( ESNUG 469 Item 2 ) -------------------------------------------- [09/27/07]
Subject: ( ESNUG 467 #6 ) TSMC DFM Reference Flow 8.0 is a marketing joke
> I felt compelled to comment on this announcement you referred to. First,
> TSMC is announcing *everyone* to be qualified in Ref 8.0, *everyone* in
> the field. Some companies don't even have some capabilities in the flow.
> So TSMC announces partnerships formed to address some features. For
> example Cadence/ClearShape for LPC and fixing, and Mentor/Sierra before
> their acquisition. TSMC announced *everyone* when their 45 nm process
> is not even ready yet (announced to be in production in Sep. 2007). How
> could TSMC qualify a tool to predict silicon contour accurately when the
> contour is still changing? This is just a marketing scheme. ClearShape
> being in the Ref 8.0 is yawn news if you ask me.
From: [ The Bionic Man ]
Hi, John,
Anon, please.
After 5+ years of DFM'ing here is the bottom line: It is nice of TSMC and
other semiconductor manufacturers to publish reference flows and process
models, but what they will NOT do is guarantee that the DFM changes you
make to the design using these so called "qualified" tools and models will
get better yield!
The cost of DFM in both area and resources (Man/CPU/Months) is extremely
invasive. For example, a standard 65 nm run of Mentor LFD or ClearShape
Inshape takes DAYS on many CPU's and it will have to be run before and
after DFM optimization. Taking the extra time to DFM small IP blocks and
cells is fine, which is why an IP team may buy a single copy of InShape
or LFD. DFM'ing a design during P&R with no extra yield guarantee is not
fine, which is why companies have not deployed multiple licenses of
InShape or LFD.
That is why the commercial DFM market has failed thus far, the semicon
manufacturers will not put money where their reference flows are!
Good post, John. Turn on the heat. TSMC, UMC, and Charter need to put up
or shut up when it comes to DFM!!!!!!
- [ The Bionic Man ]
Index
Next->Item
|
|
