (http://www.xilinx.com/ipcenter)
Xilinx and Alliance partner core solutions.
• Xilinx Software Documentation
(http://www.xilinx.com/support/software_manuals.htm)
Libraries Guide descriptions, Synthesis and Simulation Design Guide instantiation
examples for HDL.
• XAPP195: Implementing Barrel Shifters Using Multipliers
8-bit and 32-bit barrel shifter examples.
• XAPP284: Matrix Math, Graphics, and Video
Uses one multiplier running at 9x the clock rate to provide the nine results for a 3x3
matrix multiplication in one system clock cycle.
• UG431: XtremeDSP Slice for Spartan-3A DSP FPGAs User Guide
Describes the DSP48A block in the Spartan-3A DSP FPGA platform.
• XAPP467: Using Embedded Multipliers in Spartan-3 FPGAs
Details how to use the MULT18X18 and MULT18X18S for the Spartan-3 family.
• XAPP636: Optimal Pipelining of the I/O Ports of Virtex-II Multipliers
Describes a high-speed, optimized implementation of the dedicated multiplier
resulting from pipelined inputs and outputs and effective placement and routing
constraints.
• WP277: Expanding Dedicated Multipliers
This white paper describes methods for expanding the natural bit-width capability of
dedicated multipliers in a way that will make best use of the complete FPGA
resources.

Conclusion
FPGAs have a significant advantage over general-purpose DSP chips because their logic
can be customized for the specific application. Some functions can run over 100 times
faster and require much less expense in an FPGA. A key feature to take advantage of is the
dedicated multiplier block. Take advantage of the automatic optimization of
multiplication logic, and the user controls when necessary to get the exact results desired.
The CORE Generator system can create simple multipliers or combine them into more
complex functions such as MACs.

394 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Appendix A: Two's-Complement Multiplication

Appendix A: Two's-Complement Multiplication

Two’s-complement representation allows the use of binary arithmetic operations on
signed integers, yielding the correct two’s-complement results. Positive two’s-complement
numbers are represented as simple binary. Negative two’s-complement numbers are
represented as the binary number that when added to a positive number of the same
magnitude equals zero. To calculate the two's complement of an integer, invert the binary
equivalent of the number by changing all of the ones to zeros and all of the zeros to ones
(also called one’s complement), and then add one. The MSB (left-most) bit indicates the
sign of the integer; therefore it is sometimes called the sign bit. If the sign bit is zero, the
number is positive. If the sign bit is one, the number is negative. To extend a signed integer
to a larger width, duplicate the MSB on the left side of the number.
Two’s-complement multiplication follows the same rules as binary multiplication, which
are the same as the truths of the AND gate:
0x0=0
0x1=0
1x0=0
1 x 1 = 1, and no carry or borrow bits
For example,
1111 1100 = -4
× 0000 0100 = +4
1111 0000 = -16

Spartan-3 Generation FPGA User Guide www.xilinx.com 395

UG331 (v1.7) August 19, 2010
 R

Chapter 11: Using Embedded Multipliers

396 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 12

Using Interconnect
Interconnect is the programmable network of signal pathways between the inputs and
outputs of functional elements within the FPGA, such as IOBs, CLBs, DCMs, and block
RAM.

Overview
Interconnect, also called routing, is segmented for optimal connectivity. There are four
kinds of interconnect: long lines, hex lines, double lines, and direct lines. The Xilinx ISE®
Place and Route (PAR) software exploits the rich interconnect array to deliver optimal
system performance and the fastest compile times. Knowledge of the interconnect details
can help guide design techniques but is typically not necessary to efficient FPGA design.
Some types of global interconnect are controlled by the design. These include the clock
routing, selected via the use of global clock buffers, and discussed in more detail in
Chapter 2, “Using Global Clock Resources.” Two other global signals, GTS (Global Three-
State) and GSR (Global Set/Reset), are selected via the use of the STARTUP component,
which is described at the end of this chapter.

Interconnect Differences between Spartan-3 Generation Families

Functionally, interconnect resources are almost identical between the Spartan®-3,
Spartan-3E, and Extended Spartan-3A families. Although the Spartan-3E and Extended
Spartan-3A families have the DCM and block RAM/Multiplier resources “embedded” in
the array, the long routing resources extend across those elements. The Spartan-3A/3AN
platforms offer additional routing connections between the block RAM and the
multipliers. This additional routing provides a fast path from the block RAM into the
multiplier, useful for storing multiplicands in adjacent block RAM. The additional routing
also allows Port A of the block RAM to be used in full 36-bit mode even while the adjacent
multiplier is used. This is only supported on Port A for the Spartan-3A/3AN platforms.
See “Multiplier/Block RAM Routing Interaction,” page 378 for more details. The
Spartan-3A DSP platform replaces the multiplier with the DSP48A block. Although the
primary global control signals GSR and GTS are identical in functionality for each family,
there are different primitives per family. See “Global Controls,” page 401 for more
information.

Switch Matrix
The switch matrix connects to the different kinds of interconnects across the device. An
interconnect tile, shown in Figure 12-1, is defined as a single switch matrix connected to a
functional element, such as a CLB, IOB, or DCM. If a functional element spans across
multiple switch matrices such as the block RAM or multipliers, then an interconnect tile is
defined by the number of switch matrices connected to that functional element. A device

Spartan-3 Generation FPGA User Guide www.xilinx.com 397

UG331 (v1.7) August 19, 2010
 R

Chapter 12: Using Interconnect

can be represented as an array of interconnect tiles where interconnect resources are for the
channel between any two adjacent interconnect tile rows or columns as shown in
Figure 12-2.

Switch
Switch
CLB Matrix
Matrix

Switch
Matrix
Switch 18Kb MULT
IOB
Matrix Block 18 x 18
Switch RAM
Matrix
Switch
DCM
Matrix Switch
Matrix

DS312_08_020905

Figure 12-1: Four Types of Interconnect Tiles (CLBs, IOBs, DCMs, and Block
RAM/Multiplier)

Switch Switch Switch Switch Switch

IOB IOB IOB IOB
Matrix Matrix Matrix Matrix Matrix

Switch Switch Switch Switch Switch

IOB CLB CLB CLB
Matrix Matrix Matrix Matrix Matrix

Switch Switch Switch Switch Switch

IOB CLB CLB CLB
Matrix Matrix Matrix Matrix Matrix

Switch Switch Switch Switch Switch

IOB CLB CLB CLB
Matrix Matrix Matrix Matrix Matrix

Switch Switch Switch Switch Switch

IOB CLB CLB CLB
Matrix Matrix Matrix Matrix Matrix

DS312_09_020905

Figure 12-2: Array of Interconnect Tiles in an FPGA

398 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Switch Matrix

The four types of general-purpose interconnect available in each channel, shown in

Figure 12-3, are described below.

Horizontal and 24
Vertical Long Lines
(horizontal channel
shown as an example)
CLB CLB CLB CLB CLB CLB CLB CLB CLB CLB

6 6 6 6 6
DS312-2_10_071906

Horizontal and 8
Vertical Hex Lines
(horizontal channel
shown as an example) CLB CLB CLB CLB CLB CLB CLB

DS312-2_11_020905

Horizontal and 8
Vertical Double Lines
(horizontal channel
shown as an example) CLB CLB CLB

DS312-2_15_022305

Direct Connections
CLB CLB CLB

CLB CLB CLB

CLB CLB CLB

DS312-2_12_020905

Figure 12-3: Interconnect Types between Two Adjacent Interconnect Tiles

Long Lines
Each set of 24 long line signals spans the die both horizontally and vertically and connects
to one out of every six interconnect tiles. At any tile, four of the long lines drive or receive
signals from a switch matrix. Because of their low capacitance, these lines are well-suited
for carrying high-frequency signals with minimal loading effects (e.g. skew). If all global
clock lines are already committed and additional clock signals remain to be assigned, long
lines serve as a good alternative.

Hex Lines
Each set of eight hex lines are connected to one out of every three tiles, both horizontally
and vertically. Thirty-two hex lines are available between any given interconnect tile. Hex
lines are only driven from one end of the route.

Spartan-3 Generation FPGA User Guide www.xilinx.com 399

UG331 (v1.7) August 19, 2010
 R

Chapter 12: Using Interconnect

Double Lines
Each set of eight double lines are connected to every other tile, both horizontally and
vertically. in all four directions. Thirty-two double lines available between any given
interconnect tile. Double lines are more connections and more flexibility, compared to long
line and hex lines.

Direct Connections
Direct connect lines route signals to neighboring tiles: vertically, horizontally, and
diagonally. These lines most often drive a signal from a "source" tile to a double, hex, or
long line and conversely from the longer interconnect back to a direct line accessing a
"destination" tile.

Viewing Interconnect Details with FPGA Editor

The FPGA Editor can be used to view the interconnect of a blank device or to view the
interconnect used in an implemented design. FPGA Editor is a graphical application for
displaying and configuring FPGAs. The FPGA Editor requires a Native Circuit Description
(.ncd) file. This file contains the logic of your design mapped to components (such as
CLBs and IOBs). In addition, the FPGA Editor reads from and writes to a Physical
Constraints File (PCF).
The following list summarizes some functions you can perform on your designs in the
FPGA Editor:
• Place and route critical components before running the automatic place and route
tools.
• Finish placement and routing if the routing program does not completely route your
design.
• Add probes to your design to examine the signal states of the targeted device. Probes
are used to route the value of internal nets to an IOB for analysis during the
debugging of a device.
• Cross-probe your design with Timing Analyzer.
• Run the BitGen program and download the resulting BIT file to the targeted device.
• Create an entire design by hand (advanced users).
To access the FPGA Editor, first run place and route on your design. Then double-click on
the View/Edit Routed Design (FPGA Editor) process to open FPGA Editor.
For details on using FPGA Editor, see the on-line help within the FPGA Editor application.

400 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Global Controls

Global Controls
In addition to the general-purpose interconnect, Spartan-3 generation FPGAs have two
global logic control signals, as described in Table 12-1.

Table 12-1: Global Logic Control Signals

Global Control
Description
Input
Global Set/Reset: When High, asynchronously places all registers and flip-
flops in their initial state (see “Initialization,” page 213). Asserted
GSR
automatically during the FPGA configuration process (see “Start-Up” in
UG332, the Spartan-3 Generation Configuration User Guide).
Global Three-State: When High, asynchronously forces all I/O pins to a
GTS
high-impedance state (Hi-Z, three-state).

The Global Set/Reset (GSR) signal replaces the global reset signal included in many ASIC-
style designs. Use the GSR control instead of a separate global reset signal in the design to
free up CLB inputs, resulting in a smaller, more efficient design. However, the GSR signal
always re-initializes every flip-flop. The GSR signal is asserted automatically during the
FPGA configuration process, guaranteeing that the FPGA starts-up in a known state.

STARTUP_SPARTAN3 Primitives
The GSR and GTS signal sources are defined and connected using a special primitive for
each family: STARTUP_SPARTAN3, STARTUP_SPARTAN3E, or STARTUP_SPARTAN3A
(used for Spartan-3A, Spartan-3AN, and Spartan-3A DSP FPGAs). GSR and GTS are active
during configuration, and connecting signals to them on the STARTUP primitive defines
how they are controlled after configuration. By default, they are disabled on a selected
clock cycle of the start-up phase, enabling the flip-flops and I/Os in the device. The
primitives also include one or two other signals used specifically during configuration.
Each family has a CLK input that is an alternate clock for the start-up process (see the
“Sequence of Events” chapter in UG332, Spartan-3 Generation Configuration User Guide). The
Spartan-3E family has an additional input MBT for the MultiBoot Trigger (see the
“Reconfiguration and MultiBoot” chapter in UG332, Spartan-3 Generation Configuration User
Guide).

STARTUP_SPARTAN3 STARTUP_SPARTAN3E STARTUP_SPARTAN3A

GSR GSR GSR

GTS GTS GTS
MBT

CLK CLK CLK

UG331_c14_04_071906

Figure 12-4: STARTUP Primitives for Spartan-3 Generation FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 401

UG331 (v1.7) August 19, 2010
 R

Chapter 12: Using Interconnect

Summary
The flexible interconnect resources of the Spartan-3 generation FPGA families allow
efficient implementation of almost any configuration of the logic and I/O resources. The
Xilinx ISE software automatically places and routes designs to take best advantage of these
resources. Customers can control the usage of the global clock signals by the use of global
clock buffers. The global set/reset and global three-state signals are controlled by the use
of the STARTUP component.

402 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Section II: Design Software

“Using ISE Design Tools”

“Using IP Cores”

“Embedded Processing and Control Solutions”

Spartan-3 Generation FPGA User Guide www.xilinx.com 403

UG331 (v1.7) August 19, 2010
 R

Chapter :

404 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 13

Using ISE Design Tools

Summary
Software is critical to the effective use of programmable logic. The Spartan®-3 generation
is supported by the complete set of Xilinx Integrated Software Environment (ISE®) design
tools, with additional support available from a variety of partners. This chapter provides
an overview of those design tools. It is intended primarily for the user who is new to the
Xilinx development system. This chapter can be used to get a better understanding of the
specific tools mentioned elsewhere in the Spartan-3 generation literature. The first half
provides an overview of the general design flow, while the second half describes the
specific tools used at the different steps in the flow. Use the Xilinx development system
documentation for detailed information and introductory tutorials:
http://www.xilinx.com/support/documentation/dt_ise.htm
This chapter applies to all Spartan-3 generation FPGA families: Spartan-3, Spartan-3E,
Spartan-3A, Spartan-3AN, and Spartan-3A DSP FPGA platforms.

Introduction
Combined with the Spartan-3 generation FPGA family, the ISE optimized design tools
help you finish faster and lower your project costs. The ISE package is a collection of Xilinx
software design tools that concentrate on delivering the most productivity available for
your Spartan-3 generation logic performance. With ProActive Timing Closure technology,
you get the fastest runtimes in programmable logic ensuring you reach your performance
goals quicker. Incremental Design delivers faster re-compile times with guaranteed
performance, and the optional Xilinx ChipScope™ Pro verification tools provide real-time
debug with advantages that are not possible in ASIC designs. The ISE development system
makes sure you get through the logic design process faster, saving both time and project
costs, and getting you to market ahead of your competition.

Design Flow
The standard design flow for Spartan-3 generation FPGAs consists of the following three
major steps. The entire design implementation flow is run simply by selecting the desired
result in the Xilinx Graphical User Interface (GUI). The tools automatically determine
which programs and files are needed to bring the appropriate output up to date.
1. Design Entry and Synthesis
In this step of the design flow, you create your design using a Xilinx-supported
schematic editor, a Hardware Description Language (HDL) for text-based entry, or
both. If you use an HDL for text-based entry, you must synthesize the HDL file into an
industry-standard Electronic Data Interchange Format (EDIF) file. If you use the Xilinx

Spartan-3 Generation FPGA User Guide www.xilinx.com 405

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

Synthesis Technology (XST) tool, a Xilinx-specific NGC netlist file is created, which can
be converted to an EDIF file.
2. Design Implementation
By implementing the specific Xilinx Spartan-3 generation architecture, you convert the
logical design file format, such as EDIF, that you created in the design entry or
synthesis stage into a physical file format. The physical information is contained in the
Native Circuit Description (NCD) file. Then you create a bitstream file from these files
and optionally program a PROM for subsequent programming of your
Spartan-3 generation device.
3. Design Verification
Using a gate-level simulator, you ensure that your design meets your timing
requirements and functions properly. In-circuit verification can be performed by
downloading your design to the device using Xilinx iMPACT Programming Software.
Design verification can begin immediately after design entry and can be repeated after
various steps of design implementation.
Figure 13-1 shows the general overall design flow for Spartan-3 generation FPGAs.

406 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Flow

Design
Design Entry Verification

Functional
Simulation

Design
Synthesis

Design Static Timing

Implementation Analysis

Optimization

FPGAs
- Mapping
- Placement
- Routing
Back Timing
Annotation Simulation
Bitstream
Generation

Download to a In-Circuit
Xilinx Device Verification

x473_01_062103

Figure 13-1: Design Flow

Design Entry and Synthesis

You can enter a design with a schematic editor or a text-based tool for HDL code. Design
entry begins with a design concept, expressed as a drawing or functional description. From
the original design, a generic EDIF netlist is created, then synthesized and translated into a
Xilinx netlist file. This file is fed into a program called NGDBuild, which produces a logical
Native Generic Database (NGD) file. Xilinx libraries provide access to features specific to
the Spartan-3 generation architecture.
Figure 13-2 shows the design entry and synthesis flow.

Spartan-3 Generation FPGA User Guide www.xilinx.com 407

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

CORE Generator System

Schematic Synthesis
Libraries Libraries
HDL

Schematic Capture Synthesis

EDIF and NGC

UCF Constraints/NCF (XST Netlist)

NGDBuild

x473_02_061703

Figure 13-2: Design Entry and Synthesis Flow

Hierarchical Design
Design hierarchy is important in both schematic and HDL entry for the following reasons:
• Helps you conceptualize your design
• Adds structure to your design
• Promotes easier design debugging
• Makes it easier to combine different design entry methods (schematic, HDL, or state
editor) for different parts of your design
• Makes it easier to design incrementally, which consists of designing, implementing,
and verifying individual parts of a design in stages
• Reduces optimization time
• Facilitates concurrent design, which is the process of dividing a design among a
number of people who develop different parts of the design in parallel, such as in
Modular Design
Xilinx strongly recommends that you name the components and nets in your design. These
names are preserved and used by the Xilinx tools. These names are also used for back-
annotation and appear in the debug and analysis tools. If you do not name your
components and nets, the tools automatically generate the names, making it difficult to
analyze circuits.

408 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Flow

Schematic Entry
Schematic tools provide a graphical interface for design entry. You can use these tools to
connect symbols representing the logic components in your design. You can build your
design with individual gates, or you can combine gates to create functional blocks.
Primitives and macros are the “building blocks” of a device library. The Xilinx
Spartan-3 generation libraries provide primitives as well as common high-level macro
functions, all optimized for the Spartan-3 generation architecture. Primitives are basic
circuit elements, such as AND and OR gates, and special device resources, such as the
DCM and block RAM. Each primitive has a unique library name, symbol, and description.
Macros contain multiple library elements, which can include primitives and other macros.
Soft macros have pre-defined functionalities, but have flexible mapping, placement, and
routing. Relationally Placed Macros (RPMs) have fixed mapping and relative placement.
Macros are not available for synthesis because synthesis tools have their own module
generators and do not require RPMs. If you wish to override the module generation, you
can instantiate Xilinx-provided CORE Generator™ modules, which include pre-built
optimization for the Spartan-3 generation architecture. For most leading-edge synthesis
tools, this is not needed unless it is for a module that cannot be inferred.

HDL Entry and Synthesis

A typical Hardware Description Language (HDL) supports a mixed-level description in
which gate and netlist constructs are used with functional descriptions. This mixed-level
capability enables you to describe system architectures at a high level of abstraction, then
incrementally refine a design’s detailed gate-level implementation. HDL descriptions offer
the following advantages:
• You can verify design functionality early in the design process. A design written as an
HDL description can be simulated immediately. Design simulation at this high level
— at the gate-level before implementation — allows you to evaluate architectural and
design decisions.
• An HDL description is more easily read and understood than a netlist or schematic
description. HDL descriptions provide technology-independent documentation of a
design and its functionality. Because the initial HDL design description is technology
independent, you can use it again to generate the design in a different technology,
without having to translate it from the original technology.
• Large designs are easier to handle with HDL tools than schematic tools.
After creating your HDL design, you must synthesize it. During synthesis, behavioral
information in the HDL file is translated into a structural netlist, and the design is
optimized for the Spartan-3 generation architecture. Xilinx supports HDL synthesis tools
for several third-party synthesis vendor partners. In addition, Xilinx offers its own
synthesis tool, Xilinx Synthesis Technology (XST).
Functional simulation tests the logic in your design to determine if it works properly. You
can save time during subsequent design steps if you perform functional simulation early
in the design flow.
Although HDL entry offers the advantage of technology independence, it is helpful to
understand the available resources in the Spartan-3 generation architecture and design to
take advantage of those resources. For example, the abundance of registers at every I/O
and following every look-up table encourages pipelining. Most synthesis tools
automatically infer Xilinx-specific resources and optimize for the architecture. Simple

Spartan-3 Generation FPGA User Guide www.xilinx.com 409

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

ways to specify implementation requirements are to instantiate Spartan-3 generation

library components or add constraints.

Constraints
You might want to constrain your design within certain timing or placement parameters to
specify your required pin locations or timing requirements. You can specify logic mapping,
block placement, and timing specifications. Constraints can be entered as parameters or
attributes on library components. You can enter constraints by hand or use one of several
graphical tools for generating constraint files and evaluating the results. Constraints found
in the design are written to an NCF file (Netlist Constraints File). Constraints created
separately are written to a UCF file (User Constraints File).

Design Implementation
Design Implementation begins with the translating and then mapping of a logical design
file to a specific Spartan-3 generation device. It is complete when the physical design is
successfully routed and a bitstream is generated. You can alter constraints during
implementation in the same way as during the Design Entry step.
Figure 13-3 shows an overall view of the design implementation flow for
Spartan-3 generation FPGAs.

410 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Flow

UCF NGDBuild

Constraints Editor NGD

Floorplanner
MAP

FPGA Editor NCD & PCF

TRACE &
Timing Analyzer PAR

NCD

BitGen

BIT

iMPACT
X473_03_082206

Figure 13-3: Design Implementation Flow

Translating
NGDBuild performs all the steps necessary to read a netlist file in EDIF or NGC format and
create an NGD file describing the logical design. A logical design is in terms of logic
elements, such as AND gates, OR gates, decoders, flip-flops, and RAMs. The NGD file
resulting from an NGDBuild run contains both a logical description of the design reduced
to Xilinx primitives and a description in terms of the original hierarchy expressed in the
input netlist. The output NGD file then can be mapped to the Spartan-3 generation
resources.
NGDBuild performs the following steps to convert a netlist to an NGD file:
1. Reads the source netlist(s). NGDBuild invokes the Netlister Launcher. The Netlist
Launcher determines the type of the input netlist and starts the appropriate netlist
reader program. The netlist readers incorporate NCF files associated with each netlist.
NCF files contain timing and layout constraints for each module.

Spartan-3 Generation FPGA User Guide www.xilinx.com 411

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

2. Reduces all components in the design to NGD primitives. NGDBuild merges

components that reference other files. NGDBuild also finds the appropriate system
library components, physical macros, and behavioral models.
3. Checks the design by running a Logical Design Rule Check (DRC) on the converted
design. The Logical DRC is a series of tests on the logical design.
4. Writes an NGD file as output.

Mapping
The MAP program maps a logical design to a Spartan-3 generation FPGA. The input to
MAP is an NGD file, which contains a logical description of the design in terms of both the
hierarchical components used to develop the design and the lower-level Xilinx primitives.
Additionally, it contains any number of hard placed-and-routed physical macro files. MAP
then maps the logic to the components (logic cells, I/O cells, and other components) in the
Spartan-3 generation architecture. The output design is a Native Circuit Description
(NCD) file, which is a physical representation of the design mapped to the components in
the Spartan-3 generation architecture. The NCD file then can be placed and routed.
MAP performs the following steps when mapping a design:
1. Selects the target Xilinx device, package, and speed.
2. Reads the information in the input design file.
3. Performs a Logical DRC (Design Rule Check) on the input design. If any DRC errors
are detected, the MAP run is aborted. If any DRC warnings are detected, the warnings
are reported, but MAP continues to run.
4. Removes unused logic, where all unused components and nets are removed.
5. Maps pads and their associated logic into IOBs.
6. Maps the logic into Xilinx components (IOBs, CLBs, etc.). If any Xilinx mapping
control symbols appear in the design hierarchy of the input file, MAP uses the existing
mapping of these components in preference to re-mapping them. The mapping is
influenced by various constraints.
7. Updates the information received from the input NGD file and writes this updated
information into an NGM file. This NGM file contains both logical information about
the design and physical information about how the design was mapped. The NGM file
is used only for back-annotation.
8. Creates a physical constraints (PCF) file. This text file contains any constraints
specified during design entry. If no constraints were specified during design entry, an
empty file is created so that you can enter constraints directly into the file using a text
editor.
9. Runs a physical Design Rule Check (DRC) on the mapped design. If DRC errors are
found, MAP does not write an NCD file.
10. Creates an NCD file, which represents the physical design. The NCD file describes the
design in terms of Xilinx components (CLBs, IOBs, and so forth).
11. Writes a MAP report (MRP) file, which lists any errors or warnings found in the
design, details how the design was mapped, and supplies statistics about component
usage in the mapped design.

412 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Flow

Placing and Routing

After creating a mapped NCD file, you can place and route the file using the automatic
Place And Route (PAR) tool. PAR accepts an NCD file as input, places and routes the
design, and outputs an NCD file to be used by the bitstream generator (BitGen). You can
use the output NCD file as a guide file for additional runs of PAR after making minor
changes to your design.
PAR places and routes a design based on the following considerations:
• Cost-Based: Placement and routing are performed using various cost tables that
assign weighted values to relevant factors such as constraints, length of connection,
and available routing resources.
• Timing-Driven: The Xilinx timing analysis software enables PAR to place and route a
design based upon your timing constraints.

Placing
The PAR placer executes multiple phases of the placer. PAR writes the NCD after all the
phases are completed. During placement, PAR places components into sites based on
factors such as constraints specified in the PCF file, the length of connections, and the
available routing resources. Timing-driven placement is automatically invoked if PAR
finds timing constraints in the physical constraints file.

Routing
The next stage is routing the placed design. PAR writes the NCD file when the design is
fully routed. At this point the design can be analyzed against timing. A new NCD is
written as the routing improves. The router performs a procedure to converge on a
solution that routes the design to completion and meets timing constraints. Timing-driven
routing is automatically invoked if PAR finds timing constraints in the physical constraints
file.

Floorplanning
Floorplanning is the process of specifying user placement constraints. The PlanAhead tool
provides a graphical view of placement, while the FPGA Editor provides a graphical view
of both placement and routing. Both tools can be used before or after PAR to analyze or
constrain the design.

Bitstream Generation
After the design has been completely routed, it is necessary to configure the device so that
it can execute the desired function. This configuration is done using files generated by
BitGen, the Xilinx bitstream generation program. BitGen takes a fully routed NCD file as
its input and produces a configuration bitstream (binary BIT file).
The BIT file contains all of the configuration information from the NCD file defining the
internal logic and interconnections of the Spartan-3 generation FPGA, plus device-specific
information from other files associated with the target device. The binary data in the BIT
file then can be downloaded into the FPGA memory cells or it can be used to create a
PROM file.

Design Verification
Design verification is the process of testing the functionality and performance of your
design. You can verify Xilinx designs in the following ways:

Spartan-3 Generation FPGA User Guide www.xilinx.com 413

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

• Simulation (functional and timing using back-annotation)

• Static timing analysis
• In-circuit verification
Design verification procedures should occur throughout your design process, as shown in
Figure 13-4.

Simulation

Input Stimulus

Basic Design Flow

Integrated Tool
Simulation Design Entry
Functional Simulator Paths

Translate to
Simulation Netlist NGD
Simulator Format

Translate to Mapping, Placement,

Simulator Format and Routing

Static Timing
Timing Simulation Path

Translation NCD Static Timing Analysis

BitGen

In-Circuit Verification
Back-Annotation

BIT iMPACT In-Circuit

Verification

NGA

Xilinx FPGA
x473_04_062103

Figure 13-4: Design Verification Flow

Simulation
Design simulation involves testing your design using software models. It is most effective
when testing the functionality of your design and its performance under worst-case
conditions. You can easily probe internal nodes to check your circuit’s behavior, and then
use these results to make changes in your design. Simulation is performed using Xilinx or
third-party tools that are linked to the Xilinx Development System. The software models
provided for your simulation tools are designed to perform detailed characterization of
your design. You can perform functional or timing simulation.

414 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Design Flow

Functional Simulation
Functional simulation determines if the logic in your design is correct before you
implement it in a device. Functional simulation can take place at the earliest stages of the
design flow. Because timing information for the implemented design is not available at this
stage, the simulator tests the logic in the design using unit delays.

Timing Simulation
Timing simulation verifies that your design runs at the desired speed for your device
under worst-case conditions. This process is performed after your design is mapped,
placed, and routed. At this time, all design delays are known. Timing simulation is
valuable because it can verify timing relationships and determine the critical paths for the
design under worst-case conditions. It also can determine whether or not the design
contains setup or hold violations. Before you can simulate your design, you must go
through the back-annotation process, as described below. During this process, the Xilinx
netlist writers create suitable formats for various simulators.
Note that naming the nets during your design entry is important for both functional and
timing simulation because it allows you to find the nets in the simulations more easily than
looking for a software-generated name.

Back-Annotation
Before timing simulation can occur, the physical design information must be translated
and distributed back to the logical design. This back-annotation process is done with a
program called NGDAnno. These programs create a database for the netlist writers, which
translate the back-annotated information into a netlist format that can be used for timing
simulation.
NGDAnno is a command line program that distributes information about delays, setup
and hold times, clock to out, and pulse widths found in the physical NCD design file back
to the logical NGD file. NGDAnno reads an NCD file as input. The NCD file can be a
mapped-only design, or a partial or fully placed and routed design. An NGM file, created
by MAP, is an optional source of input. NGDAnno merges mapping information from the
NGM file with placement, routing, and timing information from the NCD file. NGDAnno
outputs a Native Generic Annotated (NGA) file, which is a back-annotated NGD file. This
file is input to the appropriate netlist writer, which converts the binary Xilinx database
format back to an ASCII netlist.
Netlist Writers (NGD2EDIF, NGD2VER, or NGD2VHDL) take the output of NGDAnno
and create a simulation netlist in the specified format. An NGD or NGA file is input to each
of the netlist writers. The NGD file is a logical design file containing primitive components,
while the NGA file is a back-annotated logical design file.

