09 November 2010
Mike Muller, chief technology officer, ARM
Increasing software complexity means more mcu performance is required. ARM's chief technology officer tells Graham Pitcher how the company is pushing the boundaries.
One of the founders of ARM, Mike Muller has seen the electronics industry develop over almost three decades. And there has been significant change.
"At the high end," he noted, "the industry is now driven by companies building software around complex products and hardware is probably the easiest part of the problem. But you can now see this happening at the other end of the market."
At the entry level, software complexity is increasing as designers build more functionality into their products. "This is becoming a challenge for 8 and 16bit mcus," Muller noted, "and is great news for ARM, because more software needs more performance from the mcu. To meet this need, we've extended the ARM architecture to provide more sophistication."
As an example, Muller pointed to the recently launched Cortex-A15 multicore processor, which ARM describes as the 'highest performance licensable processor' yet launched. "We're also pushing virtualisation and extending addressing ranges," he added.
ARM doesn't make semiconductors; instead, it develops IP that can be applied by other companies in their products. The mcus recently introduced by NXP and Freescale are examples; while both have launched mcus featuring ARM's Cortex-M3 core, their products are different.
Although linked historically with the mobile phone world, mobile now accounts for only half of ARM's shipments. "We've shipped about 20billion cores so far," Muller noted, "but we're projected to ship another 100bn in the next 10 years."
He expects this huge number of cores to find application in the 'internet of things', in embedded control and in sensors. "Digital consumer also has a long way to go," he added, "and there is a big push to the higher end, where ARM powered servers will be used in the 'cloud'."
Muller said ARM's processor road map is 'well plotted' for the next couple of years. "Processors will roll forward and we know what the technology ingredients will be. But things always change your focus and we need to maintain a watchful eye on the underlying technology trends."
ARM remains a relatively small company. "So we use our business model to focus on microprocessors, graphics and so on. Our partners then focus on building mobile phone chips or devices for set top boxes or whatever. By doing this, we have reduced the space we need to investigate."
But he acknowledges the need to ensure that it's easy for programmers to bring together the appropriate software and to bring together open source contributions. "ARM has done this," he claimed, "while staying outside of the politics of distribution. We have lots of answers and lots of agendas, but need to remain agnostic."
One of Muller's roles as cto is to drive ARM's technology road map. What are the challenges he's addressing? "Everyone is trying to solve basic problems," he said, "such as how to provide more functionality within a constrained power budget. In the past, power was less of an issue. Now, it's a serious topic ranging from energy scavenging to server farms."
He contended that, until recently, ARM was the only company thinking about power. "The industry has managed to make major savings in power from shrinking voltages. But, while Moore's Law continues in terms of getting more on a chip, semiconductor processes don't support voltage scaling."
ARM's R&D has, unsurprisingly, become more complex. "In the early days," Muller recalled, "one core was good for everything. We've moved beyond that to three product divisions: home; mobile; and embedded and enterprise.
But, because we don't build, for example, mobile phones, there's a whole chunk of problems and we need to know what the developers' problems are. Effectively, we're solving 20% of the product design problem, but ensuring we spend time with our partners."
Each division has an Advanced Product Development group doing work which is close to a product announcement. Then there's a group that reports to Muller directly. "This separate R&D group, which isn't affiliated to any division, allows us to do cross divisional work. We can try extreme things with the microprocessor architecture that may require strange things to be done with physical implementation and with libraries."
This group also works on a different time line to the divisions and, said Muller, has 'ring fenced' resources. "The group also engages externally, for instance with universities like Southampton and Michigan. The group is doing things that will take at least three years before they get into a product division and then a further three to five years before they get into volume."
Like most technology companies, ARM finds recruitment a challenge. "It's why we have major engineering centres in the UK, in France, in India and in the US. We've also just started a small R&D group in Shanghai," he pointed out. "These regional offices are also about having engineering talent close to customers."
And he admitted that different work practices are emerging. "Not everyone wants to work in the UK or in San Jose. For example, with Linaro, we've started to pull together digital vertical communities; seas of individuals and islands of engineering. While we can manage work in a distributed way, we still need to move people around the world, no matter how good the technology," he concluded.