Market And Tech Inflections Ahead

Aart de Geus, chairman and co-CEO of Synopsys, sat down with Semiconductor Engineering to talk about the path to autonomous vehicles, industry dis-aggregation and re-aggregation, security issues, and who’s going to pay for chips at advanced nodes.

SE: All of a sudden we have a bunch of new markets opening up for electronics. We have assisted and autonomous driving, AI and machine learning, virtual and augmented reality, and the cloud. This all came about immediately after the mobile market began flattening. Did you expect it all to happen like this?

de Geus: Actually, if you look at mobility, the technology really never slowed down. It’s that the economic return of advances was not as high, and the number of people that already had one or two mobile phones was starting to be pretty large. So there was a slowdown from an economic point of view. But the fact that technology kept continuing to advance at a high speed is what actually opened the door for what I like to call the ‘smart-everything’ age, because it literally touches everything in a way that’s still to be determined where the values are going to be. It is a very strong technology characteristic that when technology advances on a logarithmic graph with a straight line, you get these inflection points. Something may have been considered impossible and sci-fi yesterday, and tomorrow is everybody is going to have it.

SE: You mean like AI?

de Geus: AI is a fabulous example of that. But that example is no different than, yesterday nobody had mobile phones. And not only just mobile phones, but phones that can also access all knowledge in the world. Today, if you don’t have that, you’ve probably been in hibernation. Thirty years ago, nobody had a computer at home. Now everyone has multiple old ones laying around. We’ve now seen the third wave of that acceleration. Computation changed productivity in the world, mobility changed connectivity and access to knowledge in the world, and now AI is going to change the very notion of devices suddenly adding some smarts. It will be interesting to see what the definition of that is. The examples of those are already appearing out of sci-fi. The car that is autonomous is really an unbelievable thing. Except that now we see those bubble tops driving around in Silicon Valley and we understand the technology, but we also understand how far this technology still has to go.

Fig. 1: “Bubble top” car from Google/Waymo.

SE: Three years ago no one thought Tesla was going to have its self-driving car and the big automakers were thinking it was a 15-year cycle, right?

de Geus: Yes, I remember being in Germany at one of the automotive companies and they had just come back from a Tesla test drive in Silicon Valley and they couldn’t believe it. They thought, ‘It’s still a hobby and diesel is going to be great.’ And then suddenly diesel was not great—really not great. Overnight, the entire German automotive industry was on this switch to electrify massively, and simultaneously toward autonomous vehicles One shouldn’t underestimate how big the combination of those changes are, but also how fast this is moving.

SE: The electrification is a massive problem, isn’t it. You can’t just take an existing service station and add a 440-volt line. You have to figure out an alternative system, with possibly multiple different types of fuel.

de Geus: The jury is still very much out where that lands. What’s clear is that if you can have, on the utilization side of the car, an electric engine you have actually simplified a lot of problems. It’s interesting to watch the unbelievable sophistication that has gone into the electronic controls for combustion engines and how, if you apply this to electric engines, you can get away with something dramatically simpler.

SE: Something like 2,000 to 200 parts, right?

de Geus: In those orders of magnitude. Of course, we’re quickly adding another 1,900 parts with cameras, sensors, and a boatload of electronics to learn, interpret and actively make the car more and more autonomous. We’ve already seen the benefits of that from a defensive driving point of view. If you look at the improvements in the number of accidents, they’ve gone way down with the exception of last year, when the number went back up. Texting while driving is a killer, and it’s not just in cars. People are texting while crossing the street. The advances in technology are at an inflection point where the sci-fi of yesterday is actually being put into place. With electrification, the big challenge is infrastructure. Along with that is an unbelievable push on battery technology, where it’s not only about the amount of storage and therefore the distance you can drive, but also the charging time. If you can fix these things, which we will, this opens up many doors. If, in addition, the car is autonomous, it will charge itself somewhere. If it’s somewhere that’s just fine, because somewhere we can put cables to them. It’s just not every gas station that is already in your neighborhood.

SE: An electric car is fine if you have a garage and can plug it in, but a person who is renting probably doesn’t have that capability. If they are in a high-rise building, where do they plug in their car?

de Geus: Even at work, the number of chargers at our work is growing a night somehow. They seem to multiply in the dark. Ease of use in all of this and automation is actually a great answer. Automation over time becomes better than what a human can do.

