ARM’s CEO digs into the future of mobile and the IoT, why ecosystems are so important, and what’s changing in the data center.
Simon Segars, CEO of ARM, examines the future of mobile computing, how it intersects with the IoT, why ecosystems are vital, and how computing is evolving. What follows are excerpts of that conversation.
SE: Most analysts say the growth rate of mobile is slowing. The big buzz phrase now is Internet of Things. How does ARM’s role change with that shift?
Segars: Mobile is still changing and evolving, but it’s key to IoT, as well. A lot of IoT will be embedded things and smart phones and tablets, all with a very convenient mobile interface. So mobile has a big role to play in IoT. But in terms of IoT, it can mean lots of things to lots of people. Some people say a connected car is not IoT. Other people say it is. But if you boil it down to fundamentals, you have something connected on a network that isn’t a conventional computer. It can be a connected car, a connected watch, some sort of connected health care. It can be a sensor in the middle of a field determining whether a crop needs watering.
Fundamentally there is a sensor, some processing, some connectivity, and a security requirement. The connectivity could be over a myriad of standards. It could be directly on the Internet or going through a gateway. But there is a handful of architectures. We’re looking at it from the point of view that processing will be required everywhere.
Efficiency will be required, too. That will include power efficiency, radio efficiency, processing efficiency to do encryption. And efficiency will be required everywhere to get the cost down. We’re trying to broadly look at those problems and understand how they can get solved in an industry-efficient way.
SE: Initially everyone thought these IoT devices would be simple. That’s turning out not to be the case. How do you see that evolving?
Segars: The initial deployment of IoT is going to be quite vertically oriented. You can draw a picture of everything being connected everywhere, and that would be glorious, but when it comes down to solving problems it’s easier to get one thing to work. You’ll see vertical implementations of IoT.
The sophistication of the IoT device in some cases is quite high. In others, it’s still quite small. You’ll see more variance of the initial model, which was data being gathered and all getting pushed to the cloud. It won’t be everything going into the cloud, because if you put everything into the cloud your network falls over. It won’t be able to handle enough data.
What’s important is to take data that’s gathered and try and work out what information is useful to pass upwards in the chain. It’s energy-inefficient to move all the data around, throw a lot away when it’s in the cloud, and then send a very small piece of information back. If you can gather it and process it as you go, you get better efficiency. So as much as we’re focusing on IoT, we’re also thinking about how the network evolves to enable distributed computing. You don’t just have a device-sophisticated switch-cloud model. You have a more smeared computing environment through the network itself, and you have local data processing where it’s gathered, only passing information upwards if that’s useful.
SE: That’s one of the shifts in dealing with big data, in general. You keep the data close to the processing rather than moving the data to the processing, right?
Segars: Yes, and if you look at the specs of 5G, they’re about making video downloaded to your phone run faster. But they’re also about dealing with IoT and a number of connections, and lowering latency. If you have a model where all requests for data go back to a remote data center, which could be in Iceland, then having a millisecond data delay is going to be pretty difficult. You’re going to have to process it locally and get an answer locally in order to improve the effective bandwidth of the network and address the latency requirements. Moving all the data around is costly. That’s going to lead to more distributed computing. You’re going to want to do that in an energy-efficient way. You’re going to have some small servers sitting next to a base station. Energy efficiency is going to be important for that.
SE: Let’s go back to mobile. While growth is flattening, prospects for mobile devices in the IoT world are still pretty good, right?
Segars: Yes, there will be more devices. And critically for us, the devices are getting more sophisticated. The opportunity for ARM content goes up. From a business point of view, mobile is still a very important and vibrant part of our market. Handset volume growth may well be going down, but sophistication of content is going up and will continue to do so.
SE: Does it matter if sales are a mix of lower-end and high-end smart phones?
Segars: No. If you look at a low-end device, it’s what was a high-end device a few years ago. The technology trickles down very quickly. As 28nm processes become very cost-effective, you can pack a whole lot of transistors into an economically viable chip—and therefore deliver what was a premium experience a few years ago in a handset that costs a couple hundred dollars.
SE: ARM has been pitching ecosystems for years, but it has a lot more of them to manage these days. How do you deal with that?
Segars: We are very focused on ecosystems, and our ecosystems are expanding. We expect that to continue. Way back, our ecosystem was the semiconductor companies we were working with. We focused on working with our licensees—our partners—to create a gearing effect for how broadly ARM was used. Over time, we started working with more and more people up and down the supply chain. All that time we’ve been expanding that ecosystem as the use cases evolve and the range of the technology evolves. So we talk to more and more people, and when you get into IoT, there seems to be an almost limitless ecosystem you need to develop in these different verticals. There are all sorts of companies that wouldn’t have talked to each other before. As the data is moving around, people can make a benefit out of the data and have a role to play in deployment of IoT. The volume of that ecosystem continues to expand.
