Get email delivery of the Cadence blog featured here
First, some sad news. Bob Gardner, for many years the executive director of the ESD Alliance (although then called EDAC) passed away on April 11. I didn't know him that well, mostly as the guy running EDAC and as one of the members of the band of EDA professionals who would regularly play at EDA events such as the Sunday evening before DAC. There is a memorial page on the ESD Alliance website that gives a brief bio:
Over a 50-year career, Mr. Gardner held senior management positions at semiconductor and EDA companies including Signetics/Philips, AMD, Exemplar Logic, Design Acceleration, Bridges2Silicon, and ITeX. A 1965 graduate of California Polytechnic College in Pomona, he earned a Bachelor of Science degree in Electrical and Electronics Engineering. Mr. Gardner was active in instrumental music education and performances in Silicon Valley. He also was an avid sailor/yacht racer.
EDAC, as it then was, used to run a CEO panel early in the year with the outlook for the year. The audience would also be able to estimate the stock prices of the public EDA companies for the end of the year. But in the modern legal climate, that didn't work that well since the CEOs were basically told by their legal department not to say anything, except about the past. So now it's later in the year with more of an outlook of the industry as a whole. Ed Sperling was the moderator, and our panel consisted of what is now two public companies, and two private subsidiaries. Aart de Geus of Synopsys and Lip-Bu Tan of Cadence are the public companies. Wally Rhines runs Mentor, which is now a subsidiary of Germany's Siemens, and Simon Segars runs ARM, which is now a subsidiary of Japan's Softbank.
The panel opened by giving each CEO an opportunity to make a brief statement before getting to the Q&A. They went from right to left on the above picture, meaning Aart went first.
Aart said that the industry is going through a phase shift with customers consolidating, which can either be regarded as maturation or rejuvenation. He is firmly in the rejuvenation camp. The first phase of EDA was around computers and computation in the 80s and 90s, which drove performance/area optimization and initial cores. The second wave was mobile, which added the big constraint of low power. He has high hopes for a new wave of digital intelligence, and machine learning, "adding value to things that are already valuable." What's better than something a bit smart? Something smarter. Algorithms are important but hardware will make the hunger for silicon boundless. The semiconductor markets are already starting to look up this year.
Simon, fresh off a plane from London, said that there seemed to be more optimism. Yes, mobile is slowing, but the next big things are IoT, automotive, and the growth of AI. There is a need for more silicon, more devices, and not necessarily coming from the big semiconductor companies. There is demand for intelligence everywhere, leading to the triangle of silicon/software/data. Any semiconductor company needs to worry about at least two of those if not all three. It's not enough to just worry about the device. Simon thinks that there might be increased regulation around data since it is about people, the environment, medical, and may not be a free-for-all any more. So there are lots of problems and there are opportunities in new ways to solve them.
Wally pointed out that at the old forecast meeting there was good turnout since anyone who wanted to sell their startup to one of the big companies showed up, but now there aren't really any. But it is a great time for our industry. He didn't have any slides but he had all the numbers anyway. There were $210B in acquisitions in 2015 and 2016, and on average they said they would cut operational expenses by 25% as a result of the mergers (aka layoffs). So you'd think things would be terrible but we just had the best quarter in six years, with 15% overall growth (19% if IP is included). US broke $1B for the first time, EDA employment grew 6.6%. What us IC guys think of as the sleepy backwater of PCB grew 45% year on year. Even with all those "synergies" in the mergers, semiconductor R&D grew. The reason is that people who never did design before are now designing chips, people like Google, Facebook, Amazon. Tier 1 automotives are doing unsophisticated design, starting small, but they are moving into the market. China is investing at least $20B a year, initially in wafer fabs but now starting to create a fabless ecosystem (1300 fabless companies there). It is most important that we are doing systems and not just ICs. The EDA industry has pretty much automated ICs, nobody would do it manually. But systems such as cars, planes, and trains are still done manually. But they have reached the limit where it is no longer possible to do it manually. It is a huge market compared to semiconductor. EDA has traditionally been 2% of semiconductor. Wally, wearing his new Siemens hat, knows that 2% of the $2.5T electronics industry is $50B.
Lip-Bu pointed out that there are a couple of waves happening in our area. It is a more data-centric environment. So more infrastructure in HPC, storage, networking. IoT, especially industrial IoT, needs to make decisions locally and can't push it all into the cloud. Machine learning is going to be huge, with Google, Microsoft, Amazon, Facebook, and others all pouring money into it. Automotive is level 2 today and needs to do a lot to get to level 5. As a result of all this, semiconductor is not a sunset industry. There is still lots of innovation required, especially for some IoT areas to get the design cost down, support energy harvesting (and year-long battery life). He feels we are really undervalued. Siemens just bought Wally (Mentor) and are driving system design. I think Lip-Bu may just be jealous of the markup Wally got from the Germans!
Ed took that ball and ran with it, asking how we need to evolve to this system area.
Simon pointed out that EDA and IP has disaggregated the customer industries so that they can invest in whichever bits have the most value and buy the rest. The complexity of the problem is going up and up and it is not possible for any one company to do it all. Aart said that we don't know how to do systemic complexity, where hardware and software are really one thing. Every parameter that matters has to get optimized simultaneously. This is great for our industry since "complexity is our middle name." We have been the other half of what has been called Moore's Law, an exponential change that continues on from scale complexity to systemic complexity.
Wally reiterated something that he has said for years: revenue growth only comes from solving new problems with new tools (I found one of his charts from an old keynote above). For example, last 15 years, RTL has had 2% growth, P&R has had 2% growth...there just are not many engineers being added there. Over 100% of growth in that period has been in Asia Pacific. But everything requires new algorithms and new techniques and those grow faster than the mature parts of our industry. We have lots of capabilities but designers continue to use what they used to use as long as it gets the next design done. They only change when they cannot do the next one the way they did before.
