PCI Express PIPE Overview

MindShare, Inc.

Dave Dzatko

March 2004

Introduction4
PIPE Architecture..4
MAC Architecture.6
PHY/MAC Interface..7
PCS Architecture...9
PMA Architecture12
PIPE Functionality and Features..13
PIPE Signal Organization14
PIPE PLL.15
Transmitter Sub-block.16
Receiver Sub-block.16
PIPE Signal Descriptions18
PIPE Summary20

PCI Express PIPE

Introduction
PIPE, which stands for the Physical Interface for PCI Express Specification
developed by Intel, has the stated intent of providing a standard interface between
the internal logic of a PCI Express design and the analog and high-speed circuitry
required to implement the serial link. This purpose of this functional separation is
to allow ASIC and integrated circuit designers to focus on the PCI Express device
core, Transaction, Data Link and logical Physical Layers, while relying on the
PIPE-compliant physical design (PHY) for the electrical interface of the design.
Other vendors can then provide macrocells to handle this high-speed analog
interface layer, resulting in reduced time and risk for the overall design cycle.
The PIPE spec defines standard functionality that a PIPE-compliant PHY needs to
implement, as well as a standard parallel interface between the PHY and the
internal logic referred to in the spec as the Media Access Layer (MAC). The
MAC in turn connects to the PCI Express Data Link Layer logic. The PIPE spec
builds on the PCI Express base spec, so it should be noted that a working
knowledge of that document is essential for a good understanding of the PIPE
spec. This paper is based on the 1.0 version of the PIPE spec, and provides a brief
introduction only. For more information on PIPE, refer to Intels web site1. For
more information on PCI Express, refer to Mindshares web site2 or the PCI
Special Interest Group site3.

PIPE Architecture
The PIPE architecture block diagram is shown in Figure 1. As mentioned earlier,
the Data Link Layer (DLL) logic interfaces with the MAC, although the spec
doesnt define this interface or any others except the one between the PHY and
the MAC. All the other interconnects are understood to be implementation
specific. The interface between the MAC and the PCS (Physical Coding Sublayer) is a dual simplex, parallel bus called the PHY/MAC interface. The PCS and
PMA (Physical Media Attachment Layer) interface is not defined by the spec.
The functionality of the PCS and PMA are described conceptually in the PIPE
and PCI Express specifications but, again, no specific implementations are
implied. The PMA contains the high-speed analog and digital circuitry that
connects to the PCI Express link via differential transmitters and receivers. The
PIPE spec notes that there is overlap between it and the PCI Express spec, and
states that in case of any conflict between them, the PCI Express spec takes
precedence.
1

http://www.pciexpressdevnet.org/apps/org/workgroup/devnetgf/pipe/
http://www.mindshare.com/knowledge/pciexpress_overview.html
3
http://www.pcisig.com/home
2

PCI Express PIPE

TO
DLL

MAC
(Media
Access
Layer)

PHY/
MAC

PCS
(Physical
Coding
Sub-layer)

PMA
(Physical
Media
Attachment
Layer)

RX

TX

PIPE SPEC
PCI-E PHY SPEC
Figure 1. PIPE Architecture

MAC Architecture
The MAC contains many of the PCI Express logical Physical Layer circuits (such
as the Link Training and Status State Machine (LTSSM), data scrambling, 8b/10b
encoding, and byte striping), and functions as the bridge between the DLL and the
PHY/MAC interface.

MAC
(Media
Access
Layer)
TO
DLL

PHY/
MAC

PCS
(Physical
Coding
Sub-layer)

Implementation
Specific
(Not Specified)
Figure 2. MAC Architecture
4

PCI Express PIPE

.

PHY/MAC Interface
The PHY/MAC interface is the major normative area of the PIPE spec, as shown
in Figure 3. The pins that make up the interface and their functionality are
described in the spec, and timing diagrams are provided to show the synchronous
timing relationships, but no detailed timing parameters for the signals are given.
Instead, the spec provides a description of implementation-specific timings that a
vendor of a PIPE-compliant PHY macrocell or discrete chip must specify. Much
of the functionality of the PHY/MAC Interface is described in the spec using
timing diagrams, which are not repeated here.

