A microchip that outpaces supercomputers

Meet the Neurogrid It can do the work of one million cortex neurons Courtesy Rodrigo Alvarez/Stanford University

In 2005, IBM’s $2-million BlueGene supercomputer took 80 minutes to process the same data that eight million cerebral-cortex neurons—a fraction of the brain’s total—handle in one second. Now bioengineer Kwabena Boahen of Stanford University has built a microchip that could help computers catch up.

Penny For Your Thoughts: The Neurogrid can reveal how the brain processes data  Courtesy Rodrigo Alvarez/Stanford University; iSTOCK

BlueGene’s downfall was that it ran data through each of its 1,000 chips before computing even the simplest command. When completed this month, Boahen’s device, called the Neurogrid, will contain one million simple silicon circuits working in parallel. When data hits one "neuron," it relays the info to all of the circuits, and the best neuron for the job generates the response. Each neuron is slower than BlueGene’s chips, but this approach will allow the Neurogrid to bear the same workload as one million cortex neurons. More important, it does so in real time, which could help scientists follow how brains afflicted with epilepsy or schizophrenia process information and to then develop treatments. Boahen says it will cost $60,000-—cheap enough to put one on every lab bench.

2 Comments

This story was confusing. It starts off by making comparisons of these new style computer chips processing power to that of cortext neurons in the brain. Which was great but then near the end jumps track and sounds more like it can be use to literally process information from the brain that can be used to develope treatments. I think there needed to be a little better clarification or expansion of the article there near the end.

That's because the article is poorly written. If you want to understand this story, check out Kwabena Boahen's Ted talk ( http://www.ted.com/index.php/talks/kwabena_boahen_on_a_computer_that_works_like_the_brain.html ).

This is really just an advance in thinking about parallel processing that is heavily informed (or perhaps entirely informed) by our knowledge of artificial and biological neural networks.

I like to think of it as a programmable hardware neural network.



Download Our iPhone App

Stay up to date on the latest news of the future of science and technology from your iPhone with full articles, images and offline viewing



Follow Us On Twitter

Featuring every article from the magazine and website, plus links from around the Web. Also see our PopSci DIY feed



Become a Fan On Facebook

Share links with friends, comment on stories and more


December 2009: Best of What's New

In our December issue, Popular Science names the 100 best innovations of the year: bombproof wallpaper, self-parking cars, the fastest helicopter, and 97 more. Plus inventor profiles and videos.

Check out the best of what's new here.

Popular Science Photo Pool


Share your photos in the Pop Sci pool at www.flickr.com!
tags_sprite.png
POP_embeddedForm_cover_May09.jpg