Top 100: The most desirable cars of all time

"Neurogrid" circuit modeled on the human brain is the fastest, most energy efficient of its kind


May 2, 2014

A group of engineers at Stanford have developed an iPad-sized, highly power-efficient way ...

A group of engineers at Stanford have developed an iPad-sized, highly power-efficient way of simulating a million neurons and billions of synapses for as low as US$400 (Image: Stanford University)

Image Gallery (4 images)

A group of engineers at Stanford has developed an iPad-sized, highly power-efficient way of simulating a million neurons and billions of synapses for as low as US$400. The breakthrough could both help our understanding of the brain and help develop a new generation of bionic limbs that are controlled by the patient's brain in real time with little effort at all.

With its hundred billion neurons, the human brain, possibly the most complex object in the known universe, can fully operate our bodies on only 20 W of power. Our measly man-made microprocessors are still decades away from coming close, despite the fact that performance per watt has been increasing exponentially over the years,

Simulating a brain, and a human brain in particular, is a very enticing prospect. It would allow neuroscientists to learn a lot more about how our minds work, and it could have very interesting repercussions in both artificial intelligence and in creating ultra-low power biomedical implants, among other things.

Initiatives like the Human Brain Project in Switzerland are attempting to simulate the human brain using the most powerful computers in the world, while others, like SyNAPSE at IBM, are tasking themselves with reinventing a completely new computer architecture built around very low power consumption.

On the same vein as SyNAPSE, the Neurogrid project at Stanford led by associate professor of bioengineering Kwabena Boahen has built an ultra-low-power circuit board that simulates a million neurons and billions of synapses on less power than it takes to run a laptop.

Neurogrid treats all signals as analog, leading to massive power savings (Image: Stanford ...

Neurogrid is a circuit board about as big as an iPad and made op of 16 custom designed "Neurocore" chips, which have been designed to run on as little power as possible. The signals handled by each microchip in the board are treated as analog rather than digital, and with levels of voltage and current matching those sent by the neurons inside our brains.

These characteristics make Neurogrid both the most energy-efficient and the most economical way for researchers to simulate a brain, to the point that it might become a staple in the labs of researchers in the field. The device cost US$40,000 to build, but the team believes that switching to a more modern manufacturing process could cut costs by a factor of 100.

A $400 machine that can simulate a million neurons on so little power could find use as a way to control prosthetic limbs via the brain with the speed and sophistication of a real limb, but without being tethered to a power source. A chip could be implanted in a paralyzed patient's brain, interpret the signals, and translate them into commands for the artificial limb in real time.

Boahen and colleagues are now working on the next steps, which will be to lower costs by improving the manufacturing process and, secondly, to create compiler software that will make it much easier for software creators to operate it and build their own applications.

A paper describing the circuit was published in the journal Proceedings of the IEEE and the circuit board is described in the following video.

Source: Stanford University

About the Author
Dario Borghino Dario studied software engineering at the Polytechnic University of Turin. When he isn't writing for Gizmag he is usually traveling the world on a whim, working on an AI-guided automated trading system, or chasing his dream to become the next European thumbwrestling champion.   All articles by Dario Borghino

This is really interesting research. I think the brain is estimated to be equivalent to about 36.8 PFlop/s. There was a simulation done a couple years ago on the K computer which is 10.5 PFlop/s and the current 4th fastest supercomputer in the world. They simulated 1% of 1 second of the brains processing power and it took 45 minutes to do it. The cost of the electricity (not counting cooling) at 12,660 kW and ~15 cents/watt would have been about $1,500 for that calculation.

This is part of the reason for their research. With the Status quo even for all the advancements we have made we are an extremely long way away from being able to simulate the brain efficiently.

Robots are projected to displace some of the workforce over the next 20 years but once computers can match the processing power of humans I predict that they will take up things like watching reality TV and using narcotics to deal with depression. At first their productivity will continue but then their work and relationships will start to suffer and the job market will swing the other direction.

2nd May, 2014 @ 08:12 pm PDT

This is it. This is how it begins.

Joel Detrow
3rd May, 2014 @ 01:26 am PDT

Yes, bring it on, do a Kickstarter!

Kris Lee
4th May, 2014 @ 05:46 pm PDT

dashl, interesting thoughts on sentient machines.

perhaps a new, improved Touring Test , in fact.

even using analog models, they can only model known pathways and intetactioms on them. meaning i think the DARPA random walkabout Beyesian studies will be more fruitfull.

but then, moving a trigger finger is not even close to being the agent in charge of the meatsuit with that finger, is it?

Walt Stawicki
5th May, 2014 @ 09:32 am PDT

Does this mean we can have a $400 bionic man? How long will it take an army to build a space elevator, satellite city, starship ark, and populate our galaxy? Can we finally escape the womb? (Birth of a sentient species?)

Don Duncan
5th May, 2014 @ 05:34 pm PDT
Post a Comment

Login with your gizmag account:

Or Login with Facebook:

Related Articles
Looking for something? Search our 29,897 articles
Recent popular articles in Electronics
Product Comparisons