Today’s computers excel at mining vast amounts of data on scales humans can only fantasize about. But human brains are still better than computers at many tasks–we can instinctually recognize a person’s face and words, or interpret meaning from a stranger’s actions.
IBM scientists announced in a publication in the journal Science last week that it has developed an ultra low-power computer chip, named TrueNorth, that thinks like a human brain, complete with 1 million programmable “neuron” connections.
The brain-inspired chip is about the size of a postage stamp, but comes packed with 5.4 billion transistors but consumes the same amount of power as a hearing aid battery, an amount far lower than today’s standard processors. While it can’t process data nearly as quickly as a regular chip, it’s able to perform well on other tasks at which traditional chip struggle, especially related to computer vision, language, and other applications that mimic the processing of a person’s five senses.
Speaking to the New York Times, Horst Simon, deputy director of the Lawrence Berkeley National Laboratory, called the invention a “remarkable achievement in terms of scalability and low power consumption,” one that’s similar to the advent of parallel supercomputers in the 1980s.
For years, researchers have been working on similar “cognitive computing” technologies. The IBM work was funded by the military’s R&D branch, DARPA and was the result of nearly a decade of research.
IBM believes the neuromorphic chip architecture will complement–rather than replace–today’s traditional designs. In a statement, IBM chief scientist Dharmendra S. Modha said the chips could “transform mobility, via sensory and intelligent applications that can fit in the palm of your hand,” but don’t need an Internet connection. That’s because device could compute far more contextually-dependent details in real-time without needing to connect to powerful, faraway cloud data centers.
The chips aren’t available yet and require more R&D work. IBM says it will work to integrate this new processing into computing at many scales, including mobile devices, cloud services, and supercomputers.