( ESNUG 297 Item 10 ) --------------------------------------------- [7/30/98]
Subject: (ESNUG 295 #3 296 #9) Cadence's PB-OPT Trounces reoptimize_design
> With regard to Cadence's PB-OPT, I implemented this tool flow (in a former
> life) and used it on very high performance designs. The results seemed to
> match Cadence's claims and this option enabled me to achieve a one-pass,
> timing-convergent, design flow from RTL->layout. This additional feature
> allowed me to replace my previous methodology, which was to iterate
> (sometimes > 15 times) through Design Compiler's reoptimize_design and
> Cell3/Silicon Ensemble's ECO place and route.
>
> - Brian Arnold
> Fusion Networks Corp. Longmont, CO
From: Daniel Leduc <Daniel.Leduc@matrox.com>
To: briana@innie.sitera.com (Brian Arnold)
Brian,
Your experience with PB-Opt (as described in ESNUG #296) corresponds
to the "holy grail" of optimization that we can never get our hands on.
We evaluated PB-Opt last fall, and it did not seem stable enough at that
point to be used in our design flow. More importantly, feedback we had
from many people at Cadence is that the GCF-based (as opposed to the older
way of doing timing-driven layout) is still not stable enough to be
incorporated in our design flow, and that very few people within the
Cadence organization are knowledgeable with GCF.
When I read the comments from the Cadence marketing guy in ESNUG 295 #3,
I was sure it was total hype, if not outright a lie, but your comments
give me some hope...
Could you tell us what version of PB-Opt your worked with, and whether
it was GCF-based (GCF being that new scheme whereby you specify global
constraints for a design, such as clock period and input arrival times,
just as you would do with a synthesis tool, as opposed to using the
older path-based constraining) ?
Any additional insight (e.g. pitfalls to avoid, etc.) would be greatly
appreciated.
- Daniel Leduc
Matrox Graphics Dorval, Quebec, Canada
---- ---- ---- ---- ---- ---- ----
From: briana@innie.sitera.com (Brian Arnold)
To: Daniel Leduc <Daniel.Leduc@matrox.com>
Daniel,
I was using the latest version of PBOPT, I don't remember the exact
version number. I have since left the company I was using it on and I
don't have access to the bits any more. I agree, the version Cadence
had last fall was brain dead, and it was called PBS (Placement Based
Synthesis) and not PBOPT. The latest version has changed substantially
and is much better. The timing reports actually correlated to what I
saw both in Synopsys and PathMill.
I was using the complete timing-based flow that took advantage of the
GCF format. However, to do this you need a tlf format for your synopsys
library. I created my own Synopsys library, so this wasn't a problem.
To do so, Cadence has a program, syn2tlf, that will create a .tlf for
your library, but you need to have the .lib for your library. To create
a gcf file for a given block, I wrote a translator that converted a
synopsys constraints file to a .gcf file and it worked quite well.
Another thing, I didn't use Cadence's CTGEN tool either as I couldn't
tolerate the amount of skew that CTGEN introduced. Therefore, I created
my own clock tree solution that worked much better and introduce near
zero skew.
As for pitfalls, it did take me a while to get the whole flow up and
running. Read the literature Cadence has on its Timing driven flow and
make sure you input the gcf files in the appropriate manner as there are
different options to reading them in.
- Brian Arnold
SiTera Corp. Longmont, CO
|
|