Static Timing Analysis

Static timing analysis is best for quick timing checks of a design after it is placed and
routed. It also allows you to determine path delays in your design. Following are the two
major goals of static timing analysis:
• Timing verification is the process of verifying that the design meets your timing
constraints.
• Reporting is the process of enumerating input constraint violations and placing them
into an accessible file. You can analyze partially or completely placed and routed
designs. The timing information depends on the placement and routing of the input
design.

Spartan-3 Generation FPGA User Guide www.xilinx.com 415

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

You can run static timing analysis using the Timing Reporter And Circuit Evaluator
(TRACE) program, which is accessible through the Timing Analyzer GUI. Use either tool
to evaluate how well the place and route tools met the input timing constraints.

In-Circuit Verification
As a final test, you can verify how your design performs in the target application. In-circuit
verification tests the circuit under typical operating conditions. Because you can program
your Xilinx devices repeatedly, you can easily load different iterations of your design into
your device and test it in-circuit. To verify your design in-circuit, download your design
bitstream into a device using the iMPACT programming software with the Parallel Cable
IV or Platform Cable USB.

ISE Development Environment

Introduction to ISE Tools

Xilinx development systems are available in a number of easy to use configurations,
collectively known as the Integrated Software Environment (ISE) Series. Creating
Spartan-3 generation designs is easy with Xilinx ISE development systems, which support
advanced design capabilities, including ProActive Timing Closure, integrated logic
analysis, and the fastest place and route runtimes in the industry. ISE solutions enable
designers to get the performance they need, quickly and easily.
Note: To get the full details on ISE tools for Spartan-3 generation devices, go to
http://www.xilinx.com/ise/ise_promo/ise_spartan3.htm.
Project Navigator is the user interface that helps you manage the entire design process
including design entry, simulation, synthesis, implementation and finally configuration of
your device.
The following is an outline of the features offered in the ISE tools:

Design Entry
• HDL Editor
• Schematic Editor - Engineering Capture System (ECS)
• CORE Generator system

Synthesis
• XST - Xilinx Synthesis Technology
• Integration with Precision synthesis from Mentor Graphics
• Integration with Synplify/Pro and Amplify synthesis from Synplicity

Simulation
• ISE Simulator
• Integration with ModelSim Simulator from Model Technology

Implementation
• Translate

416 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

ISE Development Environment

• Map
• Place and Route (PAR)
• Floorplanner
• FPGA Editor
• Timing Analyzer
• XPower Power Analysis

Device Download
• BitGen Bitstream Generator
• iMPACT Configuration Tool
• ChipScope Pro Logic Analyzer

ISE Versions
The ISE development systems are available in the following configurations.
• ISE WebPACK™ Tool
The ISE WebPACK tool is the easiest development system to get. This free tool is
downloadable from the Xilinx website at: (http://www.xilinx.com/webpack).
ISE WebPACK software combines support for advanced HDL entry, synthesis, and
verification capabilities for all Xilinx CPLDs and lower-density FPGAs. All
Spartan-3 generation FPGA families are supported. The original Spartan-3 family is
supported up to the XC3S1500 density, while all densities are supported for the
Spartan-3E and Spartan-3A/3AN families.
• ISE Foundation™ Tool
The ISE Foundation tool is a complete, ready-to-use design environment that
integrates schematic, synthesis, and verification technologies into an intuitive, yet
highly advanced design solution. The tool has full device support as well as the full
suite of tools. See more information at
http://www.xilinx.com/ise/logic_design_prod/foundation.htm
To see a table comparison of these versions, see the Development Systems Overview at
http://www.xilinx.com/ise/devsys_feature_guide.pdf.
Development system updates are provided on a regular basis. These are available as
Service Packs that can be downloaded from the Xilinx website
(http://www.xilinx.com/support/download/index.htm). Always use the latest
development system update for the best results.

Project Navigator
Project Navigator is the primary user interface for the Xilinx ISE tools. You can create,
define, and compile your Spartan-3 generation design using a suite of tools accessible from
Project Navigator. Each step of the design process, from design entry to downloading the
design to the device, is managed from Project Navigator as part of a project. These include:
• Design Entry
• Constraint Entry
• Synthesis
• Simulation

Spartan-3 Generation FPGA User Guide www.xilinx.com 417

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

• Implementation
• Device Programming

Project Navigator Main Window

The Project Navigator workspace is made up of a title bar, a status bar, a menu bar,
toolbars, and windows.

X473_05_082206

Figure 13-5: Project Navigator Main Window

Project
The ISE development system organizes and tracks your design as a project. A project is a
collection of all files necessary to create and download your design to the selected device.
The following information is required for each project:
• A unique project name
• A specified target device family (architecture)
• A specified target device
• A specified design flow
Each project has a directory, device family, device, and design flow associated with it as
project properties. The project properties enable Project Navigator to display and run only
those processes appropriate for the targeted device and design flow.

418 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

ISE Development Environment

Sources
A source is any element that contains information about a design. In Project Navigator, you
can create and add sources to your project. Each project can contain many sources, each
one representing a different part of the overall design. Sources can include the description
of circuits (as represented by schematics and hardware description language files), state
diagrams, simulation models, test files, and documentation of the design.

Source Hierarchy
One source file in a project is the top-level source for the design. The top-level source
defines the inputs and outputs to be mapped into the device, and references the logic
descriptions contained in lower-level sources in a hierarchical design. A project must
contain at least one source as the top-level source. All source files and their accompanying
icons are displayed in the Sources in Project window below the project file.
The term instantiation describes when one source references another. Lower-level sources
also can instantiate sources to build as many levels of logic hierarchy as necessary to
describe your design.
Valid top-level source types include the following:
• Schematics
• HDL files (VHDL or Verilog)
• EDIF
For more information on the Project Navigator, see
http://www.xilinx.com/products/design_tools/logic_design/design_entry/projnav.htm.

ISE Tools
The ISE development system includes a number of individual tools and capabilities that
can be accessed standalone or within the Project Navigator.

Engineering Capture System (ECS)

The Engineering Capture System (ECS) allows you to create, view, and edit schematics and
symbols. You can use ECS to create a top-level schematic and use any of the following to
define the lower levels of the design: ECS, CORE Generator System, or HDL code. Then
you can translate the schematics created by ECS to a structural HDL for simulation and
synthesis, or use the schematics solely for documentation purposes.

HDL Editor
The HDL Editor is a text editor designed especially for editing HDL source files. In
addition to regular editing features, the editor provides syntax coloring. The syntax-
coloring feature supports both VHDL and Verilog. The HDL Editor operates as a standard
text editor as well. The ISE HDL Editor provides optimized, ready-to-use language and
synthesis templates for easy insertion into an HDL source file.

Xilinx Synthesis Technology (XST)

Xilinx Synthesis Technology (XST) provides cutting edge design optimization techniques
from a Xilinx-developed synthesis tool. XST supports the Verilog and VHDL design
languages. RTL Viewer displays the results of XST synthesis in a schematic view.

Spartan-3 Generation FPGA User Guide www.xilinx.com 419

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

For more information on XST, see

http://www.xilinx.com/products/design_tools/logic_design/synthesis/xst.htm.

HDL Advisor
The HDL Advisor gives advisory messages in the XST synthesis report files. The messages
are designed to make suggestions on how code can be changed to reduce design size and
meet timing requirements. These HDL advisors allow designers to produce better code
earlier, reducing design time, and resulting in better space utilization in the
Spartan-3 generation FPGA.

Partner Tools
The Xilinx tools provide easy integration with third-party tools, including Precision
synthesis from Mentor Graphics and Synplify/Pro and Amplify synthesis from Synplicity.
These tools can be purchased separately from the vendor.
ModelSim simulators from Model Technology can provide the simulation functions for an
ISE development system. ModelSim Xilinx Edition III (MXE-III) is available as an option
from Xilinx. It offers a complete PC HDL simulation environment that enables you to
verify the HDL source code as well as the functional and timing models of your designs.

Intellectual Property (IP)

Get to market faster and less expensively using the latest pre-verified, pre-optimized
Intellectual Property (IP) cores, reference designs, and design services for Xilinx FPGAs.
Xilinx-created LogiCORE™ products form the most successful core program in the
programmable logic industry, including PCI bus interfaces and MicroBlaze™ soft
processors. As a result, Xilinx has gained considerable experience developing and selling
cores, and servicing FPGA core customers. Through the AllianceCORE program, Xilinx is
expanding the availability of quality cores for programmable logic by sharing what has
been learned with leading third-party core developers. The AllianceCORE program is a
cooperative effort between Xilinx and independent third-party core developers. It is
designed to produce a broad selection of industry-standard solutions dedicated for use in
Xilinx programmable logic. Xilinx also provides many reference designs and design
examples provided “as-is” to help get you started with your own designs.

CORE Generator System

The Xilinx CORE Generator System provides a catalog of ready-made functions, ranging
in complexity from simple arithmetic operators like adders, accumulators, and multipliers,
to system-level building blocks such as filters, transforms, and memory resources. Cores
are organized by functional type into folders that expand or contract on demand.
The Xilinx CORE Generator System produces an EDIF netlist, schematic symbol, Verilog
template file with a Verilog wrapper file, and a VHDL template file with a VHDL wrapper
file. The Electronic Data Netlist (EDN) file contains the information for implementing the
module. The template files contain code that can be used as a model to instantiate a CORE
Generator module in a Verilog or VHDL design so that it can be simulated and integrated
into a design.
For more information on the CORE Generator system, see
http://www.xilinx.com/products/design_tools/logic_design/design_entry/coregenerator.htm.

420 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

ISE Development Environment

System Generator for DSP

The System Generator for DSP software enables electronic designs to be created, tested,
and translated into hardware for Spartan-3 generation FPGAs. The tool extends Simulink
(from The MathWorks, Inc.) to support bit- and cycle-accurate system-level simulation,
and automatic code generation for Xilinx FPGAs. System Generator co-simulation
interfaces extend Simulink to incorporate FPGA hardware and HDL simulation into the
system-level environment as naturally as other library blocks. System Generator presents a
high-level and abstract view of the design, but also exposes key features in the underlying
silicon, making it possible to build extremely high-performance FPGA implementations.
For more information on System Generator for DSP, see
http://www.xilinx.com/ise/optional_prod/system_generator.htm.

Embedded Development Kit and Platform Studio

The Embedded Development Kit (EDK) bundle is an integrated software solution for
designing embedded processing systems. This preconfigured kit includes the award-
winning Platform Studio tool suite as well as all the documentation and IP required for
designing Xilinx Platform FPGAs with embedded MicroBlaze soft processor cores. In
addition to the flexible MicroBlaze 32-bit soft processors, Xilinx processing solutions
include small footprint PicoBlaze™ 8-bit soft processors, along with a broad range of IP
and processing peripherals with robust third-party ecosystem support.
For more information on the Embedded Development Kit and Platform Studio, see
http://www.xilinx.com/edk.

Clocking Wizard
To reduce the complexities of new device technologies like Digital Clock Managers (DCM),
ISE tools include Architecture Wizards, allowing users access through an intuitive easy-to-
use dialog. Through the use of the ISE Architecture Wizards, designers can access these
leading edge technologies quickly by creating the component through a push-button flow
rather than learning all the attributes in HDL. Then the component simply can be
instantiated in the user’s design by copying the instantiation template created by the
Architecture Wizard. The Clocking Wizard supports all the capabilities of the
Spartan-3 generation DCMs.

Data2MEM Tool
Data2MEM is fundamentally a data translation tool. It translates contiguous fragments of
data into the proper initialization records for Block RAMs. It automates distribution of that
data across multiple physical Block RAMs that constitute a contiguous logical data space.
Data2MEM is also a simplified means for initializing block RAMs.

Automatic Implementation Tools

The automatic implementation tools (synthesis, translation, mapping, placement, and
routing) provide the best results for any design. ProActive Timing Closure technologies
deliver the industry's highest performance in programmable logic designs, quickly and
efficiently. The technologies include:
• Physical Synthesis
• Includes place and route information to work on the real critical paths first
• Achieves better quality of results of 5 to 20%

Spartan-3 Generation FPGA User Guide www.xilinx.com 421

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

• Supported through Synplicity's Amplify, Mentor Graphic's Time Closer, and

Xilinx's own XST synthesis tool
• Timing optimization prior to physical place and route
• Macro Builder
• Lets you freeze placement information for a given design
• You can then re-use that macro in future designs using relative placement
• Performance preservation
• Advanced Place and Route Algorithms
• Critical Path Placement first
• Xplorer script automates multiple implementation runs
• Directed Routing that lets the designer specify routing with IP
• Timing Improvement Wizard
• Interactively helps designer improve design
• Click on a timing problem and receive suggestions that can improve design
timing
• Timing Cross-Probing
• Decreases debug time by cross-probing from the timing report directly to
Floorplanner
• Click on the error, path, or net in the timing report and instantly see it in
Floorplanner or Synthesis Source Tool
• HDL Advisors
• Included in XST synthesis reports, clicking on an error or warning suggests
changes to HDL to improve the implementation

Incremental Design
Incremental Design gets your overall design to market faster by minimizing the impact
from late-arriving design changes. The Incremental Design flow facilitates more debug
cycles in a day when making small design changes. A designer quickly and easily can
floorplan design areas along hierarchy boundaries, and then finish the design as normal.
Later, if a design change is required, Incremental Design ensures that only the area of the
design change need be re-implemented; the rest of the design stays locked and intact,
delivering overall design completion faster.
For more information on Incremental Design, see
http://www.xilinx.com/products/design_tools/logic_design/advanced/incrementaldesign.htm.

Modular Design
Modular Design lets you implement a “divide and conquer” approach to multi-million
gate FPGA designs. Partitioning a design into smaller functional modules reduces the
complexities of design, implementation, and verification. These design modules then can
be brought through the design flow independently, leveraging all of the powerful tools
within the Xilinx FPGA design flow. Once completed, a module's implementation is
preserved, guaranteeing the timing in the finished device. This technology is a
requirement for any organization employing a team design methodology for the design of
a multi-million gate FPGA.

422 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

ISE Development Environment

Constraints Editor
Constraints are user instructions placed on elements of a schematic or HDL design, either
in the design itself or in a separate file. They can indicate a number of things such as
placement, implementation, naming, signal direction, and timing considerations. In the
Xilinx development system, logical constraints are placed in a file called the UCF (User
Constraints File). The Constraints Editor is a graphical program that can be used to create
and modify those constraints.

PlanAhead Tool
The optional PlanAhead™ tool provides an intuitive environment that delivers a faster,
more efficient design solution, allowing designers to find and fix problems early and
helping to achieve performance goals. The PlanAhead tool provides hierarchical, block-
based, modular and incremental design methodologies, enabling designers to change only
part of the design, leaving placement of the rest intact, thus shortening design iterations. It
helps designers consistently maintain the required performance, even while making
frequent changes.
For more information on the PlanAhead tool, see http://www.xilinx.com/planahead.

FPGA Editor
The FPGA Editor is a graphical application for displaying and configuring FPGAs. The
FPGA Editor requires an NCD file. This file contains the logic of your design mapped to
components such as CLBs and IOBs. In addition, the FPGA Editor reads from and writes to
a Physical Constraints File (PCF).
The following is a list of a few of the functions you can perform on your designs in the
FPGA Editor:
• Place and route critical components before running automatic place and route
• Fine-tune placement and routing after running automatic place and route
• Add probes to design to examine the signal states of the targeted device
• Run the Bitstream Generator and download the resulting file to the targeted device
• Create an entire design by hand (for advanced users)
For more information on the FPGA Editor PROBE tool, see
http://www.xilinx.com/products/design_tools/logic_design/verification/fpgaeditorprobe.htm.

Interactive Timing Analyzer

The Interactive Timing Analyzer provides a powerful, flexible, and easy way to perform
static timing analysis. With Timing Analyzer, analysis can be performed immediately after
mapping, placing, or routing a Spartan-3 generation FPGA design.
Timing Analyzer verifies that the delay along a given path or paths meets specified timing
requirements. It organizes and displays data that allows you to analyze critical paths in a
circuit, the cycle time of the circuit, the delay along any specified path(s), and the path with
the greatest delay. It also provides a quick analysis of the effect different speed grades have
on the same design.
Timing Analyzer creates timing analysis reports based on existing timing constraints or
user specified paths within the program. Timing reports have a hierarchical browser to
quickly jump to different sections of the reports. Timing paths in reports can be cross-
probed to synthesis tools (Exemplar and Synplicity) and the Floorplanner.

Spartan-3 Generation FPGA User Guide www.xilinx.com 423

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

ISE Simulator
ISE Simulator provides a complete, full-featured HDL simulator integrated within the ISE
development system. ISE Simulator comes in two versions:
• Free ISE Simulator Lite, included with all ISE configurations, is ideal for low-density
FPGA designs and is limited to 15,000 lines of HDL source code.
• ISE Simulator full version supports any design density and is a low-cost optional
add-on to ISE Foundation.
For more information on the ISE Simulator, see
http://www.xilinx.com/products/design_tools/logic_design/verification/ise_simulator.htm.

iMPACT Configuration Tool

The iMPACT configuration tool, a command line and GUI based tool, allows you to
configure your PLD designs using Boundary Scan, Slave Serial, SelectMap, and Desktop
Configuration modes. It also allows you to do the following:
• Download
• Read back and verify design configuration data
• Debug configuration problems
• Create PROM, SVF, STAPL, and System ACE™ CF programming files

ChipScope Pro Analyzer

The ChipScope Pro analyzer delivers in-circuit real-time debugging with shorter
verification cycles and lower project costs. By inserting special low-impact IP debugging
cores directly into your HDL code or design netlist, you can debug and verify FPGA logic
and system bus activity, capturing signals at or near system operating speeds. You easily
can change your trace points without having to recompile your design. The ChipScope Pro
analyzer embeds Integrated Logic Analyzer (ILA) and Integrated Bus Analyzer (IBA) cores
into your design. These cores allow the user to view all the internal signals and nodes
within the Spartan-3 generation FPGA.
For more information on the ChipScope Pro analyzer, see
http://www.xilinx.com/chipscope.

Power Analysis Tools

The ISE development system provides the tools to help understand power issues and
reduce the dynamic power in a Spartan-3 generation design. Advanced synthesis and
implementation algorithms help reduce dynamic power by an average of 10%.
The XPower Estimator spreadsheet is a pre-implementation analysis tool for estimating
power consumption in Spartan-3 generation FPGAs. Design requirements can be entered
directly or imported from the ISE Map results. The user specifies toggle rates for each part
of the design.
The XPower Analyzer is a post-route analysis tool for interactively and automatically
analyzing power consumption for Xilinx devices. XPower Analyzer reads VCD simulation
data to set estimation stimulus, reducing setup time. XPower Analyzer accepts VCD files
from Mentor (ModelSim), Cadence, and Synopsys. XPower Analyzer uses device
knowledge and design data to estimate device power and by-net power utilization.
Information is presented in both HTML and ASCII (text) report formats.
For more information on the XPower tools, see http://www.xilinx.com/power.

424 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

Related Materials and References

The following documents provide supplementary information useful with this chapter:
• Xilinx Design Tools Center
http://www.xilinx.com/products/design_resources/design_tool/
• ISE Software for Spartan-3 generation Designs
http://www.xilinx.com/ise/ise_promo/ise_spartan3.htm
• Software Download Center
http://www.xilinx.com/support/download/index.htm
• Software Manuals
http://www.xilinx.com/support/documentation/dt_ise.htm

Conclusion
The ISE design environment brings you the fastest, most complete family of design tools
available. The ISE tools are available in multiple configurations with various optional tools
and interfaces to third-party tools, allowing you to customize the set of tools for your own
needs. The ISE development system combines advanced technologies such as ProActive
Timing Closure with a flexible, easy-to-use graphical interface to help you achieve the best
possible designs with the least time and effort, regardless of your experience level.

Spartan-3 Generation FPGA User Guide www.xilinx.com 425

UG331 (v1.7) August 19, 2010
 R

Chapter 13: Using ISE Design Tools

426 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 14

Using IP Cores
Summary
This chapter provides an overview of the Xilinx CORE Generator™ System and the Xilinx
Intellectual Property (IP) offerings that facilitate the Spartan®-3 generation design process.
For more detailed and complete information, consult the CORE Generator System on-line
help available at http://toolbox.xilinx.com/docsan/xilinx92/help/iseguide/
mergedProjects/coregen/coregen.htm, and the Xilinx IP Center available at
http://www.xilinx.com/ipcenter/index.htm.
This chapter applies to all Spartan-3 generation FPGA families.

The CORE Generator System

The Xilinx CORE Generator System is the cataloging, customization, and delivery vehicle
for IP cores targeted to Xilinx FPGAs. The CORE Generator provides centralized access to
a catalog of ready-made IP functions ranging in complexity from simple arithmetic
operators, such as adders, accumulators, and multipliers to system-level building blocks,
such as filters, transforms, and memories. Cores can be displayed alphabetically, by
function, by vendor, or by type. Each core comes with its own data sheet, which documents
the core’s functionality in detail.
The CORE Generator user interface makes it very easy to access the latest
Spartan-3 generation IP releases and to get helpful, up-to-date information. Links to
partner IP providers also are built in for the various partner-supplied AllianceCORE
products. The use of CORE Generator IP cores in Spartan-3 generation designs enables
designers to shorten design time, and it also helps them realize high levels of performance
and area efficiency without any special knowledge of the Spartan-3 generation
architecture.
When installing the CORE Generator software, the designer gains immediate access to
dozens of cores supplied by the LogiCORE™ program. In addition, data sheets are
available for all AllianceCORE products, and additional, separately licensed, advanced
function LogiCORE products are also available. New and updated Spartan-3 generation IP
for the CORE Generator can be downloaded from the IP Center and added to the CORE
Generator catalog.

Xilinx IP Solutions and the IP Center

The CORE Generator tool works in conjunction with the Xilinx IP Center
(www.xilinx.com/ipcenter). To make the most of this resource, Xilinx highly recommends
that whenever starting a design, one first does a quick search of the IP Center to see
whether a ready-made core solution is already available.

Spartan-3 Generation FPGA User Guide www.xilinx.com 427

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

A complete catalog of Xilinx cores and IP tools resides on the IP Center, including:
• LogiCORE Products
• AllianceCORE Products
• Candidate Core Products
• Design Files
• Alliance Program Partner Services

LogiCORE Products
LogiCORE products are designed, sold, licensed, and supported by Xilinx. LogiCORE
products include a wide selection of generic, parameterized functions, such as muxes,
adders, multipliers, and memory cores, which are bundled with the Xilinx CORE
Generator software at no additional cost to licensed software customers. System-level
cores, such as PCI, Reed-Solomon, ADPCM, HDLC, POS-PHY, and Color Space
Converters are also available as optional, separately licensed products. The CORE
Generator commonly is used to quickly generate Spartan-3 generation block and
distributed memories. A more detailed listing of available Spartan-3 generation
LogiCORE products is available in Table 14-1, page 429 and on the Xilinx IP Center website
(http://www.xilinx.com/ipcenter).
Types of IP currently offered by the Xilinx LogiCORE program include:
• Basic Elements: logic gates, registers, multiplexers, adders, multipliers
• Communications and Networking: ADPCM modules, HDLC controllers, ATM
building blocks, forward error correction modules, cable modem solutions, 10/100
Ethernet MAC, SPI-4.2, and POS-PHY interfaces
• DSP and Video Image Processing: cores ranging from small building blocks (e.g., Time
Skew Buffers) to larger system-level functions (e.g., FIR Filters and FFTs)
• System Logic: accumulators, adders, subtracters, complementers, multipliers,
integrators, pipelined delay elements, single and dual-port distributed and block
RAM, ROM, and synchronous and asynchronous FIFOs
• Standard Bus Interfaces: PCI Interfaces, PCI Express PIPE Endpoint, I2C, CAN
• Processor Solutions: MicroBlaze™ 32-bit soft processor, PicoBlaze™ 8-bit soft
processor, and peripherals

AllianceCORE Products
AllianceCORE products are intellectual property (IP) cores developed, sold, and
supported by third-party Xilinx Alliance Program members. AllianceCORE certification
provides a showcase for the most popular IP cores offered.
To receive the AllianceCORE designation, members must submit netlist deliverables of the
core in the form of an ISE® project that includes both VHDL and Verilog wrapper files for
the current devices recommended for new designs. Xilinx then performs a “flow check” on
these deliverables to verify they run through the Xilinx design tools, and the reported
implementation results (device utilization and clock rates) are repeatable. Xilinx does not
verify core functionality or compliance with specific standards.

428 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Candidate Core Products

Candidate core products are developed, sold, and supported by third-party Xilinx Alliance
Program partners. To receive the Candidate core designation, the partner must submit ISE
report files to demonstrate that they have run the core through the Xilinx tools to target
specific Xilinx programmable logic devices.

Design Files
Xilinx offers two types of design files: XAPP application notes developed by Xilinx and
reference designs developed by Xilinx and its partners. Both types are extremely valuable
to customers looking for guidance when designing systems. Application notes developed
by Xilinx usually include supporting design files. They are supplied free of charge, without
technical support or warranty. Reference designs often can be used as starting points for
implementing a broad spectrum of functions in Xilinx programmable logic.

Xilinx Alliance Program Partner Services

Xilinx established the Xilinx Alliance Program to provide customers with access to a
worldwide network of certified design consultants who are proficient with Xilinx FPGAs,
software, and IP core integration. All members are certified and have extensive expertise
and experience with Xilinx technology in various vertical applications, such as
communications and networking, DSP, video and image processing, system I/O
interfaces, and home networking. Partners are an integral part of the Xilinx strategy to
provide customers with cost-efficient design solutions, while accelerating time to market.

SignOnce
The SignOnce IP License has been the industry’s first and only set of common license terms
for programmable logic soft IP cores. Xilinx and leading third-party providers have agreed
to offer cores to FPGA customers under a common set of terms known as the SignOnce IP
License, simplifying the process by which customers can access IP from multiple suppliers.
For more information see http://www.xilinx.com/ipcenter/signonce.htm.

Spartan-3 Generation IP Cores

Table 14-1 provides a partial listing of cores available for Spartan-3 generation designs. For
a complete catalog of Spartan-3 generation IP solutions and details on supported FPGA
families, visit the Xilinx IP Center parametric search engine and search for the latest
Spartan-3 generation core solutions.

Table 14-1: Spartan-3 Generation IP Cores Support

Product Name Vendor Type SignOnce Supported Devices

Audio Video and Image Processing

Spartan-3E, Spartan-3
1394a Link Layer Controller (C1394A) CAST, Inc. Candidate Core X
FPGAs

Accelerated Display Graphics Engine

BitSim AB Candidate Core Spartan-3 FPGAs
(BADGE)

Spartan-3E, Spartan-3
Adaptive Image Enhancement (Iridix) Apical Limited Candidate Core
FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 429

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Amphion Spartan-3, Spartan-II
ADPCM, 128 Simplex (CS4125) AllianceCORE X
Semiconductor, Ltd. FPGAs

Amphion Spartan-3, Spartan-IIE

ADPCM, 16 Simplex (CS4110) AllianceCORE X
Semiconductor, Ltd. FPGAs

Audio Sample Rate Converter, Low Spartan-3E, Spartan-3,

Coreworks Candidate Core X
Cost (CWda50) Spartan-IIE FPGAs

Pinpoint Solutions, Spartan-3, Spartan-IIE

Burst Locked PLL (BURST_PLL) AllianceCORE X
Inc. FPGAs

Spartan-3E, Spartan-3,
Compact Video Controller (logiCVC) Xylon d.o.o. AllianceCORE X
Spartan-IIE FPGAs

Discrete Wavelet Transform

Barco-Silex AllianceCORE X Spartan-3 FPGAs
(BA113FDWT)

Digital Design
Gamma Correction, Dynamic Candidate Core Spartan-3 FPGAs
Corporation

H.264 / MPEG4 Part 10, Decoder, 4i2i Communications Spartan-3E, Spartan-3

Candidate Core
Baseline (4i2i) Ltd. FPGAs

4i2i Communications
H.264 Encoder, Baseline (4i2i) AllianceCORE X Spartan-3 FPGAs
Ltd.

H.264 Encoder, Baseline Multichannel Spartan-3E, Spartan-3

CAST, Inc. Candidate Core X
HD (H264-E) FPGAs
Spartan-3, Spartan-IIE,
I2S Receiver (CWda10) Coreworks Candidate Core X
Spartan-II FPGAs
Spartan-3, Spartan-IIE,
I2S to SPDIF Converter (CWda04) Coreworks Candidate Core X
Spartan-II FPGAs

Spartan-3, Spartan-IIE
I2S Transmitter (CWda06) Coreworks Candidate Core X
FPGAs

JPEG, 2000 Decoder

Barco-Silex AllianceCORE X Spartan-3 FPGAs
(BA111JPEG2000D)

JPEG, 2000 Encoder

Barco-Silex AllianceCORE X Spartan-3 FPGAs
(BA112JPEG2000E)

Spartan-3E, Spartan-3
JPEG, Codec (JPEG_C) CAST, Inc. AllianceCORE X
FPGAs

Spartan-3E, Spartan-3
JPEG, Decoder (JPEG_D) CAST, Inc. AllianceCORE X
FPGAs

Spartan-3E, Spartan-3
JPEG, Encoder (JPEG_E) CAST, Inc. AllianceCORE X
FPGAs

Amphion Spartan-3, Spartan-IIE

JPEG, Motion Codec V1.0 (CS6190) AllianceCORE X
Semiconductor, Ltd. FPGAs

Amphion
JPEG, Motion Decoder (CS6150) AllianceCORE X Spartan-3 FPGAs
Semiconductor, Ltd.

