17 May 2014

Sandia National Labs is stepping up its neuro-inspired computer chip design

Human ingenuity has given birth to incredibly powerful computers that can plow through more calculations in a second than most people could in their entire lives, but computers still aren’t terribly adaptable. The human brain is a very different kind of computer a massively parallel processor that has been shaped by millions of years of evolution to recognize patterns and adjust to changing situations. This is the kind of capability computer science researchers are now trying to unlock, and scientists at Sandia National Laboratories are stepping up their game to design neuro-inspired, or neuromorphic, computer systems.

Sandia isn’t just attracted to the idea of computers designed like brains because of the capabilities, but the human brain is also incredibly efficient. A computer has trouble telling the difference between a picture of a dog and a cat, but it eats up hundreds of watts of power simply trying. A brain, by contrast, operates continuously for decades and only consumes roughly the same power as a 20-watt light bulb.

Clarity brain
The brain is a pretty complex thing

A modern computer is essentially a very powerful calculator linear and much faster than humans at running the numbers. Scandia is spinning up this long term project to work on the hardware and software that could make a computer more like a biological brain. The first step in doing so, according to Sandia, is to look at the physical architecture of transistors.

A single neuron, like the ones seen above, can be connected to as many as 10,000 other neurons. Then each of those neurons could be connected to 10,000 more, and so on until you’ve got the bundle of nerve tissue between your ears that makes you who you are. Researchers believe this parallel system is what makes the brain so flexible, or plastic in neurology lingo. Most conventional transistors only have connections to four other transistors, but making a computer more interconnected is just the first step.

Sandia isn’t embarking on a quest to crank out an artificial intelligence platform right away, which is actually the approach many researchers take. A neuro-inspired parallel computer isn’t one and the same with AI, but they are closely linked. At Sandia National Labs, scientists are going to work on improved hardware and processing algorithms, and see if artificial intelligence can be developed as a consequence of that.

Sandia chip

Computers designed from the ground up like a brain could have important applications in robotics and remote sensors. These devices could benefit from the common sense and adaptability the brain imbues us with. They could learn from data and experiences to make predictions about what a human is going to want to know or do without being explicitly programmed for it. This is a bit like taking baby steps toward true AI.

Right now, your computer, phone, and everything else with a microchip works a certain way. There is very little room for error either you enter commands correctly, and click on the right things, or you don’t. Regular computer systems are beginning to show a little flexibility, for example Google Now can often figure out what data you might want to see. It’s really cool when it works, but that’s because we’re still at the point where a computer guessing anything right is amazing. Sandia researchers want to make that common place.

Courtesy Extremetech


Post a Comment

Get every new post delivered to your Inbox.


Copyright © 2015 Tracktec. All rights reserved.

Back to Top