( ESNUG 459 Item 2 ) -------------------------------------------- [12/14/06]
Subject: ( ESNUG 458 #7 ) We don't want to have to support both CCS & ECSM
> The compactness of its models is the one practical advantage ECSM has over
> CCS. CCS takes approximately 3X more storage compared to ECSM.
>
> - Mehmet Cirit
> Lib Tech, Inc. Saratoga, CA
From: Philippe Moyer <philippe.moyer=user domain=soisic spot mom>
Hi, John,
We're a foundation IP provider company for SOI technologies (standard
cell, IO, memory compiler). We work at 65 nm, and support ECSM in our
characterization flow for best results and accuracy for our customers.
Emmanuel Pacaud in my group developed most of our ECSM characterization
environment. He does not anticipate any impact on our users' design flow,
because we have both NLDM and ECSM characterizations in our .lib file.
If a tool supports ECSM it picks up the ECSM characterization, otherwise
NLDM will be used. Here are some of his findings:
- ECSM timing does not increase the characterization time vs. NLDM
characterization -- we find we are able to do both in roughly the
same time.
- The .lib file significantly increases in size as compared to NLDM
characterization. ECSM + NLDM size is around 160 MB for 600 cells
and NLDM only is around 33 MB.
- The main issue we see with ECSM today is that so far only a few tools
support ECSM characterization -- only some Cadence and Magma tools.
We do not yet have any visibility on the tool run time differences
when using ECSM data vs. NLDM data.
The ECSM voltage characterization is really straightforward. Our only
concern is the sampling point numbers and positions must be well-defined.
We would like to see the coming "Statistical Modeling Standard" for
statistical static timing analysis as soon as possible. It's unclear what
the impact on the characterization side will be.
We just downloaded ECSM's newly released power modeling extension which we
will implement this year, but we have not yet worked with it enough to give
specific feedback.
We also plan to implement CCS timing characterization in the coming months,
and then look at power and noise implementation. With CCS, we are concerned
about the size of the .lib files. One single .lib file supporting NLDM
characterization + ECSM + CCS + PrimeTime-SI characterization will likely be
greater than 500 MB just for one single corner (PVT)! Also, supporting both
standards will double our characterization & validation/qualification work.
We would much prefer having 1 single standard adopted by all EDA vendors.
- Philippe Moyer
SoisIC Grenoble, France
Index
Next->Item
|
|
