Friday, June 26, 2009
Thursday, June 25, 2009
Rajeev founded several significant EDA companies, was apparently out-manuvered in the board room at times, and is forthcoming with what he's learned about the EDA industry and building companies.
Tuesday, June 23, 2009
I glanced at a few presentations and was most surprised to see predictions of the resurgence of ASIC vendors (vs. today's popular "COT" model). I'm not sure I agree, but it makes a certain amount of sense. It takes a lot of tools, people, and expertise to implement 45 nanometer chips. The proposition is that if it is possible to cleanly hand off at RTL, then the chip designer can focus on functionality, and let an implementation house focus on the tricks and traps of nanometer-scale design closure. But that's a big if, to be confident that the hand-off is of a properly constrained and realizable design!
Wednesday, June 17, 2009
You could find kindred spirits of any interest available for text-based correspondence and enlightenment. I solved countless software problem (including both Windows and Linux) by searching through these groups. Who could forget groups like
Oh, and the flame wars! (I remember there was a character outraged over the Ottoman Empire who sought to cancel every post containing "turkey", which swept up Thanksgiving recipes, as well.)
The beginning of the end was when Web access took off, epitomized by "AOL newbies" pouring onto Usenet without regard to the collegial etiquette that previously existed. After AOL, there was overwhelming growth of users, which strained the scalability of worldwide discussion forums. Finally, the death knell: Spam. When I peek at Usenet groups today, they're full of the most crude and amateurish spam. It appears that posts are not run through filters as is all of our email, and this makes the noise/signal ratio unbearable.
R.I.P., Usenet. You were one of the forefathers of what we enjoy today through the Web, forums, IM, and social networking.
Tuesday, June 16, 2009
There's been some griping that there's no "Free Monday" this year. Instead, there's an all-days Exhibits Pass available for $50. To me, that seems like a very reasonable proposition. It remains to be seen if this will cause a significant drop in attendance, and if those who won't pay $50 are the ones EDA vendors need to see at their booths.
As the conference approaches, there will be a number of "must-see" lists, and I hope to compile my own. Right now, I don't have a very clear idea about companies to see. I do have some ideas about what may be hot, from my ASIC design implementation-centric point of view.
What Should Be Hot
- Multi-threaded & Multi-core software. Initially, these trails were blazed by startups such as Extreme DA and Sierra Design Automation (now part of Mentor Graphics). Now, all the major EDA vendors are working hard at either developing new products or retrofitting established ones. Richard Goering has had a number of interesting technical posts about how it's going at Cadence.
- Multi-Corner Multi-Mode (MCMM) analysis and optimization. This has also been talked about for a long time. It's always seemed like a good idea, but is becoming more critical at the most advanced processes and design sizes. A key question is how practical this is. What is the performance penalty to go to MCMM? Helping to solve this will be multi-threaded & multi-core software.
- Low Power. Of all the "next big thing" areas of EDA, this has struck me as the most real. There are several viable verification products based on simulation or static analysis. For implementation, there's combinational and sequential clock gating, multi-Vth optimization, and support for voltage and power domain design driven by UPF (or in Cadence's case, CPF ;-)
Ready for Prime-Time?
- I've been doing a lot of reading about variation, and various techniques to account for this. It's a fascinating new way to think about semiconductor performance. But, there is so much that needs to fall into place for statistical techniques to be used in production: tools, libraries, and a new way of thinking about design analysis.
Count Me Skeptical
- ESL. Maybe it's the market I'm involved in, where "QoR", especially performance, trump potential productivity gains at higher-abstraction design levels. This may be more attractive when time to market is everything and the QoR tradeoff not so great.
- RTL analysis & design planning. Again, maybe it's because of where I sit, but you can argue about getting too carried away with analysis at RTL. RTL is the functional description of the design. Implementation can be done "downstream" of that.