Amphion Spartan-3, Spartan-IIE

JPEG, Motion Encoder (CS6100) AllianceCORE X
Semiconductor, Ltd. FPGAs

430 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3, Spartan-IIE
Longitudinal Time Code Generator Deltatec S.A. AllianceCORE X
FPGAs

Motion JPEG Decoder 4i2i Communications Spartan-3, Spartan-IIE

AllianceCORE X
(JPEG Decoder) Ltd. FPGAs

Motion JPEG Encoder 4i2i Communications Spartan-3, Spartan-IIE

AllianceCORE X
(JPEG Encoder) Ltd. FPGAs

Amphion
MPEG-2 Video Decoder (CS6651) AllianceCORE X Spartan-3 FPGAs
Semiconductor, Ltd.

Spartan-3, Spartan-3A
MPEG4 Simple Profile Decoder Xilinx, Inc. LogiCORE X
FPGAs

Spartan-3, Spartan-3A
MPEG4 Simple Profile Encoder Xilinx, Inc. LogiCORE X
FPGAs

4i2i Communications
MPEG-4 Video Compression Decoder AllianceCORE X Spartan-3 FPGAs
Ltd.

4i2i Communications
MPEG-4 Video Compression Encoder AllianceCORE X Spartan-3 FPGAs
Ltd.

NTSC Color Separator Pinpoint Solutions, Spartan-3, Spartan-IIE

AllianceCORE X
(NTSC-COSEP) Inc. FPGAs
Spartan-3, Spartan-IIE,
SPDIF to I2S Converter (CWda03) Coreworks Candidate Core X
Spartan-II FPGAs

SPDIF-AES/EBU Feed-Forward Spartan-3, Spartan-IIE,

Coreworks Candidate Core X
Receiver (CWda01) Spartan-II FPGAs

SPDIF-AES/EBU Feed-Forward Spartan-3, Spartan-IIE,

Coreworks Candidate Core X
Transmitter (CWda02) Spartan-II FPGAs
SPDIF-AES/EBU Transmitter Spartan-3E, Spartan-3
Coreworks Candidate Core X
(CWda05) FPGAs

VGA Compatible Cores Spartan-3E, Spartan-3

Intrinsix Corporation Candidate Core
(INT416-SM, INT416-SXGA) FPGAs

Basic Logic

Spartan-3, Spartan-IIE,
8b/10b Decoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
8b/10b Encoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3E, Spartan-3
Binary Counter Xilinx, Inc. LogiCORE
FPGAs

Spartan-3E, Spartan-3,
Binary Decoder Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
BUFE-based Multiplexer Slice Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3
Comparator Xilinx, Inc. LogiCORE
FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 431

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3E, Spartan-3,
FD-based Parallel Register Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
FD-based Shift Register Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Floating Point Comparator Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DFPCOMP) FPGAs

Spartan-3E, Spartan-3,
LD-based Parallel Latch Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
Linear Feedback Shift Register (LFSR) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3E, Spartan-3
RAM-based Shift Register Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-IIE,
8b/10b Decoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
8b/10b Encoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3E, Spartan-3
Binary Counter Xilinx, Inc. LogiCORE
FPGAs

Spartan-3E, Spartan-3,
Binary Decoder Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
BUFE-based Multiplexer Slice Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3
Comparator Xilinx, Inc. LogiCORE
FPGAs

Spartan-3E, Spartan-3,
FD-based Parallel Register Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
FD-based Shift Register Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Floating Point Comparator Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DFPCOMP) FPGAs

Spartan-3E, Spartan-3,
LD-based Parallel Latch Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
Linear Feedback Shift Register (LFSR) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3E, Spartan-3
RAM-based Shift Register Xilinx, Inc. LogiCORE
FPGAs

Bus Interface and I/O

Silicon Interfaces
1394 Link Layer Controller (SI16FW10) Candidate Core Spartan-3 FPGAs
America Inc.

Spartan-3E, Spartan-3
1394a Link Layer Controller (C1394A) CAST, Inc. Candidate Core X
FPGAs

432 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Spartan-3E, Spartan-3,
CAN (DO-DI-CAN) Xilinx, Inc. LogiCORE Spartan-3A, Spartan-3A
DSP FPGAs

Spartan-3E, Spartan-3,
CAN 2.0 B Compatible Network
Xylon d.o.o. AllianceCORE X Spartan-IIE, Spartan-II
Controller (LogiCAN)
FPGAs
Intelliga Integrated Spartan-3, Spartan-IIE
CAN 2.0B Bus Controller (iCAN) AllianceCORE X
Design, Ltd. FPGAs

CAN Bus Controller (MC-XIL-OPB- Spartan-3, Spartan-IIE

Avnet Design Services AllianceCORE X
XCAN) FPGAs

Spartan-3E, Spartan-3
CAN Bus Controller 2.0B CAST, Inc. AllianceCORE X
FPGAs
Spartan-3, Spartan-IIE
I2C Bus Controller (I2C) CAST, Inc. AllianceCORE X
FPGAs
Spartan-3E, Spartan-3
I2C Bus Controller Master (DI2CM) Digital Core Design AllianceCORE X
FPGAs

Spartan-3E, Spartan-3
I2C Bus Controller Slave (DI2CS) Digital Core Design Candidate Core X
FPGAs

J1850 Class B Interface

Drivven, Inc. Candidate Core X Spartan-3 FPGAs
(J1850_BDLC_VPW)

LIN - Local Interconnect Network Bus Intelliga Integrated Spartan-3, Spartan-IIE

AllianceCORE X
Controller (iLIN) Design, Ltd. FPGAs

Spartan-3E, Spartan-3
LIN Controller CAST, Inc. AllianceCORE X
FPGAs

MIL-STD-1553 Bus Interface GE Fanuc Embedded Spartan-3E, Spartan-3,

Candidate Core
(FlightCORE-1553) Systems, Inc. Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
OPB Arbiter Xilinx, Inc. LogiCORE
FPGAs

OPB IIC Master and Slave Bus Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE
Controller (DO-DI-IIC-SD) Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
OPB IPIF Architecture Xilinx, Inc. LogiCORE Spartan-3A, Spartan-IIE
FPGAs

Spartan-3, Spartan-IIE
OPB to OPB Bridge (Lite Version) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-3A,
OPB/PLB PCI32 Interface Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
PCI 32-bit Master Interface (PCI-M32) CAST, Inc. AllianceCORE X
FPGAs

PCI 32-Bit Multifunction Target Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Interface (PCI-TMF) FPGAs

Spartan-3, Spartan-IIE
PCI 32-bit Target Interface (PCI-T32) CAST, Inc. AllianceCORE X
FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 433

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
NitAl Consulting
PCI Express (PExCore-254) Candidate Core X Spartan-3 FPGAs
Services, Inc.

PCI Express PIPE Endpoint LogiCORE Spartan-3E, Spartan-3

Xilinx, Inc. LogiCORE X
(DO-DI-PCIE-PIPE) FPGAs

Spartan-3, Spartan-IIE
PCI Host Bridge (PCI-HB) CAST, Inc. AllianceCORE X
FPGAs

PCI Interface (NWL PCI) Northwest Logic AllianceCORE X Spartan-3 FPGAs

Spartan-3E, Spartan-3,
PCI32 Single Project Spartan Series
Xilinx, Inc. LogiCORE X Spartan-IIE, Spartan-II
(DO-DI-PCI32-SP)
FPGAs

Spartan-3A DSP,
PCI32 Spartan Series Interface Spartan-3A, Spartan-3E,
Xilinx, Inc. LogiCORE X
(DO-DI-PCI32-IP) Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
PCI64 Virtex® and Spartan Series Spartan-3A, Spartan-3E,
Xilinx, Inc. LogiCORE X
Interface, IP Only (DO-DI-PCI64-IP) Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A, Spartan-3A
PCI64 Interface Design Kit (DO-DI-
Xilinx, Inc. LogiCORE X DSP, Spartan-3, Spartan-
PCI64-DKT)
IIE, Spartan-II FPGAs

Spartan-3, Spartan-IIE
PCI-PCI Bridge (EP440) Eureka Technology AllianceCORE X
FPGAs

Spartan-3, Spartan-IIE
PowerPC® Bus Master (EP201) Eureka Technology AllianceCORE X
FPGAs

Spartan-3, Spartan-IIE
PowerPC Bus Slave (EP100) Eureka Technology AllianceCORE X
FPGAs

Serial ATA I/II Host Controller ASICS World Service,

Candidate Core X Spartan-3 FPGAs
(SATA_H1) Ltd.

HCL Technologies Spartan-3, Spartan-IIE

Smart Card Interface (HCLT SCI) Candidate Core X
Ltd. FPGAs

HCL Technologies Spartan-3, Spartan-IIE

SPI Master Interface (HCL_SPI_M) Candidate Core X
Ltd. FPGAs

Spartan-3E, Spartan-3,
SPI-Master/Slave (DSPI) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
SPI-Slave (DSPIS) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

Two-Wire Serial Interface - I2C Spartan-3, Spartan-IIE,

Avnet Design Services AllianceCORE X
(MC-XIL-TWSI) Spartan-II FPGAs

Spartan-3, Spartan-IIE
USB 1.1 Function Controller (CUSB) CAST, Inc. AllianceCORE X
FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

USB 1.1 Host (USB1_HOST) Candidate Core X
Ltd. FPGAs

434 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3, Spartan-IIE
USB 2.0 Function Controller (CUSB2) CAST, Inc. AllianceCORE X
FPGAs

USB 2.0 Function Controller ASICS World Service, Spartan-3E, Spartan-3

Candidate Core X
(USB2_DEV) Ltd. FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

USB 2.0 On-The-Go (USB2_OTG) Candidate Core X
Ltd. FPGAs

USB Function Controller with On-Chip Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Peripheral Bus (CUSB_OPB) FPGAs

Spartan-3A DSP,
V3.0 PCI 32-bit/(66-33 MHz) Core and Spartan-3A, Spartan-3,
Xilinx, Inc. LogiCORE X
Design Kit Spartan-IIE, Spartan-II
FPGAs
Spartan-3E, Spartan-3,
V3.0 PCI 64-bit/(66-33 MHz) and PCI
Xilinx, Inc. LogiCORE X Spartan-IIE, Spartan-II
32-bit/(66-33 MHz) Core
FPGAs

V5.0 PCI 64-bit/33 MHz, V3.0 PCI Spartan-3, Spartan-IIE,

Xilinx, Inc. LogiCORE X
64-bit/(66-33MHz) and PCI 32-bit Spartan-II FPGAs

VLYNQ Interface for Texas Instruments Spartan-3E, Spartan-3

Xilinx, Inc. LogiCORE X
Processors FPGAs

VME Address and Data Interface Spartan-3E, Spartan-3,

Millogic Ltd Candidate Core
(ML7C964) Spartan-II FPGAs

Communications and Networking

Silicon Interfaces
1394 Link Layer Controller (SI16FW10) Candidate Core Spartan-3 FPGAs
America Inc.

Spartan-3E, Spartan-3
1394a Link Layer Controller (C1394A) CAST, Inc. Candidate Core X
FPGAs

Spartan-3E, Spartan-3,
16550 UART w/ FIFOs (D16550) Digital Core Design AllianceCORE X
Spartan-IIE FPGAs
Spartan-3E, Spartan-3,
16750 UART w/ FIFOs (D16750) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

16750 UART w/FIFOs and sync CPU Spartan-3E, Spartan-3

CAST, Inc. Candidate Core X
Interface (H16750S) FPGAs

85C30 Serial Communication Spartan-3E, Spartan-3

Millogic Ltd Candidate Core
Controller (ML85C30) FPGAs

Spartan-3, Spartan-IIE,
8b/10b Decoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
8b/10b Encoder Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Additive White Gaussian Noise Xilinx, Inc. LogiCORE X Spartan-3 FPGAs

Amphion Spartan-3, Spartan-IIE

ADPCM, 128 Simplex (CS4125) AllianceCORE X
Semiconductor, Ltd. FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 435

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Amphion Spartan-3, Spartan-IIE
ADPCM, 16 Simplex (CS4110) AllianceCORE X
Semiconductor, Ltd. FPGAs

Spartan-3, Spartan-IIE
AES Encryption CAST, Inc. AllianceCORE X
FPGAs

AES Encryptor/Decryptor Spartan-3E, Spartan-3,

Algotronix, Ltd. Candidate Core X
(AES_CORE_G2_XE) Spartan-IIE FPGAs

Helion Technology Spartan-3E, Spartan-3

AES Fast Encryptor (Helion) AllianceCORE X
Limited FPGAs

Helion Technology Spartan-3E, Spartan-3

AES Standard Encryptor/Decryptor AllianceCORE X
Limited FPGAs

Helion Technology Spartan-3E, Spartan-3

AES Tiny Encryptor/Decryptor AllianceCORE X
Limited FPGAs

HCL Technologies Spartan-3, Spartan-IIE

ATA Host Controller (HCL ATA) AllianceCORE X
Ltd. FPGAs

ATA-4/UDMA-33 IDE Host Nuvation Research Spartan-3E, Spartan-3

Candidate Core
(Nuvation) Corporation FPGAs

ATA-5/UDMA-66 IDE Host Nuvation Research Spartan-3E, Spartan-3

Candidate Core
(Nuvation) Corporation FPGAs

ATM Adaptation Layer 5 (AAL5) ModelWare, Inc. AllianceCORE X Spartan-3 FPGAs

Spartan-3A DSP,
Spartan-3E, Spartan-3,
Convolutional Encoder Xilinx, Inc. LogiCORE
Spartan-IIE, Spartan-II
FPGAs

CRC-32 for 10 Gbps OC192 systems

Calyptech Pty Ltd. AllianceCORE X Spartan-3 FPGAs
(CORE-CRC-128)

Deframer, STM0/OC1, STM1/OC3,

Aliathon Ltd. Candidate Core X Spartan-3 FPGAs
STM4/OC12 (sdh.dfrm.001)

Demapper, STM0/OC1, STM1/OC3,

Aliathon Ltd. Candidate Core X Spartan-3 FPGAs
STM4/OC12 (sdh.dmap.001)

Spartan-3E, Spartan-3,
DES Encryption CAST, Inc. Candidate Core X
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
DES3 Encryption CAST, Inc. AllianceCORE X
FPGAs

Digital Up Converter Xilinx, Inc. LogiCORE Spartan-3 FPGAs

Ethernet 1000BASE-X Spartan-3E, Spartan-3,

Xilinx, Inc. LogiCORE X
PCS/PMA or SGMII Spartan-3A FPGAs

Spartan-3, Spartan-IIE
Ethernet MAC, 10/100 (MAC) CAST, Inc. AllianceCORE X
FPGAs

Spartan-3A DSP,
Ethernet Statistics Xilinx, Inc. LogiCORE X Spartan-3E, Spartan-3,
Spartan-3A FPGAs

436 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Helion Technology Spartan-3, Spartan-IIE
Fast AES decryptor Candidate Core X
Limited FPGAs

Spartan-3E, Spartan-3
Framer, SONET OC-3 to OC-768 (Xelic) Xelic, Inc. Candidate Core X
FPGAs

Framer, STM0/OC1, STM1/OC3,

Aliathon Ltd. Candidate Core X Spartan-3 FPGAs
STM4/OC12 (sdh.frm.001)

Generic Framing Procedure Spartan-3E, Spartan-3

Xilinx, Inc. LogiCORE X
(DO-DI-GFP) FPGAs

Gigabit Ethernet Media Access Spartan-3E, Spartan-3,

Xilinx, Inc. LogiCORE X
Controller (GEMAC) Spartan-3A FPGAs

HDLC, Single-Channel
Avnet Design Services AllianceCORE X Spartan-3 FPGAs
(MC-XIL-HDLC)

Inverse Multiplexing for ATM (IMA)

Phystream Ltd. Candidate Core X Spartan-3 FPGAs
for PhyCore (005-002-002)

Helion Technology Spartan-3E, Spartan-3

IPsec ESP Security Engine (Helion) AllianceCORE X
Limited FPGAs

Mapper, STM0/OC1, STM1/OC3,

Aliathon Ltd. Candidate Core X Spartan-3 FPGAs
STM4/OC12 (sdh.map.001)

Spartan-3E, Spartan-3,
MD5 Message Digest Algorithm CAST, Inc. AllianceCORE X
Spartan-IIE FPGAs

Multi Rate Performance Monitor,

Calyptech Pty Ltd. AllianceCORE X Spartan-3 FPGAs
2.5 Gb/s (CORE-PM-MR)

OPB 10/100 Ethernet MAC Lite Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE X
(OPB EMAC Lite) Spartan-IIE FPGAs

Spartan-3, Spartan-3A,
OPB 10/100 Ethernet Media Access
Xilinx, Inc. LogiCORE X Spartan-IIE, Spartan-II
Controller (EMAC)
FPGAs

OPB ATM Utopia Level 2 Master and Spartan-3, Spartan-IIE

Xilinx, Inc. LogiCORE
Slave FPGAs

Spartan-3, Spartan-II
OPB Single Channel HDLC Controller Xilinx, Inc. LogiCORE
FPGAs

Spartan-3A DSP,
Packet Queue Xilinx, Inc. LogiCORE X Spartan-3A, Spartan-3E,
Spartan-3 FPGAs

NitAl Consulting
PCI Express (PExCore-254) Candidate Core X Spartan-3 FPGAs
Services, Inc.

Pseudorandom Number Generator Helion Technology Spartan-3E, Spartan-3,

AllianceCORE X
(ANSI X9.17/X9.31 PRNG) Limited Spartan-IIE FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

Reed-Solomon Decoder (RSDX1) Candidate Core X
Ltd. FPGAs

Spartan-3A DSP,
Reed-Solomon Decoder Xilinx, Inc. LogiCORE X Spartan-3A, Spartan-3E,
Spartan-3 FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 437

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
ASICS World Service, Spartan-3, Spartan-IIE
Reed-Solomon Encoder (RSEX1) Candidate Core X
Ltd. FPGAs

Spartan-3E, Spartan-3,
Reed-Solomon Encoder Xilinx, Inc. LogiCORE X
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
SDLC Controller (SDLC) CAST, Inc. AllianceCORE X
FPGAs

Serial ATA I/II Host Controller ASICS World Service,

Candidate Core X Spartan-3 FPGAs
(SATA_H1) Ltd.

SHA-1, SHA-224, SHA-256 and MD5, Helion Technology Spartan-3E, Spartan-3

AllianceCORE X
Tiny with HMAC (Helion) Limited FPGAs

SHA-1,SHA-256 and MD5 Hashing, Helion Technology Spartan-3E, Spartan-3

AllianceCORE X
Fast (Helion) Limited FPGAs

HCL Technologies Spartan-3, Spartan-IIE

SPI Master Interface (HCL_SPI_M) Candidate Core X
Ltd. FPGAs
Spartan-3A DSP,
SPI-3 Link Layer Interface,
Xilinx, Inc. LogiCORE X Spartan-3E, Spartan-3
Multi-channel
FPGAs

Spartan-3A, Spartan-3A
SPI-3 Physical Layer Interface,
Xilinx, Inc. LogiCORE X DSP, Spartan-3E,
Multi-channel
Spartan-3 FPGAs

Spartan-3A DSP,
SPI-4 Phase 2 Interface Solutions
Xilinx, Inc. LogiCORE X Spartan-3A, Spartan-3E,
(DO-DI-POSL4MC)
Spartan-3 FPGAs

Spartan-3E, Spartan-3,
SPI-Master/Slave (DSPI) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
SPI-Slave (DSPIS) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
TDES AMBA Platform SoC Solutions, LLC AllianceCORE X
FPGAs

TDM Interface, Multi-port

Phystream Ltd. Candidate Core X Spartan-3 FPGAs
Programmable

Spartan-3, Spartan-II
TDM Switch (TDM_H110) Calyptech Pty Ltd. AllianceCORE X
FPGAs

Spartan-3A DSP,
Tri-Mode Ethernet Media Access
Xilinx, Inc. LogiCORE X Spartan-3E, Spartan-3,
Controller (TEMAC)
Spartan-3A FPGAs

Spartan-3A,
Turbo Convolutional Code Decoder, Spartan-3A DSP,
Xilinx, Inc. LogiCORE
CDMA2000/3GPP2 Spartan-3E, Spartan-3,
Spartan-II FPGAs

Spartan-3A DSP,
Turbo Convolutional Code Encoder, Spartan-3A, Spartan-3E,
Xilinx, Inc. LogiCORE
CDMA2000/3GPP2 Spartan-3, Spartan-II
FPGAs

438 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Turbo Product Code (TPC) Decoder Xilinx, Inc. LogiCORE Spartan-3 FPGAs

Turbo Product Code (TPC) Encoder Xilinx, Inc. LogiCORE Spartan-3 FPGAs

UART, Generic Compact

Memec Design AllianceCORE X Spartan-3 FPGAs
(MC-XIL-UART)
Spartan-3E, Spartan-3,
UMTS/3GPP Turbo Convolutional
Xilinx, Inc. LogiCORE Spartan-3A,
Decoder
Spartan-3A DSP FPGAs

UMTS/3GPP Turbo Convolutional Spartan-3E, Spartan-3,

Xilinx, Inc. LogiCORE
Encoder Spartan-3A,

Spartan-3, Spartan-IIE
USB 1.1 Function Controller (CUSB) CAST, Inc. AllianceCORE X
FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

USB 1.1 Host (USB1_HOST) Candidate Core X
Ltd. FPGAs

Spartan-3, Spartan-IIE
USB 2.0 Function Controller (CUSB2) CAST, Inc. AllianceCORE X
FPGAs

USB 2.0 Function Controller ASICS World Service, Spartan-3E, Spartan-3

Candidate Core X
(USB2_DEV) Ltd. FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

USB 2.0 On-The-Go (USB2_OTG) Candidate Core X
Ltd. FPGAs

USB Function Controller with On-Chip Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Peripheral Bus (CUSB_OPB) FPGAs

UTOPIA, Multi-port Serial Bridge Phystream Ltd. Candidate Core X Spartan-3 FPGAs

Spartan-3E, Spartan-3,
Viterbi Decoder Xilinx, Inc. LogiCORE X Spartan-3A, Spartan-IIE
FPGAs

Viterbi Decoder, Spartan-3, Spartan-IIE

Xilinx, Inc. LogiCORE X
(IEEE 802-Compatible) FPGAs

Spartan-3, Spartan-IIE
Z80 Serial I/O Controller (CZ80SIO) CAST, Inc. AllianceCORE X
FPGAs

Digital Signal Processing

Additive White Gaussian Noise Xilinx, Inc. LogiCORE X Spartan-3 FPGAs

Amphion Spartan-3, Spartan-IIE

ADPCM, 128 Simplex (CS4125) AllianceCORE X
Semiconductor, Ltd. FPGAs

Amphion Spartan-3, Spartan-IIE

ADPCM, 16 Simplex (CS4110) AllianceCORE X
Semiconductor, Ltd. FPGAs

Spartan-3E, Spartan-3,
Convolutional Encoder Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
CORDIC Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Digital Up Converter Xilinx, Inc. LogiCORE Spartan-3 FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 439

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Spartan-3E, Spartan-3,
Direct Digital Synthesizer Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Discrete Wavelet Transform

Barco-Silex AllianceCORE X Spartan-3 FPGAs
(BA113FDWT)

Spartan-3E, Spartan-3,
Divider Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3A DSP,
DVB-S.2 FEC Encoder Xilinx, Inc. LogiCORE X Spartan-3E, Spartan-3
FPGAs

Spartan-3E, Spartan-3,
Fast Fourier Transform Xilinx, Inc. LogiCORE
Spartan-3A

FFT, Pipelined (Vectis HiSpeed) RF Engines, Ltd. AllianceCORE Spartan-3 FPGAs

FFT, Pipelined (Vectis-QuadSpeed) RF Engines, Ltd. AllianceCORE Spartan-3 FPGAs

Sundance
FFT/IFFT (FC100) Multiprocessor Candidate Core Spartan-3 FPGAs
Technology Ltd

Spartan-3E, Spartan-3,
FIR Compiler Xilinx, Inc. LogiCORE Spartan-3A,
Spartan-3A DSP

Floating-Point Operators Xilinx, Inc. LogiCORE Spartan-3E, Spartan-3,

High Data Rate Demodulator Silicon Infusion
Candidate Core X Spartan-3 FPGAs
(Zaltys HDRM-D) Limited

Spartan-3E, Spartan-II
IEEE 802.16E CTC Decoder Xilinx, Inc. LogiCORE X
FPGAs

Spartan-3A DSP,
IEEE 802.16e CTC Encoder Xilinx, Inc. LogiCORE Spartan-3E, Spartan-3,
Spartan-3A FPGAs

Spartan-3E, Spartan-3
IEEE 802.16e LDPC Encoder Xilinx, Inc. LogiCORE
FPGAs

J.83 Universal Modulator Annex A/C Xilinx, Inc. LogiCORE X Spartan-3 FPGAs

J.83 Universal Modulator Annex B Xilinx, Inc. LogiCORE X Spartan-3 FPGAs

Spartan-3, Spartan-IIE
Linear Feedback Shift Register (LFSR) Xilinx, Inc. LogiCORE
FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

Reed-Solomon Decoder (RSDX1) Candidate Core X
Ltd. FPGAs

Spartan-3E, Spartan-3,
Reed-Solomon Decoder Xilinx, Inc. LogiCORE X
Spartan-3A FPGAs

ASICS World Service, Spartan-3, Spartan-IIE

Reed-Solomon Encoder (RSEX1) Candidate Core X
Ltd. FPGAs

440 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Spartan-3A DSP,
Reed-Solomon Encoder Xilinx, Inc. LogiCORE X Spartan-3E, Spartan-3,
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
Sine Cosine Look Up Table Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs
Spartan-3, Spartan-IIE
TMS32025 DSP Processor (C32025) CAST, Inc. AllianceCORE X
FPGAs

TMS32025TX Digital Signal Processor Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
(C32025TX) FPGAs

Turbo Product Code (TPC) Decoder Xilinx, Inc. LogiCORE Spartan-3 FPGAs

Turbo Product Code (TPC) Encoder Xilinx, Inc. LogiCORE Spartan-3 FPGAs
Spartan-3E, Spartan-3,
Viterbi Decoder Xilinx, Inc. LogiCORE X Spartan-3A, Spartan-IIE
FPGAs

Spartan-3A DSP,
Viterbi Decoder, Spartan-3A, Spartan-3E,
Xilinx, Inc. LogiCORE X
(IEEE 802-Compatible) Spartan-3, Spartan-IIE
FPGAs

Spartan-3E, Spartan-3
WiMAX FEC Pack Xilinx, Inc. LogiCORE X
FPGAs

Math

Spartan-3E, Spartan-3,
Accumulator Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
Adder/Subtracter Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3
Comparator Xilinx, Inc. LogiCORE
FPGAs

Spartan-3A, Spartan-3A
Complex Multiplier Xilinx, Inc. LogiCORE DSP, Spartan-3E,
Spartan-3 FPGAs

Spartan-3E, Spartan-3,
CORDIC Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Discrete Wavelet Transform

Barco-Silex AllianceCORE X Spartan-3 FPGAs
(BA113FDWT)

FFT, Pipelined (Vectis HiSpeed) RF Engines, Ltd. AllianceCORE Spartan-3 FPGAs

FFT, Pipelined (Vectis-QuadSpeed) RF Engines, Ltd. AllianceCORE Spartan-3 FPGAs

Sundance
FFT/IFFT (FC100) Multiprocessor Candidate Core Spartan-3 FPGAs
Technology Ltd

Spartan-3 Generation FPGA User Guide www.xilinx.com 441

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3E, Spartan-3,
Floating Point Adder (DFPADD) Digital Core Design AllianceCORE X
Spartan-IIE FPGAs

Floating Point Comparator Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DFPCOMP) FPGAs

Spartan-3E, Spartan-3,
Floating Point Divider (DFPDIV) Digital Core Design AllianceCORE X
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
Floating Point Multiplier (DFPMUL) Digital Core Design AllianceCORE X
Spartan-IIE FPGAs

Floating Point Square Root Operator Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DFPSQRT) FPGAs

Floating Point to Integer Converter Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DFP2INT) FPGAs

Floating-Point Operators Xilinx, Inc. LogiCORE Spartan-3E, Spartan-3

Integer to Floating Point Converter Spartan-3E, Spartan-3

Digital Core Design AllianceCORE X
(DINT2FP) FPGAs

Spartan-3E, Spartan-3,
Multiplier Accumulator Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
Multiplier Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
Sine Cosine Look Up Table Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Memory Interface and Storage Element

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Asynchronous FIFO Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Block Memory Generator Xilinx, Inc. LogiCORE X
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
BUFT-based Multiplexer Slice Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Content Addressable Memory (CAM) Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Distributed Memory Generator Xilinx, Inc. LogiCORE X
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

442 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Distributed Memory Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
Dual-Port Block Memory Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
FD-based Parallel Register Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
FD-based Shift Register Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
FIFO Generator Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
LD-based Parallel Latch Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3A DSP,
Spartan-3A, Spartan-3E,
RAM-based Shift Register Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

SDRAM Controller, DDR

Array Electronics AllianceCORE X Spartan-3 FPGAs
(DDR-XS-XILINX)

Spartan-3E, Spartan-3,
SDRAM Controller, DDR (NWL) Northwest Logic AllianceCORE X
Spartan-IIE FPGAs

