IBM Reveals Chip That Acts Like Human Brain

IBM is testing a computer chip modeled on human neural processing, as it tries to create silicon better suited to real-world, multi-source data processing at low power.
100 Years Of IBM: 25 Historic Milestones
Slideshow: 100 Years Of IBM: 25 Historic Milestones
(click image for larger view and for slideshow)
Despite having already outperformed humans at Jeopardy using brute force computing, IBM researchers see something worth emulating in the human brain.

Hoping to make computer chips more adept at tasks at which the human brain excels, IBM researchers on Thursday introduced an experimental computer chip architecture modeled on human neural systems for perception and cognition.

As if to reassure the technophobic that this research doesn't involve organ theft, IBM notes that the chips contain no biological elements.

These neurosynaptic computing chips use algorithms and electronics to emulate the behavior of neurons and synapses in the human mind. In so doing, IBM's scientists anticipate that so-called cognitive computers will be able to more easily handle tasks like recognizing faces and objects while rivaling the brain's compactness and low power usage.

According to the company, researchers have already demonstrated that the experimental chip design can perform tasks like navigation, machine vision, pattern recognition, associative memory, and classification. The company also expects that cognitive chips will be useful for computer security and complex processes like climate modeling.

IBM has fabricated two prototype chips, which are currently being tested as part of a multi-year cognitive computing initiative. IBM has been awarded $21 million in funding from the Defense Advanced Research Projects Agency (DARPA) for the second phase of what the computer company is calling its Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project.

Dharmendra Modha, project leader for IBM Research, characterizes the SyNAPSE initiative as an effort to move beyond the von Neumann paradigm that has dominated computer architecture for the past 50 years. Von Neumann's approach involved storing program instructions and data in random-access memory, an improvement over the prior method of altering electronic circuits.

"Future applications of computing will increasingly demand functionality that is not efficiently delivered by the traditional architecture," said Modha in a statement. "These chips are another significant step in the evolution of computers from calculators to learning systems, signaling the beginning of a new generation of computers and their applications in business, science and government."

The two chip prototypes were fabricated using 45-nm SOI-CMOS technology and contain 256 computational "neurons." One "neurosynaptic" core has 262,144 programmable synapses while the other has 65,536 learning synapses. These chips will not be programmed in the traditional sense. Rather, IBM's scientists say they will learn by experience, remembering data, correlating it, and forming hypotheses in a manner similar to the way the human brain operates.

IBM says its long-term goal is to build a chip that has ten billion neurons and one hundred trillion synapses, requires one kilowatt of power, and occupies less than two liters of volume. According to the company, the human brain uses less energy than a 25-watt light bulb while occupying less space than a two-liter bottle of soda.

InformationWeek Analytics is conducting a survey on the current state of server technology within the enterprise. Take the survey now. Survey ends Aug. 19.