Brain-Inspired Chips Will Allow Smartphones to Understand Us
December 17th, 2013 by admin

We should look to biology to figure out how to make smartphones more ­helpful, says M. Anthony Lewis.

By M. Anthony Lewis on December 17, 2013

A modern smartphone is the most powerful information portal the world has known, integrating a traditional telephone with a powerful Internet-connected computer capable of navigating, playing multimedia, and taking photos. I think the next major step in smartphone evolution is obvious: the devices will become intelligent assistants that can perceive the environment and follow our commands. This will become possible thanks to progress in building chips inspired by the functioning of mammalian brains (see “Thinking in Silicon”).

We hope to achieve what I call embedded cognition—intelligence that resides on the mobile handset itself rather than on a distant server. We want devices that are always listening, watching, and paying attention to us, without compromising battery life. We need new kinds of algorithms to process streams of sensory data from sights, sounds, physical sensations, and more. We need our phones to be capable of learning so that they can come to understand their owner. And we need to stuff this intelligence inside compact, power-efficient hardware because we don’t want to transmit data off the smartphone for processing—a requirement that causes delays for users of Apple’s Siri and the Google Now app for Android phones.

 

A team of engineers and neuroscientists at Qualcomm Research is working on a new type of processor to meet those challenges. It takes design cues from the human brain, which despite using only about 20 watts of power is the most impressive and efficient “computer” that we know of at processing data from the real world—the kind we want smartphones to handle too.

The Zeroth processor, as it is called, works on data using silicon “neurons” that are linked into networks and communicate via electrical spikes. A system with a Zeroth processor can learn. In one test, researchers trained a wheeled robot to favor certain areas of a room by rewarding it when it was in the correct place. We also envision sensors modeled on the nervous system. They would conserve energy by reporting only when the environment had changed, instead of transmitting data constantly at all times.

This biologically inspired approach to computing should pave the way for the next major upgrade to the 130-gram marvel we call the smartphone.