SDRAM Controller, DDR/DDR2 Spartan-3E, Spartan-3

Eureka Technology Candidate Core X
(EP532) FPGAs

Spartan-XL, Spartan-3
SDRAM Controller, DDR2 (NWL) Northwest Logic AllianceCORE X
FPGAs

Spartan-3E, Spartan-3,
SDRAM Controller, SDR (NWL) Northwest Logic AllianceCORE X
Spartan-IIE FPGAs

SDRAM Controller, SDR Spartan-3E, Spartan-3

Array Electronics Candidate Core X
(SDR-XS-XILINX) FPGAs

Spartan-3A, Spartan-3A
DSP, Spartan-3E,
Single-Port Block Memory Xilinx, Inc. LogiCORE
Spartan-3, Spartan-IIE,
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
Synchronous FIFO Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 443

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Embedded Processing

Spartan-3E, Spartan-3,
16550 UART w/ FIFOs (D16550) Digital Core Design AllianceCORE X
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
16750 UART w/ FIFOs (D16750) Digital Core Design Candidate Core X
Spartan-IIE FPGAs

16750 UART w/FIFOs and sync CPU Spartan-3E, Spartan-3

CAST, Inc. Candidate Core X
Interface (H16750S) FPGAs

2910A Microprogram Controller Spartan-3E, Spartan-3

CAST, Inc. AllianceCORE X
(C2910A) FPGAs

68000 Compatible Microprocessor Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
(C68000) FPGAs

Spartan-3E, Spartan-3
68681 DUART (ML68681) Millogic Ltd Candidate Core
FPGAs

8051 High-speed 8-bit RISC Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Microcontroller (R80515) FPGAs

8051 Microcontroller, PicoBlaze Spartan-3E, Spartan-3,

Roman-Jones, Inc. AllianceCORE X
Emulated (PB8051-MX/TF) Spartan-IIE FPGAs

80C51 Compatible RISC Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Microcontroller (R8051) FPGAs

Spartan-3E, Spartan-3
82C51 USART (ML82C51) Millogic Ltd Candidate Core
FPGAs

85C30 Serial Communication Spartan-3E, Spartan-3

Millogic Ltd Candidate Core
Controller (ML85C30) FPGAs

HCL Technologies Spartan-3, Spartan-IIE

ATA Host Controller (HCL ATA) AllianceCORE X
Ltd. FPGAs

ATA-4/UDMA-33 IDE Host Nuvation Research Spartan-3E, Spartan-3

Candidate Core
(Nuvation) Corporation FPGAs

ATA-5/UDMA-66 IDE Host Nuvation Research Spartan-3E, Spartan-3

Candidate Core
(Nuvation) Corporation FPGAs

Spartan-3, Spartan-IIE,
Block RAM (BRAM) Block (v1.00a) Xilinx, Inc. LogiCORE
Spartan-II FPGAs

CAN Bus Controller Spartan-3, Spartan-IIE

Avnet Design Services AllianceCORE X
(MC-XIL-OPB-XCAN) FPGAs

Channelized Direct Memory Access Spartan-3, Spartan-IIE,

Xilinx, Inc. LogiCORE
and Scatter Gather Spartan-II FPGAs

Spartan-3, Spartan-IIE,
ChipScope OPB IBA Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
ChipScope PLB IBA Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3E, Spartan-3,
Compact Video Controller (logiCVC) Xylon d.o.o. AllianceCORE X
Spartan-IIE FPGAs

444 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3, Spartan-IIE,
DCR Interrupt Controller (DO-EDK) Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Digital Clock Manager (DCM) Module Xilinx, Inc. LogiCORE Spartan-3 FPGAs

Spartan-3, Spartan-3A,
Fixed Interval Timer (FIT) Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3, Spartan-3A,
FSL_V20 Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Helion Technology Spartan-3E, Spartan-3

IPsec ESP Security Engine (Helion) AllianceCORE X
Limited FPGAs

Spartan-3, Spartan-3A,
LMB BRAM Interface Controller Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

MC68230 Parallel Interface/Timer Spartan-3E, Spartan-3

Millogic Ltd Candidate Core
(ML68230) FPGAs

Spartan-3, Spartan-IIE,
MicroBlaze Parameterized Netlist Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-3A
Multi Channel OPB DDR Controller Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-3A
Multi Channel OPB DDR2 Controller Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-3A
Multi Channel OPB EMC Controller Xilinx, Inc. LogiCORE
FPGAs
Spartan-3, Spartan-3A,
Multi-Channel-OPB (MCH_OPB)
Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
SDRAM Controller
FPGAs

Spartan-3, Spartan-3A,
OPB 16550 UART Controller Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs
Spartan-3, Spartan-IIE
OPB Arbiter Xilinx, Inc. LogiCORE
FPGAs

OPB ATM Utopia Level 2 Spartan-3, Spartan-IIE

Xilinx, Inc. LogiCORE
Master and Slave FPGAs

Spartan-3, Spartan-3A,
OPB BRAM Controller Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
OPB Bus Structure Xilinx, Inc. LogiCORE Spartan-3A, Spartan-IIE
FPGAs

Spartan-3, Spartan-3A,
OPB Central DMA Controller Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3E, Spartan-3,
OPB DDR SDRAM Controller Xilinx, Inc. LogiCORE Spartan-3A, Spartan-IIE
FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 445

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

Spartan-3, Spartan-3A,
OPB Delta-Sigma Analog to Digital
Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
Converter (ADC)
FPGAs

Spartan-3, Spartan-3A,
OPB Delta-Sigma Digital to Analog
Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
Converter (DAC)
FPGAs

Spartan-3, Spartan-3A,
OPB EMC (DO-EDK) Xilinx, Inc. LogiCORE Spartan-IIE, Spartan-II
FPGAs

Spartan-3, Spartan-3A,
OPB GPIO Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

OPB IIC Master and Slave Bus Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE
Controller (DO-DI-IIC-SD) Spartan-IIE FPGAs
Spartan-3, Spartan-3A,
OPB Interrupt Controller Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3E, Spartan-3,
OPB IPIF Architecture Xilinx, Inc. LogiCORE Spartan-3A, Spartan-IIE
FPGAs

Spartan-3, Spartan-3A,
OPB PCI Arbiter Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-3A,
OPB SDRAM Controller Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-II
OPB Single Channel HDLC Controller Xilinx, Inc. LogiCORE
FPGAs

OPB SPI Master and Slave Bus Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE
Controller Spartan-IIE FPGAs

OPB System ACE Interface Controller Spartan-3, Spartan-IIE

Xilinx, Inc. LogiCORE
(DO-EDK) FPGAs

OPB TimeBase/WatchDog Timer Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE
(WDT) Spartan-IIE FPGAs

Spartan-3, Spartan-3A,
OPB Timer/Counter Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
OPB to OPB Bridge (Lite Version) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-3A,
OPB UART 16450 Controller Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-3A,
OPB UART Lite Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
OPB ZBT Controller (DO-EDK) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-3A,
OPB/PLB PCI32 Interface Xilinx, Inc. LogiCORE
Spartan-IIE FPGAs

446 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Spartan-3 Generation IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices
Spartan-3, Spartan-IIE
PCI Host Bridge (PCI-HB) CAST, Inc. AllianceCORE X
FPGAs

PIC1655x Fast RISC Microcontroller Spartan-3E, Spartan-3,

Digital Core Design AllianceCORE X
(DFPIC1655X) Spartan-IIE FPGAs

PIC16C55x Fast RISC Microcontroller Spartan-3E, Spartan-3,

Digital Core Design AllianceCORE X
(DRPIC1655X) Spartan-IIE FPGAs

PIC16C6x Fast RISC Microcontroller Spartan-3E, Spartan-3,

Digital Core Design AllianceCORE X
(DRPIC166X) Spartan-IIE FPGAs

PicoBlaze for Spartan-3, Virtex-II, and

Xilinx, Inc. Spartan-3 FPGAs
Virtex-II Pro FPGAs

Spartan-3, Spartan-IIE
PLB ATMC Xilinx, Inc. LogiCORE
FPGAs

PLB Ethernet 10/100 Mbps Media Spartan-3, Spartan-3A,

Xilinx, Inc. LogiCORE
Access Controller (PLB EMAC) Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
PLB to OPB Bridge (DO-EDK) Xilinx, Inc. LogiCORE
FPGAs

Spartan-3, Spartan-IIE
PowerPC Bus Master (EP201) Eureka Technology AllianceCORE X
FPGAs

Spartan-3, Spartan-IIE
PowerPC Bus Slave (EP100) Eureka Technology AllianceCORE X
FPGAs

SD/SDIO/MMC Host Controller ASICS World Service, Spartan-3E, Spartan-3,

Candidate Core X
(SDMMCH1) Ltd. Spartan-IIE FPGAs

Spartan-3, Spartan-IIE
SDRAM Controller, DDR (EP525) Eureka Technology AllianceCORE X
FPGAs

Spartan-3E, Spartan-3,
SDRAM Controller, DDR (NWL) Northwest Logic AllianceCORE X
Spartan-IIE FPGAs

SDRAM Controller, DDR/DDR2 Spartan-3E, Spartan-3

Eureka Technology Candidate Core X
(EP532) FPGAs

Spartan-XL, Spartan-3
SDRAM Controller, DDR2 (NWL) Northwest Logic AllianceCORE X
FPGAs

Spartan-3E, Spartan-3,
SDRAM Controller, SDR (NWL) Northwest Logic AllianceCORE X
Spartan-IIE FPGAs

SDRAM Controller, SDR Spartan-3E, Spartan-3

Array Electronics Candidate Core X
(SDR-XS-XILINX) FPGAs

HCL Technologies Spartan-3, Spartan-IIE

Smart Card Interface (HCLT SCI) Candidate Core X
Ltd. FPGAs
Spartan-3, Spartan-IIE
TDES AMBA Platform SoC Solutions, LLC AllianceCORE X
FPGAs

Spartan-3, Spartan-IIE
TMS32025 DSP Processor (C32025) CAST, Inc. AllianceCORE X
FPGAs

TMS32025TX Digital Signal Processor Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
(C32025TX) FPGAs

Spartan-3 Generation FPGA User Guide www.xilinx.com 447

UG331 (v1.7) August 19, 2010
 R

Chapter 14: Using IP Cores

Table 14-1: Spartan-3 Generation IP Cores Support (Cont’d)

Product Name Vendor Type SignOnce Supported Devices

UART, Generic Compact (MC-XIL-

Memec Design AllianceCORE X Spartan-3 FPGAs
UART)

USB Function Controller with On-Chip Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Peripheral Bus (CUSB_OPB) FPGAs

Spartan-3, Spartan-IIE,
Util Bus Split Operation Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
Util Flop-Flop Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
Util Reduced Logic Xilinx, Inc. LogiCORE
Spartan-II FPGAs

Spartan-3, Spartan-IIE,
Util Vector Logic Xilinx, Inc. LogiCORE
Spartan-II FPGAs

VGA Compatible Cores Spartan-3E, Spartan-3

Intrinsix Corporation Candidate Core
(INT416-SM, INT416-SXGA) FPGAs

Z80 Compatible Programmable Spartan-3, Spartan-IIE

CAST, Inc. AllianceCORE X
Counter/Timer (CZ80CTC) FPGAs

Spartan-3, Spartan-IIE
Z80 Serial I/O Controller (CZ80SIO) CAST, Inc. AllianceCORE X
FPGAs

Related Materials and References

• Xilinx IP Center:
http://www.xilinx.com/ipcenter
• IP Parametric Search
http://www.xilinx.com/products/ipcenter/advancedsearch.htm
• DSP Design Tools
http://www.xilinx.com/ise/dsp_design_prod/index.htm
• Embedded Design Tools
http://www.xilinx.com/ise/embedded_design_prod/index.htm
• IP Updates (Download Center)
http://www.xilinx.com/support/download/index.htm
• CORE Generation System Online Help
http://toolbox.xilinx.com/docsan/xilinx92/help/iseguide/mergedProjects/coregen
/coregen.htm

448 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 15

Embedded Processing and Control

Solutions
Introduction
In a variety of applications, an embedded processor or controller is key to system
flexibility, maintainability, and low cost. Spartan®-3 generation FPGAs support two
powerful yet flexible Field Programmable Controller solutions, shown in Table 15-1. The
PicoBlaze™ processor is a simple, highly efficient 8-bit RISC controller optimized for the
Spartan-3 generation FPGA architecture (see Figure 15-1). The MicroBlaze™ processor is a
powerful, full-featured, high-performance 32-bit RISC processor offering high-level
language and real-time operating system (RTOS) support (see Figure 15-2).
Program Counter

CALL/RETURN

1Kx18 PORT_ID
31x10

Stack

64-Byte
(PC)

Instruction Scratchpad RAM

PROM
OUT_PORT

Flags
Instruction Constants Z Zero
Decoder
C Carry

INTERRUPT
16 Byte-Wide Registers
Enable Operand 1 ALU
IE s0 s1 s2 s3
s4 s5 s6 s7
IN_PORT s8 s9 sA sB
sC sD sE sF
Operand 2

UG331_c17_01_082406

Figure 15-1: PicoBlaze Embedded Microcontroller Block Diagram

Spartan-3 Generation FPGA User Guide www.xilinx.com 449

UG331 (v1.7) August 19, 2010
 R

Chapter 15: Embedded Processing and Control Solutions

Figure 15-2: MicroBlaze Core Block Diagram

Table 15-1: Embedded Processing/Control Solutions for Spartan-3 Generation FPGAs

Function/Feature PicoBlaze Processor MicroBlaze Processor
Processor Architecture 8-bit RISC controller 32-bit RISC CPU
Typical Applications Embedded control, state machines, Embedded computation and control
I/O processing
Memory Architecture Harvard Harvard
(separate data/code data paths) (separate data/code data paths)
ALU/register width 8 bits (byte) 32 bits (word)
Registers 16 byte-wide 32 word-wide
Pipeline Stages 0 3
Code Address Space 512 or 1K instructions 512 to 4G bytes
Code Storage Block RAM (internal) Block RAM (internal)
External memory
Data Address Space 64 bytes (internal) 0 to 4G bytes
Data Storage Distributed RAM (internal) Block RAM (internal)
External memory
I/O Address Space 256 locations N/A
Processor Instructions 57 106
Operands per Instruction 2 3

450 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Introduction

Table 15-1: Embedded Processing/Control Solutions for Spartan-3 Generation FPGAs (Cont’d)
Function/Feature PicoBlaze Processor MicroBlaze Processor
Clocks per Instruction 2 1 to 3, 34 for integer divide
Call/Return/Interrupt Stack 31 locations (internal) Variable size, in data memory
Interrupts 1, Expandable 1, Expandable
Maximum Interrupt Latency 4 clock cycles 7 to 40 clock cycles
(46 ns at maximum clock rate) (application dependent)
Instruction Cache N/A 64 to 64K
Data Cache N/A 64 to 64K
Floating-Point Unit N/A Optional, up to 120X performance
improvement
Hardware Multiplier N/A 32x32 = 32 in 3 cycles
Hardware Divider N/A Optional, up to 20% performance
improvement
Hardware Barrel Shifter N/A Optional, up to 15X performance
improvement
Hardware Debugger Support N/A XMD
LocalLink Direct Processor N/A 200 MB/sec communication
Interface

The PicoBlaze processor is always fully embedded within a Spartan-3 generation FPGA
using on-chip block RAM and distributed RAM for code and data storage. The MicroBlaze
processor optionally uses internal FPGA memory resources or interfaces to external
memory to support larger code or data storage requirements. The Embedded
Development Kit (EDK) for the MicroBlaze processor includes hardware IP cores to
support external Flash, SRAM, SDRAM, DDR DRAM, and ZBT SRAM memory. Similarly,
the MicroBlaze processor supports both instruction and data caches, each up to 64 Kbytes,
to increase performance when connected to external memory.

Table 15-2: PicoBlaze and MicroBlaze Resource Requirements and Performance

Function/Feature PicoBlaze Processor MicroBlaze Processor
Resource Requirements
Slices (4 slices = 1 CLB) 96 525
Block RAMs 0.5 or 1 2+
Effective cost in high-volume From US$0.40 From US$1.40
applications (250Ku)
Spartan-3A DSP FPGAs
Percent of XC3SD1800A 0.5% – 1% 3%+
Percent of XC3SD3400A 0.5% – 1% 2%+
Spartan-3A/3AN FPGAs
Percent of XC3S50A/AN 17%-33% 75%+

Spartan-3 Generation FPGA User Guide www.xilinx.com 451

UG331 (v1.7) August 19, 2010
 R

Chapter 15: Embedded Processing and Control Solutions

Table 15-2: PicoBlaze and MicroBlaze Resource Requirements and Performance (Cont’d)
Function/Feature PicoBlaze Processor MicroBlaze Processor
Percent of XC3S200A/AN 5%-6% 29%+
Percent of XC3S400A/AN 3%-5% 15%+
Percent of XC3S700A/AN 3%-5% 10%+
Percent of XC3S1400A/AN 2%-3% 6%+
Spartan-3E FPGAs
Percent of XC3S100E 10%-25% 55%+
Percent of XC3S250E 4%-8% 21%+
Percent of XC3S500E 3%-5% 11%+
Percent of XC3S1200E 2%-4% 7%+
Percent of XC3S1600E 2%-3% 6%+
Spartan-3 FPGAs
Percent of XC3S50 13% – 25% 68%+
Percent of XC3S200 5% – 8% 27%+
Percent of XC3S400 3% – 6% 15%+
Percent of XC3S1000 2% – 4% 8%+
Percent of XC3S1500 2% – 3% 6%+
Percent of XC3S2000 1.3% – 3% 5%+
Percent of XC3S4000 0.5% – 1% 2%+
Percent of XC3S5000 0.5% – 1% 2%+
Performance (Spartan-3 FPGA –5 speed grade)
Maximum clock frequency 87 MHz 100 MHz
Instructions per second 43.5M 92M
Dhrystone MIPS (D-MIPS) N/A 92

Using Spartan-3 generation FPGAs, both MicroBlaze and PicoBlaze processors consume
minimal FPGA resources and are highly cost effective, as shown in Table 15-2. Complete
PicoBlaze solutions cost as little as $0.40 in high-volume applications. MicroBlaze solutions
start from $1.40 in volume.
Both the MicroBlaze and PicoBlaze processors provide significant numbers of flexible I/O
at much lower cost than off-the-shelf controllers. Similarly, the peripheral set for both
processors can be customized to meet the specific feature, function, and cost requirements
of the target application. Because both processors are delivered in synthesizable HDL, both
cores are future-proof, safe from any possible product obsolescence. Being integrated into
the FPGA, both processors reduce board space, design cost, and inventory.

452 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

PicoBlaze Application Development Support

PicoBlaze Application Development Support

The PicoBlaze processor solution is a simple 8-bit RISC controller with an easy-to-use
assembler. The PicoBlaze core has no direct support for in-system debugging although it
can be debugged using the standard Xilinx JTAG-based interface. A simple instruction-set
simulator is available.
The PicoBlaze reference design also includes UART transmitter and receiver macros with
integrated 16-byte FIFOs. The UART supports 8-bit data, no parity, with one stop bit.

MicroBlaze Application Development Support

The MicroBlaze processor offers complete application development support, including a
full suite of software development tools, an IP library of processor hardware peripheral
functions, plus in-circuit hardware debugger/emulation support.

Embedded Development Kit (EDK)

The Embedded Development Kit (EDK) is an all-encompassing solution for creating
embedded programmable systems design. The EDK includes and supports the MicroBlaze
soft processor core.

Xilinx Platform Studio (XPS)

• Tools for editing software; creating hardware and software platforms
• Runs library generation, and compiler tool chains; generates implementation and
simulation netlists for use with ISE® Logic Design Tools

GNU Software Development Tools

• C/C++ compiler for MicroBlaze cores (GNU gcc)
• Debugger for MicroBlaze cores (GNU gdb)
• Other GNU utilities

Hardware/Software Development Tools

• XMD - Xilinx Microprocessor Debug engine for MicroBlaze cores
• System ACE™ tools
• Data2MEM – Updates internal block RAM contents without recompiling the FPGA
design

Board Support Packages (BSPs)

• Stand Alone BSP - For non-RTOS systems (MicroBlaze cores)

Operating Systems
Many embedded processing applications require operating system capabilities. The
following operating systems and real-time operating systems (RTOS) have ports to the
MicroBlaze processor.
• Micriμm μC/OS-II Real-Time Operating System
http://www.micrium.com/products/rtos/kernel/rtos.html

Spartan-3 Generation FPGA User Guide www.xilinx.com 453

UG331 (v1.7) August 19, 2010
 R

Chapter 15: Embedded Processing and Control Solutions

• μClinux Operating System

http://www.uclinux.org
http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux/
• ATI Nucleus Real-Time Operating System
http://www.mentor.com/nucleus
• Xilkernel Libraries
• Highly modular scheduler, network stack, and file system
• Minimal resource requirements and footprint size
• Royalty-free license included with EDK purchase
• Fully supported by Xilinx

Processor Peripheral IP Functions

The EDK includes the following processor IP cores that support the MicroBlaze processor.
The IP cores also include device drivers and RTOS adaptation layers. Add one or more IP
cores to create a custom processor to meet specific application requirements.

Processor Peripherals
• Timer/Counter
• Timebase/Watchdog Timer
• UART-Lite
• Interrupt Controller
• General-Purpose I/O port (GPIO)

Serial I/O
• SPI Master and Slave
• JTAG UART
• 16450 UART*
• 16550 UART*
• I2C two-wire serial Master and Slave*

Memory Interfaces
• SDRAM controller and interface
• DDR SDRAM controller and interface
• Flash memory interface
• SRAM memory interface
• Block RAM interface

Networking Interfaces
• Single-channel HDLC controller*
• ATM Utopia L2 master and slave controller*
• 10/100 Ethernet Media Access Controller (MAC)* (Full and Lite versions)
* IP core available as a separate product. Plugs into EDK. Evaluation versions available.

454 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

In-Circuit Hardware Debugger Support

• EDK Software Debugger
• Requires MicroBlaze Hardware Debug Module
• Connects via FPGA JTAG port using Xilinx Parallel Cable IV

Related Materials and References

• Xilinx Embedded Processing
http://www.xilinx.com/products/design_resources/proc_central/index.htm
• MicroBlaze 32-bit RISC Processor
http://www.xilinx.com/microblaze
• PicoBlaze 8-bit RISC Controller
http://www.xilinx.com/picoblaze
• Platform Studio and the Embedded Development Kit (EDK)
http://www.xilinx.com/ise/embedded_design_prod/platform_studio.htm
• Spartan-3E FPGA MicroBlaze Development Kit
http://www.xilinx.com/products/devkits/DO-SP3E1600E-DK-UNI-G.htm
• Embedded Systems Development Training Course
http://www.xilinx.com/support/training/abstracts/embedded-systems.htm

Spartan-3 Generation FPGA User Guide www.xilinx.com 455

UG331 (v1.7) August 19, 2010
 R

Chapter 15: Embedded Processing and Control Solutions

456 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Section III: PCB Design Considerations

“Packages and Pinouts”

“Package Drawings”

“Powering Spartan-3 Generation FPGAs”

“Power Management Solutions”

“Using IBIS Models”

“Using Boundary-Scan and BSDL Files”

Spartan-3 Generation FPGA User Guide www.xilinx.com 457

UG331 (v1.7) August 19, 2010
 R

Chapter :

458 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 16

Packages and Pinouts

Summary
This section describes the various pin functions on Spartan®-3 generation FPGAs and
general information about the packages. For specific information on connections within
the supported component packages, see Module 4 of the individual data sheets. For
package drawings, see Chapter 17, “Package Drawings.”

Differences in Pinouts Between Spartan-3 Generation FPGAs

The Spartan-3A and Spartan-3AN platforms have identical pinouts for common
part/package offerings, and there can be upward migration to the XC3SD1800A FG676 in
the Spartan-3A DSP platform. See the data sheets for specific part/package offerings and
information on pinout compatibility.
There are significant differences in pinouts between the Extended Spartan-3A family,
Spartan-3E, and Spartan-3 families. A device from one of those three groups cannot be
used as a drop-in replacement for a device in another group because of the functional
differences. Each group has a pinout optimized to its own unique features.
Within a family, however, there is pin compatibility, allowing the user to easily migrate to
a larger density or optimize to a smaller device as required. Minor differences between
densities might exist in a given package, and they are shown in the pinout tables found in
Module 4 of the data sheet for each family.
There are also some differences in pin naming between the families due to improved
naming conventions and improvements made in functionality. Table 16-1 shows the pin
naming differences.

Table 16-1: Differences in Pin Names Between Families

Extended Spartan-3A Spartan-3E Spartan-3
Family FPGAs FPGAs FPGAs
Configuration
Pull-Ups During IO/PUDC_B IO/HSWAP HSWAP_EN
Configuration
Low During Configuration IO/LDC2-LDC0 IO/LDC2-LDC0 N/A
High During Configuration IO/HDC IO/HDC N/A
SPI Variant Select IO/V2-V0 IO/V2-V0 N/A
SPI Output IO/MOSI MOSI N/A
Chip Select Input IO/CSI_B IO/CSI_B IO/CS_B

Spartan-3 Generation FPGA User Guide www.xilinx.com 459

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

Table 16-1: Differences in Pin Names Between Families (Cont’d)

Extended Spartan-3A Spartan-3E Spartan-3
Family FPGAs FPGAs FPGAs
Chip Select Output IO/CSO_B IO/CSO_B N/A
Parallel Address IO/A25-A0 IO/A23-A0 N/A
Parallel Status N/A IO/BUSY IO/BUSY
Parallel Read/Write IO/RDWR_B IO/RDWR_B/ IO/RDWR_B
GCLK0
Configuration Clock IO/CCLK IO/CCLK CCLK
Mode IO/M2-M0 IO/M2/GCLK1, M2-M0
IO/M1, IO/M0
User Function
Suspend SUSPEND N/A N/A
Awake IO/AWAKE N/A N/A
Input-Only IP_# IP N/A
Global Clock IO/GCLK15-GCLK0 IO/GCLK15- IO/GCLK7-
GCLK0/D7-D1 GCLK0
Left Half Clock IO/LHCLK7-LHCLK0 IO/LHCLK7- N/A
LHCLK0
Right Half Clock IO/RHCLK7-RHCLK0 IO/RHCLK7- N/A
RHCLK0/A10-A3
DCI Reference Inputs N/A N/A IO/VRN,
IO/VRP

Notes:
1. # = I/O bank number, an integer between 0 and 3 (7 for the Spartan-3 family).

460 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Pin Types

Pin Types
Most pins on a Spartan-3 generation FPGA are general-purpose, user-defined I/O pins.
There are, however, up to 12 different functional types of pins on Spartan-3 generation
packages, as outlined in Table 16-2. The color coding is used in the package footprint
drawings that are found in Module 4 of each data sheet.

Table 16-2: Types of Pins on Spartan-3 Generation FPGAs

Type / Color
Description Pin Name(s) in Type
Code
I/O Unrestricted, general-purpose user-I/O pin. Most pins can be IO_#
paired together to form differential I/Os. IO_Lxxy_#
INPUT Unrestricted, general-purpose input-only pin. This pin does not IP_#
have an output structure. IP_Lxxy_#
DUAL Dual-purpose pin used in some configuration modes during the M[2:0]
configuration process and then usually available as a user I/O PUDC_B
after configuration. If the pin is not used during configuration, (Extended Spartan-3A family
this pin behaves as an I/O-type pin. See UG332: Spartan-3 FPGA)
Generation Configuration User Guide for additional information on HSWAP (Spartan-3E FPGA)
these signals. HSWAP_EN (Spartan-3 FPGA)
CCLK
MOSI/CSI_B
D[7:1]
D0/DIN
CSO_B
RDWR_B
BUSY/DOUT
INIT_B
A[25:0]
VS[2:0]
LDC[2:0]
HDC
VREF Dual-purpose pin that is either a user-I/O pin or Input-only pin, IP/VREF_#
or, along with all other VREF pins in the same bank, provides a IP_Lxx_#/VREF_#
reference voltage input for certain I/O standards. If used for a IO/VREF_#
reference voltage within a bank, all VREF pins within the bank IO_Lxx_#/VREF_#
must be connected.
CLK Either a user-I/O pin or an input to a specific clock buffer driver. IO_Lxx_#/GCLK[15:0],
Every package has 16 global clock inputs that optionally clock the IO_Lxx_#/LHCLK[7:0],
entire device. The RHCLK inputs optionally clock the right-half of IO_Lxx_#/RHCLK[7:0]
the device. The LHCLK inputs optionally clock the left-half of the
device. See Chapter 2, “Using Global Clock Resources,” for
additional information on these signals.
CONFIG Dedicated configuration pin. Not available as a user-I/O pin. DONE, PROG_B
Every package has three dedicated configuration pins. These pins
are powered by VCCAUX. See UG332: Spartan-3 Generation
Configuration User Guide for additional information on these
signals.
PWRMGMT Control and status pins for the Extended Spartan-3A family SUSPEND, AWAKE
power-saving Suspend mode. SUSPEND is a dedicated pin.
AWAKE is a Dual-Purpose pin and an I/O if Suspend mode is not
enabled.

