CEO Outlook On Chip Industry (Part 1)

Semiconductor Engineering sat down with Wally Rhines, president and CEO of Mentor, a Siemens Business; Simon Segars, CEO of Arm; Grant Pierce, CEO of Sonics; and Dean Drako, CEO of IC Manage. What follows are excerpts of that conversation.

L-R: Dean Drako, Grant Pierce, Wally Rhines, Simon Segars. Photo: Paul Cohen/ESD Alliance

SE: What are the big changes ahead, and where do you see the potential pitfalls?

Segars: It’s a really exciting time to be in chip design. If you go back a few years, everyone you spoke to in the chip industry was wondering what comes next. They had gone through many years where mobile was driving everything and there was tons of design work to do. And then mobile growth rates started flattening and everyone was pulling their hair out, wondering what is the next big thing. Now we have so many next big things it’s hard to know where to start. There are new communications protocols, whether it’s 5G, LoRA, Narrowband IoT, and new technologies which in themselves require a lot of innovation in semiconductor devices. You’ve got the world of AI driving chips in the cloud. There is inferencing at the edge, which is driving innovation in designs that eventually will underpin all of these technologies. The cloud itself is exploding, and there seems to be no end in sight there. And that in itself is changing who is doing these leading-edge designs.

SE: Where do you see the negatives?

Segars: The cost of building a 7nm chip is not for the faint of heart. Managing complexity gets worse at every node.

Rhines: I agree. It’s a unique period. We have these every once in a while where things come together for an outstanding time. One thing that the EDA industry does have a correlation with is the semiconductor industry. So we had four years of 2% average growth in semiconductors, and then it took off and in the past year we’ve had 20% growth. Now half of that is memory, and we don’t do that much in terms of tools for memory, but if you take out memory the rest of the industry grew 9.5% this past year, and the average forecast for the industry is about 9% to 9.5% for the coming year. That depends a lot on DRAM pricing, but no matter what it is, it will be a continued show of strength. Another thing that is really interesting is the entry of new people buying and using design tools. We have the Googles, Facebooks and Amazons, which were doing their own printed circuit boards, and now they’re doing their own integrated circuits. They become new customers, and they bring a lot of money to the table, and so do all of the little companies doing all of their own IoT sensors and actuators. Those are people who are not traditional customers of EDA, and they’re getting into design for the first time. In the automotive industry, 338 companies have announced they’re developing electric cars, and 127 have said they’re developing driverless cars. Ten years from now most of those will not be in business, or at least they will not be doing what they are doing today.

Drako: Do you know how many companies were manufacturing cars at the turn of the last century?

Rhines: In 1909 in the United States there were 285 car companies. By 1930 there were 27. And then by 1960, there were basically three—Ford, Chrysler and GM. In 2000, the largest DRAM producers had 50% of the market. Today, they have 90% of the market. But just wait until the Chinese come into that market. So we’ve got new entrants, and we’ve got new customers in industrial IoT who previously weren’t buying anything at all. The next thing that’s driving discontinuity is this whole AI thing. It’s causing a frenzy of chip activity. The number of venture capital-funded startups has been on a pretty steady decline for the past 15 years. Then, all of a sudden in the fourth quarter of 2017, there were $900 million of venture-funded startups. A very high percentage are AI driven, but there are all sorts of new processor architectures. And then there is the whole entry of new businesses.

SE: What can go wrong?

Rhines: The lawyers. Nasty things are going on in trade negotiations that don’t appear to have any effect on semiconductors because they would hurt both sides, but they could affect the overall economy.

Pierce: Back in the embryonic days of my career, I worked at MIPS. It felt very exciting then because we had this hope for new ideas that were coming to the market and completely changing how people wrote software, what applications were available, how this was done. It might scale all the way up to workstations and supercomputers, which were very graphically oriented at the time. Graphics were just emerging. And then we discovered later on that the technology could be ubiquitous. It went into everything. Probably every device we’re using today is based on a RISC architecture. Fast forward to today, and it’s very exciting again with the emergence of cognitive computing, machine learning, and AI. Some of these things go way back into the 1980s, when people were already working on AI. A lot of these things have been with us for a long time. But now we have huge data sets available to us in the cloud. You can go and develop AI applications in the cloud, never having to have those resources in-house, with access to an Amazon server farm. That’s pretty incredible. And as an IP provider, it means there will different architectures running different neural nets everywhere. They’re going to be close to where all this data resides. We have a customer that has more than a quarter of a million processors concentrated into a single machine, they’re feeding it with 4.5 terabytes of data per second in order to do the training. And then we’re working with a customer that ships a lot of sensors. Those sensors are gathering an alarming amount of data, which will be collected, analyzed and stored somewhere. We’re going to see new things every day.

SE: Is there a downside to this?

Pierce: All of the people who want to develop those devices are not chip designers. They’re system architects and software developers. That’s terrifying, because the level of support that the industry will demand or the level of integration that will be required at some higher level of abstraction.

