Fall 2015: CMPE 315 Project Specification

Project Specification: Cache Design

(Please refer to the webpage for any changes to this specification over the next couple of weeks).
Assigned: Oct 23th
Due: Last day of classes, Dec 8th.
Code Submission: In three weeks, details will be provided on the webpage and discussed in class.

Description:

CD[0] CD[1] CD[2] CD[3] GND CD[4] CD[5] CD[6] CD[7]

BUSY

CA[0]

CLK
Output Enable

CA[1]

Register

Register

CA[2]

RESET

Mux

MD[7]

Decoder

CA [3]

State Machine

MD[6]
1 3

CA [4]
CA [5]

8
8
Number of bits

8
blocks

MD[4]

Cache

CA [6]

Valid Tag

CA [7]

MD[3]
4 words/ block

MD[2]

Hit/Miss

RD/WR

MD[5]

MD[1]

Output Enable

MD[0]

START

MA[0] MA[1] MA[2] MA[3] VDD MA[4] MA[5]MA[6] MA[7] ENABLE

Input Output

Input

Output

Core GND

Core VDD

Corner

Fall 2015: CMPE 315 Project Specification

Details
Your assignment is to design, implement and simulate in VHDL, and do the layout for a cache
block described below. The details of a cache and its use will be discussed in the class.

Direct Mapped Cache (see extensions for extra credit)

Byte addressable
4 byte blocks, requires 2 address bits to decode
8 blocks, requires 3 address bits to decode
8 bit address, 3 bit tag
Write through with no write allocate
Read hit: 2 clocks
Write hit/miss: 3 clocks
Read miss: 19 clocks

Chip interface
A cache usually needs two address and data interfaces, one to the upper level cache/processor and
another to lower level cache/memory. In your design, you have both these interfaces, 8-bit address
and data connections to both the CPU and the memory.
CPU Address (CA):8-bit address input from the CPU for both read and write requests. Address
will be provided by the CPU on the rising edge of clock along with the start signal
CPU Data (CD): 8-bit input/output data bus. Used as input for write requests and output for read
requests. Data will be provided on the rising edge of clock for write requests along with the start
signal
Start: Handshaking signal indicating the start of a read/write request from the CPU, goes high on
a positive edge. (Internally your chip works on the negative edge. The address, start, read_write
control and data, will be setup half a clock cycle before you start operating on the negative edge)
Busy: Handshaking signal indicating to the CPU that your chip is processing the previous request.
Read_Write (RD/WR): If signal is high CPU is requesting a read operation, if low a write operation
Reset: Master reset to the chip. A high on reset should invalidate all the entries in the cache and
reset your state machine to its reset state. Reset any other registers as required.
Clock (clk): Clock input to the chip. The CPU inputs/outputs and the memory inputs/outputs will
be synchronized with this clock signal also.
Memory Address (MA): Address output to the memory in case of read miss. The last two bits of
the address should always be 00 i.e. you provide the address for the last word in the block. The
memory controller will automatically increment the address four times and provide you data for
the whole block (4 bytes).
Memory Data (MD): Data input from the memory. It takes the memory 8 clock cycles after getting the enable signal to provide the first byte of data. It will sequentially provide four bytes of
data required for the whole block. The first data byte will become valid on the 8th negative edge
after asserting enable and will stay stable for 2 clock cycles. The next byte will be provided on the

Fall 2015: CMPE 315 Project Specification

10th negative edge after asserting enable and will stay stable for 2 clock cycles. The last two data
bytes will have similar timing and will be provided on the 12th and 14th negative edges.
Enable: The enable signal is an output requesting the memory to perform a read operation starting
at the address provided on MA. The memory returns four bytes as explained above starting at the
8th clock cycle after asserting this signal.

Operations
The cache has 8 blocks, with four bytes per block. Therefore, you need 3 bits to select one of the
8 blocks and 2 bits to select the correct byte from the block. The remaining 3 bits in the address
are used as a tag.
3
Tag