MAC
(Media
Access
Layer)

TxData (8 or 16)
TxDataK (1 or 2)
Command (7)

PCS
(Physical
Coding
Sub-layer)

RxData (8 or 16)
RxDataK (1 or 2)
Status (6)
PCLK

Figure 3. PHY/MAC Interface

The PHY/MAC Interface is a parallel interface for transferring data to be
transmitted on the PCI Express bus. The width of this parallel interface for bytes
of data is shown as either be 8 or 16 bits in each direction. In addition to the data
path, the PHY/MAC Interface includes one Control (K) lines in each direction for
each byte to indicate whether the corresponding byte is a data or a control
character. If the interface is 16 bits wide, there will be two Control lines, but only
one is needed if the interface is 8 bits wide.
Seven signals from the MAC to the PCS make up the command bus that allows
the MAC to tell the PHY to begin performing a variety of tasks, including:
Begin Receiver Detection
Enter the External Loop-back state
Enter the Electrical Idle state
Set the current running disparity to positive or negative
Invert the polarity of the received data
5

PCI Express PIPE

Reset the PHY

Power down the PHY into various Power Management (PM) states

The PHY/MAC Interface also includes a 6-bit status bus for communication of
PHY status from the PCS to the MAC. The status bus has codes and signals that
indicate conditions including:
The PHY has obtained symbol lock
The received data is valid
The PHY has completed various PM state transitions
The PHY has detected that a receiver is attached
The PHY has detected an Electrical Idle state on the Link
There are 8 additional status and error codes defined
The PHY includes a clock output called PCLK which is used to synchronize data
transfers across the parallel PHY/MAC Interface. If the parallel interface is 16
bits wide, then PCLK runs at 125 MHz, but if the interface is 8 bits wide, then the
PCLK runs at 250 MHz so as to maintain the data rate with respect to the serial
link.

PCS Architecture
The Physical Coding Sub-layer (PCS), although part of the Logical Sub-block of
the PHY, is included within a PIPE-compliant discrete device or macrocell. The
PCS supports the PHY side of the standard PHY/MAC Interface as well as an
implementation-specific interface between it and the Physical Media Attachment
Layer (PMA). No specifications are provided in the PIPE document for this
interface.
The clock reference input (CLK) is used by the PHY to generate the internal bit
rate clocks for transmitting and receiving PCI Express data. Specifications for this
implementation-specific clock, which will be used internally to generate the bitrate clock for the PHY transmitter and receiver as well as the PCLK for the
PHY/MAC interface, must be provided by PIPE-compliant PHY vendors. Spread
spectrum modulation that matches the system reference clock modulation is
permitted for this signal.
The 8B/10B logic required by PCI Express resides in the PCS. The input to the
8B/10B encoder is the data or control character from the PHY/MAC Interface, the
value of the current running disparity, and the TxDataK signal that is needed
because the encoding for a data character is different from that of a control
character.
The PCS also contains the elastic buffer used to compensate for the slight
variations in frequency between the transmitters clock and the clock used by the
receiver to process the incoming data.

PCI Express PIPE

CLK (Implementation Specific)
PHY/
PCS
MAC
(Physical
Standard Coding
Interface Sub-layer)

Implementation
Specific
(Not Specified)

Figure 4. PCS Architecture

After reset, the first state of the link initialization and training process is the
receiver detect, during which the transmitter detects whether a receiver is attached
to it at the other end of the link. The PCS contains the logic that controls the
receiver detection and reports the detect status to the MAC via the PHY/MAC
Interface.
The PCS also contains a Phase Lock Loop (PLL) to generate the internal, high
speed clocks used for the PHY based on the CLK input.