Spartan-3 Generation FPGA User Guide www.xilinx.com 461

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

Table 16-2: Types of Pins on Spartan-3 Generation FPGAs (Cont’d)

Type / Color
Description Pin Name(s) in Type
Code
JTAG Dedicated JTAG pin. Not available as a user-I/O pin. Every TDI, TMS, TCK, TDO
package has four dedicated JTAG pins. These pins are powered by
VCCAUX.
GND Dedicated ground pin. The number of GND pins depends on the GND
package used. All must be connected.
VCCAUX Dedicated auxiliary power supply pin. The number of VCCAUX VCCAUX
pins depends on the package used. All must be connected. See
Chapter 18, “Powering Spartan-3 Generation FPGAs,” for
additional information on this signal. VCCAUX is 2.5V in Spartan-3
and Spartan-3E families. Extended Spartan-3A family devices can
have VCCAUX at either 2.5V or 3.3V and the user should set
CONFIG VCCAUX = 2.5 or 3.3. The value should be 3.3V in the
Spartan-3AN platform when using the In-System Flash memory.
VCCINT Dedicated internal core logic power supply pin. The number of VCCINT
VCCINT pins depends on the package used. All must be connected
to +1.2V. See Chapter 18, “Powering Spartan-3 Generation
FPGAs,” for additional information on this signal.
VCCO Along with all the other VCCO pins in the same bank, this pin VCCO_#
supplies power to the output buffers within the I/O bank and sets
the input threshold voltage for some I/O standards. See
Chapter 18, “Powering Spartan-3 Generation FPGAs,” for
additional information on these signals.
N.C. This package pin is not connected in this specific device/package N.C.
combination but might be connected in larger devices in the same
package.

Notes:
1. # = I/O bank number, an integer between 0 and 3 (7 for the Spartan-3 family).

Pin Labeling
The pin label is abbreviated but descriptive for each pin. All I/O pins begin with IO, while
the input-only pins begin with IP. If a pin can be used as a differential signal, the name
includes an L followed by the pair number and the bank number (see “Differential Pair
Labeling”). Bank numbers are also indicated on single-ended pins and on the voltage
inputs that are bank-specific, VCCO and VREF . Dual-purpose pins have a forward slash
separating the two functions. _B is used as the active-Low designator, as in CSI_B.

Differential Pair Labeling

A pin supports differential standards if the pin is labeled in the format Lxxy_#. The pin
name suffix has the following significance. Figure 16-1 provides a specific example
showing a differential input to and a differential output from Bank 1.
L indicates that the pin is part of a differentiaL pair.
xx is a two-digit integer, unique for each bank, that identifies a differential pin-pair.
y is replaced by P for the true signal or N for the inverted. These two pins form one
differential pin-pair.

462 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Pin Types

# is an integer, 0 through 3 (7 in the Spartan-3 family), indicating the associated I/O

bank.

Pair Number

Bank 0 Bank Number

IO_L38P_1
IO_L38N_1

Bank 1
Positive Polarity,
Bank 3 True Driver
IO_L39P_1
Spartan-3E
FPGA IO_L39N_1
Negative Polarity,
Inverted Driver
Bank 2 DS312-4_00_111105

Figure 16-1: Differential Pair Labeling

Pinout Files
The pinouts are found in the data sheets for each family. Comma-delimited text files and
Excel graphical footprints for the pinouts specific to each Spartan-3 FPGA family are
available from the data sheets on xilinx.com. Using a spreadsheet program with the
comma-delimited CSV files, the data can be sorted and reformatted according to any
specific needs. Similarly, the ASCII text file is easily parsed by most scripting programs.
For information on how to use these files to create OrCAD symbols, see Answer Record
10078 at http://www.xilinx.com/support/answers/10078.htm.
The following subsections provide additional details on using the downloadable pinout
files.

Pinout Tables
The comma-delimited ASCII text files located in the /tables directory list pinout
information for a specific package type. Each line represents one pin on the package.
Pinout information for all devices available in the package for the family appears on the
line. This subsection provides brief descriptions of the fields available on each line.

SORT_PIN (QFP Packages Only)

Sorting by SORT_PIN orders the pins sequentially on the quad flat pack style packages,
such as VQ100, TQ144, and PQ208.

SORT_ROW (BGA Packages Only)

Sorting by SORT_ROW orders the pins alphabetically on the ball grid array packages.
Sorting by SORT_ROW is sufficient for the smaller BGA packages. However, the larger
BGA packages have ROW indices such as AA and AB. An additional field, called
SORT_ROW_#, is provided on the large BGA packages to aid sorting:

SORT_ROW_# (Larger BGA Packages Only)

The SORT_ROW_# field is similar to SORT_ROW, except that SORT_ROW_# is an integer
value instead of an alphabetic value and is used for sorting pins on the BGA package.

Spartan-3 Generation FPGA User Guide www.xilinx.com 463

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

SORT_COLUMN (BGA Packages Only)

SORT_COLUMN is an integer value indicating the column number of the pin on a BGA
package.

PIN_NUMBER
PIN_NUMBER is the pin identifier for each pin on the package.
For a particular package and family, there can be multiple FPGAs available in that package.
For each pin, all possible FPGAs are listed. Each device is represented by two fields on each
line, XC3S**_PIN and XC3S**_TYPE, described below.

XC3S**_PIN
The XC3S**_PIN field indicates the name for a particular package pin and for a particular
Spartan-3 generation FPGA in that package. The ** characters here indicate a wildcard
character. In the pinout table file, the ** characters are replaced by an actual part number,
such as XC3S250E.

XC3S**_TYPE
The XC3S**_TYPE field indicates the pin type for a particular package pin and for a
particular Spartan-3 generation FPGA in that package. The listed type matches those
described in Module 4 of the data sheet. The ** characters here indicate a wildcard
character. In the pinout table file, the ** characters are replaced by an actual part number,
such as XC3S250E.

BANK
Sorting by BANK orders the pins by their associated I/O bank. The possible values for
BANK include integers between 0 and 3 (0 and 7 for the Spartan-3 family), VCCAUX, and
N/A. N/A indicates that the pin is not associated with a specific bank.

DIFFERENCE
Sorting by DIFFERENCE in descending order highlights any pinout differences between
Spartan-3E FPGAs in the same package. A period (.) indicates that the pins match
identically. DIFF indicates that the pins are different between packages.
To locate unconnected pins in a package type, sort by the TYPE of the smallest device
offered in the package footprint. Any unconnected pins on larger devices are a subset of
those on the smallest device.

Footprint Diagrams
The files in the \footprints directory are all Microsoft Excel spreadsheet files. These
files present a common footprint for each package type and show the pins on the package
as viewed from the top (QFP packages) or through the top of the package (BGA packages).
Note the location of the pin 1 indicator on QFP packages.
Each pin is labeled and color -coded according to Module 4 of the data sheet. No Connect
(N.C.) pins are also indicated with special symbols.
Most footprints were saved as 50% to 75% of normal size so that the entire footprint is
visible on the screen. To change the magnification, select View --> Zoom from the Excel top
menu, then select the desired magnification factor.

464 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

PlanAhead Design Analysis Tool

Excel might issue a warning when you open the file, indicating that the file might contain
macros. Select either Disable Macros or Enable Macros. There are no active macros in the
Excel files.

PartGen
The pinout files can also be generated from the Xilinx ISE® development system by using
the PartGen program. To create pinout files in PartGen, go to a command prompt and type,
for example, partgen -p xc3s50atq144. That command writes a text file called
xc3s50atq144.pkg to the current directory, which contains a list of the pin names and
pin numbers. Using the -v option (verbose) generates a more detailed .pkg file that
includes the bank number and nearest CLB, among other information. Details on using
PartGen are found in the PartGen chapter of the Development System Reference Guide, found
at: http://www.xilinx.com/support/documentation/dt_ise.htm.

ISE Development System Pin Assignment Reports

The ISE development system can also be used to list or view the pinout for a design,
showing the actual placement of signals resulting from Place & Route.
The Place and Route (PAR) program generates three reports showing the actual pin
assignments:
• a PAD file, containing I/O pin assignments in a parsable database form
• a CSV file, containing I/O pin assignments in a format supported by spreadsheet
programs, delimited by the “|” character
• a TXT file, containing I/O pin assignments in an ASCII text version for viewing in a
text editor

PlanAhead Design Analysis Tool

The PlanAhead™ tool streamlines the design step between synthesis and place and route,
allowing you to divide a larger design up into smaller, more manageable blocks and focus
efforts toward optimization of each module. This methodology results in improved
performance and quality of the entire design. PlanAhead includes PinAhead Technology
to help users better deal with the complexities of pin assignments. PinAhead offers an
environment for fully automatic or semi-automated assignment of I/O ports to physical
package pins. All ISE® Design Suite configurations now include PlanAhead Lite,
providing the I/O pin planning capabilities of the PinAhead technology. It also includes
design analysis and floorplanning capabilities as well as implementation control with the
ExploreAhead environment. For more information, see
http://www.xilinx.com/planahead.

Spartan-3 Generation FPGA User Guide www.xilinx.com 465

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

Packages
Table 16-3 shows the low-cost, space-saving production package styles for the
Spartan-3 generation families.

Table 16-3: Spartan-3 Generation Package Options

Lead
Spartan-3A Spartan-3AN Spartan-3A Spartan-3E Spartan-3 Max Footprint Height Mass(1)
Package Leads Type Pitch
DSP FPGAs FPGA FPGA FPGA FPGA I/O Area (mm) (mm) (g)
(mm)

X X X VQ100/ 100 Very Thin Quad 66 0.5 16 x 16 1.20 0.6

VQG100 Flat Pack (TQFP)

X CP132/ 132 Chip-Scale Ball 92 0.5 8x8 1.10 0.1

CPG132 Grid Array (CS)

X X X X TQ144/ 144 Thin Quad Flat 108 0.5 22 x 22 1.60 1.4

TQG144 Pack (TQFP)

X X PQ208/ 208 Quad Flat Pack 158 0.5 30.6 x 4.10 5.3
PQG208 (QFP) 30.6

X X X X FT256/ 256 Fine-pitch, Thin 195 1.0 17 x 17 1.55 0.9

FTG256 Ball Grid Array
(FBGA)

X X X FG320/ 320 Fine-pitch Ball 251 1.0 19 x 19 2.00 1.4

FGG320 Grid Array
(FBGA)

X X X FG400/ 400 Fine-pitch Ball 311 1.0 21 x 21 2.43 2.2

FGG400 Grid Array
(FBGA)

X FG456/ 456 Fine-pitch Ball 333 1.0 23 x 23 2.60 2.2

FGG456 Grid Array
(FBGA)

X X X FG484/ 484 Fine-pitch Ball 376 1.0 23 x 23 2.60 2.2

FGG484 Grid Array
(FBGA)

X CS484/ 484 Chip-Scale Ball 309 0.8 19 x 19 1.80 1.4

CSG484 Grid Array (CS)

X X X X FG676/ 676 Fine-pitch Ball 519 1.0 27 x 27 2.60 3.4

FGG676 Grid Array
(FBGA)

X FG900/ 900 Fine-pitch Ball 633 1.0 31 x 31 2.60 4.2

FGG900 Grid Array
(FBGA)

Notes:
1. Package mass is ±10%.

Pb-Free Packages
Each package style is available as a standard and an environmentally friendly lead-free
(Pb-free) option. The Pb-free packages include an extra G in the package style name. For
example, the standard TQ144 package becomes TQG144 when ordered as the Pb-free
option. The mechanical dimensions of the standard and Pb-free packages are similar, as
shown in the mechanical drawings provided in Table 17-1, page 471. The materials listed
in the Material Declaration Data Sheet and the thermal characteristics will be different. The

466 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Packages

pinouts are always identical between the standard and Pb-free packages. For more
information on Pb-free packages, see Xilinx Pb-Free and RoHS-compliant Products.

Differences in Packages Between Spartan-3 Generation Families

Not all Spartan-3 generation devices are available in all packages. For a specific package,
however, there is a common footprint within the associated family that supports all the
devices available in that package for the family. See the footprint diagrams in the data
sheets for details and for any exceptions. There is no footprint compatibility between
families.
For additional package information, see UG112: Device Package User Guide.

Selecting the Right Package Option

Spartan-3 generation FPGAs are available in both quad-flat pack (QFP) and ball grid array
(BGA) packaging options. While QFP packaging offers the lowest absolute cost, the BGA
packages are superior in almost every other aspect, as summarized in Table 16-4.
Consequently, Xilinx recommends using BGA packaging whenever possible.

Table 16-4: QFP and BGA Comparison

Quad Flat Pack Ball Grid Array
Characteristic
(QFP) (BGA)
Maximum User I/O 158 633
Packing Density (Logic/Area) Good Better
Signal Integrity Fair Better
Simultaneous Switching Output (SSO) Support Fair Better
Thermal Dissipation Fair Better
Minimum Printed Circuit Board (PCB) Layers 4 4-6
Hand Assembly/Rework Possible Difficult

Package Thermal Characteristics

The power dissipated by an FPGA application has implications on package selection and
system design. The power consumed by a Spartan-3 generation FPGA is reported using
either the XPower Estimator worksheet or the XPower Analyzer integrated in the Xilinx
ISE development software. For more information on these tools, see the Power Solutions
page.
The power is then combined with the thermal resistance to calculate the resulting
temperature. Module 4 of each family’s data sheet provides the thermal characteristics for
the specific package offerings within each family. The Package Thermal Data Query tool
allows the user to use the Device Family (for example, Spartan-3E family, Device name (for
example, XC3S100E), and Package code (for example, FT256) to obtain product-specific
Thermal Data. Three types of thermal resistance are provided as shown in Table 16-5. All
resistance values are relative to the die junction temperature and are measured in °C per
watt.

Spartan-3 Generation FPGA User Guide www.xilinx.com 467

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

Table 16-5: Thermal Characteristics

Symbol Characteristic Description
Temperature difference between ambient environment and
junction. Drops with increasing air flow. Specified at air flow
θ JA Junction to Ambient
rates of 0 (still air), 250, 500, and 750 Linear Feet Per Minute
(LFPM)
θJB Junction to Board Temperature difference between board and junction.
Temperature difference between the package body (case) and
θJC Junction to Case
junction.

Related Materials and References

The following list provides additional information related to pinouts and packages:
• Chapter 17, “Package Drawings”
Mechanical drawings for each package and links to the Material Declaration Data
Sheets for each package.
• DS610, Spartan-3A DSP Pinouts
Module 4 contains detailed information on the pinouts specific to the Spartan-3A DSP
FPGAs.
• Spartan-3A DSP ASCII Pinouts and Excel Footprints
Comma-delimited text files and Excel graphical footprints for the pinouts specific to
the Spartan-3A DSP FPGAs.
• DS557, Spartan-3AN Pinouts
Module 4 contains detailed information on the pinouts specific to the Spartan-3AN
FPGAs.
• DS529, Spartan-3A Pinouts
Module 4 contains detailed information on the pinouts specific to the Spartan-3A
FPGAs.
• Spartan-3A/3AN ASCII Pinouts and Excel Footprints
Comma-delimited text files and graphical footprints for the pinouts specific to the
Spartan-3A/3AN FPGAs.
• DS312, Spartan-3E Pinouts
Module 4 contains detailed information on the pinouts specific to the Spartan-3E
FPGAs.
• Spartan-3E ASCII Pinouts and Excel Footprints
Comma delimited text files and Excel graphical footprints for the pinouts specific to
the Spartan-3E FPGAs
• DS099, Spartan-3 Pinouts
Module 4 contains detailed information on the pinout specific to the Spartan-3 FPGAs.
• Spartan-3 ASCII Pinouts and Excel Footprints
Comma-delimited text files and Excel graphical footprints for the pinouts specific to
the Spartan-3 FPGAs.
• UG112, Device Package User Guide

468 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

Description and specifications for packages, pack and ship, thermal characteristics,
electrical characteristics, PCB design rules, moisture sensitivity, and reflow soldering
guidelines.
• Device Packaging Application Notes
Application notes on board routability, solder reflow, and related topics.

Spartan-3 Generation FPGA User Guide www.xilinx.com 469

UG331 (v1.7) August 19, 2010
 R

Chapter 16: Packages and Pinouts

470 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 17

Package Drawings
Summary
This chapter provides mechanical drawings of the Spartan®-3 generation packages listed
in Table 17-1. These drawings are also available on the Xilinx website at
http://www.xilinx.com/support/documentation/package_specifications.htm. Also
found on the Xilinx website is the Material Declaration Data Sheet for the standard and Pb-
free versions of each package.

Table 17-1: Spartan-3 Generation Mechanical Drawings

Package Drawing
MDDS URL
Spartan-3A DSP Spartan-3AN Spartan-3A Spartan-3E Spartan-3 Xilinx.com URL (http://www.xilinx.com/
Package UG331 (http://www.xilinx.com/ support/documentatio
FPGA FPGA FPGA FPGA FPGA
Page support/documentation n/package_specs/)
/package_specs/)
VQ100 √ √ √ pk173_vq100.pdf
page 473 vq100.pdf
VQG100 √ √ √ pk130_vqg100.pdf
CP132(1) √ pk147_cp132.pdf
page 474 cp132.pdf
CPG132(1) √ pk101_cpg132.pdf
TQ144 √ √ √ pk169_tq144.pdf
page 475 tq144.pdf
TQG144 √ √ √ √ pk126_tqg144.pdf
PQ208 √ √ pk166_pq208.pdf
page 476 pq208.pdf
PQG208 √ √ pk123_pqg208.pdf
FT256 √ √ √ pk158_ft256.pdf
page 477 ft256.pdf
FTG256 √ √ √ √ pk115_ftg256.pdf
FG320 √ √ √ pk152_fg320.pdf
page 478 fg320.pdf
FGG320 √ √ √ pk106_fgg320.pdf
FG400 √ √ pk182_fg400.pdf
page 479 fg400.pdf
FGG400 √ √ √ pk108_fgg400.pdf
FG456 √ pk154_fg456.pdf
page 480 fg456.pdf
FGG456 √ pk109_fgg456.pdf
FG484 √ √ pk183_fg484.pdf
page 481 fg484.pdf
FGG484 √ √ √ pk110_fgg484.pdf
CS484 √ pk230_cs484.pdf
page 482 cs484.pdf
CSG484 √ pk231_csg484.pdf
FG676 √ √ √ pk155_fg676.pdf
page 483 fg676.pdf
FGG676 √ √ √ √ pk111_fgg676.pdf
FG900 √ pk186_fg900.pdf
page 484 fg900.pdf
FGG900 √ pk114_fgg900.pdf

Spartan-3 Generation FPGA User Guide www.xilinx.com 471

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

Table 17-1: Spartan-3 Generation Mechanical Drawings

Package Drawing
MDDS URL
Spartan-3A DSP Spartan-3AN Spartan-3A Spartan-3E Spartan-3 Xilinx.com URL (http://www.xilinx.com/
Package UG331 (http://www.xilinx.com/ support/documentatio
FPGA FPGA FPGA FPGA FPGA
Page support/documentation n/package_specs/)
/package_specs/)
FG1156(2) √ N/A
page 485 fg1156.pdf
FGG1156(2) √ N/A

Notes:
1. The CP132 and CPG132 packages are being discontinued and are not recommended for new designs. See
http://www.xilinx.com/support/documentation/customer_notices/xcn08011.pdf for details.
2. The FG1156 and FGG1156 packages are being discontinued and are not recommended for new designs. See
http://www.xilinx.com/support/documentation/customer_notices/xcn07022.pdf for details.

472 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

VQ100/VQG100 Very Thin QFP Package (pk012)

VQ100/VQG100 Very Thin QFP Package (pk012)

Figure 17-1: VQ100/VQG100 Very Thin QFP Package (pk012)

Spartan-3 Generation FPGA User Guide www.xilinx.com 473

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

CP132/CPG132 Chip Scale BGA Package (pk500)

Figure 17-2: CP132/CPG132 Chip Scale BGA Package (pk500)

474 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

TQ144/TQG144 Thin QFP Package (pk009)

TQ144/TQG144 Thin QFP Package (pk009)

Figure 17-3: TQ144/TQG144 Thin QFP Package (pk009)

Spartan-3 Generation FPGA User Guide www.xilinx.com 475

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

PQ208/PQG208 QFP Package (pk007)

Figure 17-4: PQ208/PQG208 QFP Package (pk007)

476 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

FT256/FTG256 Fine-Pitch Thin BGA Package (pk053)

FT256/FTG256 Fine-Pitch Thin BGA Package (pk053)

Figure 17-5: FT256/FTG256 Fine-Pitch Thin BGA Package (pk053)

Spartan-3 Generation FPGA User Guide www.xilinx.com 477

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

FG320/FGG320 Fine-Pitch BGA Package (pk071)

Figure 17-6: FG320/FGG320 Fine-Pitch BGA Package (pk071)

478 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

FG400/FGG400 Fine-Pitch BGA Package (pk083)

FG400/FGG400 Fine-Pitch BGA Package (pk083)

Figure 17-7: FG400/FGG400 Fine-Pitch BGA Package (pk083)

Spartan-3 Generation FPGA User Guide www.xilinx.com 479

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

FG456/FGG456 Fine-Pitch BGA Package (pk034)

Figure 17-8: FG456/FGG456 Fine-Pitch BGA Package (pk034)

480 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

FG484/FGG484 Fine-Pitch BGA Package (pk081)

FG484/FGG484 Fine-Pitch BGA Package (pk081)

Figure 17-9: FG484/FGG484 Fine-Pitch BGA Package (pk081)

Note: The Spartan-3 generation FPGAs use the 4-layer version of the FG484 package.

Spartan-3 Generation FPGA User Guide www.xilinx.com 481

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

CS484/CSG484 Chip-Scale BGA Package (pk223)

Figure 17-10: CS484/CSG484 Fine-Pitch BGA Package (pk223)

482 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

FG676/FGG676 Fine-Pitch BGA Package (pk035)

FG676/FGG676 Fine-Pitch BGA Package (pk035)

Figure 17-11: FG676/FGG676 Fine-Pitch BGA Package (pk035)

Spartan-3 Generation FPGA User Guide www.xilinx.com 483

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

FG900/FGG900 Fine-Pitch BGA Package (pk038)

UG331_c17_12_111809

Figure 17-12: FG900/FGG900 Fine-Pitch BGA Package (pk038)

484 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

FG1156/FGG1156 Fine-Pitch BGA Package (pk039)

FG1156/FGG1156 Fine-Pitch BGA Package (pk039)

Figure 17-13: FG1156/FGG1156 Fine-Pitch BGA Package (pk039)

Spartan-3 Generation FPGA User Guide www.xilinx.com 485

UG331 (v1.7) August 19, 2010
 R

Chapter 17: Package Drawings

486 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 18

Powering Spartan-3 Generation FPGAs

Introduction
FPGA designers are faced with a unique task when it comes to designing power supplies
and power distribution systems. Most other ICs have very specific requirements. Because
FPGAs can implement a countless number of applications at undetermined frequencies
and in multiple clock domains, and have multiple selectable power supplies, it is
important to carefully determine the requirements for a specific application and design to
those requirements.

Differences between Spartan-3 Generation Families

The Spartan®-3, Spartan-3E, and Extended Spartan-3A families share the same 90 nm
process technology and core VCCINT voltage of 1.2V, and have similar architectures.
Therefore they have similar power consumption characteristics. Unlike the other families,
the Spartan-3A and Spartan-3A DSP platforms allow VCCAUX to be 3.3V (the user should
set the CONFIG VCCAUX constraint to match the value used). In the Spartan-3AN
platform, VCCAUX must be set to 3.3V. Using a VCCAUX of 3.3V can eliminate the 2.5V
power rail by providing full 3.3V compliance and by being compliant to all aspects of hot-
swap applications. Table 18-1 highlights significant differences between the families.

Table 18-1: Differences between Spartan-3 Generation Families

Spartan-3AN Spartan-3A/3A DSP Spartan-3E Spartan-3
Feature
FPGA FPGA FPGA FPGA
VCCINT 1.2V 1.2V 1.2V 1.2V
VCCAUX 3.3V 2.5V or 3.3V 2.5V 2.5V
VCCO 1.2V to 3.3V + 10% 1.2V to 3.3V + 10% 1.2V to 3.3V + 5% 1.2V to 3.3V + 5%
VIN Max Recommended 4.6V 4.6V VCCO + 0.5V 3.75V or VCCO + 0.3V
Hot Swap Full Support Full Support Sequenced Connector Sequenced Connector
VCCO Banks 4 4 4 4-8
VCCINT, VCCAUX, VCCINT, VCCAUX, VCCINT, VCCAUX, VCCINT, VCCAUX,
Power-On Reset Inputs
VCCO Bank 2 VCCO Bank 2 VCCO Bank 2 VCCO Bank 4
Power-Off Mode with VCCO can be VCCO can be VCCO must be VCCO must be
Active Inputs removed removed maintained maintained
Suspend Mode Supported Supported N/A N/A

Specific requirements for power supplies, including ramp rates and quiescent current, are
different for each family and are specified in the FPGA data sheets. Dynamic power

Spartan-3 Generation FPGA User Guide www.xilinx.com 487

UG331 (v1.7) August 19, 2010
 R

Chapter 18: Powering Spartan-3 Generation FPGAs

consumption also varies by family and can be estimated using the Xilinx power estimator
tools.

Voltage Supplies
Spartan-3 generation FPGAs have multiple voltage supply inputs, as shown in Table 18-2.
There are two supply inputs for internal logic functions, VCCINT and VCCAUX. In the
Spartan-3A/3A DSP platforms, the VCCAUX level is programmable as either 2.5V (default)
or 3.3V. The user specifies the value in the software through the CONFIG VCCAUX=2.5 or
CONFIG VCCAUX=3.3 constraint. In the Spartan-3AN platform, the user must set CONFIG
VCCAUX=3.3 (default) for using the In-System Flash.

Table 18-2: Spartan-3 Generation Voltage Supplies

Supply
Description(1) Nominal Supply Voltage
Input
Internal core supply voltage. Supplies all internal logic functions, such as CLBs,
VCCINT block RAM, and multipliers. Input to the Power-On Reset (POR) circuit. Powers 1.2V
input signals for standards at 1.2V, 1.5V, and 1.8V.
Auxiliary supply voltage. Supplies Digital Clock Managers (DCMs), differential 2.5V; 3.3V option in
drivers, dedicated configuration pins, JTAG interface. Powers 2.5V and 3.3V Spartan-3A/3A DSP
VCCAUX input signals. Input to the POR circuit. platforms;
3.3V requirement in
Spartan-3AN platform
Supplies the output buffers in I/O Bank 0, the bank along the top edge of the Selectable: 3.3V, 3.0V,
VCCO_0
FPGA. 2.5V, 1.8V, 1.5V, or 1.2V
Supplies the output buffers in I/O Bank 1, the bank along the right edge of the
Selectable: 3.3V, 3.0V,
VCCO_1 FPGA. In Byte-Wide Peripheral Interface (BPI) Parallel Flash Mode, connects to
2.5V, 1.8V, 1.5V, or 1.2V
the same voltage as the Flash PROM.
Supplies the output buffers in I/O Bank 2, the bank along the bottom edge of the
Selectable: 3.3V, 3.0V,
VCCO_2 FPGA. Connects to the same voltage as the FPGA configuration source. Input to
2.5V, 1.8V, 1.5V, or 1.2V
the POR circuit.
Supplies the output buffers in I/O Bank 3, the bank along the left edge of the Selectable: 3.3V, 3.0V,
VCCO_3
FPGA. 2.5V, 1.8V, 1.5V, or 1.2V

Notes:
1. The VCCO designations apply to Spartan-3E and Extended Spartan-3A family FPGAs. The Spartan-3 family has eight VCCO supplies
numbered 0 to 7 clockwise, starting from the top left half-edge. The Spartan-3 devices in the TQ144 and CP132 packages have four
VCCO supplies as shown, but are connected together to make the equivalent of VCCO_TOP, VCCO_RIGHT, VCCO_BOTTOM, and
VCCO_LEFT.

Each of the I/O banks has a separate VCCO supply input that powers the output buffers
within the associated I/O bank. All of the VCCO connections to a specific I/O bank must be
connected to the same voltage. The VCCO voltage can be 1.2V to 3.3V, depending on the
output standard specified for a given bank.
Most devices have four I/O banks. The Spartan-3 family offers eight I/O banks in most
packages, one for each half-edge, except the TQ144 and CP132 packages, which have one
VCCO level per side. In those packages, the VCCO signals are connected together to form the
equivalent of VCCO_TOP, VCCO_RIGHT, VCCO_BOTTOM, and VCCO_LEFT.
In a 3.3V-only application, all VCCO supplies and VCCAUX in the Extended Spartan-3A
family, connect to 3.3V. However, Spartan-3 generation FPGAs allow bridging between

488 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Power Estimation

different I/O voltages and standards by applying different voltages to the VCCO inputs of
different banks. Refer to “I/O Banking Rules,” page 363 for which I/O standards can be
intermixed within a single I/O bank.
In the Extended Spartan-3A family, the 3.3V supplies support the full ±10% range from
3.0V to 3.6V, simplifying the selection of the 3.3V power supply. The Spartan-3/3E families
support –10% to + 5%, or 3.0V to 3.45V.
The Extended Spartan-3A family also allows input voltages (VIN) of up to 4.1V,
independent of the VCCO level. The Spartan-3/3E families restrict VIN to no more than
0.3V/0.5V above VCCO (or VCCAUX for dedicated pins). For applications requiring higher
voltages, see XAPP459, “Eliminating I/O Coupling Effects when Interfacing Large-Swing
Single-Ended Signals to User I/O Pins on Spartan-3 Generation FPGAs”.

VREF
Each I/O bank also has a separate, optional input voltage reference supply, called VREF. If
the I/O bank includes an I/O standard that requires a voltage reference such as HSTL or
SSTL, then all VREF pins within the I/O bank must be connected to the same voltage. The
VREF pins are available as I/O pins if no standards within a bank require them.
Xilinx recommends to always separate VREF from VTT as the VTT supply is very noisy. A
stable VREF using a small LDO is the desirable implementation. A voltage divider
implementation is also possible. Knowledge of the PCB environment, such as frequency of
coupled noise, is required to correctly calculate the resistance and capacitance values of the
divider circuit. As a result, an isolated reference supply is usually a more robust and
simpler approach.

