IBM's experimental chip 'thinks' like a human brain
- By Kevin McCaney
- Aug 22, 2011
Computers, even supercomputers, are still basically processing machines, programmed to perform tasks. They perform those tasks more quickly than humans do and, unlike humans, they don’t get distracted and they don’t forget what they’re supposed to do.
But they can’t think like humans. They don’t draw on sights, sounds, smells and other stimuli all at once, mix it with remembered facts and events, apply it to goals, and interpret, or project, events.
Not yet, anyway. IBM says it has taken computers one step closer to brain-like processing with its Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project, which seeks to ape, as it were, the functions of the brain on a new type of highly efficient processing chip.
IBM’s Watson vs. the human brain: Tale of the tape
Why IBM’s Watson is good news for government
The project, which has received funding from the Defense Advanced Research Projects Agency, combines elements of neuroscience, supercomputing and nanotechnology into to what it calls cognitive computing, Dharmendra Modha, manager of cognitive computing at IBM’s Almaden Research Center, writes on a company blog.
Modha writes that the SyNAPSE project uses advanced algorithms and silicon circuitry to create computers that could function without set programming but could “learn through experiences, find correlations, create hypotheses, and remember — and learn from — the outcomes.”
Such a system could, for example, monitor the world’s waters via a network of sensors monitoring temperature, water pressure, wave heights and other factors, and use that information to predict tsunamis, Modha writes.
In another example, “imagine traffic lights that can integrate sights, sounds and smells and flag unsafe intersections before disaster happens,” Modha said in a company release.
IBM unveiled two prototypes of the new chips Aug. 18. Called neurosynaptic computing chips, the processors are designed to work in a way similar to neurons and synapses in a biological brain, the company said.
The cores of each 45-nanometer chip seek to emulate synapses with integrated memory, neurons with computational components and axons, which conduct electrical impulses, the company said. Each chip contains 256 neurons. One chip has 262,144 programmable synapses; the other has 65,536 learning synapses.
The goal is to not only emulate the brain but to take up about the same amount of space and use as little power as possible, the company said. IBM said the project’s long-term goal is a chip system with 10 billion neurons and 100 trillion synapses that takes up less than two liters of space (about the same as the human brain) and uses one kilowatt of power (more than the brain’s 20 watts, but still efficient for that much processing).
The company, which drew attention to pioneering work in natural language processing earlier this year when the company’s Watson computer won big on “Jeopardy!”, says it has completed phase 0 and 1 of the project, and DARPA has put up $21 million in funding for phase 2.
The SyNAPSE team includes IBM researchers and scientists from Columbia University, Cornell University, the University of California, Merced, and the University of Wisconsin-Madison.
Kevin McCaney is a former editor of Defense Systems and GCN.