Arm CEO interview: Little low-power processors and big powerful dreams

VentureBeat: It could one day? [laughs] There are other edges to your market, like the Arm servers. What do you think of the progress there? I saw Renee James’s company got to market. That’s a milestone.

Segars: Yeah, what Renee is doing with Ampere is great. With AWS deploying the Annapurna Labs device, those AI devices, a 41 percent cost savings is really significant. And then just this Monday, Huawei just launched their server device, which again, you look at the stats on the performance they’re achieving, it’s right up there. Quite a lot going on in the Arm server space.

VentureBeat: How much importance do you put on that particular market, strategically?

Segars: It’s important for us. When we think about how computing is evolving, there’s a plethora of different end devices, whether it’s the Procter & Gamble thing or handsets. That’s just growing like wildfire. The network and the data center end of that is an important class of computing to us. We’ve been investing in that road map. At TechCon, where we last saw each other, we launched Neoverse, which is a lineup specifically for that. We see that as a big growing market. We want to participate in it.

I can see a convergence of the devices we might deploy in a data center based on Arm with the devices we might deploy in a fully autonomous vehicle. Again, it’s foundational technology that we’ll see in other places.

VentureBeat: You grew by thousands of people since the acquisition. Where are they all going?

Segars: It really is across the board. We have more people working on our core road map, delivering processes and variants of our processes for different markets. Just the other day we launched another part specifically designed for automotive, the A65AE. It’s a processor designed with safety features and split-lock features for automotive. That’s come from our core engineering growing and being able to take a core processor and create variants for mobile, variants for automotive, and variants for enterprise, with the right features and performance levels.

A lot of our investment has gone into that capability, along with some of the specific IP and software that we’ll need for different markets, whether it’s ISPs — we just launched a couple of new ISPs — or security features for IOT devices, or software. We’re doing a lot of work around software and services for managing and securing IOT devices. This is where our engineering growth is going.

VentureBeat: The usual way I had of describing the competition with Intel was, Intel comes from the top on down and you guys go from the bottom on up. Do you have any observation about who wins in that kind of competition?

Segars: Well, I don’t know who wins, but we look at that and see a lot of opportunity to put highly energy-efficient processing into silicon devices that our partners are going to make. Building around standard architecture, you can get lots of software reuse. We’re integrating all the specialist know-how our partners have. The applications for that seem to grow and grow.

Every year CES seems more diverse than the year before. Some years you get a big step change and other years it’s more incremental, but every year I come here I see different ways in which people are putting that together and just integrating more and more. I don’t see that trend slowing down any time soon.

VentureBeat: Trying to solve the hardest problems with this approach — is it more trying to figure out ways to make small things work together? Like bees in a hive? Or is it some other approach?

Segars: Our approach has been that high performance, energy-efficient processors are foundational to many things. We have a suite of other IP we can put around that for where those computing elements need to interact. We do a lot of work on the integration of those, a CPU and a GPU and a neural network accelerator, things like that. We’ll think about how the data flows within that. Security elements, interconnected elements, I/O elements. That creates a fabric that our licensees can then take and mix and match and build around with standardized interfaces.

We’re trying to provide the things which are very commonly found in a lot of devices. I don’t really differentiate the end product. That enables our partners to spend their R&D on the things that will create the greatest differentiation. But really what we’re trying to is enable that and then get out of the way of it. The hundreds of companies we work with, the people there will come up with ideas we can’t possibly think of.

VentureBeat: The initiative you were starting to take to drive the manufacturing technology forward was interesting. You were commenting that, once upon a time, Intel was supposed to be the one to do the heavy lifting for that, and now the world is different. Somebody has to shoulder some of this responsibility, going from 10nm to 7nm or whatever. Is that what some of your thinking is, trying to make that happen?

Segars: When was it, 2004, when we acquired Artisan Components? That got us into the physical IP business. I used to run that business. The reason we made that acquisition was because we knew that going from the RTO of a microprocessor that you shipped to somebody to actually implementing it on silicon — as the processor technology got more complex, that was going to be really hard. We wanted to sit in the middle of that. We wanted to develop relationships with the foundries. Over the long term we knew that would become a really difficult problem.

Sure enough, it has. We’re in a position today where we’re partnering with the likes of TSMC and Samsung. I was just at lunch with E.S. Chung, who runs Samsung’s foundry business, talking about what’s coming down the pipe in terms of new transistor structures. How are we going to optimize our designs around those? How do we rebalance the pipeline when the performance ratios between interconnect and memory and transistors change as you go through these transitions? How are we going to run early devices together to get the learning of that?

We’re doing that with Samsung and we’re doing that with TSMC to kind of trailblaze, so that when our licensees are then looking to build a chip, we’ve proven the route. We’ve worked with the EDA companies, Cadence and Synopsys in particular, on the flow issues and trying to deal with the friction that slows down a designer going from design to manufactured chip. Those relationships, as the industry has shaken down and you see both Samsung and TSMC on the leading edge — our relationships with them are deep. Samsung is one of our first licensees ever. It’s a multi-decade relationship in that case. It’s very important.

VentureBeat: Did it seem like something that was maybe more the responsibility of an integrated device manufacturer?