Power Estimation
Xilinx provides a number of spreadsheet and Internet-based power estimation tools,
power analyzers, and power-related documentation to meet all power solutions needs.
The Power Solutions page on xilinx.com provides access to these tools, documentation,
news, and supply solutions.
There are two recommended ways to estimate the total power consumption (quiescent
plus dynamic) for a specific design:
• The XPower Power Estimator spreadsheet provides quick, approximate, typical
estimates, and does not require a netlist of the design. (The Spartan-3 family uses the
“Web Power Tool”.)
• The XPower Analyzer is delivered with ISE® software and uses a netlist as input to
provide maximum estimates as well as more accurate typical estimates.

Voltage Regulators
The choice of a voltage regulator depends on system requirements and the estimated
power consumption requirements for the FPGA. Use the XPower tools to calculate the
requirements for a specific device and design. If the design is not complete or the XPower
tool does not support the target device, use the closest match in the Spartan-3 generation
families that is supported. Then choose a regulator from a pin-compatible family so the
current capability can be adjusted up or down. External power FETs are easy to upgrade. A
softstart feature that controls output ramp time is useful.

Spartan-3 Generation FPGA User Guide www.xilinx.com 489

UG331 (v1.7) August 19, 2010
 R

Chapter 18: Powering Spartan-3 Generation FPGAs

With care, use of overcurrent protection is possible, such as foldback or fuses. In this case,
apply VCCAUX no later than VCCINT to avoid the surplus ICCINT current (see “Surplus
ICCINT if VCCINT is Applied before VCCAUX,” page 491). Also be aware that capacitors will
be charging at power-on and might draw a significant amount of current for a short time.
If necessary, slow the supply voltage ramp to control the charge current. If foldback is not
a design requirement, it is best to avoid it, keeping the power supply design simple.
Various power supply manufacturers offer complete power solutions for Xilinx FPGAs
including some with integrated three-rail regulators specifically designed for
Spartan-3 generation FPGAs. The Xilinx Power Solutions website provides links to vendor
solution guides and Xilinx power estimation and analysis tools.

Power-On Behavior
Spartan-3 generation FPGAs have a built-in Power-On Reset (POR) circuit that monitors
the three power rails required to successfully configure the FPGA (see Figure 18-1). At
power-up, the POR circuit holds the FPGA in a reset state until the VCCINT, VCCAUX, and
VCCO Bank 2 supplies reach their respective input threshold levels (see the respective
FPGA data sheets). In the Spartan-3 family, the POR input is VCCO Bank 4, the equivalent
half-edge to VCCO Bank 2 in the Spartan-3E and Spartan-3A/3AN/3A DSP devices.
References to POR and VCCO_2 apply to VCCO_4 in the Spartan-3 family. After all three
supplies reach their respective thresholds, the POR reset is released and the FPGA begins
its configuration process.

Spartan-3 Generation FPGA

Glitch
VCCINT Filter
Power-On
Reset

Glitch
VCCAUX
Filter Reset

Configuration
Glitch Logic
VCCO_2 Filter

Glitch
PROG_B
Filter

UG331_c18_01_081806

Figure 18-1: Simplified POR Circuit Diagram

Supply Sequencing
Because the three FPGA supply inputs must be valid to release the POR and can be
supplied in any order, there are no FPGA-specific voltage sequencing requirements.
Applying the FPGA’s VCCINT supply last uses the least ICCINT current.
Although the FPGA has no specific voltage sequence requirements, be sure to consider any
potential sequencing requirement of the configuration device attached to the FPGA, such
as an SPI serial Flash PROM, a parallel NOR Flash PROM, or a microcontroller. For
example, Flash PROMs have a minimum time requirement before the PROM can be

490 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Hot Swap

selected, and this time must be considered if the 3.3V supply is the last in the sequence. See
the Spartan-3 Generation Configuration User Guide (UG332) for more details.
For the Spartan-3AN devices, when configuring from the In-System Flash, VCCAUX must
be in the recommended operating range. On power-up make sure VCCAUX reaches at least
3.0V before INIT_B goes High to indicate the start of configuration. VCCINT, VCCAUX, and
VCCO supplies to the FPGA can be applied in any order if this requirement is met.
When all three supplies are valid, the minimum current required to power-on the FPGA
equals the worst-case quiescent current, specified in the FPGA data sheets.
Spartan-3 generation FPGAs do not require Power-On Surge (POS) current to successfully
configure.
Independent of the hot-swap pin, if VCCO is applied after VCCINT and VCCAUX, the internal
pull-ups are enabled for all I/Os from the time VCCO reaches approximately 0.4V until
VCCO exceeds VCCINT. If this pull-up is not desired, the user should avoid this power
sequence or place pull-down resistors to hold the pin at a Low logic level. Selection of a
pull-down value should be based on the minimum resistor value of the FPGA data sheet
(RPU, VCCO = 1.14V) and the VIL maximum specification of the downstream device.

Surplus ICCINT if VCCINT is Applied before VCCAUX

If the VCCINT supply is applied before the VCCAUX supply, the FPGA might draw a surplus
ICCINT current in addition to the ICCINT quiescent current levels specified in the FPGA data
sheets. The momentary additional ICCINT surplus current might be a few hundred
milliamperes under nominal conditions, significantly less than the instantaneous current
consumed by the bypass capacitors at power-on. However, the surplus current
immediately disappears when the VCCAUX supply is applied, and, in response, the FPGA’s
ICCINT quiescent current demand drops to the levels specified in the data sheets. The FPGA
does not use or require the surplus current to successfully power-on and configure. If
applying VCCINT before VCCAUX, ensure that the regulator does not have a foldback
feature that could inadvertently shut down in the presence of the surplus current. To avoid
the surplus current, apply VCCINT after VCCAUX has reached its minimum recommended
operating condition and is stable. For the lowest power-on current, apply VCCINT last, after
both VCCAUX and VCCO have been applied.

Ramp Rate
To ensure successful power-on, VCCINT, VCCO Bank 2, and VCCAUX supplies must rise
through their respective threshold-voltage ranges with no dips. The FPGA data sheets
specify any ramp rate requirements. The Spartan-3 family has no ramp rate requirements
for the current revision; refer to the Spartan-3 FPGA Family Data Sheet (DS099) for
specifications for earlier versions. The Spartan-3E family has ramp rate requirements from
0.2 to 50 ms. The Extended Spartan-3A family device ramp rate requirements are from 0.2
to 100 ms.

Hot Swap
Hot swap, also known as hot plug or hot insertion, refers to plugging an unpowered board
into a powered system. To support hot swap, an unpowered board or device must be able
to be plugged directly into a powered system or backplane without affecting or damaging
the system or the board/device. Devices that support hot swap include the following I/O
features:
• Signals can be applied to I/O pins before powering the device

Spartan-3 Generation FPGA User Guide www.xilinx.com 491

UG331 (v1.7) August 19, 2010
 R

Chapter 18: Powering Spartan-3 Generation FPGAs

• I/O pins are high-impedance (that is, 3-stated) before and during the power-up and
configuration processes
• There is no current path from the I/O pin back to the voltage supplies
While all Spartan-3 generation families can be used in hot-swap applications, they do not
offer the same levels of support. The Extended Spartan-3A family is fully hot-swap
compliant to the definition provided above. The Spartan-3/3E families require sequenced
connectors to make sure power is applied to the FPGA before the I/Os receive signals.
During the power-down sequence, to cleanly transition from valid signals to the disabled
state, remove VCCO first (VCCO < 0.5V) in the Extended Spartan-3A family, and remove
VCCAUX before VCCINT in the Spartan-3 and Spartan-3E families. See UG332, the Spartan-3
Generation Configuration User Guide for more details.

Configuration Data Retention, Brown-Out

The FPGA’s configuration data is stored in robust CMOS configuration latches. The data in
these latches is retained even when the voltages drop to the minimum levels necessary to
preserve RAM contents, as specified in the FPGA data sheets.
After configuration, if the VCCAUX or VCCINT supply drops below its minimum data
retention voltage, the integrity of the CMOS configuration latches is no longer guaranteed,
and the current device configuration must be cleared using one of the following methods:
• Force the VCCAUX or VCCINT supply voltage below the minimum POR voltage
threshold (as shown in the FPGA data sheets), then raise the voltage above the
maximum threshold requirement.
• Assert PROG_B Low.
The POR circuit does not monitor the VCCO_2 supply after configuration. Consequently,
dropping the VCCO_2 voltage does not reset the device by triggering a POR event. The
PROG_B input bypasses the POR circuit (see Figure 18-1) and therefore can be used as an
independent means to initialize the FPGA.
After the INIT_B signal goes High to indicate successful clearing of the FPGA, reconfigure
the FPGA.

Saving Power
Lower power consumption not only reduces power supply requirements but also reduces
heat, which increases reliability and might allow for smaller form factor packaging and
eliminate heat sinks and fans. Xilinx FPGAs are designed to minimize power consumption
without sacrificing high performance and low cost.
Dynamic power consumption can be reduced by reducing the number or frequency of
nodes and I/O toggling in a design. The lowest power state is the quiescent state with no
inputs toggling, all outputs disabled, and no pull-up or pull-down resistors in use. In this
state, the power consumption equals the sum of the power required for each power supply.
Consider the following techniques to eliminate any unnecessary switching in a design and
reduce dynamic power:
• Bring all incoming signals to a static state
• Apply rail-to-rail levels to inputs wherever possible
• Use signals that swing from GND to VCCO
• Turn off as many outputs as possible

492 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Saving Power

• Tie all unused inputs to VCCO or GND outside device

• Avoid instantiating pull-up and pull-down resistors on I/Os
• Disable as many internal oscillating circuits as possible
• Assign signal standards with small swings to outputs
• Have block RAMs operate in "No read on write" mode to reduce toggling of the
outputs of the block RAM
• Reduce the total length of heavy loaded signals to reduce capacitance

Saving Clock Routing Power

Clocks are a significant aspect of power consumption because of their high fanout nets and
also because controlling them limits the number of logic elements toggling in a design. If
possible, stop the clock where it enters the FPGA, so that it will not consume any FPGA
power. If you cannot gate the clock externally, then disable it inside the FPGA using the
BUFGCE component (see Figure 18-2). The BUFGCE functions as an AND gate for the
clock.

Spartan-3 Generation FPGA

BUFGCE
Clock

Enable

Global Clock Line

UG331_c18_02_081906

Figure 18-2: Using BUFGCE to Disable a Clock

Avoid using logic to generate gated or multiple clocks. Using CLB logic on a clock signal
introduces route-dependent skew and makes the design sensitive to the timing hazards of
lot-to-lot variations.
Minimizing the amount of routing a clock net uses is helpful. The Xilinx software
automatically disables clock nets in unused columns of CLBs, so reduce the number of
clock columns in use by concentrating the clocked logic in the fewest possible columns of
CLBs. Also reduce the number of rows that the clock is driving.
Floorplanning can be helpful to minimize clock power. Partition logic driven by global
clocks into clock regions and reduce the number of clock regions to which each global clock
is routed. Organize the design into independent clock domains, and clock each domain at
the lowest possible frequency.
Even if the clock cannot be manipulated, the activity on the loads can be controlled
through the use of clock enables to reduce switching activity on the outputs of flip-flops.

Spartan-3 Generation FPGA User Guide www.xilinx.com 493

UG331 (v1.7) August 19, 2010
 R

Chapter 18: Powering Spartan-3 Generation FPGAs

Power-Off Mode
In some cases, the device can simply be powered off to save power. This is useful for
designs where the FPGA has lengthy periods of non-operation and the power
consumption must be as low as possible. For the Extended Spartan-3A family, all three
supplies can be removed even if signals are still toggling on the inputs (see Figure 18-3).
For the Spartan-3E and Spartan-3 families, VCCO must be kept at a valid level to keep the
power diodes off (see Figure 18-4). VCCINT and VCCAUX can still be removed, reducing the
total static current to the typical ICCO quiescent value. External FETs can be used to switch
power. Configuration and memory data will be lost, so the FPGA must be reconfigured
after powering on again.

VCCO
2.5V/3.3V 1.2V Supply

Power
Switch

VCCAUX VCCINT VCCO

Active Device
Connected on
Board
Spartan-3A/3AN
FPGA

PROG_B
UG331_c18_03_022507

Figure 18-3: Extended Spartan-3A Family FPGA Power-Off Diagram

494 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Board Design and Signal Integrity

VCCO
2.5V 1.2V Supply

Power
Switch

VCC
VCCAUX VCCINT VCCO

I/O Bank
Device

Spartan-3/3E
FPGA

PROG_B
UG331_c18_04_081806

Figure 18-4: Spartan-3/3E FPGA Power-Off Diagram

To enter the Power-Off state, first pull PROG_B Low to turn the outputs off and initialize
the configuration memory to all zeros. After INIT_B and DONE go Low, switch off VCCINT
and VCCAUX. To restore power, reapply VCCINT and VCCAUX and then pull PROG_B back
High. After INIT_B goes High, reconfigure the FPGA and return to user mode.

Suspend Mode
The Extended Spartan-3A family FPGA Suspend Mode reduces power to below quiescent
current levels while saving the state of the device, including all configuration and user
data. For details on the Suspend Mode, see Chapter 19, “Power Management Solutions.”

Board Design and Signal Integrity

Building a working system today requires knowledge of a great deal more than just
Boolean logic and HDL code. Feature size reduction and the need for reduced power
consumption has driven core voltages down from the old standard of 5V to the 1.0V range.
This change in voltage and signal frequency content requires the use of new design
practices that take into account electrical effects that previously could be ignored. The
documents and links on the Xilinx Signal Integrity website provide everything needed to
achieve reliable PCB designs the first time.

Simultaneously Switching Outputs

Ground or power bounce occurs when a large number of outputs simultaneously switch in
the same direction. The output drive transistors all conduct current to a common voltage
rail. Low-to-High transitions conduct to the VCCO rail; High-to-Low transitions conduct to
the GND rail. The resulting cumulative current transient induces a voltage difference
across the inductance that exists between the die pad and the power supply or ground
return. The inductance is associated with bonding wires, the package lead frame, and any
other signal routing inside the package. Other variables contribute to SSO noise levels,

Spartan-3 Generation FPGA User Guide www.xilinx.com 495

UG331 (v1.7) August 19, 2010
 R

Chapter 18: Powering Spartan-3 Generation FPGAs

including stray inductance on the PCB as well as capacitive loading at receivers. Any SSO-
induced voltage consequently affects internal switching noise margins and ultimately
signal quality.
The number of SSOs allowed for quad-flat packages (VQ, TQ, PQ) is lower than for ball
grid array packages (FG) due to the larger lead inductance of the quad-flat packages. The
results for chip-scale packaging are better than quad-flat packaging but not as good as for
ball grid array packaging. Ball grid array packages are recommended for applications with
a large number of simultaneously switching outputs.
Each FPGA data sheet provides guidelines for the recommended maximum allowable
number of Simultaneous Switching Outputs (SSOs). These guidelines describe the
maximum number of user I/O pins of a given output signal standard that should
simultaneously switch in the same direction, while maintaining a safe level of switching
noise. Meeting these guidelines for the stated test conditions ensures that the FPGA
operates free from the adverse effects of ground and power bounce.

Power Distribution System Design and Decoupling/Bypass Capacitors

Good power distribution system (PDS) design is important for all FPGA designs,
especially for high-performance applications greater than 100 MHz. Proper design results
in better overall performance, lower clock and DCM jitter, and a generally more robust
system. Before designing the printed circuit board (PCB) for the FPGA design, please
review XAPP623, Power Distribution System (PDS) Design: Using Bypass/Decoupling
Capacitors.

No Internal Charge Pumps or Free-Running Oscillators

Some system applications are sensitive to sources of analog noise. Spartan-3 generation
FPGA circuitry is fully static and does not employ internal charge pumps.
The CCLK configuration clock is active during the FPGA configuration process. After
configuration completes, the CCLK oscillator is automatically disabled unless the
Bitstream Generator (BitGen) option Persist=Yes.

Large-Swing Signals
Under recommended operating conditions, the User I/O and Dual-Purpose pins of
Spartan-3 generation FPGAs handle signals that swing anywhere from 1.2V to 3.3V. The
Dedicated pins of these FPGAs normally use the LVCMOS25 standard. To handle signals
with a larger swing than is ordinarily recommended, see the guidelines in XAPP459,
Eliminating I/O Coupling Effects when Interfacing Large-Swing Single-Ended Signals to User I/O
Pins on Spartan-3 Generation FPGAs.

Related Documents
• Power Solutions (http://www.xilinx.com/power)
• Signal Integrity
(http://www.xilinx.com/products/design_resources/signal_integrity)
• XAPP453, The 3.3V Configuration of Spartan-3 FPGAs
• XAPP623, Power Distribution System (PDS) Design - Using Bypass/Decoupling Capacitors
• UG332, Spartan-3 Generation Configuration User Guide
• Spartan-3 Generation Data Sheets

496 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 19

Power Management Solutions

Overview
While some applications require the lowest possible system cost or highest performance,
still other applications require the lowest possible standby power. Spartan®-3 generation
FPGAs offer low-power options that balance cost and performance trade-offs.
All Spartan-3 generation FPGAs offer a power management option called Hibernate,
which essentially allows all or most of the FPGA logic to be turned off to save power. This
option also requires that the FPGA be reconfigured before returning to normal operation,
and it does not preserve the state of the FPGA application.
The Extended Spartan-3A family (including the Spartan-3A, Spartan-3AN, and
Spartan-3A DSP platforms) offers an advanced power management feature called Suspend
mode, which reduces FPGA power consumption while retaining the FPGA’s configuration
data and application state. While Hibernate provides the most power savings, Suspend
retains all data and offers fast wake-up times.
Table 19-1 summarizes the difference between Suspend and Hibernate.

Table 19-1: Spartan-3 Generation Power-Saving Options

Extended Spartan-3A Extended Spartan-3A Spartan-3/3E
Family Suspend Family Hibernate Family Hibernate
Configuration data retained Yes No
Application state retained
Yes No
(flip-flops, RAM, SRL16)
Time to exit from Approximately 500 μs + DCM FPGA configuration time
power-saving mode lock time + programmable timing (tens of milliseconds)
Power consumption while
Low Lowest
in power-saving mode
Power supplies Maintained or scaled down to Removed except for
Removed
save additional power VCCO

Spartan-3 Generation FPGA User Guide www.xilinx.com 497

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Extended Spartan-3A Family Suspend Mode

The Extended Spartan-3A family introduces an advanced power management option
called Suspend mode. This section describes the system advantages of Suspend mode.
More details on Suspend mode can be found in XAPP480, Using Suspend Mode in Spartan-3
Generation FPGAs.
Figure 19-1 graphically demonstrates the effect that Suspend mode has on some example,
representative designs measured on a typical XC3S1400A FPGA. The results for other
array sizes roughly scale with device density. The Suspend mode primarily affects current
consumption on the VCCINT and VCCAUX power rails, there are also power savings for the
VCCO rail, depending on the how the user-programmable SUSPEND constraints are
defined in the application (see “Define the I/O Behavior During Suspend Mode,”
page 506).
Figure 19-1 includes three example designs that highlight the Suspend mode behavior:
• Blank: A blank FPGA design. No logic is used in this application. A blank design
provides the lowest quiescent current and establishes the baseline power
consumption.
• 32LVDS: A design that includes 32 LVDS differential input channels (64 pins)
connected to 32 LVDS differential output channels (64 pins). On
Extended Spartan-3A family FPGAs, the differential I/O buffers are powered by the
VCCAUX voltage supply.
• 32LVDS+8DCM: A design that includes the circuitry described for 32LVDS plus eight
Digital Clock Managers (DCMs). On Extended Spartan-3A family FPGAs, the
differential I/O buffers and DCMs are powered by the VCCAUX voltage supply.

100 400

3V)
90 er (VCCAUX=3.
Quiescent Pow 350
Bars: Quiescent Current (mA)

80
r (VCCAUX=2.5V)

Lines: Quiescent Power (mW)

300
70 Quiescent Powe

250
60
Quiescent Power
(Suspend)

50 200
ICCINT (Suspend)
ICCINT (Suspend)
ICCINT (Suspend)

40
ICCAUX

150
ICCAUX

ICCINT
ICCINT

30
ICCAUX
ICCINT

100
20

ICCAUX ICCAUX ICCAUX 50

10
(Suspend) (Suspend) (Suspend)

0 0
Blank 32 LVDS 32 LVDS + 8 DCM
UG331_c19_10_022507

Figure 19-1: Effects of Suspend Mode on Example Designs Measured on Typical XC3S1400A

498 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Figure 19-1 also shows four bars, indicating the typical quiescent current on the VCCINT
and VCCAUX supplies under normal quiescent conditions with all clocks stopped and
during Suspend mode. The associated current measurement, in mA, appears along the
left-side vertical axis. Note that the current on VCCAUX during Suspend mode is near the
base of the chart, highlighted in burgundy.
Furthermore, Figure 19-1 shows the quiescent power (current multiplied by the voltage
applied to each power rail). The associated resulting power measurement, in mW, appears
on the right-side vertical axis. On Spartan-3A/3A DSP FPGAs, VCCAUX can be either 2.5V
or 3.3V nominally. By physics, the quiescent power is lower when VCCAUX = 2.5V. Note the
significant reduction in total power consumption when the Spartan-3A FPGA is in
Suspend mode. Although the total power savings is design dependent, Suspend mode
typically reduces power consumption by 40% or more, with a minimum power savings of
about 20%.
During Suspend mode, some of the circuitry powered by the VCCAUX supply is switched
over to the VCCINT supply. Note the Blank design example in Figure 19-1. The current on
the VCCINT supply actually increases while the current on the VCCAUX supply drops
significantly! Fortunately, the total VCCINT current during Suspend remains below that
used in an active FPGA application. Furthermore, despite the increased VCCINT current,
the overall system power is reduced because current is being switched from the 2.5V or
3.3V VCCAUX supply to the 1.2V VCCINT supply.
The power savings are more pronounced in the 32LVDS and 32LVDS+8DCM examples
and both designs use circuitry that consumes current on the VCCAUX supply.

Suspend Features and Benefits

• Quiescent current is reduced by 40% or more and active current is significantly
reduced.
• FPGA configuration data and the present state of the FPGA application during
Suspend mode is retained.
• Fast, programmable FPGA wake-up time from Suspend mode, in as little as 500 μs.
• Each user-I/O pin has an individual control that defines how the pin behaves during
Suspend mode.
• When enabled in the FPGA bitstream, Suspend mode is externally activated by the
system using a single dedicated control pin called SUSPEND. Note that the SUSPEND
pin, and therefore Suspend mode, is not available in the VQ100 package.
• The FPGA’s AWAKE pin indicates the present Suspend mode status. AWAKE is
automatically dedicated when SUSPEND is enabled in the FPGA bitstream.

Design Preparation for Suspend Mode

To use the Suspend feature in an Extended Spartan-3A family FPGA application, follow
these steps:
1. Define the I/O Behavior During Suspend Mode in the source design or in a user
constraints file (UCF).
2. Define the AWAKE Pin Behavior when Suspend Feature Enabled.
3. Define the SUSPEND Input Glitch Filter setting.
4. Define the Suspend Mode Wake-Up Timing Controls.
5. Enable the Suspend Feature.

Spartan-3 Generation FPGA User Guide www.xilinx.com 499

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

6. Generate the FPGA bitstream.

500 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Entering Suspend Mode

Figure 19-2 provides a block diagram of how an Extended Spartan-3A family FPGA enters
Suspend mode. Figure 19-3 provides example waveforms.

FPGA Application Logic

FPGA FPGA
Inputs Writable Clocked Elements Outputs
Flip-Flops SRL16 Block RAM

SUSPEND
Latches LUT RAM Constraint

SUSPEND
Constraint

Block FPGA
Inputs

Write-Protect Writable
Clocked Elements

Apply SUSPEND Constraints

to FPGA Outputs

SUSPEND AWAKE
Glitch Filter

Suspend Enable
en_suspend Filter Select
suspend_filter UG331_c19_01_113006

Figure 19-2: Entering Suspend Mode

The FPGA can only enter Suspend mode if enabled in the configuration bitstream (see
“Enable the Suspend Feature”). Once power is applied to the system, the FPGA always
powers up and configures regardless of the value applied to the SUSPEND pin. Once
enabled via the bitstream, the FPGA unconditionally and quickly enters Suspend mode if
the SUSPEND pin is asserted. If Suspend is not enabled in the bitstream, the SUSPEND
input will have no effect and the AWAKE pin will be usable as a general-purpose I/O.
When the FPGA enters Suspend mode, all nonessential FPGA functions are shut down to
minimize power dissipation. The FPGA retains all application state and configuration data
while in Suspend mode. All writable clock elements are write-protected against spurious
write operations. All FPGA inputs and interconnects are shut down.
Each FPGA output pin or bidirectional I/O pin assumes its defined Suspend mode
behavior, which is described as part of the FPGA design using a “SUSPEND Constraint”.
The AWAKE pin goes Low, indicating that the FPGA is in Suspend mode. The DONE pin
remains High while the FPGA is in Suspend mode because the FPGA does not lose its
configuration data.

Spartan-3 Generation FPGA User Guide www.xilinx.com 501

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Entering Suspend Mode Exiting Suspend Mode

sw_gwe_cycle
1 6 sw_gts_cycle
SUSPEND Input

tSUSPENDHIGH_AWAKE tSUSPENDLOW_AWAKE
3 8
AWAKE Output
tSUSPEND_GWE tAWAKE_GWE
2
10
Flip-Flops, Block RAM, Write Protected
Distributed RAM
tSUSPEND_GTS tAWAKE_GTS
4 9
FPGA Outputs Defined by SUSPEND constraint
tSUSPEND_DISABLE tSUSPEND_ENABLE
5 7
FPGA Inputs,
Blocked
Interconnect
UG331_c19_02_011209

Figure 19-3: Suspend Mode Waveforms (Entering and Exiting)

Items 1 through 5 in Figure 19-3 are described below:

1. An external signal drives the FPGA’s SUSPEND pin High, unconditionally forcing the
FPGA into the power-saving Suspend mode. Data values are captured for I/O pins
with a SUSPEND constraint set to DRIVE_LAST_VALUE; however, this value is not
presented until Step 4.
2. In response to the SUSPEND input going High, the FPGA immediately write protects
and preserves the states of all clocked elements. The states of all flip-flops, block RAM,
distributed RAM (LUT RAM), shift registers (SRL16), and I/O latches are preserved
during Suspend mode.
3. The FPGA drives the AWAKE output Low to indicate that it is entering SUSPEND
mode.
4. The FPGA switches the normal behavior of all outputs over to the Suspend mode
behavior defined by the SUSPEND constraint assigned to each I/O. See “Define the
I/O Behavior During Suspend Mode,” page 506.
5. FPGA inputs are blocked and the interconnects shut off to prevent any internal
switching activity.

502 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Exiting Suspend Mode

There are two possible ways to exit Suspend mode in a powered system:
1. Drive the SUSPEND input Low, exiting Suspend mode normally.
2. Pulse the PROG_B input Low, resetting the FPGA and causing the FPGA to reprogram.

FPGA Application Logic

FPGA FPGA
Inputs Writable Clocked Elements Outputs
Flip-Flops SRL16 Block RAM

SUSPEND
Latches LUT RAM Constraint

SUSPEND
Constraint

Re-enable
FPGA Inputs

Set/Reset
Flip-Flops

Enable
en_sw_gsr
sw_gwe_cycle
Unlock Clocked
Elements

1 5 1,024
sw_gts_cycle
Wake-Up Activate Outputs
Timing Clock
Source
SUSPEND 1 4 1,024
sw_clk
Glitch Filter AWAKE

Suspend Enable
Filter Select drive_awake
en_suspend
suspend_filter UG331_c19_03_092606

Figure 19-4: Exiting Extended Spartan-3A Family FPGA Suspend Mode

Figure 19-4 is a block diagram showing how to exit Suspend mode using the SUSPEND
pin.
When SUSPEND goes Low, the FPGA automatically re-enables all inputs and
interconnects.
If enabled in the FPGA bitstream, all flip-flops are optionally, globally set or reset
according the FPGA design description. By default, the flip-flops are not globally set or
reset, which preserves the state of the FPGA application before entering Suspend mode.
The remaining wake-up process depends on the logic value applied to the AWAKE Pin.
Once AWAKE goes High, two user-programmable timers define when FPGA outputs are

Spartan-3 Generation FPGA User Guide www.xilinx.com 503

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

re-enabled and when the write-protect lock is released from all writable clocked elements.
The wake-up timing clock source is also programmable.
Items 6 through 10 correspond to the markers in Figure 19-3, page 502:
6. The system drives the FPGA’s SUSPEND input Low, causing the FPGA to exit
Suspend mode.
7. The FPGA releases the inputs and interconnect, allowing signals to propagate
internally. There is no danger of corrupting the internal state because all clocked
elements are still write protected.
8. The FPGA asserts the AWAKE signal with the bitstream option drive_awake:yes. If the
option is drive_awake:no, then the FPGA releases AWAKE to become an open-drain
output. In this case, an external pull-up resistor is required or an external signal must
drive AWAKE High before the FPGA continues to awaken. All subsequent timing is
measured from when the AWAKE output goes High.
9. The FPGA switches output behavior from the specified SUSPEND Constraint to the
function specified in the FPGA application. The timing of this switch-over is controlled
by the Suspend/Wake sw_gts_cycle bitstream generation setting, which defines when
the FPGA’s internal Global Three-State (GTS) control is released. After the specified
number of clock cycles, the outputs are active according to normal FPGA application.
By default, the outputs switch over four clock cycles after AWAKE goes High. The
outputs are generally released before the clocked elements to allow signals to
propagate out of the FPGA.
10. The writable, clocked elements are released according to the Suspend/Wake
sw_gwe_cycle bitstream generator setting, which defines when the FPGA’s internal
Global Write Enable (GWE) control is asserted. After the specified cycle, it is again
possible to write to flip-flops, block RAM, distributed RAM (LUT RAM), shift registers
(SRL16), and I/O latches. By default, the clocked elements are released five clock
cycles after AWAKE goes High. Generally, the write-protect lock should be held until
after outputs are enabled.
It is good design practice to apply a Reset to any design DCMs after exiting the Suspend
mode so that the DCM will re-acquire lock. See “Momentarily Stopping CLKIN,” page 152.