PMA Architecture
The Physical Media Attachment (PMA) Layer implements the high-speed analog
and digital circuitry for PCI Express signaling, including the differential drivers
and receivers for each lane of a link. Although the data is serially transmitted over
the link, the connection from the PCS to the PMA is a ten bit wide,
implementation-specific parallel interface. The high-speed serialization and deserialization logic (SERDES), needed to create the serial data stream of PCI
Express, resides in the PMA.
7

PCI Express PIPE

PCS
(Physical
Coding
Sub-layer)

PMA
(Physical
Media
Attachment
Layer)

RX
One Lane
of the Link
TX

Figure 5. PMA Architecture

PHY Functionality and Features

A PIPE-compliant PHY discreet or macrocell, as shown in Figure 6, is designed
to handle all the low-level PCI Express protocol and high-speed PCI Express
signaling. To review, the required functionality provide by the PIPE-compliant
PHY includes:
The SERDES or serialization/de-serialization circuitry
The 8B/10B Encoder and Decoder
Analog buffering for the received data and the data to be transmitted
Elastic buffering for clock tolerance compensation
Receiver detection circuitry
Disparity control for use in transmitting the PCI Express compliance
pattern
Beacon transmission and reception circuitry for implementing remote
wakeup capability
The PIPE spec does not define the specific internal architecture or design of the
PCS and PMA functional blocks. These functional blocks simply represent highlevel functionality that must exist in a PIPE-compliant PHY. The PIPE
specification does describe, in general, the architecture and behavior of the PCS
and PMA.

PCI Express PIPE

PHY/
MAC

PCS
(Physical
Coding
Sub-layer)

PMA
(Physical
Media
Attachment
Layer)

RX

TX

Figure 6. PIPE Functionality

PIPE Signal Organization

A PIPE-compliant PHY will have signal groups as shown in Figure 7. There are
four basic signal groups:
1. Clocks to and from the PLL
2. Transmitter Data and Control
3. Receiver Data and Control
4. Command and Status

PCI Express PIPE

CLK

PLL

PCLK
TxData (8 or 16)

Transmitter
Sub-block

TxDataK (1 or 2)

TX+, TX-

Command (7)
Status (6)
RxData (8 or 16)
RxDataK (1 or 2)

Receiver
Sub-block

RX+, RX-

Figure 7. PIPE Signal Organization

PIPE PLL
The PLL generates the PCLK used in synchronizing the parallel PHY/MAC
Interface based on the CLK input. The PCLK frequency is 125 MHz for 16-bit
implementations and 250 MHz for 8-bit implementations. The PLL will produce a
250 MHz clock used as an input to the 8B/10B encoder and decoder and the
elastic buffer. The PLL will produce a 2.5 GHz clock that is used as an input to
the SERDES and the clock recovery circuitry.

CLK

PLL
(Phase
Lock
Loop)

PCLK

250 MHz
2.5 GHz

Figure 8. PIPE PLL

10

PCI Express PIPE

Transmitter Sub-block
The Transmitter Sub-block of the PHY has 4 distinct sections. Data to be
transmitted comes from the PHY/MAC Interface at the rate of either 8 or 16 bits
per rising edge of the PCLK into an optional 16-to-8 bit converter. This converter
would, of course, be unnecessary if the PHY/MAC interface is already 8 bits
wide, but it would ensure delivery of 8 bits per tick of a 250 MHz clock to the
8B/10B Encoder if a 16-bit interface is used.
The 8B/10B Encoder converts the 8-bit character into a 10-bit symbol at 250
MHz, using the TxDataK input to determine whether it is encoding a control or
data character. The TxCompliance input signal is used to force the current
running disparity at the encoder to be negative, permitting forced error conditions
for compliance testing.
The Loop-back path is shown to define the flow of data when the PHY is placed
into the loop-back mode. In that case, the data to be looped back from the
receiver bypasses most of the transmitter logic and goes directly to the parallel-toserial converter.
The parallel to serial converter creates a serial bit stream running at 2.5 GHz
using a clock derived from the internal PLL and the differential driver sends the
bit stream out in accordance with the PCI Express specification. The TxElecIdle
signal informs the driver that it should enter the Electrical Idle state. If this is not
set, the driver should be sending something either data during normal operation,
or else the beacon signal if operating in lower power states. If it is set, the
transmitter should switch to electrical idle as quickly as possible, though it is
required to send at least three bytes of the electrical ordered set first.
TxDetectRx
PCLK 250 MHz
TxData
(8 or 16)

