( DAC 04 Item 44 ) --------------------------------------------- [ 02/09/05 ]
Subject: Libtech and Z Circuit
EBERT DOES LIBRARIES -- I like small EDA companies. They listen to you
and they provide great support. Mehmet Cirit, the CEO of Libtech (a 3 man
company), is the king of small. But don't confuse small with unstable;
he's been doing library characterization tools since 1994 -- outlasting all
his rivals. (And for those who want to compare two libraries, you should
take a look at the Z Circuit tool in the lower part of this section.)
We had a cell library characterization tool, until Silvaco sued
Circuit Semantics and won. Then we were left without a tool. Silvaco
informed us that for $150 K we could legally use the tool that we had
already purchased from Circuit Semantics. So basically, these two
companies left us out in the cold without solution!
We had to go out and find another one. We recently evaluated several
of the Library Characterization tools, and chose Libtech. Since then
we have characterized an entire library and we are very happy with our
decision to go with Libtech. We received on-site training, and the
customer support is great, every question is answered promptly and to
our satisfaction. Their tool does a great job with verifying the
models up-front so that errors in the models will be caught sooner
rather than later. All of the models are verified after they are
generated to ensure that they are error-free.
I'm an ASIC designer, so I was very happy to find that the models
worked the first time and I didn't have to do any debugging like
I used to with Circuit Semantics.
- [ An Anon Engineer ]
I've been using the Libtech software for a little over 2 years to do
cell characterization and build timing libraries. The software is very
typical of a small shop where functionality is more important than form
(which I like). Flashy GUIs are nice but they can drive me crazy.
Probably the most important feature of the software is the user-support.
They came down to help me install the software and set it up. When I
have problems running the software or run into some obscure corner case,
I can call or send an email to them and Mehmet (the president) gets back
to me, usually right away. Given that this is only a small part of my
job and I don't want to have to become an expert on the tool, that's
pretty important to me.
The Libtech software runs from a Perl configuration script so, if you're
a Perl person you'll have no problem figuring out how to configure it.
If you're not a Perl expert, why aren't you? :-) The licensing is
custom, which can drive me crazy. Then again, pretty much all licensing
is annoying sooner or later. Some of the characterization can take a
very long time so I wish there was an easy (cheap?) way to multi-thread
them. I understand Libtech's point of view where if you want to run
multiple machines you should have multiple licenses. However, my use
doesn't justify the expense and I end up with characterization times on
the order of one or two weeks.
- Tim Morse of TDK Semiconductor
We have been using Libtech's bundled tool package for over a year now.
The package comprises of libopt and libchar. We are extremely pleased
with the tool and the results we get. The tool offers flexibility
when setting up the flow. Company support is extremely good, which
makes setting up flows for single event upset protected (SEUP) flops,
or other non-standard digital cells trouble free.
Their optimizer is essential to us in ensuring low-power optimization
at cryogenic operation. The libchar characterization tool provides
output in easy text-base, Verilog, Synopsys, etc., formats. (Output
formats are extremely configurable allowing us to implement specific
timing and corner checks for each cell.)
We have been able to depend on Libtech over the past year in producing
reliable digital libraries for our cryogenic operation environments.
- Raphael Ricardo of Rockwell Scientific
We used Z Circuit services for lib re-characterization to a different
operating point, to basically reduce some of the over-margining that
normally happens during design. Our experience was very smooth and
pleasant. No surprises. Z Circuit provided us with detailed
comparison reports for each of the timing arcs, for each of the cells
between the two operating points, in a consolidated form to:
1. Validate the correctness of the re-characterization, i.e. make
sure that everything that's unexpected or outside of expected
range is explained for completely.
2. Give us a good feel as to what we can expect in terms of the
performance of the library.
The Z Circuit Library Analyzer in general is very versatile. It can
be used to compare libraries to understand their impact on a given
design. Edmond and Fred (the two key people at Z Circuit that I had
the opportunity to work with) are very knowledgeable, and we used
them as sounding boards for some of the issues we had to deal with
during our design. They provided valuable feedback.
- [ An Anon Engineer ]
I have been using Z Circuit tools for couple of months. We have a an
internal set of scripts to do the same job, because comparing timing
library at the cell level can give a good insight to ASIC designers in
terms of using the new library release compared to the older releases.
Z Circuit can compare two timing libraries and gives some statistical
information about the differences.
Strengths:
- Z Circuit has a good theory and insight regarding averaging for lib
differences and all the statistical information it gives. It shows
developers are well informed about the timing shift and how it may
be manifested in the libraries. It does a good job in comparing two
libraries from one IP vendor; it can compare libraries from different
vendors if you map equivalent cells properly.
- Z Circuit makes intermediate cells based on available drive strengths
in the libraries, which helps to compare two libraries with different
range of drive strength.
- Ed/Fred are two knowledgeable supports behind their tools. I really
appreciate their support and help.
Short comings:
- It is not strong enough to detect "cell functionality" definition,
all it does is a simple text comparison to find the equivalent cells
function (or "when" conditions). This text comparison is not enough
because different IP vendors may have different nomenclatures.
- In terms of mapping equivalent cells in two libraries (when cells are
named differently) it is quite cumbersome to map them using cell
footprint.
- It uses the same scale factor used to derive delay/transition values
for intermediate drive strength for leakage/power values. This is
not a right assumption because leakage is not a linear function of
input_slew/output_load.
As part of good support, all the above drawbacks are supposed to be
resolved. I hope Z Circuit keeps its good effort to enhance Z Package.
- Rubil Ahmadi of ATI
With regards to the Z Circuit Library Analyzer, we didn't really
compare it to the Libtech or Magma CellRater tools. What we did see
in the Z Circuit tool is that it provided us with an easy means to
quickly and easily compare the timing numbers between cells in 2
libraries to one another. From the comparisons of the cells, we
were able to determine that strengths and weaknesses of the cells
in the library, which allowed us to better tune the cells that had
issues with it.
Z Circuit's strength is that the software can perform its comparison
of the library data very quickly. However, the process can slow down
if the designer relies on the program to create the mapping table
between the 2 libraries. Also, the amount of data that is created in
the comparison is large and can be a bit overwhelming. Luckily, there
is a summary of the key comparison points at the end of the output file.
Overall, we are satisfied with the Z Circuit results and operation.
It's easy usage, and summary of the key points has made it easier for
us to point out issues between libraries.
- [ An Anon Engineer ]
Index
|
|