PROG_B Programming Pin Always Overrides Suspend Mode

Pulsing the PROG_B programming pin Low always overrides Suspend mode and forces
the FPGA to restart configuration. Likewise, power-cycling the FPGA also restarts
configuration.

Suspend Mode Timing Example

Table 19-2 provides example, typical timing for the Extended Spartan-3A family FPGA
Suspend feature. Refer to the Spartan-3A FPGA Family Data Sheet (DS529), the Spartan-3AN
FPGA Family Data Sheet (DS557), and the Spartan-3A DSP FPGA Family Data Sheet (DS610)
as the official sources of these values.

504 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Table 19-2: Suspend Mode Timing Parameters

Symbol Description Min Typ Max Units
Entering Suspend Mode
TSUSPENDHIGH_AWAKE Rising edge of SUSPEND pin to falling edge of
– 7 – ns
AWAKE pin without glitch filter (suspend_filter:No)
TSUSPENDFILTER Adjustment to SUSPEND pin rising edge parameters
+160 +300 +600 ns
when glitch filter enabled (suspend_filter:Yes)
TSUSPEND_GWE Rising edge of SUSPEND pin until FPGA output pins
– 10 – ns
drive their defined SUSPEND constraint behavior
TSUSPEND_GTS Rising edge of SUSPEND pin to write-protect lock on
– <5 – ns
all writable clocked elements
TSUSPEND_DISABLE Rising edge of the SUSPEND pin to FPGA input pins
– 340 – ns
and interconnect disabled
Exiting Suspend Mode
TSUSPENDLOW_AWAKE Falling edge of the SUSPEND pin to rising edge of the
– 4 to 108 – μs
AWAKE pin. Does not include DCM lock time.
TSUSPEND_ENABLE Falling edge of the SUSPEND pin to FPGA input pins
– 3.7 to 109 – μs
and interconnect re-enabled
TAWAKE_GWE1 Rising edge of the AWAKE pin until write-protect
lock released on all writable clocked elements, using – 67 – ns
sw_clk:InternalClk and sw_gwe_cycle:1.
TAWAKE_GWE512 Rising edge of the AWAKE pin until write-protect
lock released on all writable clocked elements, using – 14 – μs
sw_clk:InternalClk and sw_gwe_cycle:512.
TAWAKE_GTS1 Rising edge of the AWAKE pin until outputs return to
the behavior described in the FPGA application, – 57 – ns
usingsw_clk:InternalClk and sw_gts_cycle:1.
TAWAKE_GTS512 Rising edge of the AWAKE pin until outputs return to
the behavior described in the FPGA application, – 14 – μs
usingsw_clk:InternalClk and sw_gts_cycle:512.

Enable the Suspend Feature

The Suspend power-saving feature must first be enabled in the FPGA bitstream before it
can be used. By default, the Suspend feature is disabled, SUSPEND has no effect, and the
AWAKE pin is usable as a general-purpose I/O. When Suspend is enabled, the software
will not allow use of the AWAKE pin for I/O.

Via User Constraints File (UCF)

Suspend mode is enabled and the SUSPEND Input Glitch Filter option is defined using a
CONFIG statement in a user constraints file (UCF). Table 19-3 shows the available options.
This is the recommended method for enabling Suspend mode as this constraint also
automatically reserves the AWAKE pin.

CONFIG ENABLE_SUSPEND = “FILTERED” ;

Figure 19-5: UCF Constraint Defining Suspend Mode Behavior for an I/O pin

Spartan-3 Generation FPGA User Guide www.xilinx.com 505

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Table 19-3: Available Options for the ENABLE_SUSPEND Constraint

Option Suspend Mode SUSPEND Pin Filter AWAKE Pin
NO Suspend mode is disabled Not applicable. Connect Available as a user I/O pin
SUSPEND pin to GND. in the FPGA application
FILTERED Suspend mode is enabled Glitch filter is enabled AWAKE status indicator
UNFILTERED Glitch filter is bypassed

Via BitGen
Note: Setting the en_suspend bitstream option is an alternate way to enable the Suspend mode.
However, this method is not recommended because it does not automatically reserve the AWAKE pin
in the application.
bitgen -g en_suspend:Yes

Define the I/O Behavior During Suspend Mode

Use a SUSPEND Constraint to define the behavior of each pin to be programmed
differently than the default, 3STATE.

Single-Ended I/O Standards

Each output, open-drain output, or bidirectional I/O pin in the FPGA application that uses
a single-ended I/O standard can be individually programmed for one of the Suspend
mode behaviors shown in Table 19-4. The default behavior is for the pin to be high
impedance during Suspend mode although other options are available.

Table 19-4: Output Behavior Options during Suspend Mode

SUSPEND Attribute Function
DRIVE_LAST_VALUE The output continues to drive the level that was last stored in the
output latch, according to the chosen standard. Requires VCCO to
remain at the recommended operating conditions for the bank.
3STATE The output is in the high-impedance state with no active internal
(default) pull-up or pull-down resistor. Results in the lowest possible I/O
current draw.
3STATE_PULLUP The output is in the high-impedance state with an internal pull-up
resistor to the associated VCCO supply. Requires VCCO to remain at
the recommended operating conditions for the bank.
3STATE_PULLDOWN The output is in the high-impedance state with an internal pull-
down resistor to GND.
3STATE_KEEPER The output is high impedance. The internal bus keeper circuit is
active. Requires VCCO to remain at the recommended operating
conditions for the bank.

Differential I/O Standards

The differential output drivers and input receivers consume static power when used in an
FPGA application. In Suspend mode, differential inputs and outputs are disabled to save
power.

506 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

The output drivers for the “true” differential I/O standards (LVDS, RSDS, mini-LVDS,
PPDS, TMDS) are high impedance, using one of the 3STATE attributes described in
Table 19-4. The DRIVE_LAST_VALUE attribute is not supported for differential output
drivers.
Treat the pseudo-differential I/O standards, such as BLVDS, LVPECL, DIFF_HSTL, and
DIFF_SSTL, as two single-ended I/O pins. All the attributes apply as for “Single-Ended
I/O Standards” although the settings must be set appropriately for the complementary
pair.
When in the high-impedance state, the differential driver pair does not conduct current to
the power or ground rails, or between adjacent pins.
Differential input receivers are disabled in Suspend mode.
Differential input termination (DIFF_TERM) is disabled when in Suspend mode.

SUSPEND Constraint
The SUSPEND constraint allows each pin to have an individually defined behavior during
Suspend mode. The available options are in Table 19-4, page 506.

UCF Example
Figure 19-6 shows an example UCF constraint that defines the Suspend mode behavior for
a specific pin. The SUSPEND constraint can be included on the same UCF line as other
constraints for a pin.

NET "<net_name>" SUSPEND = “<io_type>” ;

Figure 19-6: UCF Constraint Defining Suspend Mode Behavior for an I/O pin

More Information
For additional information on the SUSPEND constraint, see the Constraints Guide for the
latest software version (http://www.xilinx.com/support/documentation/dt_ise.htm).
• Constraints Guide for ISE® Software
http://www.xilinx.com/support/documentation/dt_ise.htm

Application State Retained during Suspend Mode

When entering Suspend mode, all writable clocked elements are write-protected. The state
of all clocked memory elements is retained during Suspend mode. Internal signals are
forced High during Suspend mode. To avoid an unwanted transition on a clock or control
signal, set clock and control signals High when entering Suspend mode.
• Logic block flip-flops
• I/O block latches and flip-flops
• Logic block distributed RAM (LUT RAM)
• Logic block shift registers (SRL16)
• Block RAM and registers
When exiting Suspend mode, all writable clocked elements are re-enabled, controlled by
the sw_gwe_cycle setting.
An additional bitstream option called en_sw_gsr controls whether all clocked elements are
globally set or reset when the FPGA awakens from Suspend mode. By default,

Spartan-3 Generation FPGA User Guide www.xilinx.com 507

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

en_sw_gsr:No, which means that clocked elements are not set or reset when the FPGA
awakens and all states are preserved.

Suspend Mode Wake-Up Timing Controls

When exiting Suspend mode, the wake-up sequence for the FPGA is programmable and
controlled by a single clock.

Wake-Up Timing Clock Source (sw_clk)

The wake-up timing when exiting Suspend mode is controlled by a selectable clock source
as shown in Figure 19-7 and described in Table 19-5. The clock source is defined by up to
two bitstream generator options, sw_clk and sometimes StartupClk.
The internal oscillator is disabled during Suspend mode to conserve power.

CCLK input only available for

applications that configure in Slave
mode. Persist = Yes required.

CCLK Cclk Suspend

Wake-Up
TCK Jtag StartupClk Timing
Control
User Clock from FPGA Fabric
UserClk
InternalClk

~50 MHz
STARTUP_SPARTAN3A Internal
Oscillator
StartupClk sw_clk

CLK
UG331_c19_06_092606

Figure 19-7: Suspend Mode Wake-Up Timing Control Clock Selection

• The sw_clk option is specific to the Suspend feature. By default, sw_clk:InternalClk.

• The StartupClk option is available on every application. By default StartupClk:Cclk.
Consequently, the CCLK pin is the default clock source when exiting Suspend mode.

508 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Table 19-5: Clock Sources to Wake-Up from Suspend Mode

sw_clk StartupClk
Setting Setting Clock Source Restriction
InternalClk -- Internal Oscillator The oscillator has an imprecise frequency of about 50 MHz.
StartupClk Cclk CCLK pin on FPGA This option is only available for FPGAs using Slave
configuration mode. The bitstream option Persist:Yes must
be set. This option is not available for FPGAs using the
Master configuration mode; use InternalClk instead.
JtagClk TCK pin on FPGA The JTAG interface must be active to exit Suspend mode.
UserClk CLK input on the The clock input to the STARTUP design primitive can
STARTUP_SPARTAN3A originate from any nonclocked signal in the FPGA. It cannot
design primitive originate from a flip-flop source because all clocked elements
are write-protected while in Suspend mode.

Switch Outputs from Suspend to Normal Behavior (sw_gts_cycle)

The Suspend/Wake sw_gts_cycle bitstream option controls when I/O pins are released
from their SUSPEND constraint settings and returned to normal operation. The timing is
controlled by the “Wake-Up Timing Clock Source (sw_clk)” described above. The default
sw_gts_cycle setting is 4 cycles, but this control can be set for any value between 1 and
1,024 clock cycles.
This control becomes active after the AWAKE pin goes High. After the specified number of
clock cycles, all output, open-drain output, and bidirectional I/O pins transition from their
Suspend behavior, individually specified using a SUSPEND Constraint, back to the normal
behavior specified in the original FPGA application.
It is best to release the outputs before releasing the write-protect lock on all clocked
elements.

Release Write Protect on Clocked Elements (sw_gwe_cycle)

The Suspend/Wake sw_gwe_cycle bitstream option controls when the write-protect lock is
released on all clocked elements.
The timing is controlled by the Wake-Up Timing Clock Source (sw_clk) described above.
The default sw_gwe_cycle setting is 5 cycles, but this control can be set for any value
between 1 and 1,024 clock cycles.
This control becomes active after the AWAKE pin goes High. After the specified number of
clock cycles, the write-protect lock is released from all writable, clocked elements such as
flip-flops, block RAM, etc.
If the en_sw_gsr:yes option was set, then the clocked elements are already globally set or
reset to the value specified in the original FPGA design before the write-protect lock is
released. If en_sw_gsr:no, then the state of the FPGA before entering Suspend mode is
preserved.
It is best to release the outputs before releasing the write-protect lock on all clocked
elements.

Spartan-3 Generation FPGA User Guide www.xilinx.com 509

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Dedicated Configuration Pins Unaffected During Suspend Mode

The following dedicated configuration pins are unaffected when the FPGA is in Suspend
mode:
• JTAG pins TDI, TMS, TCK, and TDO
• DONE pin
• PROG_B pin

SUSPEND Pin
When the Suspend feature is enabled (see “Enable the Suspend Feature,” page 505), the
SUSPEND pin controls when the FPGA enters Suspend mode. During normal FPGA
operation, the SUSPEND pin must be Low. When High, the SUSPEND pin forces the FPGA
into the low-power Suspend mode. Table 19-6 describes the functionality of the SUSPEND
pin.
If the Suspend feature is not enabled for an application (the application never enters low-
power mode), then connect the SUSPEND pin to GND.

Table 19-6: SUSPEND Pin Functionality

en_suspend SUSPEND
Setting Pin Function
no The suspend feature is disabled. The SUSPEND pin is
(default) unused and ignored. Connect the SUSPEND pin to GND.
X
Suspend mode
disabled
The FPGA performs the application described in the
yes bitstream loaded into the FPGA during configuration. When
0
the SUSPEND pin changes from High to Low, wake the
Suspend mode
FPGA from Suspend mode.
enabled
1 Force the FPGA to enter power-saving Suspend mode.

Characteristics
The SUSPEND pin is an LVCMOS/LVTTL receiver, and power to the input buffer is
supplied by the VCCAUX power rail. The SUSPEND pin has no pull-up resistors during
configuration, and the PUDC_B control has no affect on the SUSPEND pin.

SUSPEND Input Glitch Filter

The SUSPEND pin has a programmable glitch filter to guard against short pulses, which
could cause the FPGA to spuriously enter Suspend mode. Turning off the filter allows the
FPGA to enter or exit SUSPEND mode more quickly, but the application must guard
against spurious pulses.

Via User Constraints File (UCF)

The SUSPEND filter is set as part of the ENABLE_SUSPEND constraint, as described in
“Via User Constraints File (UCF),” page 505.

510 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Bitstream Generator (BitGen) Option

The filter can also be enabled via a bitstream generator option:
bitgen -g suspend_filter:Yes

Effect on FPGA Configuration

Suspend mode is activated by an FPGA configuration bitstream option. Consequently, the
SUSPEND pin has no effect on configuration.
If Suspend mode is enabled in the bitstream and the SUSPEND pin is High, the FPGA
successfully configures and then immediately enters Suspend mode. The FPGA’s DONE
pin will be High, but the AWAKE pin will be Low.

Tie SUSPEND to GND if not Using Suspend Mode

If not using Suspend mode, connect the SUSPEND pin to GND. Do not leave the pin
floating.

AWAKE Pin
The AWAKE pin optionally provides status on the Suspend power-savings mode.

General Behavior (Suspend Feature Disabled)

Unless the Suspend feature is enabled, the AWAKE pin is a general-purpose user-I/O pin.

AWAKE Pin Behavior when Suspend Feature Enabled

If the Suspend feature is enabled, then the AWAKE pin indicates the present state of the
FPGA, as summarized in Table 19-7. The AWAKE pin cannot be used by the FPGA
application as a general-purpose I/O pin.

Table 19-7: AWAKE Pin Status

AWAKE Pin Indication
0 The FPGA is presently in the low-power Suspend mode.
1 The FPGA is active.

The AWAKE pin can further be configured as an open-drain output (the default) or a full-
swing output driver, as shown in Figure 19-8. This behavior is controlled by a bitstream
generator (BitGen) option:
bitgen -g drive_awake:no

Spartan-3 Generation FPGA User Guide www.xilinx.com 511

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

drive_awake=yes drive_awake=no External Pull-up

FPGA FPGA Resistor Required
VCC
Internal
Awake
Internal Signal 20 kΩ
Awake (OE)
Signal T
O AWAKE O AWAKE

LVCMOS LVCMOS
12 mA 12 mA
FAST FAST
UG331_c19_07_011209

Figure 19-8: AWAKE Output Drive Options if Suspend Mode Enabled

The AWAKE output pin is supplied by the VCCO power rail on bank 2 when Suspend
mode is enabled.
When drive_awake:yes, the AWAKE pin is an active output driver, equivalent to a user
I/O configured as LVCMOS, with 12 mA output drive and a Fast slew rate.

Controlling Wake-Up from an External Source

By default drive_awake:no. When drive_awake:no, the AWAKE pin is an open-drain
output capable of sinking 12 mA. In this case, an external pull-up resistor is required to exit
Suspend mode. The resistor value should be high to minimize the amount of current flow
during Suspend mode. The resistor needs to be strong enough to overcome the I/O pin
leakage. A large resistor value also equates to a longer AWAKE rise time. The FPGA does
not exit Suspend mode until AWAKE goes High.
Holding the AWAKE pin Low delays the transition from Suspend mode to Active mode
and allows an external controller to decide when to awaken the FPGA.

JTAG Operations Allowed During Suspend Mode

Table 19-8 shows the JTAG operations permitted when the FPGA is in Suspend mode.
Executing these JTAG operations increases the FPGA’s power consumption while in
Suspend mode.

Table 19-8: JTAG Operations Allowed during Suspend Mode

Boundary Scan
Command Description
Read the JTAG ID code that describes the
Extended Spartan-3A family FPGA array type in the JTAG chain.
IDCODE
This value is different from the Device DNA identifier, which is
unique to every device.
BYPASS Enables BYPASS.
USERCODE Read the user-defined code embedded in the FPGA bitstream.

Do not use any other JTAG instructions when in Suspend mode or while transitioning into
and out of Suspend Mode. Furthermore, do not enter Suspend mode when performing a
Readback operation.

512 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Extended Spartan-3A Family Suspend Mode

Post-Configuration CRC Limitations When Using Suspend Mode

If an application uses the post-configuration CRC feature and an error occurs, do not enter
Suspend mode. The FPGA will not wake from Suspend mode without reprogramming,
such as asserting PROG_B or power-cycling the FPGA.
Several design options are possible:
1. Do not use the post-configuration CRC feature when the Suspend mode feature is
enabled and vice versa.
2. If the post-configuration CRC feature is enabled, externally gate the SUSPEND pin
input with the INIT_B pin. The post-configuration CRC feature signals an error by
driving the INIT_B pin Low. The external gate ensures that the SUSPEND pin cannot
drive High when the INIT_B pin is Low. Enable the SUSPEND Input Glitch Filter to
avoid a possible race condition between the SUSPEND and INIT_B pins.
For more information, see the “Configuration CRC” chapter in UG332: Spartan-3
Generation Configuration User Guide.

Spartan-3 Generation FPGA User Guide www.xilinx.com 513

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Suspend Mode Bitstream Generator Options

Table 19-9 summarizes the various bitstream options associated with Suspend mode.

Table 19-9: Suspend Mode Bitstream Generator Options

Suspend Mode
Bitstream Options
Options (default) Description
en_suspend Suspend mode is not used in this application. Connect the
No
SUSPEND pin to GND.
Enables the power-saving Suspend feature, controlled by
Yes
the SUSPEND pin.
drive_awake If Suspend mode is enabled, indicates the present status on
No AWAKE using an open-drain output. An external pull-up
resistor or High signal is required to exit SUSPEND mode.
If Suspend mode is enabled, indicates the present status by
Yes
actively driving the AWAKE output.
suspend_filter Yes Enables the glitch filter on the SUSPEND pin.
No Disables the glitch filter on the SUSPEND pin.
en_sw_gsr No The state of all clocked elements in the FPGA is preserved.
Pulses the GSR signal during wake-up, setting or resetting
all clocked elements, as originally specified in the FPGA
Yes application. The GSR pulse occurs before the AWAKE pin
goes High and before the sw_gwe_cycle and sw_gts_cycle
settings are active.
sw_clk Uses the clock defined by the StartupClk bitstream
StartupClk
generator setting to control the Suspend wake-up timing.
Uses the internally generated 50 MHz oscillator to control
InternalClk
the Suspend wake-up timing.
sw_gwe_cycle After the AWAKE pin is High, indicates the number of clock
cycles as defined by the sw_clk setting, when the global
write-protect lock is released for writable clocked elements
1,..,5,...,1024
(flip-flops, block RAM, etc.). The default value is five clock
cycles after the AWAKE pin goes High. Generally, this value
is equal to or greater than the sw_gts_cycle setting.
sw_gts_cycle After the AWAKE pin is High, indicates the number of clock
cycles as defined by the sw_clk setting, when the I/O pins
switch from their SUSPEND Constraint settings back to
1,..,4,...,1024
their normal functions. The default value is four clock cycles
after the AWAKE pin goes High. Generally, this value is
equal to or less than the sw_gwe_cycle setting.

FPGA Voltage Requirements During Suspend Mode

During Suspend mode, the VCCINT and VCCAUX rails must remain powered at their
specified data sheet levels. VCCO for bank 2 should also be maintained since it powers the
AWAKE pin. However, the VCCO supply to the other three I/O banks can be potentially
turned off to conserve additional power, depending on system requirements. Optionally,

514 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Hibernate

VCCO can be reduced to 1.0V during SUSPEND mode, but this also affects the voltage
levels for any output pin with a SUSPEND=”DRIVE_LAST_VALUE” constraint.
The FPGA’s power-on reset (POR) circuit continues to monitor the VCCINT and VCCAUX
supplies. The POR circuit does not monitor the VCCO supplies after configuration. By
default, if the VCCINT or VCCAUX supply dips below the minimum specified data sheet
voltage limit, then the FPGA restarts configuration.

Supply Requirements During Suspend Mode

When entering Suspend mode, the FPGA exhibits the following characteristics on the
VCCINT and VCCAUX power rails:
• The current required on the VCCAUX supply drops significantly as the internal FPGA
circuits powered by VCCAUX are internally switched over to the VCCINT supply
during Suspend mode.
• The current required on the VCCINT supply increases slightly from its quiescent
current level.

Hibernate
Hibernate provides the maximum possible power savings for applications that can be
turned off for long periods of time and that can afford to lose the present application state.

Forcing FPGA to Quiescent Current Levels

Pulse PROG_B Low to achieve the quiescent current levels. Driving PROG_B Low forces
all I/Os into a high-impedance state, ceases all internal switching, and converts the
bitstream held in internal memory to all zeros. During and after the Low pulse on
PROG_B, disable the internal pull-up resistors on all I/Os by driving the pull-up resistor
control input High. The specific signal name varies by product family: PUDC_B for the
Extended Spartan-3A family, HSWAP for the Spartan-3E family, or HSWAP_EN for
Spartan-3 devices. Holding PROG_B Low continues clearing the configuration memory. To
minimize quiescent current, release PROG_B High but hold off configuration by setting
the Mode pins to a slave or JTAG configuration mode and disabling the external
configuration clock (CCLK or TCK).
To restart the application, release PROG_B High and in slave or JTAG modes, enable the
external configuration source. The FPGA must reconfigure before the application restarts.
No state information is preserved.
If the application must retain the FPGA configuration bitstream, then there are a few
options. If using Spartan-3A, Spartan-3AN, or Spartan-3A DSP FPGAs, use the
Extended Spartan-3A Family Suspend Mode. If using Spartan-3E or Spartan-3 FPGAs, do
not assert PROG_B. If all other test conditions are met (for example, no internal switching,
I/Os are off), quiescent current levels are close to or slightly above the data sheet quiescent
levels. Ensure that internal pull-up and pull-down resistors on I/O pins are disabled.

Entering Hibernate State

Hibernate starts with the approach described in “Forcing FPGA to Quiescent Current
Levels.” Hibernate provides further power savings by switching off power rails. This state
reduces quiescent power consumption to the lowest possible level. The FPGA enters
Hibernate by switching off the VCCINT (core) and VCCAUX (auxiliary) power supplies.

Spartan-3 Generation FPGA User Guide www.xilinx.com 515

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

Power is supplied to VCCO lines throughout the hibernation period. Figure 19-9 shows
how to put Spartan-3 generation FPGAs into Hibernate.
During the Hibernation period, the VCCINT and VCCAUX rails are turned off. Power FETs
with low “on” resistance are recommended to perform the switching action. Configuration
data is lost upon entering Hibernate; therefore, the device will reconfigure after exiting the
state.

VCCO
2.5V 1.2V Supply

Power
Control

VCCAUX VCCINT

I/O Bank
Inactive
Low Device
Spartan-3E Hi-Z
FPGA

HSWAP

I/O Bank
Active,
Powered
PROG_B Device

VCCO
Supply UG331_c19_08_092506

Figure 19-9: Spartan-3E FPGA Hibernate Example

Holding the PROG_B input Low during the transition into Hibernation period keeps all
FPGA output drivers in a high-impedance state. Release PROG_B after re-applying power
to the VCCINT and VCCAUX rails. See “Design Considerations,” page 518 for recommended
levels on Dedicated and Dual-Purpose pins.

Extended Spartan-3A Family FPGA: Turn Off VCCO

Extended Spartan-3A family I/O pins have a floating-well structure, providing full hot-
swap/hot-insertion capability. When an Extended Spartan-3A family FPGA is in the
Hibernate state, the VCCO supply can be safely turned off without adversely affecting
either the FPGA or the external application. When entering Hibernation, turn off VCCO
first to disable the outputs and prevent any unwanted transitions.

Spartan-3E and Spartan-3 FPGAs: Maintain VCCO on I/O Banks Connected

to Powered External Devices
Each user I/O or input-only pin on Spartan-3E and Spartan-3 FPGAs has a power diode
between the pin and the associated VCCO rail. The power diodes are present on all signal-
carrying pins all of the time. In general, it is safest to maintain VCCO power for all banks
throughout the Hibernation period to keep the power diodes inside the I/O block turned
off when signals are applied to the pins. In Hibernate, the powered VCCO rails account for
little current because the I/Os are in a high-impedance state.
Under certain conditions, it is also possible to switch off the VCCO rail to a particular bank.
This action eliminates the VCCO current for those banks (a few milliamperes). There are
various ways to achieve this, as shown for the “Inactive Device” in Figure 19-9.

516 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Hibernate

1. Turn off power to any external devices connected to a particular FPGA I/O bank. If
both the FPGA I/O bank and the external device are unpowered, there is no current
flow.
2. If the external device is powered but the FPGA I/O bank is not, then ensure that all
signals driving into the FPGA are either high-impedance (Hi-Z) or that they are under
0.5V. Both cases ensure that there is no current flow through the FPGA’s power diodes.
Voltages higher than 0.5V can turn on the power diodes. Keep the diodes off to prevent
"reverse current" from flowing into the VCCO rail.

Hibernate

PROG_B

I/Os

VCCINT

VCCAUX

VCCO
(per I/O Bank)

INIT_B

DONE Undefined

Undefined
CCLK in Master Mode
STARTUP cycles
UG331_c19_09_022507

Figure 19-10: Hibernate Waveform

Figure 19-10 shows the waveforms for entering and exiting Hibernate. The steps for
entering Hibernate are as follows:
1. Pull the PROG_B pin Low to force all user-I/O pins and Input-only pins into a high-
impedance state.
2. The FPGA drives the INIT_B and DONE pins Low.
3. External switches turn off the VCCINT and VCCAUX supply rails to the FPGA.
Depending on the FPGA product family and the application, it might also be possible
to turn off power to the VCCO supply rail.
• See “Extended Spartan-3A Family FPGA: Turn Off VCCO,” page 516
• See “Spartan-3E and Spartan-3 FPGAs: Maintain VCCO on I/O Banks Connected
to Powered External Devices,” page 516
4. The FPGA is now in Hibernate. While the FPGA is kept in this state, power
consumption rests at the lowest possible level.

Exiting Hibernate
The steps for exiting Hibernate are as follows.

Spartan-3 Generation FPGA User Guide www.xilinx.com 517

UG331 (v1.7) August 19, 2010
 R

Chapter 19: Power Management Solutions

1. Reapply power to all rails that were switched off. Apply power in any sequence.
2. Before FPGA initialization can begin, deassert PROG_B to a High logic level. The
rising transition on PROG_B must occur after turning all three power supplies back on.
3. After logic initialization, the FPGA releases the open-drain INIT_B signal. With
INIT_B High, the FPGA starts its configuration process.
4. When configuration is complete, the FPGA enters the Startup phase, asserts DONE,
and enables the I/Os, according to how the BitGen options are set.
5. The FPGA is now ready for user operation.

Design Considerations
Be aware of how various pins are powered in the application. Most user-I/O pins,
including the Dual-Purpose configuration pins, are powered by a specific VCCO supply
input. The Dedicated configuration pins are powered by the VCCAUX supply. If
disconnecting power to any of these supplies, consider how that will affect FPGA
configuration when power is re-applied.
For specific information on configuration pins and their associated power rails, refer to the
“Configuration Pins and Behavior during Configuration” chapter in UG332: Spartan-3
Generation Configuration User Guide.
If disconnecting power to VCCO or VCCAUX supplies on Spartan-3 or Spartan-3E FPGAs
during Hibernate, do not apply voltages on the pins in excess of 0.5 V to ensure that the
power diodes are kept off. This restriction does not apply to Extended Spartan-3A family
FPGAs, which have a floating N-well structure for improved hot-swap performance. For
more information, see XAPP459: Eliminating I/O Coupling Effects when Interfacing Large-
Swing Single-Ended Signals to User I/O Pins on Spartan-3 Generation FPGAs.