3
Block
Offset

2
Byte
Offset

The start signal from the CPU marks the beginning of a operation. The CPU provides the address
for read operations and the address as well as data for write operations. All these signals are provided on the rising edge of clock. Your chip internally works at the falling edge of clock so the
signals are stable before you need to latch them thus avoiding setup time violations. The signals
will be removed by the CPU on the next rising edge of clock. The busy signal triggers the removal
of these signals. Busy is an output that should go high on the negative edge after receiving the
start signal from the CPU.
Read Hit
For a read operation, the CPU turns start high, provides the address and turns read_write signal
high on the positive edge of clock. On the negative edge you need to latch all these required signals and turn on the busy signal. The inputs will be removed by the CPU on the positive edge
once it receives a busy signal. You determine whether the data being referenced is in the cache, by
comparing the tag bits of the address with the tag bits stored in the block given by the block offset
in the address. Also you need to check that the block is valid. Simultaneously you need to read the
correct byte from the cache. If the tag and valid checking operation signals a hit, output enable
should go high on the next negative edge, the data should be latched in the output register so that
it stays stable for one whole clock cycle and the busy signal should be turned off. The CPU will
read the data off the data bus on the positive edge. The next operation could be requested on any
subsequent clock cycle.
Write Hit
CPU signals follow the same timing as the read, but read_write is set to low and the data to be
written is provided. The required inputs should be latched on the negative edge, busy should be
turned on and tag/valid compare should be performed. If the result is a hit, on the next negative
edge the data should be written to the correct byte in the selected block, busy should still stay
high. On the second negative edge after receiving the write request busy should go low signalling
the end of the write operation. The CPU will provide a new input on any subsequent clock cycle.

Fall 2015: CMPE 315 Project Specification

Write Miss
As we are using write through with no write allocate, you dont have to get the block from memory on a write miss. The timing and operations are similar to the write hit case. Only difference is
that as you dont have a hit, no write operation should be performed on the cache. The busy
should go low after two clocks from receiving the write request as in the hit case and the CPU can
provide a new request after busy is low.
Read Miss
The signals are provided by the CPU as described in the read hit operation. On the second negative edge of the clock, you get a miss in this case i.e. the block is not present in the cache. On this
edge you keep busy high, turn on the enable signal to the memory requesting a block read and
provide the byte address for the last byte in the block that you need. The memory latches the
address once it sees enable is high on a positive edge. Enable should be turned low after one clock
cycle (third negative edge since receiving the request). The memory needs 8 clock cycles to
access the data. After 8 clock cycles, the memory provides the data for the last byte on the negative edge. The data will stay stable for 2 clock cycles. You should setup the correct address for
write and update the cache with the new data byte. During the first write operation you should
also update the tag and turn on the valid bit. Write timing is very critical, if there are any glitches
in the address lines or the data lines while the write signal to the cache is enabled, you could
write the wrong data or overwrite another location in the cache. Your write signal should be
turned on exactly for one clock cycle from the positive edge after the data is valid to the positive
edge before the new data byte is provided. The address should be stable for 2 clock cycles, similar to the data. The address should be stable for half a clock cycle before the write is turned on and
for half a clock cycle after the write is turned off. Once the bytes are written, the correct byte
requested by the CPU should be read and latched on the output register, busy should be turned off
and the output enabled on a negative edge. Output enable should go low after one clock cycle as
in the case of read hit. CPU will read the data on the positive edge after output enable goes high
and will provide a new request on any subsequent clock cycle.

Cache Cell
Usually SRAMs and CAM (content access memory) cells are used for caches. However, extensive simulations need to be performed to evaluate the effect of loads on these designs. To avoid
such situations use the following cell for your cache, which provides, simultaneous read and write
capability at the expense of extra area.
Write Data
Write Enable

Positive level
sensitive
Latch

Tx
Gate

Read Enable
Read Enable
Read Data

Chip Waveforms
The chip timing waveforms explaining the above operations are provided as a separate document.

Fall 2015: CMPE 315 Project Specification

Extra Credit (25%)
Make a two-way set associative cache i.e. instead of a 32 byte cache you have a 64 byte cache
implementing 2 sets. For block replacement implement a least recently used (LRU) scheme.
Report Requirements and Deadlines
Detailed report requirements will be provided for both the submissions. The first submission
(VHDL code, simulation results and imported schematics) will be due in three weeks and the full
project submission (corrections required from the first submission, the chip layout, LVS and simulation results) will be due before the last lab session.
Late penalty will be applied for submissions that dont make the above deadlines. There is
an extra credit component to this project worth 25% (that translates to more than half a letter grade in the overall class grade!!!). If you have intentions of doing the extra credit think
about how you will integrate it into you design before starting the initial design process. The
extra credit part is designed to be relatively easy to integrate into your code and layout once
the primary project is done.