Arteris’ chairman and CEO talks about what’s behind the industry consolidation, what’s changing in automotive, and where the holes are in the IoT.
SE: One of the big stories these days is consolidation. What are you seeing on your side?
Janac: There are about 230 companies doing SoCs right now. Maybe 150 should be doing that. As the game gets more expensive and more difficult, some of the companies that don’t have volume may have to do something else. Consolidation is part of that. But you’re also going to see movement toward platforms. Some will move toward FPGA SoCs. Some SoC companies will move toward becoming IP companies. And in the end we’re going to be left with a lot fewer companies building SoCs, and that includes systems houses that have major SoC design efforts and large SoC companies.
SE: Do any new companies join the mix along the way?
Janac: Absolutely. What you’re seeing are the largest companies investing billions of dollars to gain SoC credibility. They’re not merchant silicon providers, but they’re so big that it doesn’t matter. They’re driving what we call software-defined hardware. They are essentially building hardware that enables their software business models, apps business models and service business models to work more efficiently. This is like a pendulum, and right now it’s swinging toward large systems houses at least partially designing their own silicon.
SE: Does that eventually swing back?
Janac: Yes, but trends are easier to predict than the timing. It ends when systems houses realize how difficult it is to build some of this advanced silicon. Then they swing back toward commercial solutions. When does that happen? No idea. These guys aren’t looking at building at optimized silicon. If you look at some of these application processors, they’re too big to compete as merchant silicon. They have different objectives.
SE: To run their apps faster or use less power?
Janac: Yes. And they’re willing to accept some sub-optimizations of the result because they have the apps to cover for the silicon and all sorts of software models.
The people that are getting swallowed up these days are the ones who have partially lost their way. If you lose a bunch of hundreds of millions of dollar investments and there isn’t anything on the back of it, you wind up selling the company.
SE: So how does Arteris maneuver around this consolidation?
Janac: We’re not tied to merchant silicon providers. We’re very happy to sell into systems houses and to people who build FPGA SoCs. We also license to people who are building platforms and to the chip divisions of systems houses. We even provide technology to other IP companies. We can be very flexible and nimble. What’s driving our business is that things are getting more complex. The interconnect could be built internally. When I started in 20005, people were spending $5 million on their internal bus groups. Now, to do the job right, they have to spend $15 million to $20 million.
SE: Are there specific markets that are more amenable to network on chip technology?
Janac: Absolutely. The market segment selection is really critical. In 2007 we went after mobility. They have the problems of area, power, cost, time to market and performance—all at the same time. The network on chip is an ideal technology to address some of those issues. It’s half the gates, half the buses, half the wires, very good performance. You can optimize latencies according to the requirements of the chips. That saves people tons of money. Network on chip is essentially a cost reduction compared to other technologies. Now what’s up and coming is solid-state storage. There the problem is how you make the flash medium, which is not completely reliable, much more reliable. You do that with very sophisticated controllers. I’m also extremely excited about automotive, even though it is slow moving, particularly the self-driving car.
SE: That includes things like adaptive distance and accident avoidance all bundled together?
Janac: Yes, and it will happen over many years. What we have today is ADAS (advanced driver assistance systems), so you have parking and curb detection, adaptive cruise control. I’m not convinced you’ll need changes in road technologies because when you have the autopilot dependent on the lines on the highway and the lines are not well maintained, that’s a problem. You’ll see some modifications of roads. Trucks will use this—particularly delivery trucks. But clearly the trend is to automate driving because even though there will be accidents at the individual level, society will be way better off. There are texting teenagers and grandmothers who can no longer see. These driving aids will make society much, much better.
SE: This already is happening with other forms of transportation, right?
Janac: Yes, in France the TGV [Train à Grande Vitesse, or high-speed train] is completely computerized. There is a driver there just for emergencies. But they’re stuffing a TGV every two minutes on a track because they’re completely computer controlled. The capacity is much, much higher with automation than when there were human drivers, who needed 15 minutes between trains. It’s a historical trend. Those kinds of electronics are safer. We have resilience that captures and detects faults. Eventually we’ll be working on fault-tolerant electronics. And the other big one, which we haven’t figured out how to exploit yet, is the data center. There are huge amounts of data center infrastructure being put in place. For every 600 phones there’s a server blade. That’s another big opportunity.
SE: This is where ARM and its partners are pushing in with a server architecture.
Janac: Yes, because there is such a large variety of infrastructures there. While the entrenched companies probably will be able to defend most of their franchise, there is so much variety of applications that other companies will get a foothold.