SE: The pieces in this progression are not moving at the same rate. You always have these disconnects, where one piece is at this level, another piece is at a different level, and we have to wait for that to catch up. But is it different now?

de Geus: Every high tech project is that. When the hardware is ready, is the software ready? In automotive we now have lots of software. If you have to wait for the software to be ready, you have a lot of CapEx that’s tied up in the parking lot. To alleviate that, we’re heading toward prototyping of the hardware before it’s actually ready so the software people can run on it. Projects very quickly find where the critical path is. Whoever fixes that first has a benefit.

SE: Because of that, do we see more consolidation in the system companies or less consolidation? Do they start spinning things off?

de Geus: You get system consortia, which are visibly developing. Let’s take Tesla, which has been a leadership example. Tesla also owns a battery factory and it’s in the solar cell business, so now they can offer you a mega-battery in your house that makes all of that smoother. The question is how long will these things last because the technology is evolving so rapidly. Nonetheless, you have those types of trends. You also have Intel buying MobilEye and wanting to be on the path of data, not because they necessarily are so much automotive-oriented, but because being on the path of data serves their natural cloud and compute center business particularly well. Everybody finds their own adjacency. If you have a certain technology capability, is the adjacency stronger with this partner or with that partner? What’s so exciting about the high tech industry is this is where you have a whole VC community that says, ‘This bet plus that bet makes a better bet.’ At the same, you have truly mega companies—Facebook, Google, Amazon, Baidu, and so on—that have so much income available that they can try things on their own. It will be interesting to see what they provide. But is this really any different than the late 1990s when networking came up? Did you need to be a networking and compute company at the same time? Did you need to be a software company?

SE: What’s changed is that in the past companies tried to own a lot more of the pieces. This goes back to the Henry Ford days, so there’s nothing new there. The difference today is systems are now connected to other systems, so it’s not so clear that you can own all those pieces and what you get if you do own all those pieces. Amazon is an example where they own the cloud, so they have a cloud leasing operation, as well. So there are some strange hybrids.

de Geus: The reason you see hybrids is because you’re optimizing for something, while at the same time having to find the right balance and compromises. Take the age of the hybrid car. A lot of people started to optimize for what electricity could do, while still needing the distance that only a combustion engine could bring. Hybrids are often either a compromise that finds a new value, or a compromise during a transition time. High tech is like a never-ending transition.

SE: But aren’t those transitions becoming more frequent and faster?

de Geus: Or is that just our perception? We’ve always been on an exponential in high-tech, and people always felt it was moving faster and faster. Linearly speaking, it is. Exponentially, it’s not. Moore’s Law now is really the continued exponential impact of technology. It may have a different form than the strict price per transistor, but it doesn’t have a different form in terms of the increased delivery of functionality.

SE: Yes, and it’s not just a shrink anymore. It’s a shrink-plus.

de Geus: Or ‘plus-plus-plus.’ Take the electric car, for example. The infrastructure for it is a ‘plus-plus-plus.’

SE: We’re starting to head there even with communication. 5G will be a massive infrastructure buildout, as well, right?

de Geus: 5G is a good example. It’s a jump from 4G, which was a big step up from 3G. You can say there are discontinuities, but they may just be a sort of exponential technology step that’s large enough to warrant people rethinking their standard.

SE: But these signals are so high frequency, particularly when we start talking about millimeter wave, that they won’t go very far. They won’t go through windows. We’ll have to put antennas basically everywhere, and the buildout of this is going to be massive. This takes time and it’s not always consistent, right?

de Geus: We love it when our devices become better, and we hate it because nothing works. It’s not that different from the past. But here you have the other challenge to all this technology, which is security. We should not underestimate that where there’s a 10X growth of complexity, there’s a 10X growth of potential cracks that malevolent people can take advantage of. I’ve argued for a number of years that the shift that we’ve already made is from scaled complexity to systemic complexity, where scale is just more transistors on a chip. Now you have many chips, many systems, many software environments all interacting, so you’re deep into systemic complexity. The very fact that system complexity itself is particularly well-suited for AI approaches—because it’s not some logic right/wrong answer to a lot of things, it’s more like ‘see the patterns’ —is also a challenge from a security point of view. These are all progress steps that bring their own challenges. We are heading into a massive wave of high-tech evolution and impact. Synopsys has been fortunate in that we came in really at the very beginning of the digital design wave. If you look at the history of design, massive automation came in with two things—digital simulation and digital synthesis. Before that, most of the design was analog, mixed-signal, and that was SPICE and a few transistors at a time. With simulation and synthesis, suddenly we really kept up completely with Moore’s Law.