250 MHz

2.5 GHz

TX+
Optional
16 to 8

8B/10B 10
Encode

Parallel 1
To
Serial

Differential
Driver

TX-

TxCompliance TxDataK
Loopback
From Receiver

TxElecIdle

Figure 9. Transmitter Sub-block

11

PCI Express PIPE

Receiver Sub-block
The Receiver Sub-block has nine distinct sections. The received differential data
is converted to a serial bit stream running at 2.5 GHz by the receiver, which can
also detect the Electrical Idle state on the Link and report that to the MAC via the
RxElecIdle status signal. Based on transitions in the incoming data stream, the
Clock Recovery Circuit recovers the frequency of the transmitted clock using the
2.5 GHz clock input provided by the PLL. The recovered clock is used to latch
data into the Data Recovery Circuit, from which it is fed into the Serial to Parallel
converter.
The Serial to Parallel converter supplies a 10-bit symbol to both the Elastic Buffer
and the K28.5 (Comma character) Detect circuitry. The Elastic Buffer is used for
clock tolerance compensation and this is explained in much more detail in a
Mindshare whitepaper4. The Comma Detect circuitry is used to establish symbol
lock on the 10-bit symbols by detecting the special Comma control character that
begins every PCI Express ordered set. This detect circuitry indicates to the MAC
that received symbols are valid with the RxValid status signal. The output of the
Elastic Buffer is fed into the 8B/10B Decoder and the Rx Status circuitry, and is
also routed back to the transmitter for use when the PHY is placed into the
External Loop-back mode.
The Receiver Status circuitry creates the RxStatus signals based on the
characteristics of the received data. If the data has an 8B/10B decode error or
experiences an Elastic Buffer overflow or underflow condition, it will report these
conditions to the MAC with the RxStatus signals. The 8B/10B Decoder converts
the 10-bit symbol back into an 8-bit character and determines the state of the
RxDataK signal based on whether the decoded character is a data or control
character. If the PHY/MAC interface is 16 bits wide, the optional 8-to-16-bit
converter transfers two bytes at the 125 MHz PCLK frequency. If the interface is
8 bits wide, the optional converter is not needed and the received data byte is sent
at 250 MHz.

http://www.mindshare.com/classroom/pciexpress/elastic_buffer_implem.pdf

12

PCI Express PIPE

RxElecIdle

RxValid

RxStatus

RxDataK

Clock
Recovery

Comma
Detect

Rx
Status
RxData
(8 or 16)

10

RX+
RX-

2.5 GHz

1
Differential
Receiver

Data
Recovery

10

Serial
To
Parallel

Elastic
Buffer

8B/10B
Decode

Optional
8 to 16

PCLK
250 MHz
Loopback
To Transmitter

Figure 10. Receiver Sub-block

PIPE Signal Descriptions

CLK
TxData (8 or 16)
TxDataK (1 or 2)
TxDetectRx/Loopback
TxElecIdle
TxCompliance
RxPolarity
Reset#
Powerdown[1:0]
RxData (8 or 16)
RxDataK (1 or 2)
RxValid
PhyStatus
RxElecIdle
RxStatus[2:0]
PCLK

PCS
(Physical
Coding
Sub-layer)

PMA
(Physical
Media
Attachment
Layer)

RX+, RXTX+, TX-

Figure 11. PHY/MAC Interface Signals; Refer to Table 1

13

PCI Express PIPE

Table 1. PHY/MAC Interface Signals
Signal Name
CLK

TxData (8 or 16)

TxDataK (1 or 2)

