Executive Insight: Simon Segars

SE: What concerns you most?

Segars: In the context of design and where chip design is going, ARM is a long-term business. We’re doing stuff now that is going to ship in five years’ time. Obviously, for everyone in this space, Moore’s Law has been a fantastic thing. It’s enabled us to achieve really fantastic scaling of transistors, and everyone knows that is getting harder and harder. There’s a bit of denial: ‘I’ve been hearing this my entire career.’ But in the long term, what happens there? It’s an issue.

SE: Do you still see people going down to 7nm, or are we going to go stick at 28nm, or 2.5D/3D?

Segars: It’s all of that, and because it is all of that there’s a growing R&D cost. In the past is was, ‘Scale, scale, scale.’ You have to deal with the physics issues that come up. Tools get more complex. There are new materials and the usual scaling issues, but also new ways of thinking about doing design—3D structures, stacking of die. There’s a multidimensional space to explore, which increases R&D. The consumer has been trained to expect more for less money each year. R&D costs with every generation goes up anyway, and now we’re trying to attack the problem on multiple fronts. That’s a different challenge.

SE: This is an opportunity for ARM, too, right? It’s no longer just a monolithic processor anymore. It’s now lots of little processors, too.

Segars: Clearly, design is a key methodology for solving this problem. We’ve gone from monolithic processors to heterogeneous and homogeneous arrays of processors in a chip doing different things. That’s been developed anyway for power management and performance, and it definitely does play to our strength because we’ve got big processors and small ones. And generally, design and system architecture are how these problems are going to get solved, which plays to our strength as an IP provider.

SE: ARM has targeted the cloud with the microserver architecture. How is that business doing and what sorts of challenges do you face there?

Segars: It became apparent early on there was an opportunity to do some disruption in mass-scale cloud and servers. It also became clear that 64-bit-architectures were critical for most server applications, so one of the reasons we licensed the architecture as early as we did was to kick start some of that. The last couple years we had people developing silicon, and now we have a number of partners sampling silicon. A software ecosystem needed to be developed in parallel with that. Linaro has been a key part of that. It’s all coming together and we expect to see some boxes shipping soon.

SE: How much of that is being driven by new opportunity, and how much is being driven by the cost of power?

Segars: The two go hand in hand. If you are building a huge data center, power and cooling are a huge issue. They want to cram in as many servers per square meter as possible, and if you get one that packs more densely, that’s a good answer. The power benefits come from lower-power processors, processors that are right-sized for the task you’re trying to run, and then integrating into everything else. So instead of having a big monolithic processor with a bunch of stuff and a big PCB, you’ve got an integrated SoC. In some of these data centers there are 100,000 boxes doing the same thing. It’s worth the effort of building a chip that’s optimized for that workload.

SE: There are some rumblings about rethinking data center architectures. But how much of that is talk and how much is real?

Segars: We’re a couple generations away from seeing that. People used to get a box, plug it in, and optimize software for the box. A lot of people we talk to are still in that mindset. A lower power chip has different levels of integration, so now you optimize your software for it. But you have more degrees of freedom now, too. You can optimize the software and the hardware. When you think about the workload and the problem you’re trying to solve, that’s not a paradigm that has existed in the past. It will take awhile before people really start thinking that way. So it will be a few years before we see the full benefit of that, but in the meantime we’re seeing chips optimized for storage workloads. If you’re a big social networking site getting photos uploaded by the terabyte every hour, people look at them for five minutes and never look at them again. But they have to keep the photos forever. So what’s the point of having a big server there when it’s just controlling a disk that’s not getting used very often? That sort of discussion is happening. In a generation or two we’ll see chips and software tailored together to make it much more efficient.

SE: What will drive data center traffic in the future to make this work?

Segars: Three things. People like using social media and it will continue to grow. That’s a trend that will go in one direction and grow storage. Then there’s IoT, which will be the source of tons of data. And then there is the growth of mobility. There are about 7 billion people on the planet, and about 2 billion smart phones in use. A few billion more smart phones will produce a huge amount of data, which will drive data center traffic. Those are all interrelated. They’re going to drive network traffic and data center traffic.

SE: Let’s talk about the IoT. What will be the effect on chips, complexity and kind of technology in use?