Drako: One of the big things coming down the pike that will affect the EDA industry is the use of cloud computing farms. It hasn’t quite happened yet, but as we engage with customers we’re hearing more about them wanting to do it, starting to do it. Gartner put out a report that the infrastructure as a service business will be $72 billion by 2021. I don’t see software as a service being used for EDA tools, but I do see infrastructure as a service being used heavily. We have Amazon and Google touting this in a big way. The buzz around this started about 10 years ago, but customers are actually talking about it. There are five points that are relevant for EDA. First, EDA is at the cusp of moving to the cloud. They really want to be able to use peak computing hours to run all those regression tests or simulation runs quickly. They’re looking at the price of their compute farms versus Amazon, and it saves time to market. The second point involves security. The economics are pushing them to do the due diligence to get through the security hoops. That box is being checked. The third major issue is that the business model of EDA has to be enhanced. Twenty years ago we moved from perpetual licenses to time-based licenses. That was a painful transition. Amazon and the infrastructure service providers sell services on a per minute or a per second basis. Revenue will go up. The fourth point is that the cloud model will be a hybrid model. Our customers have made huge investments in servers, tools, scripts, in an environment that is really complicated. They’re not going to throw that all away. Customers need shared storage in a big way, and the cloud vendors don’t do that well. The fifth point is that software development for the cloud is significantly different than software development we’ve traditionally done in the EDA industry.

Segars: When do you think it goes mainstream? For us, it can’t come soon enough. We own a couple of data centers. The biggest pain is trying to forecast who’s using it when. You can’t forecast a random event. I want to stop worrying about the scenario where someone says, ‘You can’t run your regression yet because they’ve got to finish first.’

SE: But don’t you have to plan all of this out as part of your design cycle? It’s not always as simple as just turning on extra capacity.

Drako: No, it’s pay on demand.

Segars: I want to pay for the tools I need and the cycles I need at that time. If three projects need to do RTL simulation at the same time, I want to use all the computers for that. And if something happens and a fourth one comes along, I want to be able to do that, too.

SE: Still, does it work that easily in chip design? A lot of this is being done in-house, and then you want to add extra cycles. But you still have to transfer over a lot of data.

Drako: That’s the hybrid model, and it’s a challenge. We’re hearing from everyone that they’d like to get to a 100% cloud model. They want someone else to take care of the data center. But there’s a problem because there’s an investment in the current compute farm, and you have to act in a mixed mode for awhile. But in between there’s all this data, and that’s a huge problem for us to solve.

Segars: But that’s transitory.

Drako: Yes.

SE: Is this good for EDA?

Rhines: EDA will make the transition if the customers make the transition. All three of the major EDA companies have offered everything we have in the cloud and no one is buying it. The reason it didn’t take off initially, and maybe even today, is that people thought if it’s in the cloud then they could spend two seconds on it and save money on their EDA bill. We asked them if they rent a car for a day, how does that pricing compare to a week or a year. And then they started comparing the pricing we gave them versus the servers they already had, and there wasn’t enough motivation. We offer it today, and so do Synopsys and Cadence, and eventually it will take off. I’ve never met a designer who thought they did enough simulation on a design. If you can just do 1 more, or 1,000 more simulations, you might feel a lot better. So the demand will increase with affordability. Years ago when we did computing on mainframes, the corporate gurus would come around and say, ‘I want to see a plus in your forecast.’ But the designers would run simulations until there was no more money left. The same will happen here. There is another aspect to talk about, though. You develop differently if you’re going to deploy in the cloud versus on a server farm. We have customers who have hundreds of thousands of servers, and they are used to getting the software ported for their local server farm. They don’t see the need to go to Amazon. But then we have other customers—and this may be the real kickoff point—who don’t have big server farms. All of a sudden they have a real need, and this doesn’t require lots of revenue or special algorithms.

Segars: Another side of this is that some of the big guys coming in to do chip design are the people who own the cloud.

Rhines: We saw that happen with emulation. People used to have emulators in the lab and they would plug in cables. Then they found they could do it virtually. They could have software stimulus, and just put it in a server farm and everyone else could tap into. So essentially rather than plugging in cables, they’re running it in the cloud.

SE: So what does this do in terms of the competitive landscape? There are new companies coming in with no infrastructure, and then there is massive consolidation among the biggest players, which traditionally have been the largest customers of EDA.

Segars: With a cloud model, if you make access to the building blocks easy, then you’ll get a lot more people doing it. With consolidation, there’s a lot more chip designers who don’t work at big companies anymore. They’re getting a group of people together to start building a company. One of the reasons you don’t invest in a company is the large up-front bill.

Drako: It’s kind of like the foundry model. The fab has been outsourced. Now the compute will be outsourced.

Segars: You’re lowering the barrier to entry.

Rhines: How many licensees does Arm have?

Segars: About 500.

Rhines: And we’ve just talked about another 500 companies coming in from the auto industry. And there are over 1,000 fabless companies in China that no one has heard of, all looking for tools. This model isn’t going to be like the semiconductor equipment model, where you had a handful of companies able to buy tens of billions of dollars worth of equipment. With EDA, you have a computer and a brain, you can be an EDA company. And the need to compile high-level software into FPGAs or custom chips is increasing, not decreasing.

Related Stories
CEO Outlook On Chip Industry Part 2
Concerns grow about ethical choices and liability in AI-based safety-critical systems, and who will be responsible for making those decisions.
The Bumpy Road To 5G
How pervasive will this new wireless technology actually become, and what problems still need to be solved?
Giant Auto Industry Disruption Ahead
Autonomous vehicles will cause fundamental shifts across a number of established industry segments tied to automotive, opening up big opportunities for chips and tools.