SE: Do you ultimately see it as a single ecosystem, or is it concentric rings that resemble a giant Venn diagram?
Segars: It’s more like the Venn diagram. If you take health care and smart cities, you can view them as completely different ecosystems. But there are some overlaps. People who are going to manage services in a hospital have very big equipment you want to do preventive maintenance on. That is a service that might be run by a big company. You might also have services in a smart city where the devices are coming from similar companies, but the people who are going to use the data might be completely different.
SE: What do you provide to health care versus smart cities? Do you provide the entire stack of MCUs and CPUs?
Segars: In some ways, we facilitate bringing that today. In terms of actual products, Cortex M controllers are going into lots of products, which will form devices. Our Cortex-A processors are going into gateways and the network. So in terms of a raw product offering, we’re detached from some elements of those ecosystems. But we do go out and talk to people about use cases and what they’re going to do with that. Who don’t they know that they should be talking to? Part of the role is facilitating conversations between different people in the ecosystem who didn’t necessarily have to talk before.
SE: With the diversifying markets, do you find that your products can play across all of these markets, or do you have to come up with different solutions for each of these areas?
Segars: In terms of processing, that is something that works horizontally. We have different processors with different power, performance and area and different features for different markets. But if you look at the M-class cores, for example, there’s a family of them, but there’s nothing specific for health care or smart cities. There are markets where ZigBee is really important and other markets where Bluetooth is really important. There are some standards that tend to differ.
SE: But you have to put a lot more work into how products will play in one market versus another, right?
Segars: Yes, and we have more and more people who understand those end markets. But we don’t want to make vertically oriented products. We want to make horizontally oriented products. But we will take as much learning as we can so we don’t inadvertently create something that doesn’t work for some end market.
SE: As other markets begin growing in terms of mindshare, the volume may be significantly smaller than mobile. Even with automotive, you may have 10 SoCs, but each is different. How do you deal with that?
Segars: Yes, the dynamics are different and the volumes are different. Within a high-end car there may be 100 to 150 different chips. They’re not all the same. The chip that’s in the engine management, the one that’s in the IVI, and the one doing digital processing on a camera are all different. There are some MCUs that are used in lots of places, like opening the doors and the windows and other lower-level functions. But there are a number of places where there is significant intelligence that needs to go into a car. The silicon content per car is going to be higher than a smart phone. There are many chips in a smart phone, but there will be many of those in cars, especially when you get to self-driving cars and ADAS.
SE: ARM has a significant presence in a lot of markets these days. What does the company have to do now that it didn’t have to do five years ago?
Segars: We want to see IoT deployed without a ton of fragmentation. A software platform can make it easier to take a microcontroller and connect it to the cloud and get services running. Providing software running alongside the hardware that our licensees are creating accelerates the IoT and makes the ARM pie bigger. We are in a lot more markets doing software and software-enabling work, partnering with a lot of people. We aren’t experts in all of these areas. We know a lot about a lot, but there is a whole lot of expertise we don’t have. Nor should we bring that into the company and solve all problems for all people. We’re still very much focused on partnering and creating solutions everyone can benefit from.
SE: Can you give an example?
Segars: If you take networking, we’re doing a lot of work around NFV. We’re doing a lot of work in Linaro around a whole bunch of areas. In the data center, we’re partnering with people, doing benchmarking of workloads and working out how to optimize it. Our role has moved far beyond designing microprocessors. We’re thinking about the use cases, the systems, and bridging those partnerships to create a complete solution.
SE: Is the focus less on performance than power?
Segars: It depends upon the market. If you take mobile, it’s all about delivering the most performance you can out of that battery. The demands on that battery are extreme, and you have to balance the power going into the modem and compute power. If you look ahead at 5G modems running a screen and other features, there is a lot going on. Delivering the most performance you can in a strict power budget is really important in mobile. In the embedded space, you need sensors that can run off a solar cell for a decade, or maybe forever. That’s a very different set of concerns. But in any one of those markets it’s all about power/performance tradeoffs. In data centers, systems used to be designed around how much electricity you had coming through the front door. Now it’s about delivering the most compute power you possibly can based on how much electricity you’ve got.
SE: How much efficiency is there to be gained in data centers?
Segars: If you started with a clean sheet of paper, you probably wouldn’t design data centers the way they’re designed today. There’s a rack architecture, and that’s the way it is. You have to work with that. We see an opportunity to take an SoC approach into the data center and change the way it’s put together. But there’s also a lot of infrastructure already out there. It will take time to change, but there are some clear technology benefits to be had by bringing some fresh thinking into that.