ARM to Address Devices from the Edge to the Core (and Everywhere in Between)
On my way from one room to another at ARM TechCon 2018 last week, I found ARM CEO Simon Segars in the hallway at his own show, sitting at a high-top, tapping away on a notebook. “Doing your own typing?” I enquired. He looked up, smiled, and said “Of, course.” He’s the hands-on type.
I asked Segars how many people had turned up at his conference, and he told me 5,000, a respectable lot, quite a bit beyond the 11 attendees who populated the company’s first show in 2004. And well this crowd should be there. ARM, perhaps serendipitously, has in the meantime floated to the center of the next era of computing, really because its original mandate put power budget on the first rank along with performance, rather than as an afterthought. Who knew that one day mobile computing would take over the world?
Well, arguably, Robert Noyce, an Intel founder, and Steve Jobs, who back in the 1970s learned from Noyce the expected trajectory of computing over the next several decades, an arc that would make possible ever smaller form factors. And it’s no mistake that Apple took an ARM architectural license for its iPhone processors when the right time came along. The combination of a stingy power budget and sufficient performance made it perfect for the job. Today, essentially all mobile phones — both Android and Apple — are based on ARM architecture.
At this year’s conference, the company revealed its view of a wider world, filled with a trillion interconnected devices, a pool several orders of magnitude greater than today’s billions. And its not just that ARM’s power budget fits the profile of ever smaller sensing and activating endpoints, remote nodes, and potentially far-flung fully-connected devices on a vast network. The company is also bidding for the middle, the giant ocean of servers that make up the cloud.
At the show, Drew Henry, SVP of the Infrastructure Business Unit at ARM, revealed a company initiative called “Neoverse.” Neoverse addresses the architecture of “cloud-to-edge infrastructure for 1 trillion intelligent devices,” the coming world of the Internet of Things (IoT). So, besides already having the inside track on these trillion devices, ARM also plans to tackle the cloud.
The company first started talking this way several years ago. It was already the master of mobile and seemed an obvious fit for IoT, but most analysts were skeptical of its prospects on the server side, which has been the citadel of x86 architecture and particularly Intel.
However, the world has evolved since then. Most importantly, Intel’s dominance of process-node technology, the art of making ever smaller features on silicon chips, has faded as the company struggles to get its 10nm parts to market while 10nm ARM chips are now common. ARM manufacturing partners plan to bring out 7nm silicon in 2019, potentially putting them a generation ahead of Intel.
Since every “shrink,” as these steps are called, yields benefits in both power consumption and performance, ARM can now legitimately address the server market with denser, faster chips. Henry said partners would ship 1 million 24-core servers this year and described R&D investments in a 32-core “cloud application” server chip.
Along with this expansive vision of ARM’s role in the next generation of computing, the company is also taking security seriously. Since each one of these tiny endpoints represents a potential entry into the soft underbelly of the network, even low-cost, limited-function devices, like security cameras, home thermostats, and urban infrastructure sensors and actuators, must be battle hardened against intrusion. Thus, security was a topic on everyone’s mind at the show, with many ARM partners addressing different aspects of the security problem. There’s no way to handle the security discussion briefly, but the short of it is that ARM security architecture starts right in the silicon so hackers can’t get “underneath” the base layer of trust. One company on the showroom floor, Dover Microsystems (whose company name CEO Jothy Rosenberg said comes from a mashup of the phrase Do Over), demonstrated how its CoreGuard policy enforcer technology can prevent hackers from using software bugs to attack critical hardware. Dover recently snagged NXP, the automotive silicon provider, as a marquee client.
And speaking of NXP, one change from last year to this is that the huge merger deal between Qualcomm and NXP fell apart over regulatory issues. I had a chance to sit down with Lakshmi Mandyam, ARM’s VP of automotive, and she described this huge market as “part of the fifth wave of computing” (after mainframes, PCs, the Internet, and mobile). ARM technology is licensed by both NXP and Qualcomm. The former is dominant in what she calls “under the hood” systems like drive train; the latter has a strong position in in-vehicle infotainment. She noted that it will take a billion lines of code to run a fully autonomous vehicle, which will take year to bring to market. By way of contrast, Microsoft’s Windows 10 has about 50 million. “We’re trying to make sure these platforms are ARM,” she said.
I don’t think there’s much doubt at this point that they will be. Of all the architectures that existed 14 years ago, when the first TechCon took place, only x86 and ARM remain viable in the small-device space, and x86, which missed mobile entirely and seems to be seriously flagging on IoT, doesn’t appear to be getting much traction in automotive or for that matter almost any of the trillion devices expected to populate the IoT world in the near future. At this point, these are ARM’s markets to lose.