Segars: Oh, yes, once upon a time. That shift really started in the ‘90s. The first Arm processors we built were shipped as GDS2 hard macros. They were optimized for each licensee’s process. I can remember — I think Cirrus Logic was first fabless company we worked with. I remember sitting in the room and thinking, “What do you mean, they haven’t got a fab?” It seemed like a crazy idea. But that really was the start of the foundry model as pioneered by TSMC. It’s been one of the things that’s enabled such acceleration of semiconductor technology.

VentureBeat: It always seemed like you guys were designing and the foundries were manufacturing. It was a nice way to divide up the industry. But that doesn’t seem so true anymore. You need this collaboration.

Segars: Yeah, you do. If you go back 10 years and look at the way people would describe how the industry is dis-aggregated — it went from IBM doing everything, from product design all the way down to transistors and maybe even equipment, to the likes of ASML producing equipment, and you have a TSMC that does wafer manufacturing. You’ve got Arm and the EDA companies that provide IP and design flows. And then you have a designer. That looks like really neat, partitioned areas of expertise, and you focus that expertise to get the best thing.

As it’s all become more complex, the interfaces matter. That’s where you get the partnership approach, the collaboration approach that us and the foundries and the design companies and the EDA companies all focus on. It delivers what I like to think is the best of both worlds. We’re focusing on the expertise of processing elements in our case, EDA in the case of the EDA companies, but we’re partnering to make sure that when it all comes together, you’re not losing something in the gaps. You’re really squeezing those gaps.

VentureBeat: Qualcomm, once upon a time, didn’t necessarily care so much about manufacturing. Now they have to make these investments that drive it forward.

Segars: The big fabless guys, while they are fabless, they have people that go and sit in the foundries. They care very much about yield, about power. They need to be on the bleeding edge of the process. It’s not just, “I’ve finished my design, please manufacture it and I’ll sit around for three months until I get wafer back.” It’s a much more interactive process.

VentureBeat: I always thought Masayoshi Son’s talk about the singularity was interesting. How does that affect your thinking, being part of a larger plan like that?

Segars: It’s a different way of expressing what we were doing. Like I said earlier, we see this growing set of applications and growing volume of embedded computing. Devices are getting smarter and smarter. At some point, where does the sum of that intelligence — what does that deliver? With the growth of AI, which has been very rapid when you think about it, and the advent of deep learning and the step change that caused in the deployment of AI, very quickly we’ve gone from discrete computing devices to connected computing devices to a world of data and that data training algorithms. It’s been a rapid evolution of what we were doing.

Masa has a very interesting way of looking at the world, and a lot of money to deploy to make that vision come true. It’s fascinating being part of that.

VentureBeat: I almost thought he would have kicked the tires on something like Nvidia, or whatever the supercomputing companies are now, rather than go to Arm, which didn’t necessarily seem to have this on the road map.

Segars: Part of Masa’s interest in Arm was and is the fact that the broad usage of our technology is really unprecedented. In an event like this, I can spend one meeting talking to somebody building an autonomous car and the next one talking to somebody building handsets, and the next talking to somebody building networks. We’re in a unique position in terms of understanding the future-looking road maps of all these different areas. That does give us some insight into the future.

VentureBeat: If you’re also thinking that maybe a hive mind is the way to get to the singularity, it makes sense.

Segars: Absolutely, yeah.

VentureBeat: It leads to interesting thoughts about — what are the other pieces? If you’re just a piece of something, what other pieces have to be there? Amazon buys Whole Foods, and there’s something strategic there, but I’m not sure exactly what it is. Some kind of 20-year plan to kill retail, acquire it, and do something else with it.

Segars: I don’t know if Amazon’s killed retail. They’ve changed it, and they now know an awful lot about people’s retail habits. You can’t quite buy anything edible via Amazon, but you now can through Whole Foods. It’s a source of some other data.

SoftBank has changed quite a lot over the last couple of years since we were acquired. It’s very much about, what can the group learn about what’s going on in the world? How is that used to then invest in companies that are leveraging these long-term trends? That’s more, really, the philosophy that’s driving the growth of SoftBank at the moment.

VentureBeat: What would you say affects more of your decisions about day-to-day investment, versus this 30-year investment?

Segars: Day-to-day investment is very much driven by, what do we need to have in our product portfolio so the compute platforms of these applications that are coming in front of us run on Arm? In that transition from a world driven by mobile to — we’re in a moment, in a world where the technology that’s come from mobile is being used in lots of other things. What are the next big drivers? The next big drivers are, thematically, 5G networks, AI, and autonomy. They all overlap. We’re investing in the compute platforms that will drive the growth in those areas, whether it’s data centers or autonomous cars or IOT devices. It’s about generating data. It’s about transporting data. It’s about processing data. Data is the driver behind the next generation of computing.

VentureBeat: Are there some things that capture your imagination here, that you’re very excited about? I see some progress with artificial limbs that’s exciting. It’s just cool to see.

Segars: It is. The exciting things are what people do with this technology. In my career, we’ve gone from technology being hard to use and expensive and evolving slowly. Now it’s much easier to use. Programming languages — I learned assembler 30 years ago. Now you can program in Python and express a neural net in one line of code. Productivity has increased enormously. Access to computing has come down in price enormously.

To me, the exciting thing is, what do you do with that? The barriers to innovation have gotten out of the way. It’s now about how quickly you can implement your idea. All the stuff that used to slow you down — you don’t have to build a factory anymore. You don’t have to build a server farm anymore. You can swipe your credit card and it’s all there for you. The exciting thing is what people do with all of that.