Recently, the ESD Alliance organized the annual CEO Outlook panel with Simon, Wally, Grant and Dean. I covered the opening statements yesterday , but I figured it would get too long to put everything into a single post, so here's the rest of the evening. There's a bit of false advertising in the photo below, since despite being held on the Cadence campus, and despite attending the panel, Lip-Bu was not on the panel. Bob Smith told me that they deliberately decided not to just have the CEOs of Arm, Cadence, Synopsys and Mentor (or Mentor-a-Siemens-Company) every year. Since this was the most interesting CEO panel I can remember attending, except maybe 2009 when EDA was memorably described as being "fleas on a sick dog" as the semiconductor industry retrenched during the worldwide recession. EDA in the Cloud The opening statements started a discussion. Simon couldn't wait for EDA in the cloud. He owns 2 datacenters and it is a huge pain to forecast who is using them when. They are running out of electric power in Cambridge too. He wants to be able to have 3 projects all doing simulation at the same time, and if a fourth comes along, no problem. Wally said that EDA will make the transition as the customers do. About 5 years ago we all talked about it. Aart pointed out that all 3 big guys had all offered stuff but nobody was interested. But customers seemed to be focused on the price going down down. But if you rent a car for a day, a week, a month you expect to pay more the shorter the rental. "I’ve never seen a designer who felt they’d done enough simulation on a design. If I could just do one, or one thousand, more simulations." Wally believes that demand will increase with affordability. His experience at TI was that “designers will keep running simulations until the budgets run out.” Simon pointed out something I'd not thought of, that a lot of the people moving into IC design (Google, Microsoft, Amazon) own the clouds. They are not going to be building their own datacenters too. Their datacenters are cloud datacenters. Ed wondered what this would do to the competitive landscape, especially with consolidation shrinking the number of chip makers around the world. Simon reckoned that if we make access easy, people will use it. There may be consolidation, but there are lots of startups with the same engineers. Wally pointed out that Simon has 500 licensees using Arm cores, so it's more than just a few big semis doing designs. There are 1000 fabless startups in China. All the automotive companies are coming in. We are not like the semi equipment industry where there are just a handful of companies that can afford to build a $10B fab. EDA is not like that. The need to compile algorithms into custom chips is only increasing. Artificial Intelligence Ed asked about AI, which he described as a hot growth market, but requiring extremely high performance and low power. Simon pushed back on low power. "The biggest purchasers of wafers are building huge chips for crunching data submerged in baths of coolant. On the training side anyway." Of course, Arm sees a world with inference in everything going up from small microcontrollers perhaps to something to do vector math (just based on that statement, I predict Cortex M series with neural network vector units added). Wally said that whenever we get a new wave, with new architectures, and new designs, it is good for EDA. Lots of competitors. Even the ones that fail are good for EDA, they need tools. Look at when 2G went to 3G and the number of cell phone manufacturers went from 350 to less than a tenth of that, but EDA suffered little. "Semi can go up and down 30% but that doesn't happen in EDA." Security, Reliability and Traceability Ed wondered how chip design and EDA tools would change. We've been going along with faster, smaller, low power. Are we equipped to build more in? Dean said that reliability standards are one thing we will need. We've had some of this, but historically a huge part of semi is marked that it is not for use in life-threatening situations. But now we have to learn all this new stuff for automotive. Simon pointed out that the semi industry knows how to build reliable things but it has been a niche. He thinks traceability will be come a big thing, and costs will go up to provide it. Many people are going to be selling into customers who care about reliability and audit. We may get this in EDA too. Does your SPICE model match simulation? Does your extractor produce the right values? Complexity is the friend of the EDA industry and audit and trace will be another level of complexity. Dean said that people are starting to worry about tracing for national security reasons. All the security cameras are now made in China, and the chips are all made in China. So it's a huge problem sourcing cameras for the US government where the chips are not made in China. That's new, to see a customer concerned about that. Wally agreed, and said that we have lots of technology that could be used, but is not, since unless you are designing chips for the NSA, the customers are not (yet) motivated enough. Simon said that it depends a lot on the end-market and whether a few people can quantify their loss and drive change. For example, the movie industry really cares about piracy of their movies, and insists on a special chip. The same for bank cards. There is a consortium of banking institutions who define how bank cards get built. But for IoT at large it is the wild west and it is all about cost. If the end market won't pay more money for a secure chip, then it won't get built. That has to change. When you go to Fry's and buy something, you have no idea if zero minutes or thousands of hours were spent on security. Plus, if the end-user sets their password to "password" we can't fix that in chip design or EDA. Most people think that they are of no interest to a hacker, but we in this room know everyone is being used. Just look at the Mirai botnet. Grant (having got his eyes out of his newspaper) said that he'd pay more for a secure phone, for the idea that it is private. But Dean pointed out it's a generational thing. The younger people (younger than most people in the room) don't worry about that. Everything is out there, that's just the way it is. Social media is way beyond the wild west, there is societal learning that has to happen as we tune up how the technology should be used. It's a Wrap Ed asked a last question, not much different from his first, about where the panel sees the industry moving. Simon went back to traceability, understanding the problems of every component in the system, how the design was done, how it was signed off for safety and security. The supply chain will become a lot tighter as we have to "show our working, not just the final answer." Wally returnd to something I've seen him say many times. In EDA, 80% of the growth comes from new capability, only 20% for increased sales of old capability. For EDA we can count on thermal, security, supply chain, and more, to drive future growth. But what we do in IC design is 40 years ahead of system design (and even more ahead of what they do in software, Dean added). How we automate the design of cars and trains and planes will grow at twice the rate of the underlying IC business. Dean returned to some of his earlier points. One: a slew of companies are doing neural network chips and delivering stuff that people are and will use every day. It will require new tools like neural network compilers. Two: Cloud stuff will shift the industry, but it will be gradual. But it won't be a discontinuity: I was in spam filtering biz in 2003-2005 and we were making a fortune, and the President said that they would ban spam, and Bill Gates said Microsoft would solve spam. When the President and Bill Gates come after your business, that looks like a discontinuity. But even then it never happened. TL;DR Cloudy, with no chance of meatballs, but traceabliity. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