SE: At least at the edge of the network, right?
Janac: Yes, or at the edge of the data center or in low-cost data centers or low-power data centers. It may not be the big iron data center, but others will get some share. It may take a while, but it will happen.
SE: With consolidation, what happens to volume?
Janac: That stays the same or grows. Whoever has a royalty model will be fine. The question is what happens to the licensee revenue.
SE: A year ago you were skeptical about the Internet of Things. Has your opinion changed?
Janac: I’m a huge supporter of IoT, but in my opinion the term is a marketing misnomer. There is an Internet of Cars, Internet of Watches, and they’re really different. Somebody who approaches it as a monolithic IoT market probably won’t be successful. The key is to understand what is happening in a segment at an application level, and they to build silicon for that. The chips you need to build for the Internet of Cars are very different from the ones you need for the Internet of Watches.
SE: But they are increasingly connected, right? There are issues such as security that span both of them.
Janac: That’s true, but the security you need in a car is way higher than the security you need for a watch. It all lines up in the cloud. The IoT is the devices. The cloud infrastructure doesn’t change much. It doesn’t matter much whether the cloud is accessed by a phone or a car or a watch. The data center stays the same. But the local devices are substantially different, and the people who are successful there are the ones who are going to focus on a certain set of use cases. I’m not skeptical about the IoT, but I am skeptical about it being a monolithic market.
SE: Where is your growth? Is it selling more into the same companies, or new opportunities you didn’t have before?
Janac: It’s both. By customer, we have about 23% market share out of the 230 companies doing SoCs. By individual account penetration, my guess is our penetration is probably in the low teens. So if you look at 15 billion SoCs being made, and our licensee base making about 1 billion of those, that’s about a 7% penetration. So there’s a long way to go before the network on chip has the kind of penetration you see in a mature market.
SE: Is that good or bad?
Janac: It’s an opportunity if you execute on it. But there’s an opportunity to execute on much higher levels of penetration than we’ve had so far. The other thing that happens is there are new applications. If you had talked to me about driverless cars four years ago you probably would have thought I’m a nut case. If you had asked me about DTV at that time, I probably would have said it’s a wonderful market. It may be a wonderful market yet again, when people figure out how to make DTVs into interactive terminals to the Internet. You have IoT that will have volume, such as watches and cars and several others. You have new applications. You have new technologies, such as the march from 28nm to 20 to 16/14 to 10, which heavily favors the network on chip interconnect because it’s the most efficient technology approach. And finally, you have the ability to add much more value into each product. We announced resilience last year for mission-critical electronics and critical systems. We just introduced technology for the physical part of the design cycle, which gives you things like automatic pipeline insertion and better physically aware architectures and better starting points for the layout teams. You eventually have things like 3D. There’s another set of technologies that have to built there for 3D silicon. So there is a never-ending set of technologies to develop and add value.
SE: Where can the industry go wrong?
Janac: One scenario—and one that I don’t think will happen—is that there could be a few, large, highly integrated players, which would own their own IP and an EDA company. They probably would not own a fab, but they would be highly integrated. The innovative little guys would have trouble breaking into that model. So that starts to look less like the car industry of the 1950s and more like the car industry of the 1990s, which is right before Tesla came in. You would go from a dynamic, entrepreneurial industry to an oligopoly. That’s the biggest strategic concern.
SE: What will stop that trend?
Janac: There is too much innovation and the market swings now are happening very quickly.
SE: In Internet time?
Janac: Yes. There are companies that were absolute kings now getting sued for antitrust and struggling to keep up with their competition. But it is a strategic concern for everyone. The other concern is execution. You can’t disappoint your customers. The interconnect is a critical part of the chip. It has to work. You have to be able to deliver the features ahead of when the customers need them—not too far ahead, but never behind. And you have to deliver the quality. That puts a tremendous pressure on execution, and it’s a capital-intensive process.
SE: That’s a problem these days. R&D costs are rising. How do you deal with that?
Janac: The reason we’ve done well is that we offer technology that provides a major cost reduction. You gain a couple square millimeters of silicon, 6 or 7 milliwatts of idle power, several months in time-to-market cycle time, and when you add those up it’s a huge cost reduction. It comes from a combination of IP and tools that allow that IP to be deployed quickly, and a lot of attention to internal and external verification capability.
SE: But your own R&D goes up, too. Can you recoup that?
Janac: Only if you can participate in the customer’s volume growth. If you do this as an EDA model, you cannot.
SE: You’re talking about a royalty model?
Janac: Yes. Without that, you can’t keep up.