Engineering On The Edge

Not that long ago, I wrote about computers behaving in a more intelligent manner, in the field of artificial intelligence (AI). Other computing disciplines also look to the brain for inspiration, but rather than working on a computer that thinks like a person, the scientists at Intel are working on a chip that mimics how the brain operates.

Even though you might not be able to recall the amount of information available to a supercomputer, your brain still operates in ways that can’t be duplicated by even the most sophisticated computers. You might not be aware of all the simultaneous operations your brain performs (breathing, chewing gum while walking, ignoring your nose in your field of view), but it is a fantastic organic computer.

Neuromorphic chip architecture. Courtesy of Intel.

Moreover, your brain is much more energy efficient than any computer that could come close to its computational power. Even in a household — with multiple devices that have to be recharged, TVs blaring, computers clicking — it doesn’t come close to using the amount of electricity consumed by IBM’s Watson in a day.

Even just looking at the power consumption advantage, you can understand why scientists are interested in creating a computer that acts more like the human brain than the silicon models we use now. Our brains use neurons to conduct business and, thus far, every attempt to create some sort of neuron-silicon hybrid has required excessive amounts of power and isn’t capable of the parallel processing that allows your brain to process drinking coffee and reading this blog.

Neuromorphic chips seek to solve this dilemma by copying the human brain, using new technologies, rather than just building on silicon. Intel’s Circuit Research Laboratory in Hillsboro, OR has based its design on lateral spin valves and memristors. Intel says the new architecture mimics the movement of neurons and can be used to test methods of reproducing the brain’s processing power.

Intel also claims that chips created using its new architecture use less power. From the Intel proposal:

We present a design-scheme for ultra-low power neuromorphic hardware using emerging spin-devices. We propose device models for ‘neuron’, based on lateral spin valves that constitute of nano-magnets connected through metal-channels. Such magneto-metallic neurons can operate at ultra-low terminal voltage of ~20 mV, resulting in small computation energy…We show that the spin-based neuromorphic designs can achieve 15X-300X lower computation energy for these applications, as compared to state of art CMOS designs.

All of this may or may not herald a breakthrough in technology, but it does set the groundwork for further testing in the field.

Below you’ll find a video from BrainScaleS that discusses neuromorphic tech.

Sources: Technology Review, Cornell University Library