Segars: It’s very smart sensors gathering data, processing some locally and then passing it up to the cloud. It’s not just for pure consumerism. It’s not toys. There are real products to monitor vital signs about your state of well being, partnering with hospitals to make sense of that data and use it to present your physician with a much more accurate view of your health. There’s huge potential for IoT in medical grade data collection. And that’s just one market.

SE: All of these things have the potential for massive disruption, too.

Segars: Trying to predict based upon changes in the data—who wouldn’t want that?

SE: What if the wrong person gets a hold of that information?

Segars: You need to secure the data, but you also need to authenticate the user. It’s a computation problem. If your insurance company is giving you a lower rate because you go jogging every day, they’re going to want to know you didn’t strap the watch on your dog. Some elements are scary. But having your credit card company call you up and question whether there’s fraudulent activity with your card is good. There is mass opportunity for people, and huge opportunities for businesses to be more efficient and glean information about how people use their products—and generate new products based on that information.

SE: Mobile ties all of that together, right?

Segars: Yes. There are so many screens that you don’t always need another one. The mobile device is the interface with many IoT devices. It’s hand in hand. Even in industrial settings where you’re tracking how factories are operating or goods moving through the supply chain, these devices will be the window into that.

SE: Is the future for ARM selling processor or graphics IP, or will you be selling bigger, more integrated pieces?

Segars: The complexity of putting those compute subsystems has gone up. If you look the chips in networking, there are a lot of processors and the communication between them is very important. One area we’ve invested a lot in over the year is the interconnect between the processors—maintaining coherency between processors is very important.

SE: This is AMBA and ACE?

Segars: Yes. We call it systems IP. You look at the complexity of some of that and it looks like a microprocessor. It’s not as high profile as GPUs and CPUs, but it’s as important to system-level performance as anything else.

SE: You also can build security into every step of that, too, right?

Segars: Yes, and Trust Zone does extend into that—signals in the interface to be enabled when you’re secure space versus not when you’re in the untrusted space. So you can use that to have peripherals on chip or off chip that are only enabled when you’re in the trusted mode.

SE: So do you build full platforms out of this?

Segars: We are providing all the components so people can build their own, and mix and match what they want. In some cases, people ask us to show them how to put it all together so they can concentrate on something else. But we don’t limit people’s ability to innovate. We don’t say, ‘All chips should look like this.’ That’s a bad idea. The exception is in the server area. We looked at the complexity of porting software over to a server chip. If every chip is completely different, then from a software engineer’s point of view that makes your life hard. So we focused on what areas need to be standard that don’t limit innovation. We created a single server-based architecture. We got big companies involved—Microsoft, AMD, Dell and HP—all companies with a vested interest in standardizing in the right way and leaving innovation in the right place. We’ve achieved very good progress there.

SE: You just acquired Duolog. Why?

Segars: It’s about the complexity of putting these systems together and the complexity of the interconnect, and how we make that easier to deploy. As our interconnect products have become more sophisticated, we’ve built some fixed instantiation with that. Otherwise it’s too hard to deploy and verify. It’s about making that easier. We’ve been working with them for the past couple years and we decided it was time to bring that in-house and integrated it with what we were doing. It will provide more rapid time to market for our customers.

SE: Where are acquisitions on your road map?

Segars: We’ve done three or four in the last 12 months. They’ve been small. Sensinode was one. We acquired some technology from Cadence. Our goal is to acquire some core skills and grow up from there.

SE: Is that part of the plan going forward?

Segars: We’ve continued to look for opportunities. Typically these companies want to see their technologies get to market. The benefit is the integration with the rest of ARM and our reach into the industry. ARM is a horizontal play company. CPUs and GPUs can get used in loads of places. The number of screens you’re going to see in cars will go up and up.

SE: Any new markets that will crop up?

Segars: In general, it’s mobile, servers, and the network that links it all together in between. The volume of data that needs to be processed will go up and up and up. The growth of IoT creates a new challenge for networks that isn’t immediately obvious. On a cellular network you’re looking at a video or making a call or looking at a Web page. You create a connection, download a chunk of data. With an IoT device, it wakes up every period, there’s a really small amount of data it needs to transmit, and then it shuts down again. So you have very high-frequency on and off connections that have very different profiles than what you do on a smart phone. That creates a new set of challenges for the infrastructure, which will drive change and innovation. And that, coupled with the demand for higher bandwidth and lower latency, will drive innovation in the network infrastructure, which is in itself a really interesting and valuable market.