Lip-Bu emphasized the important of IP since the trend to outsource parts of the design is continuing. Even a big company trying to catch up with all the protocols and interface standards cannot. Almost by definition, there is no differentiation in standard interfaces. The move to the cloud will happen big time, not for cost reasons but for performance, to run huge workloads especially coming up to tapeout. But some tools are old and need a complete re-write to achieve massive scale and that will be a big change for the industry.
Ed heard about all the opportunity for growth but wondered where it is. Verification? IP? Something new?
I could tell Wally was itching to throw up a slide with a bar graph, but he knew the numbers off the top of his head anyway: new technologies grow fast, he said, stuff that has been around grows slowly. Average growth for EDA is mid single digits. Adoption at system level is double digits. Mentor's automotive revenue growth has been over 20% for five years.
Aart had to get his little 3x5 cheat card out to look up the numbers. We have done really well (I think he meant the industry not specifically Synopsys). Semiconductor grew 3.2% but EDA+IP grew almost twice as fast at 6%. From an investment point of view (stock prices, not R&D investment) EDA+IP grew nearly 15% when the S&P grew less than 7% (and that doesn't count the acquisition premiums of ARM and Mentor). The opportunity is now at the intersection of hardware and software, which is why Synopsys have invested a lot in its software business. Signoff for software is becoming a big thing, especially in security.
Wally was wary of semiconductor revenue numbers due to bleeding off of reported revenue. Samsung and Apple together represent 3.5% of semiconductor revenue that used to be counted, but now just shows up as foundry wafer revenue at half the amount.
Lip-Bu visits China every month. It is a very important market not just because of the number of companies, but how fast they are scaling. There are two in the top 10 now and it will be five or six in the next five years.
Wally said there is no spurt of semiconductor revenue until a particular application takes off, and then it lasts 10 to 20 years: mainframes, PCs, smartphones (imagine Wallys' chart with the overlapping waves). We are in the lull before the next one after mobile. But that will happen for sure since we keep reducing the cost of a transistor by 30% per year (imagine a Wally learning curve chart, oh I found one for you).
Aart says that cars full of electronics are good news for us. Self-driving cars are so close he can almost taste it. He doesn't know the numbers but this will revolutionize the reasons for buying a car and how they are used. Lip-Bu can't taste it (and he drives a Tesla every day, or rather it drives him) and thinks we are still a long way from level 5.
Ed wondered how the markets aggregate. Most of them are not 1B units of a single chip, so the economics of developing them has to change.
Wally said that the actual cost of doing a design hasn't increased that much, 10-15% per year, and per transistor it is cheaper. Once you have a platform you can do a lot of derivatives. Also, you can buy more tools so it doesn't take so long. We, as an industry, have reduced the cost of EDA per transistor about 30% per year.
Aart say design hasn't been the issue for 30 years, although for 30 years people have said "it's getting too hard." The single biggest productivity boost was the use of IP in various way. But he acknowledges that volumes will not be really large for all these chips. "Smart everything will be directed to things that are already high value: cars, house, bodies (medical). Silicon cost may go up a little but the value is in the overall system."
Simon feels that more system companies will take on chip design to get differentiation, which is a good thing.
Lip-Bu thinks the balance between manufacturing and EDA is rebalancing. Mobile justifies the costs to invest in big fabs but in IoT the cost is not that high. Many people will stay on older processes like 28nm so EDA needs to do more there to drive first-time success, keep costs down, and avoid respins.
The next question was whether we are making progress in security. "Not fast enough," said Simon. "The state of the art right now is pretty grim." Wally thinks that sooner or later there will be an embedded Trojan in a chip that causes a big financial loss, and then it will be a requirement to get a purchase order. We will end up with best in class and people will be willing to pay for it, just like they pay for low power or any other feature. Aart wasn't going to underestimate the enemy. "Some of the smartest people in industry design chips, but some of the smartest people are the hackers." The biggest vulnerabilities are at the intersections (the Jeep hack got in through the entertainment system, for example). Expectations will rise and there need to be standards. Lip-Bu agreed, saying it is a big issue and a big effort but people won't pay for it until a big problem occurs.
An audience question said that EDA is driven by consumer, but future customers are automotive and stuff driven by risk mitigation, sharing liabilities, and so on. Wally has a longer memory than that and said that consumer is the current era, but originally it was mil-aero where reliability was king and not price. Aart pointed out that there was a much longer design cycle than a cellphone. Another issue is that the technology gets locked in, so that the GPS and graphics in your car is hugely inferior to your phone since it got locked in years ago.
"How can government help?" asked one of the government students touring Silicon Valley.
"H1B visa" Aart said. "You should be crying about this. Every year since 9/11, US has shut its doors to the best people in the world, and it is increasingly difficult for people who are brilliant, may have studied here even. Look at this room? How many people started here."
Wally warned about export controls. "Be careful about how you use export controls. If a company thinks it might be ITAR classified, we won’t develop it. We just are not designed for that."
Simon went for education. "We need to get people interested in science and technology early in their career. Supply and demand seem a bit out for lawyers."
Lip-Bu went for the big picture since China is doing a good job of identifying the importance of semiconductor, whereas in the US “I was at congress meetings and they were saying why are we spending time on semiconductor." It is not just in Washington though. "Professors tell kids go to software, go to social media. But we need the best talent in the semiconductor ecosystem, too."
So we need a more sane immigration policy, limited export controls, better education, and more visibility for semiconductors from our political overlords.