SE: Doesn’t EDA get credit for making Moore’s Law possible?

de Geus: We’re half of it. The other half is the technology development, which is remarkable and really nailed lithography, because that was the thing that was going to conk out quickly. It didn’t. The manufacturing and technology was clearly one of the great contributions, but then the utilization was all automation of design. The steps were clearly simulation, synthesis, and place-and-route. That automated things, and then other things came around that. We were fortunate to be part of all of that rise in the late 1980s/early 1990s. Then came a big technology setback, because there was an overshoot in 2001, the dot-com, Y2K, networking breakdown, but then came mobility. Mobility was really the same as before computation, but with wireless and low power. There’s no end to low power.

SE: In what way?

de Geus: There are two issues with power—how do you get enough of it, and how do you get rid of the heat.

SE: Then there’s rapid switching on and off which destroys the circuits, electromigration, noise, voltage-related issues.

de Geus: Those are good problems that are all solvable. That progress has continued and we’ve participated heavily. The vast majority of all of the embedded processors in phones were designed with our tools. We are now sitting in the middle of this again because we are seeing a whole bunch of new semiconductor companies that are designing chips for AI. It will be a freefall for a while, because it’s all going to be a question of, ‘Can you optimize the architecture for a specific application?’ Fundamentally, AI is an unbelievable gift because the minute something starts to work a little bit, if you’re the AI algorithm, you think, ‘Oh, that hardware is terrible, I want something 1,000X faster—not 2X faster.’ You can’t get there just by adding smaller transistors. But you can get there by narrowing the domain of applicability so you can have a focused architecture. We transition from very general-purpose computation to mobility computation, and then things like graphics processors, FPGAs, and things that are narrower.

SE: Isn’t this part of the trend toward dis-integration, where you go from a general-purpose processor to specialized processors and accelerators?

de Geus: Let’s see how long it dis-integrates. Let’s not forget the ‘techonomic’ critical mass notion, which is if you have a great idea and you can do some samples that work, can you actually scale this? It’s not trivial to build a successful larger company. It’s also not trivial to refuse an offer because large companies are interesting in buying the small ones. A lot have been bought already.

SE: And those big companies have the reach to be able to propel the marketing globally very quickly, right?

de Geus: And to go vertical. It’s the same race as we saw in the 1990s around networking, which is a lot of opportunities, and they become more specialized quickly. Having said that, we’re at the beginning of this wave where the hunger for more machine learning or AI, and the hunger of AI from more computation, and more computation enabling AI—that loop is going to go full bore for a number of years.

SE: But with a lot of changes in it, right? That’s why people have not been able to consolidate a lot of this stuff. It’s not as if you know what the best algorithm is going to be for a specific application.

de Geus: The people working on this would probably differ on this and say ‘mine is the best one.’ There’s a lot of those, which is exactly what entrepreneurial behavior is all about. When a field feels wide open, there are many ships that are going to sail from Europe to India, except they’re going to find other stuff. The same is happening here in the AI space. Over time, we’ll find out which ships are best, go in the right direction with the right timing.

SE: Then we’ll see more consolidation?

de Geus: Yes, and that’s a very healthy process.

SE: Let’s go back to security. At one point you said most of the security issues happen around where the seams of technology come together. With the latest breaches, Meltdown and Spectre, it’s not the seams. It’s the hardware itself.

de Geus: I like to use the analogy of a bank. The danger zones for any bank are the same. You see the movies and they always come in at the seams. If overnight, someone discovers that they can pulverize concrete without noise using some new gizmo, then of course now you’re going through the very basics of a what a vault is—a block of concrete that’s thick and takes several hours to drill through without a lot of noise. Therefore, it’s hard to do unnoticed. If you could with some super magnetic wave, pulverize it quickly, then all banks are in danger. So if fundamental processors have a newly discovered vulnerability, that has big impact, but it also gets fixed relatively quickly—except that you have all the existing systems. One of the key challenges with security is the timing of knowing about a vulnerability, including when you communicate that. Let’s say you find that you are delivering a system that has a vulnerability, how do you communicate that to your customers? Do you inform them all at the same time? Do you inform them loudly? Do you inform them quietly? If you inform them loudly, you’re simultaneously informing all of their enemies. A lot of enemies are very fast.

