( ESNUG 473 Item 5 ) -------------------------------------------- [05/29/08]
Subject: ( ESNUG 472 #7 ) We don't use Cadence ETS; we use TimeCraft
> We recently taped out a 65 nm wireless design with 2 million instances.
> Our Cadence ETS eval was done about 8 months ago and during it, we found
> that ETS was within 15 psec of PrimeTime.
>
> At the block level, our ETS runs on our processor core were 5 minutes
> versus around 30 minutes for PrimeTime.
>
> - Umesh Srikantiah
> Sandbridge Technologies Tarrytown, NY
From: Gabriel Chang <gchang=user domain=ambarella not calm>
Hi, John,
Our design flow is primarily Cadence-based. We use Encounter for P&R, QRC
for extraction and NDC for delay calculation. TimeCraft is well integrated
into this flow. It takes me 2-3 days to set up my TimeCraft environment
for each design when our process changes. If there is no process change,
the set up from design to design is too brief to even count.
We started using Incentia's TimeCraft static timing analyzer just over a
year ago to verify timing for our VLSI designs. We now also use TimeCraft
for sign-off.
Two years ago, we did an extensive evaluation of Incentia's TimeCraft versus
Synopsys Primetime. We picked TimeCraft over PrimeTime because:
i) TimeCraft accuracy was comparable to PrimeTime.
ii) Capabilities and features were comparable to Primetime; e.g.
report_time, report_path, and recognize false path, from the
same SDC file as PrimeTime.
iii) Learning curve was minimal, it was as user-friendly as PrimeTime.
iv) TimeCraft was cheaper than Primetime.
Speed: on a design with 4000 std cells and 60 custom macros, TimeCraft's
runtime was only 20 minutes. PrimeTime's speed for the same size design
was either same or a little bit slower for the initial run. In interactive
mode, TimeCraft ran 20% faster than PrimeTime for new constraints.
Accuracy: we did a cross check between PrimeTime and TimeCraft before tape
out and the results were comparable. We verified TimeCraft's accuracy both
against Primetime before tape-out, and then against actual silicon. We have
now taped out two designs with TimeCraft as our sign-off timing analyzer.
What TimeCraft could improve:
1) it needs to do "read_lib" and "write_lib" as same as PrimeTime.
2) it should provide default parameters in the Tcl script for each
major VLSI process.
TimeCraft's debugging was straightforward. For example, its "what if's" and
warning markers for missing or disconnecting connections in the log file
were useful in debugging.
It also has MCMM analysis, e.g. functional|scan_shift|scan_capture/min|max,
and multiple supply voltage analysis (VDD/VDDA).
TimeCraft's global slack report this is the current default report format.
I would rate TimeCraft a 9 on a scale from 1 to 10, based on both the tool
and Incentia's technical support. I recommend it to other engineers.
- Gabriel Chang
Ambarella Santa Clara, CA
Index
Next->Item
|
|
