The future according to ARM

A global umbrella-net and multicore mobiles: we look at what ARM's research team is working on

ARM-designed chips have been a key stimulus to mobile computing, influential in smartphones and tablets, but according to the British company, the benefit of its research will only really show as more devices are connected to the net.

The “Internet of Things” is a concept that was mooted as far back as 1999, but has yet to take hold. Despite countless stories about fridges talking to Waitrose and foodstuffs alerting users when they’re nearing their sell-by date, net connections on household appliances remain relatively rare.

Global umbrella-net run on low-power chips

According to ARM, that could change as lower-powered and cheaper chips become available and are shoehorned into everyday items, from light bulbs to radiator thermostats. Much of ARM’s focus is on saving energy.

“We haven’t started yet – there are already millions of device-to-device connections, but that will be dwarfed by the numbers of devices we’re going to see,” says Gary Atkinson, ARM’s head of embedded computing. “Over the next five years, it’s about saving energy – we need to take more control of that.”

ARM predicts an increase in granular controls that make homes, cars, appliances and office spaces smarter. Lights or heating will be controlled by a generic set of protocols rather than enclosed within proprietary systems. “It’s about an appropriate gateway – can you buy a sensor from one company that will work with kit from another company off the shelf?” says Atkinson.

“It’s about making everything connected, not only mobile phones and tablets; if you can provide the connections to things that you can control via apps, it makes them more powerful.”

It’s about making everything connected, not only mobile phones and tablets; if you can provide the connections to things that you can control via apps, it makes them more powerful

ARM says that lower-powered chips don’t only enable connected widgets – they can also extend the web to areas without proper power or telecoms equipment. The company says if the power usage of equipment that runs mobile networks, for example, can be throttled sufficiently, areas with no access to modern communications could be revolutionised by drop-and-go wireless hubs.

“Once you can reduce power enough, then you can start to distribute computing power and run mobile communications servers nearer the edge of the network. You can make them run at ambient temperatures, outside the data centres,” explains Atkinson. “We see a time when basestations could have caching servers built into them and they could run off solar-powered batteries.”

“If you can use wireless technology and put basestations into groups of villages, then you can provide connectivity – that’s the target power consumption point. If you can make that work, you push wireless into areas without it.”

Multicore crunching

If a global web of things is a grandiose vision, then ARM’s immediate plans are to squeeze more processing into smartphones and tablets. The company sees a move towards “heterogeneous computing” as vital to improving performance in smartphones without crippling already overworked batteries.

According to ARM, system-on-a-chip technology with multiple cores working more efficiently will start to arrive with the next generation of mobile processors. “The direction it’s going in is about true multicore and heterogeneous computing – the concept that you’re no longer running an application on a single processor. It’s about code being compiled for a platform and the platform being a mix of processor cores,” says Atkinson.

The direction it’s going in is about true multicore and heterogeneous computing – the concept that you’re no longer running an application on a single processor

The vision involves a mix of general compute cores and GPU cores, with specific roles such as hardware acceleration given to the most suitable parts of the processor. The “general-purpose computing on graphics processing units” (GPGPU) means number crunching and parallel processes could be moved off the main processor and onto the graphics chip when it isn’t in use.

“Take a Samsung Galaxy S III that has both ARM’s processor core and GPU cores inside a single chip,” says Atkinson. “Future generations of that will have an interconnect that will link the GPU cores and processing cores, and create a GPGPU or an open platform. You can compile your code to take advantage of the eight or 16 GPU cores you have, and some algorithms or parts of that app will run quicker on a GPU with multiple parallel pipelines than it would with CPU processing power.”

The chips would move data-processing tasks that weren’t time-sensitive to smaller cores that consume less power, while pushing other tasks onto cores that deal with the data quickly so that part of the processor can be turned off.

Read more

Analysis