Prospective IP Search
EECS 290A
Homework #2
4/11/99
Rhett Davis
Summary
|
Multiplier |
Viterbi Decoder |
Viterbi Decoder |
Viterbi Decoder |
Ethernet PHY Core |
Company |
NTNU |
SICAN-Microelectronics |
Mentor/Inventra &
IBM |
Hantro Products Oy |
Mysticom |
Block Type |
Soft |
Soft |
Soft |
Soft |
Hard + Soft |
Performance |
- |
- |
0.8 - 85 Msps |
45 MHz |
- |
Technology |
- |
- |
0.5 um (Mentor)
1.8V, 0.25um (IBM) |
Altera FPGA |
3.3V, 0.35 um |
Size |
to be counted |
- |
3200-34000 gates |
500 LC's,
5120 RAM bits |
- |
Cost |
free |
~$100,000 |
negotiable |
- |
negotiable |
Includes |
HDL source |
HDL source, test bench, support services |
HDL Source, targeted for IBM's 0.25 um process |
HDL Source, test-bench, test vectors, synth. scripts |
GDSII for hard portion,
Netlist & SDF for soft portion |
Contact Info |
- |
Andy Karsten
(650) 871-1494 |
Marnie Mar (at IBM)
(650) 966-6703 |
- |
Debbie Vogt
(650) 520-7656 |
Sources
Most examples were found using the Electronic Design Technology & News
Network (EDTN) Virtual
Components Yellow Pages.
Multiplier
NTNU - Norwegian University
of Science and Technology
One block which often takes a long time to build is a multiplier.
The EDTN Yellow-Pages had one reference to a Multiplier IP block, which
turned out to be a web engine that generates VHDL or Verilog source on
demand. The source for a Booth-encoded multiplier with 2 pipeline
stages is given above.
It would be helpful for ASIC designers to be able to drop multipliers
into a design in much the same way that a standard cell can be used, and
perhaps this kind of generator is a step in that direction.
Viterbi
Decoder
SICAN Microelectronics
SICAN Microelectronics offers this Viterbi decoder core and mentions
it on their web site but offers no information on it. I talked briefly
with Andy Karsten at SICAN. He informed me that they offer the source
code for around $100,000 and support services for customers (primarily
fabless semiconductor houses) to integrated it into their process and get
it functioning properly. He said that they get such different results
with each customer that posting data sheets and area/power/speed characterization
info does not give an accurate idea of the decoder's capabilities.
Viterbi
Decoder
Mentor/Inventra
Repackaged
by IBM
Mentor Graphics/Inventra offers a Viterbi decoder core which IBM repackages
and offers as a block for use with their "Blue Logic" fabrication services.
I spoke with Marnie Mar at IBM who told me that their customers are still
so few and far between that no immediate dollar value can be given to the
Viterbi decoder core. Customers generally negotiate for a "Design
Kit" which includes a software suite of design tools and agree to purchase
a "production quantity" of chips. At this point, the negotiations
for the IP core begin.
No characterization numbers were available for the IBM core, so all
numbers reflect the Mentor/Inventra predictions for a 0.5 um process.
Important Parameters:
-
Block Size - The number of bits shifted in and out for each step
of the coding. Fixed to 1 for this decoder.
-
Constraint Length - The number of blocks used for each step of the
coding. The complexity of the hardware is, in general, exponential
with respect to this value.
-
Code Wordlength - The number of bits in each code word. Code
rate = block size/code wordlength
-
Modulation Scheme - The modulation scheme assumed for computation
of Branch Metrics. Different assumptions about the modulation scheme
can have a drastic effect on the complextity of the hardware.
-
Code bit word length - The number of sampled bits per code bit.
This determines the complexity of the branch metric computation hardware.
-
Truncation Depth - the number algorithm steps to store in memory
for traceback. In general, a smaller traceback depth increases the
chance that the algorithm will have to force a decision before an optimal
answer can be found.
-
Accumulated State Metric Wordlength - The wordlength of values in
the traceback memory
In order for a fair comparison between different Viterbi decoders, all
of these parameters should be the same. Mentor characterized the
module for an ASIC implementation in a 0.5um CMOS process with the following
parameters.
Block size |
Constraint Length |
Code Wordlength |
Modulation Scheme |
Code Bit
Word Length |
Truncation Depth |
Accumulated St.
Mtr. Wordlength |
1 |
7 |
2 |
BPSK (apparently) |
3 |
48 |
not specified |
Because the block allows resource sharing of the ACS (Add/Compare/Select)
blocks, the performance and gate count vary with the number of ACS units
assumed.
# ACS units |
1 |
8 |
64 |
Speed (Msps) |
0.810 |
6.32 |
85 |
Gates |
3,200 |
19,400 |
34,900 |
Viterbi Decoder
Hantro Products Oy
Hantro also offers a parameterizable Viterbi decoder which has the same
basic architecture as the Mentor/Inventra core. The ACS units are
also assumed to be resource-shared with the Hantro Viterbi-decoder.
It has been characterized on an Altera EPF10K10 FPGA with the follow characteristics
and results:
Block size |
Constraint Length |
Code Wordlength |
Modulation Scheme |
Code Bit
Word Length |
Truncation Depth |
Accumulated St.
Mtr. Wordlength |
1 |
7 |
2 |
BPSK or QPSK (not specified) |
3 |
55 |
not specified |
No. of ACS units |
No. of Logic Cells |
Ram Bits |
Max Speed |
4 |
500 |
5120 |
45 MHz (Msps not available) |
In addition to offering HDL source code, Hantro also offers a post-route
test bench and test vectors as well as simulation and synthesis scripts.
These could reduce design time if they were compatible with the design
groups tools.
Ethernet
PHY Core
Mysticom
Mysticom offers an Ethernet PHY core which consists of an analog hard block
coupled with a digital soft block. The block is intended for
a 0.35 um digital process with 3 metal layers. From my discussions
with Debbie Vogt at Mysticom, they negotiate with specific fab houses to
get design tools which they use to create a GDSII analog block and a standard
cell netlist (w/SDF file) for the digital portion. I'm still waiting
for power and area estimates from her, but I'm not sure that they will
be forthcomming...
Rhett Davis
wrdavis@eecs.berkeley.edu