TxDetectRx/
Loopback

TxElecIdle

Description
This input to the PCS is used to generate the high speed (2.5 GHz) transmit and receive
clocks and PCLK from the PLL. Specifications for this signal are implementation specific,
and must be supplied by the PHY implementer. This signal may use spread spectrum
modulation.
These signals make up the parallel data bus from the MAC to the PCS. If implemented as a
16-bit interface then two character of data are transferred per clock. The first character is
transmitted on bits [7:0] of the parallel bus, and the second character, if there is a 16-bit
interface, is on bits [15:8] of the parallel bus.
These signal(s) indicate the type of characters on the TxData bus. Bit 0 is for the lower byte
on bits [7:0] of the parallel bus and Bit 1 is for the upper byte on bits [15:8] of the parallel
bus. A zero indicates a data character; while a one indicates a control character.
This signal indicates to the PHY to begin the receiver detect process or to begin external
loopback as described in the PCI Express Specification. Like many PIPE signals, the
meaning of this signal depends on which power state the PHY is in. When in the P1 state,
this signal informs the PHY to do a receiver detect, but when in the P0 state, this signal
informs the device to go into the loopback mode as quickly as possible.
When this signal is asserted (active high), it tells the PHY to put the Tx outputs into the
electrical idle state. This is true for all power states. When this signal is deasserted, valid
data from the TxData and TxDataK signals should be transmitted in the P0 state. If the PHY
is in the P2 state, a beacon should be transmitted, if the TxElecIdle signal is deasserted. In
the P0s and P1 states, the TxElecIdle signal must be asserted. The use of this signal is also
affected by the PHY power state, since there are some states in which the transmitter must
be electrically idle. See section 6.14 of the PIPE spec for more detail.

TxCompliance

When asserted, this ordinarily forces he current running disparity (CRD) to negative. As the
name implies, this is useful in conjunction with the transmission of the compliance pattern
to generate test data. It is also used in a multi-lane implementation to turn off any lanes that
are not going to be used, as in a x4 link that must operate as a x1. When both
TxCompliance and TxElecIdle are asserted, that informs the affected lane to turn off and
conserve as much power as possible.

RxPolarity

When asserted active high, this signal tells the PHY to do a polarity inversion on the
received data.

Reset#

When asserted, this active low input to the PHY puts the transmitter and receiver in the
default state

Powerdown[1:0]

These inputs put the transceiver into one of four power states:
1. P0 normal operational mode
2. P0s PCLK remains on, but Rx conserves power; entered when Rx detects
electrical idle. Corresponds to link state L0s.
3. P1 PCLK remains on, both Rx and Tx are in electrical idle. Corresponds to link
state L1.
4. P2 PCLK is off, PHY should minimize power consumption now, since it must
operate within the limits of Vaux. Corresponds to link state L2.
See section 6.3 of the PIPE spec for more details on PHY power management.

RxData (8 or 16)

These signals make up the parallel data bus from the PCS to the MAC. If implemented as a
16-bit interface, two symbols of data are transferred per clock. The first symbol received on
the PCI Express data lines is on bits [7:0] of the parallel bus. The second symbol received
on the PCI Express data lines is on bits [15:8] of the parallel bus.

14

PCI Express PIPE

RxDataK (1 or 2)

These signal(s) are the Data#/Control indicator(s) for the received symbols on the RxData
bus. Bit 0 is for the lower byte on bits [7:0] of the parallel bus and Bit 1 is for the upper byte
on bits [15:8] of the parallel bus. A zero indicates a byte of data; a one indicates a byte of
control.

RxValid

This signal qualifies the data on the RxData and RxDataK signals. When this signal is
asserted, the data on the receiver data bus is valid and the PHY has achieved symbol lock.

PhyStatus

Indicates completion of PHY power state transitions and receiver detection.

RxElecIdle

Indicates that the PCI Express receiver pins have detected an Electrical Idle state on the
Link.

RxStatus[2:0]

Delivers receiver status and error codes for the received data and receiver detect status from
the PHY to the MAC, noting such things as SKP symbol added or removed, disparity error,
elastic buffer overflow or underflow, 8b/10b decode error, etc.

PCLK

This clock synchronizes the parallel interface between the PHY and the MAC. The
frequency is 125 MHz for 16-bit data and 250 MHz for 8-bit data. Spread spectrum
modulation is allowed on this clock.

TX+, TX-

These are the PCI Express differential outputs.

RX+, RX-

These are the PCI Express differential inputs.

15