( ESNUG 456 Item 9 ) -------------------------------------------- [07/17/06]
Subject: ( ELSE 06 #19 ) We're happy using the Magma FixedTime methodology
> Been a Blast Fusion user for ~5 years. Glad to see a stronger partnership
> with foundries to work on smaller geometry issues. Competition seems to
> be improving in the last few years after they decided to take the last 3-4
> years off and concede the market to Magma.
>
> My biggest wish list: multi-threading (and not just the router, the entire
> tool... timer, placer, extractor, etc). Runtime and block sizes are
> getting larger, and CPU clock speeds aren't. If they want to keep up with
> our chip designs, multi-threading (and 2+ CPUs) is going to be a must
> have, and I'm not sure if they are taking this seriously yet.
>
> - from http://www.deepchip.com/items/else06-19.html
From: Evelyn Landman <evelyn=user domain=mellanox.co.il>
Hi, John,
Blast Fusion was impressive for its capacity and runtime. We were mostly
impressed was the timing predictability (their FixedTime methodology). I
could not believe that at such an early stage (fix time) a tool could give
such an accurate indication as to where the real speed paths are.
Two years ago, our dc_shell results did not match at all with our PhysOpt
results which did not match to our Astro timing-driven routing results.
At each stage new paths were appearing, and we had to go back to RTL to
fix them. We are now using Magma from synthesis to routing and we have
much better correlation between synthesis results and routing results. So
we can fix immediately after synthesis and shorten the loop. I believe
the Synopsys synthesis as a tool by itself is better than Magma, but Magma
has the advantage of good correlation with post routing results, which
shortens the loop and gives overall better results.
The last design we did was done with Magma, using TSMC's 0.13 LV OD process.
Due to the huge capacity of the tool, we merged almost 90% of the core into
a big block with 900K placeable instances and 222 memory arrays. We were
able to run it from RTL to routing in 2 days.
- Evelyn Landman
Mellanox Yokneam, Israel
Index
Next->Item
|
|