518 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 20

Using IBIS Models

Summary
Input/Output Buffer Information Specification (IBIS) models are industry-standard
descriptions used to simulate I/O characteristics in board-level design simulation. IBIS
models for Spartan®-3 generation devices are available at
http://www.xilinx.com/support/download/index.htm. Each family has its own IBIS
models because there are differences in I/O standards and device characteristics. The
models can be used with third-party simulation tools to verify proper signal integrity
characteristics in board designs.

Introduction
As I/O switching frequencies have increased and voltage levels have decreased, accurate
analog simulation of I/Os has become an essential part of modern high-speed digital
system design. By accurately simulating the I/O buffers, termination, and circuit board
traces, designers can significantly shorten their time-to-market of new designs. Identifying
signal integrity related issues at the beginning of the design cycle decreases the required
number of board fixes and increases quality.
The device data sheets provide basic information about guaranteed DC and switching
characteristics of the I/Os. However, the data sheet does not include all the information
required to determine the best board implementation for a particular application, such as
slew rates and drive strength, which are included in the IBIS model. Designers can use IBIS
models for system-level analysis of signal integrity issues, such as ringing, crosstalk, and
RFI/EMI. Complete designs can be simulated and evaluated before going through the
expensive and time consuming process of producing prototype PCBs. This type of pre-
layout simulation can reduce considerably the development cost and time to market, while
increasing the reliability of the I/O operation.

IBIS Advantages over SPICE

Traditionally SPICE analysis has been used extensively in areas like IC design, where a
high level of accuracy is required. However in the PCB and systems domain, there are
several disadvantages to the SPICE method, both for the device vendor and the user.
Since SPICE simulations model a circuit at transistor level, it is necessary for the SPICE
models to contain detailed information about the circuit and process parameters. For most
IC vendors, this type of information is regarded as proprietary.
Although SPICE simulation accuracy is typically very good, a significant limitation with
any simulation method is simulation speed. Simulation speeds are particularly slow for
transient simulation analysis, which is most often used when evaluating signal integrity

Spartan-3 Generation FPGA User Guide www.xilinx.com 519

UG331 (v1.7) August 19, 2010
 R

Chapter 20: Using IBIS Models

performance. SPICE simulation has a further disadvantage in that not all SPICE simulators
are fully compatible. Often, default simulator options are not the same in different SPICE
simulators. As there are some very powerful options that control accuracy, convergence
and the algorithm type, any options that are not consistent might give rise to poor
correlation in simulation results across different simulators. Also, because of the different
variants of SPICE, these models are often incompatible between simulators, thus models
must be extracted for a specific simulator.

IBIS Background
IBIS, originally developed by Intel, is an alternative to SPICE simulation. The IBIS
specification now is maintained by the EIA/IBIS Open Forum, which has members from a
large number of IC and EDA vendors. IBIS is the ANSI/EIA-656 and IEC 62014-1 standard.
For more information about the IBIS specification, see
http://www.eigroup.org/ibis/ibis.htm.
The core of the IBIS model consists of a table of current versus voltage and timing
information. This is very attractive to the IC vendor as the I/O internal circuit is treated as
a black box. This way, transistor-level information about the circuit and process details is
not revealed.
IBIS models can be used to model best-case and worst-case conditions (best-case = strong
transistors, low temperature, high voltage; worst-case = weak transistors, high
temperature, low voltage). The “fast/strong” model represents best-case conditions, while
the "slow/weak" model represents worst-case conditions. The "typical" model represents
typical behavior.
IBIS cannot be used for internal timing information (propagation delays and skew); the
timing models instead provide that information. IBIS also does not model power and
ground structures or pin-to-pin coupling. The implications are that ground bounce, power
supply droop, and simultaneous switching output (SSO) noise cannot be simulated with
IBIS models. Instead, Xilinx provides device/package-dependent SSO guidelines once
extensive lab measurements are completed. IBIS models also do not provide detailed
package parasitic information. Package parasitics usually are provided in the form of
lumped RLC data, which loses its accuracy at higher speeds. To model the package
parasitics accurately, include a transmission line with a delay of 25 ps to 100 ps and an
impedance of 65Ω.
Using IBIS models has a great advantage to the user in that simulation speed is
significantly increased over SPICE, while accuracy is only slightly decreased. Non-
convergence, which can be a problem with SPICE models and simulators, is eliminated in
IBIS simulation. Virtually all EDA vendors presently support IBIS models, and ease of use
of these IBIS simulators is generally very good. IBIS models for most devices are freely
available over the Internet making it easy to simulate several different manufacturers’
devices on the same board. Several different IBIS simulators are available today, and each
simulator provides different results. An overshoot or undershoot of ±10% of the measured
result is tolerable. Differences between the model and measurements occur because not all
parameters are modeled. Simulators for IBIS models are provided by Cadence, Hyperlynx,
Mentor, and Intusoft.

520 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Xilinx Support of IBIS

Xilinx Support of IBIS

Xilinx provides IBIS models for all current products; they are downloaded easily from our
website at http://www.xilinx.com/support/download/index.htm. The models also are
made available in the development system. The Preliminary models are based initially on
simulation and then verified against the silicon.
An IBIS file contains two sections, the header and the model data for each component. One
IBIS file can describe several devices. The following is the content list in a typical IBIS file:
• IBIS Version
• File Name
• File Revision
• Component
• Package R/L/C
• Pin name, model, R/L/C
• Model (for example, 3-state)
• Temperature Range (typical, minimum, and maximum)
• Voltage Range (typical, minimum, and maximum)
• Pull-Up Reference
• Pull-Down Reference
• Power Clamp Reference
• Ground Clamp Reference
• I/V Tables for:
• Pull-Up
• Pull-Down
• Power Clamp
• Ground Clamp
• Rise and Fall dV/dt for minimum, typical, and maximum conditions (driving 50Ω)
Spartan-3AN FPGA IBIS models are identical to those for the Spartan-3A FPGA platform.

IBIS I/V and dV/dt Curves

A digital buffer can be measured in receive (3-state) mode and drive mode. IBIS I/V curves
are based on the data of both these modes. The transition between modes is achieved by
phasing in/out the difference between the driver and the receiver models, while keeping
the receiver model constantly in the circuit.
The I/V curve range required by the IBIS specification is –VCC to (2x VCC). This wide
voltage range exists because the theoretical maximum overshoot due to a full reflection is
twice the signal swing. The ground clamp I/V curve must be specified over the range
– VCC to VCC, and the power clamp I/V curve must be specified from VCC to (2x VCC).
The three supported conditions for the IBIS buffer models are typical values (required),
minimum values (optional), and maximum values (optional). For CMOS buffers, the
minimum condition is defined as high temperature and low supply voltage, and the
maximum condition is defined as low temperature and high supply voltage.

Spartan-3 Generation FPGA User Guide www.xilinx.com 521

UG331 (v1.7) August 19, 2010
 R

Chapter 20: Using IBIS Models

An IBIS model of a digital buffer has four I/V curves:

• The pull-down I/V curve contains the mode data for the driver driving low. The
origin of the curve is at 0V for CMOS buffers.
• The pull-up I/V curve contains the mode data for the driver driving high. The origin
of the curve is at the supply voltage (VCC).
• The ground clamp I/V curve contains receive (3-state) mode data. The origin of the
curve is at 0V for CMOS buffers.
• The power clamp I/V curve contains receive (3-state) mode data. The origin of the
curve is at the supply voltage (VCC).

Ramp and dV/dt Curves

The Ramp keyword contains information on how fast the pull-up and pull-down
transistors turn on/off. The dV/dt curves give the same information, while including the
effects of die capacitance (C_comp). C_comp is the total die capacitance as seen at the die
pad, excluding the package capacitance.
dV/dt curves describe the transient characteristics of a buffer more accurately than ramps.
A minimum of four dV/dt curves are required to describe a CMOS buffer: pull-down ON,
pull-up OFF, pull-down OFF, and pull-up ON. dV/dt curves incorporate the clock-to-out
delay, and the length of the dV/dt curve corresponds to the clock speed at which the buffer
is used. Each dV/dt curve has t = 0, where the pulse crosses the input threshold.

IBISWriter
A Xilinx IBIS file downloaded from the Web contains a collection of IBIS models for all I/O
standards available in the targeted device. The ISE® tool can generate IBIS models specific
to your design via the IBISWriter tool, simplifying design export into signal integrity
analysis tools. IBISWriter associates IBIS buffer models to each pin of the customer design
according to the design specification for each I/O buffer. IBISWriter outputs an IBS file that
can be used directly as an input file to your signal integrity analysis tool.
Generating design-specific IBIS files requires only three easy steps:
1. Implement your design in Project Navigator.
2. In the Processes window, under Implement Design/Place & Route, select Generate
IBIS Model and click Run. A design-specific file is generated where all input/output
pins are associated with an IBIS model.
3. Incorporate this file onto your favorite signal integrity analysis tool to perform the
desired simulations.

References
IBIS files are available at the Xilinx Download Center:
http://www.xilinx.com/support/download/index.htm

522 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Chapter 21

Using Boundary-Scan and BSDL Files

Summary
Boundary Scan Description Language (BSDL) files are provided for every part and
package combination of IEEE 1149.1 (JTAG) compatible devices produced by Xilinx,
including all the Spartan®-3 generation FPGAs. Third-party Boundary-Scan tools use
these files to generate test vectors and perform the tests. Xilinx programming software also
uses BSDL files when configuring devices through Boundary-Scan.
This chapter applies to all Spartan-3 generation FPGA families: Spartan-3, Spartan-3E, and
Extended Spartan-3A family (Spartan-3A, Spartan-3AN, and Spartan-3A DSP) platforms.

Boundary-Scan Overview
Boundary-Scan testing is used to identify faulty board-level connections, such as
unconnected or shorted pins. Boundary-Scan tests allow designers to quickly identify
manufacturing or layout problems, which otherwise could be nearly impossible to isolate,
especially with high-count ball-grid packages. More recently, PLD vendors such as Xilinx
have made use of Boundary Scan as a convenient way of configuring devices, including
the Spartan-3 generation FPGA families. For details on configuration through Boundary-
Scan, see UG332, Spartan-3 Generation Configuration User Guide.

IEEE Standards
Joint Test Action Group (JTAG) is the commonly used name for IEEE standard 1149.1,
which defines a method for Boundary-Scan. JTAG compliant devices have dedicated
hardware that comprises a state machine and several registers to allow Boundary-Scan
operations. This dedicated hardware interprets instructions and data provided by four
dedicated signals: TDI (Test Data In), TDO (Test Data Out), TMS (Test Mode Select), and
TCK (Test Clock). The JTAG hardware interprets instructions and data on the TDI and TMS
signals, and drives data out on the TDO signal. The TCK signal is used to clock the process
(see Figure 21-1).

Spartan-3 Generation FPGA User Guide www.xilinx.com 523

UG331 (v1.7) August 19, 2010
 R

Chapter 21: Using Boundary-Scan and BSDL Files

IEEE Standard 1149.1 Compliant Device

TAP State Machine
Test-Logic-Reset
1
0
1 1
Run-Test/Idle Select-DR Select-IR
0
0 0
1 1
Capture-DR Capture-IR
0 0
Shift-IR/Shift-DR
Select Next State 0
Shift-DR
0
Shift-IR

TMS 1
Exit1-DR
1
1
Exit1-IR
1

0 0
Pause-DR Pause-IR
0 0
1 1
Exit2-DR Exit2-IR
0 0

TCK 1
Update-DR
1
Update-IR
1 1
0 0

TDI Instruction Register

Select Data
Instruction Decoder Register
TDO
Bypass[1] Register

IDCODE[32] Register

Boundary-Scan[N] Register

I/O I/O I/O I/O UG071_47_082306

Figure 21-1: Typical JTAG Architecture

In the Spartan-3 generation FPGAs, the four JTAG signals TDI, TDO, TMS, and TCK are on
dedicated pins powered by VCCAUX. Each can be configured with a pullup (default),
pulldown, or neither, through bitstream options. IEEE 1532 is a superset of the IEEE 1149.1
JTAG standard. IEEE 1532 provides additional flexibility for configuring programmable
logic devices. IEEE Std 1532 enables designers to concurrently program multiple devices,
minimize programming times with enhanced silicon features, and produce robust systems
that are more easily maintained. This standard defines the three additional items required
to configure in-system programmable logic devices:
• Device architectural components for configuration
• Algorithm description framework
• Configuration data file
General information on the IEEE 1532 JTAG standard is available at:
http://www.xilinx.com/products/design_resources/config_sol/isp_standards_specs.htm

Boundary-Scan Functions
Spartan-3 generation devices support the mandatory IEEE 1149.1 commands, as well as
several Xilinx vendor-specific commands. The EXTEST, INTEST, SAMPLE/PRELOAD,
BYPASS, IDCODE, USERCODE, and HIGHZ instructions are all included. The TAP also
supports internal user-defined registers (USER1, USER2) and configuration/readback of
the device.

524 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Boundary-Scan Tools

The Spartan-3 generation Boundary-Scan operations are independent of mode selection.

The Boundary-Scan mode in Spartan-3 generation devices overrides other mode
selections. For this reason, Boundary-Scan instructions using the Boundary-Scan register
(SAMPLE/PRELOAD, INTEST, and EXTEST) must not be performed during
configuration. All instructions except the user-defined instructions are available before a
device is configured. After configuration, all instructions are available.
Each Spartan-3 generation FPGA array type has a 32-bit device-specific device identifier
readable through the Boundary-Scan logic. The Boundary-Scan interface also provides the
option to store a 32-bit User ID, loaded during configuration and specified via the UserID
configuration bitstream option.
For details on the standard Boundary-Scan instructions EXTEST, INTEST, and BYPASS,
refer to the IEEE Standard.

Boundary-Scan Tools
Boundary-Scan testing requires specialized test equipment and software. The Boundary-
Scan test software is used to generate test vectors, which are typically delivered to the
Boundary-Scan chain using a test pod connected to a PC.
To develop vectors for Boundary-Scan testing, the test software must be provided with
information about the scan chain:
1. The composition of the scan chain - how many devices, what type, and so forth.
The chain composition can be either specified by the user or automatically detected by
the Boundary-Scan software.
2. The Boundary-Scan architecture of each device - the Instruction Register length,
opcodes, number of I/Os, and how each of those I/Os behaves.
The Boundary-Scan architecture of each device is defined in a Boundary Scan
Description Language (BSDL) file.
3. How the device I/Os are connected to each other.
This information typically is extracted from a board-level netlist.

BSDL Files
Any manufacturer of a JTAG-compliant device must provide a BSDL file for that device.
The BSDL file contains information on the function of each of the pins on the device -
which are used as I/Os, which are power or ground, and so forth. All Xilinx BSDL files
have file extensions of .bsd.
BSDL files for Xilinx devices are available in the development system and on the Xilinx
website at http://www.xilinx.com/support/download/index.htm. BSDL files for other
manufacturers typically can be found on the manufacturer's website.
Files for the IEEE 1532 extension to the BSDL files are also available for Xilinx products.
They are included with the other BSDL files.
IEEE 1149.1 BSDL files appear as: <device_name>.bsd
For example: xc3s50.bsd
These BSDL files are the only ones needed for programming. For JTAG testing, the
package-specific files are used.
For example: xc3s50_pq208.bsd

Spartan-3 Generation FPGA User Guide www.xilinx.com 525

UG331 (v1.7) August 19, 2010
 R

Chapter 21: Using Boundary-Scan and BSDL Files

IEEE 1532 BSDL files appear as: <device_name>_1532.bsd

For example: xc3s50_pq208_1532.bsd
IEEE Std 1532 BSDL files should not be used in place of or alongside 1149.1 BSDL files.

BSDL File Composition

BSDL files describe the Boundary-Scan architecture of a JTAG-compliant device, and are
written in VHDL. There are eight main parts of a BSDL file:
1. Entity Declaration
The entity declaration is a VHDL construct used to identify the name of the device that
is described by the BSDL file.
Example (from the xc3s50_pq208.bsd file):
entity XC3S50_PQ208 is
2. Generic Parameter
The Generic parameter specifies which package the BSDL file describes.
Example (from the xc3s50_pq208.bsd file):
generic (PHYSICAL_PIN_MAP : string := "PQ208" );
3. Logical Port Description
The Logical Port Description lists all of the pads on a device, and states whether that
pin is an input (in bit;), output (out bit;), bidirectional (inout bit;) or
unavailable for Boundary Scan (linkage bit;).
Example (from the xc3s50_pq208.bsd file):
port (
GND: linkage bit_vector (1 to 28);
CCLK_P104: inout bit;
DONE_P103: inout bit;
HSWAP_EN_P206: in bit;
M0_P55: in bit;
M1_P54: in bit;
M2_P56: in bit;
PROG_B: in bit;
TCK: in bit;
TDI: in bit;
TDO: out bit;
TMS: in bit;
VCCAUX: linkage bit_vector (1 to 8);
VCCINT: linkage bit_vector (1 to 4);
VCCO0: linkage bit_vector (1 to 2);
IO_P2: inout bit; -- PAD124
IO_P3: inout bit; -- PAD123
4. Package Pin Mapping
The Package Pin Mapping shows how the pads on the device die are wired to the pins
on the device package.
Example (from the xc3s50_pq208.bsd file):
constant PQ208: PIN_MAP_STRING:=
"GND:(P1,P8,P14,P25,P30,P41,P47,P53,P59,P66," &
"P75,P82,P91,P99,P105,P112,P118,P129,P134,P145," &
"P151,P157,P163,P170,P179,P186,P195,P202)," &

526 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

BSDL Files

"CCLK_P104:P104," &
"DONE_P103:P103," &
"HSWAP_EN_P206:P206," &
"M0_P55:P55," &
"M1_P54:P54," &
"M2_P56:P56," &
"PROG_B:P207," &
"TCK:P159," &
"TDI:P208," &
"TDO:P158," &
"TMS:P160," &
"VCCAUX:(P17,P38,P69,P89,P121,P142,P173,P193)," &
"VCCINT:(P70,P88,P174,P192)," &
"VCCO0:(P188,P201)," &
"IO_P2:P2," &
"IO_P3:P3," &
5. use statements
The use statement calls VHDL packages that contain attributes, types, constants, and
others that are referenced in the BSDL File.
Example (from the xc3s50_pq208.bsd file):
use STD_1149_1_1994.all;
6. Scan Port Identification
The Scan Port Identification identifies the JTAG pins: TDI, TDO, TMS, TCK, and TRST
(if used). TRST is an optional JTAG pin that is not used by Xilinx devices.
Example (from the xc3s50_pq208.bsd file):
attribute TAP_SCAN_IN of TDI : signal is true;
attribute TAP_SCAN_MODE of TMS : signal is true;
attribute TAP_SCAN_OUT of TDO : signal is true;
attribute TAP_SCAN_CLOCK of TCK : signal is (33.0e6, BOTH);
7. TAP description
The TAP description provides additional information on the device’s JTAG logic. Some
of this information includes the Instruction Register length, Instruction Opcodes, and
device IDCODE. These characteristics are device specific, and can vary widely from
device to device.
Examples (from the xc3s50_pq208.bsd file):
attribute COMPLIANCE_PATTERNS of XC3S50_PQ208 : entity is
"(PROG_B) (1)";
attribute INSTRUCTION_LENGTH of XC3S50_PQ208 : entity is 6;
attribute INSTRUCTION_OPCODE of XC3S50_PQ208 : entity is
"EXTEST (000000)," &
attribute INSTRUCTION_CAPTURE of XC3S50_PQ208 : entity is
"XXXX01";
attribute IDCODE_REGISTER of XC3S50_PQ208 : entity is
"XXXX" & -- version
"0001010" & -- family
"000001101" & -- array size
"00001001001" & -- manufacturer
"1"; -- required by 1149.1
8. Boundary Register description
The Boundary Register description gives the structure of the Boundary-Scan cells on
the device. Each pin on a device can have up to three Boundary-Scan cells, each cell

Spartan-3 Generation FPGA User Guide www.xilinx.com 527

UG331 (v1.7) August 19, 2010
 R

Chapter 21: Using Boundary-Scan and BSDL Files

consisting of a register and a latch. Boundary-Scan test vectors are loaded into or
scanned from these registers.
Example (from the xc3s50_pq208.bsd file):
attribute BOUNDARY_REGISTER of XC3S50_PQ208 : entity is
" 0 (BC_2, *, controlr, 1)," &
" 1 (BC_2, IO_P161, output3, X, 0, 1, PULL0)," & -- PAD30
" 2 (BC_2, IO_P161, input, X)," & -- PAD30

BSDL File Verification

Xilinx verification of the supplied BSDL files has two levels. Preliminary files are generated
using an automated, Xilinx-standard, BSDL generation process. The process is script-based
and extracts information directly from the device design files, which fully describe the
architecture and pinout. The quality of "Preliminary" BSDL files is very high, and the
syntax is always tested. Files marked "Final" have undergone all the tests for syntax and
hardware verification. To guarantee that the BSDL files exactly describe the operation of
each pin, Xilinx uses an independent third-party Boundary-Scan tool vendor — Intellitech
— to verify the actual silicon against the BSDL.
Xilinx BSDL files are checked for 1149.1 compliance with the Intellitech Eclipse product
using 'strict' BSDL syntax checking. Every semantic check described in the IEEE 1149.1b-
1994 (the standard for BSDL syntax) is performed using strict parsing. Test patterns then
are generated from the BSDL file that include unique tests for every I/O pin. Each Xilinx
device/package combination is tested on the Intellitech Reduced Contact Tester (RCT).
The test patterns include verification of Test-Logic-Reset and TAP controller operation,
BYPASS/IDCODE/USERCODE instructions and registers, and pin mapping of the
boundary register to every input/output/bidirectional/clock pin and control cell. Finally,
each device is tested for 1149.1 compliance after the device is programmed by
downloading a design and using the RCT tester to verify post configuration compliance.
For more information on BSDL files and their verification, see
http://www.xilinx.com/isp/bsdl/bsdl.htm.

Using BSDLAnno for Post-Configuration Boundary-Scan Behavior

Whenever possible, Boundary-Scan tests should be performed on an unconfigured
Spartan-3 generation device. Unconfigured devices allow for better test coverage, since all
I/Os are available for bidirectional scan vectors. Boundary-Scan tests should be performed
after configuration when configuration cannot be prevented and when differential
signaling standards are used. If the differential signals are located between Xilinx devices,
both devices can be tested pre-configuration. Each side of the differential pair will behave
as a single-ended signal.
The BSDL files provided by Xilinx reflect the Boundary-Scan behavior of an unconfigured
device. After configuration, the Boundary-Scan behavior of a device changes. I/O pins that
were bidirectional before configuration might now be input-only, output-only,
bidirectional, or unavailable. Boundary-Scan test vectors typically are derived from BSDL
files, so if Boundary-Scan tests are going to be performed on a configured Xilinx device, the
BSDL file must be modified to reflect the device's configured Boundary-Scan behavior.
The Boundary-Scan architecture changes after the device is configured because the
Boundary-Scan registers sit behind the I/O buffer and sense amplifier. The hardware is
arranged in this way so that the Boundary-Scan logic operates at the I/O standard
specified by the design. This allows Boundary-Scan testing across the entire range of
available I/O standards.

528 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Software Support

Because certain connections between the Boundary-Scan registers and pad might change,
the Boundary-Scan architecture is effectively changed when the device is configured.
These changes often need to be communicated to the Boundary-Scan tester through a post-
configuration BSDL file. If the changes to the Boundary-Scan architecture are not reflected
in the BSDL file, Boundary-Scan tests might fail.
Xilinx offers the BSDLAnno utility to automatically modify the BSDL file for post-
configuration testing. BSDLAnno obtains the necessary design information from the
routed .ncd file and generates a BSDL file that reflects the post-configuration Boundary-
Scan architecture of the device.
Use the following syntax to generate a post-configuration BSDL file with BSDLAnno:
bsdlanno [options] infile[.ncd] outfile[.bsd]
The infile is the routed (post-PAR) NCD design source file for the specified design. The
outfile[.bsd] is the destination for the design-specific BSDL file. The .bsd extension
is optional. For more details on BSDLanno including suggested user modifications to the
output file, see the Development System Reference Guide at
http://www.xilinx.com/support/documentation/dt_ise.htm

Software Support
Xilinx offers several tools for generating device files and for device programming.
Boundary-Scan test functionality is available from several third-party vendors, as noted at
http://www.xilinx.com/products/design_resources/config_sol/resource/isp_ate.htm.

iMPACT
iMPACT is a full featured software tool used for configuration and programming of all
Xilinx FPGAs, CPLDs, and PROMs. It features a series of "wizard" dialogs that easily guide
the user through the every step of the configuration process. iMPACT supports a host of
output file types including SVF. iMPACT configuration software enables users to easily
configure Xilinx FPGAs using different modes: slave serial, SPI, SelectMAP (Slave
Parallel), and JTAG IEEE 1149.1. iMPACT supports the Parallel Cable IV and Platform
Cable USB.
iMPACT features a special function in the JTAG mode to test both the operation of the cable
and the robustness of the JTAG chain. The user can test chain operation by instructing
iMPACT to write to and read back from the user code location multiple thousands of times.
It then counts the number of errors that occur in this operation. This gives the user the
opportunity to evaluate the relative robustness of the JTAG chain and the susceptibility to
noise and other influences like board layout.
For more information on iMPACT see the iMPACT help at
http://www.xilinx.com/support/documentation/dt_ise.htm

SVF Files
Serial Vector Format (SVF) is an industry-standard file format that is used to describe JTAG
chain operations in a compact, portable fashion. SVF files capture all of the device specific
programming information within the SVF instructions. SVF files are useful because
intricate knowledge of the device is not needed. The capability to create SVF files is
included in the iMPACT tool. For more information on SVF files see XAPP503 at
http://www.xilinx.com/support/documentation/application_notes/xapp503.pdf.

Spartan-3 Generation FPGA User Guide www.xilinx.com 529

UG331 (v1.7) August 19, 2010
 R

Chapter 21: Using Boundary-Scan and BSDL Files

J Drive Engine for IEEE 1532 Programming

Xilinx has developed and introduced the world's first IEEE Std 1532 Programming Engine:
J Drive Engine. Using this engine and a simple cable connected to the parallel port of any
PC, users can easily configure Xilinx IEEE Std 1532 compatible PLDs. The designer
provides J Drive Engine with the data and 1532 BSDL files for the device(s) to be
programmed using a command line interface to configure the PLDs in the JTAG chain. For
more information, see http://www.xilinx.com/products/design_resources/config_sol/
and XAPP500 at
http://www.xilinx.com/support/documentation/application_notes/xapp500.pdf.

Using the BSCAN_SPARTAN3A Macro

BSCAN_SPARTAN3A provides access to the BSCAN sites on a Spartan-3A, Spartan-3AN,
or Spartan-3A DSP platform device (see Figure 21-2). The BSCAN_SPARTAN3 macro
provides similar functionality for the Spartan-3 and Spartan-3E families, but does not
include the TCK and TMS pins. The component is primarily used to create internal
Boundary-Scan chains. The four-pin JTAG interface (TDI, TDO, TCK, and TMS) contains
dedicated pins in Spartan-3 generation FPGAs. To use normal JTAG for Boundary-Scan
purposes, just hook up the JTAG pins to the port and go. The pins on the
BSCAN_SPARTAN3A symbol do not need to be connected unless those special functions
are needed to drive an internal scan chain.

BSCAN_SPARTAN3A TCK
TDO1 TMS
CAPTURE
DRCK1
DRCK2
RESET
SEL1
SEL2
SHIFT
TDO2 TDI
UPDATE
UG331_c22_02_082306

Figure 21-2: BSCAN_SPARTAN3A Symbol

Spartan-3 generation FPGAs provide hooks for two user-definable scan chains through the
USER1 and USER2 instructions. These instructions can be used to provide access to the
user design through the JTAG interface. To take advantage of the optional USER1 and
USER2 instructions, the designer must instantiate the BSCAN_SPARTAN3A macro in the
source code, and wire it to the user-defined scan chain. Only one BSCAN_SPARTAN3A
component can be used in any single design.
The BSCAN_SPARTAN3A component is generally used with IP, such as the ChipScope™
PRO analyzer, for communications via the JTAG pins of the FPGA to the internal device
logic. When used with this IP, this component is generally instantiated as a part of the IP
and nothing more is needed by the user to ensure it is properly used. If, however, custom
access is desired, the BSCAN_SPARTAN3A component can be instantiated and connected
to the design to get this functionality. All appropriate pins should be connected to the
internal logic. For details on using boundary scan in Spartan-3 generation FPGAs, see
Chapter 9, JTAG Configuration Mode and Boundary-Scan, in UG332.

530 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010
 R

Related Materials and References

Related Materials and References

• BSDL File Download:
http://www.xilinx.com/support/download/index.htm
• ISP Standards and Specifications:
http://www.xilinx.com/products/design_resources/config_sol/isp_standards_specs.htm
• Xilinx Software Manuals and Help:
http://www.xilinx.com/support/documentation/dt_ise.htm
• Boundary-Scan and JTAG Application Notes:
http://www.xilinx.com/support/documentation/boundary_scan_and_jtag.htm
• Spartan-3 Generation FPGAs Configuration User Guide:
http://www.xilinx.com/support/documentation/user_guides/ug332.pdf
• Xilinx data sheets:
http://www.xilinx.com/support/documentation/data_sheets.htm
• Intellitech Boundary-Scan tools and tutorials:
http://www.intellitech.com

Spartan-3 Generation FPGA User Guide www.xilinx.com 531

UG331 (v1.7) August 19, 2010
 R

Chapter 21: Using Boundary-Scan and BSDL Files

532 www.xilinx.com Spartan-3 Generation FPGA User Guide

UG331 (v1.7) August 19, 2010