SE: Especially if you haven’t closed the loophole yet.

de Geus: Exactly. That brings an amendment to the question. Do you only inform once you know what to do about it? There’s no easy answer to any of this.

SE: Let’s move up a couple thousand feet. How do you characterize Synopsys these days?

de Geus: We characterize it as silicon-to-software. I absolutely believe that the center of gravity is the intersection between hardware and software. Anything from safety to security is now moving more into the hardware. That’s the concrete part of the bank. Then in the software, there are many buildings being built on top of the vault, all with their own challenges. The combination of all of that requires a discipline that’s quite substantial. One of the acquisitions we did late last year was a company named Black Duck, which was a very good acquisition because it brought another aspect to it. If you look at software, there’s a massive amount of IP reuse. In hardware, Arm and Synopsys are the two largest ones doing this. In software, there’s a whole bunch of people doing open source software that is being re-used. There’s a lot of very effective reuse. But how safe is it? Who did it? Is it secure? If there’s an issue, will anyone fix it? You get all of these issues. Black Duck can look at the executable in your product set and diagnose whether anyone used open source, and there’s a registry of any known vulnerabilities in open source that gets updated all the time.

SE: It’s too hard for a person to track?

de Geus: Impossible. It’s hard for management teams to track if they have any open source. We’re now interacting with many more car companies. We always ask if they’re using any open source. ‘Absolutely not, way too dangerous.’ We also ask if their suppliers are using open source. ‘You’re checking, right?’ ‘No, I thought you were checking.’ There’s a whole evolution now of understanding where the vulnerabilities could be. In any case, Black Duck can go through someone’s entire software and find it. By the way, when you acquire a company, it’s pretty important, too, because with different sources of software there’s not only vulnerabilities, there are also legal rules. With some open source, if you use it, anything you connect to it now becomes open source, too.

SE: That becomes really interesting as we get into automotive, because you have to establish a chain of liability.

de Geus: Yes, and the chain of liability continues through the lifecycle of the car because periodically you need to update based on new knowledge of past vulnerabilities. When you update, are you sure you did it right? The automotive industry is the most sophisticated supply chain management industry.

SE: It’s had a 120-year head start.

de Geus: Supply chains optimize themselves. Except, in the case of the automotive industry, the supply chain needs to be able to stay intact for many years. It has rules that you need to have multiple suppliers, and it has an increasing set of standards that are there specifically for safety.

SE: If we are going down to 10/7nm in some of the chips that are to be used for automotive, they were not designed for reliability in the past. Aren’t we changing the dynamics here.

de Geus: In automotive, that has been the case already for a while.

SE: But never at the most advanced nodes.

de Geus: No, because they haven’t used the most advanced nodes. Now you have another opportunity collision, which is that cars that have the autonomous driving need a lot of computation. So what nodes are we using? We want the best, lowest power. Now we have many automotive companies designing the advanced finFETs, which was unthinkable five years ago.

SE: Are finFETs that reliable under those extreme conditions for that period of time?

de Geus: It’s all how you design things. That’s why the rules of design—safety, ISO 26262, ASIL-B, C and D—these are all different categories of rules that answer that question with fairly high degree of rigor.

SE: What about the economics of doing this? The cost of designing a 7/5nm chip is very expensive. You have to worry about the massive simulations with reliability, power domains, interactions, noise and everything else. That works fine if you’re dealing with an iPhone, where you can change sell a billion units. But each one of these car companies is not on the same chip. Each one is doing their own thing and the volumes are not going to be anywhere near what an iPhone is.

de Geus: That remains to be seen. If the economics are rich enough, that differentiation allows you to either sell more cars or charge more and you’re fine. The minute that starts to not be the case, you then say, ‘Why don’t I team up with my neighbor here, and now together we’re better than that guy because we find a tradeoff.’ This is how value chains self-organize. In the past, the chip design was amortized over multiple users. This is going to be an ongoing optimization. How horizontal are the chips, or how vertical are they? Now we’re back to systemically complex differentiation. When multiple things change at the same time, different companies will try different things and keep looking sideways like crazy just in case they went in the wrong direction. These are exciting times. These are winner-and-loser times, and the losers get re-absorbed because the talent is of high value. Or they get acquired and then realign, hopefully behind the winning value proposition.