There was a recent panel discussion at Cadence on the future of EDA. The panelists all got to have their say first. You can read about that in the last two days' blog posts: The Future of EDA: The View from Academia and Future of EDA: Industry, Well Cadence, Weighs In . Read them first, if you haven't already done so. Q&A With that the audience got to ask some questions. I only know who some of the questioners were, so I think I'll leave everyone anonymous. The first question was that there is still a disconnect from the beginning of a design to the end. It is still basically point tool after point tool. Anirudh took the question and said that the databases we have are pretty good for one run, but no good at all for multiple runs, which is what you need to leverage machine learning. We have done some work with vManager, but can do more, and then build intelligence and other algorithms on top. Tantalizing, Anirudh said "I'm not ready to talk about what we are doing internally yet." The next question was for the academics. In the questioner's grad days, there was lots of CAD stuff at CMU and Berkeley. It is not just lack of interest in the incoming students, he felt it was true of the faculty, too. Kunle said that at Stanford, CS is the most popular undergrad major, but EDA isn't where students are focused. To grow in the millennial era, CAD has to become more sexy, focused on solving societal problems. Kameshwar gave the Berkeley view: we can wait for the CS bubble to burst, as it surely will. But he feels we shouldn't look for students with specific CAD training, we should look for smart people. Look for excellence in modeling, excellence in optimization. If they happen to be in CS, just go and recruit them. But we need to make it sexier, for sure. Alberto said that since Berkeley has an EECS department, they have lots of faculty who consider themselves the bridge. Everyone studying IC design takes a course on the basics of EDA, for example. There are the cyber-physical domain courses (see the video below). However, those people end up at Google or Amazon. We in EDA need to offer an environment with deep projects like the autonomous car. So don't forget it is not just making EDA sexier, it is making Cadence sexier. (Please visit the site to view this video) The next question referred back to Chi-Ping's discussion of brain versus computer efficiency, a lot of which is down to memory. What sort of memory do we need for neural computing? Chi-Ping asked if Alberto knew the answer. "Of course not, or I would be founding another company!" he replied immediately. There is a lot of research going on about merging computing and storage, but nothing conclusive. Kunle pointed out something that I've heard our Tensilica guys say before: These algorithms are fundamentally statistical, so there are lots of games we can play with precision, giving just what the algorithms need. But EDA is very exact: would you tape out a chip with an approximately good layout? Anirudh cautioned against just focusing on using the brain as the archetype. The brain is limited in size so we can be born, so digital intelligence might not evolve in the same way. We don't really model vision recognition the way that humans seem to do it. Someone reckoned that EDA is the heart of the growth of technology, but it has not been compensated fairly. What does EDA need to do differently to get fair compensation? Anirudh took the bait and said we shouldn't be too down on ourselves. People tell me all the time "You EDA guys have the wrong business model, you don't make money." But in the last 10 years, we have seen the most massive explosion of silicon in the history of mankind and the semiconductor industry is...basically flat. EDA is still growing. Another rumor Anirudh pointed out is that you can get "so much money at Google." But if you look at real data, compensation at Cadence is among the top companies in the world. Don't think you can "go to Google, walk around, have fancy food, press a keyboard once in a while, and make a lot of money." But there is more to a job than money. Even if it is true that at Google you might make less money than at Cadence, kids look at what sort of work they do. Google is more like a kid in a toy store. That is what it takes to make it sexier. Alberto suggested that we look at any industry. There are some people making money, some not so much. Automotive overall makes little money, but the semiconductor people selling into them have the highest margins. The serious problem in our industry is competition, we are cutting price every single day to get the business. There are just three players, and we play. In any business, the people who make money are people who are much better than their competition. Kameshwar said that at Berkeley, they say the same stuff about Stanford. "They are like Google, great campus, great coffee shops. But they don't teach." "It's all true," Kunle said laughing. A perfect note to end on. Happy Thanksgiving to all the readers of Breakfast Bytes. There will be a blog tomorrow, completely off topic. Normal service resumes on Monday. Previous: Future of EDA: Industry...Well, Cadence...Weighs In![]()
![]()
↧
