By Barry Levine / Top Tech News. Updated August 18, 2011.
One of computing's futures has arrived. On Thursday, IBM announced that its researchers have developed experimental computer chips that can "emulate the brain's abilities for perception, action and cognition."
Dubbed cognitive chips, the new processors are the first "neurosynaptic computing chips" from IBM, and the company said the chips emulate the way neurons and synapses work in living organisms.
Most important, they won't be programmed, as are regular processors. Rather, the company said, their job is "to learn through experiences, find correlations, create hypotheses, and remember -- and learn from -- the outcomes."
Two prototype chips have been built and are in testing.
Moving Beyond von Neumann
The chips are part of a major, multiyear cognitive computing initiative at the company and at selected universities, funded in part by $21 million from the Defense Advanced Research Projects Agency (DARPA). The project is called Systems of Neuromorphic Adaptive Plastic Scalable Electronics, or, appropriately, SyNAPSE.
The goal is nothing less than a system that takes in data from multiple sensors and then "dynamically rewires itself" as it responds to changes in the environment, all within a compact size and with a low power requirement.
Dharmendra Modha, IBM Research project leader, said in a statement that the initiative intends to "move beyond the von Neumann paradigm that has been ruling computer architecture for more than half a century." He added that the new chips are "another significant step in the evolution of computers from calculators to learning systems."
The von Neumann paradigm, named for computer pioneer and mathematician John von Neumann, refers to a design architecture in which chips store an instruction set and a data set. In processing, the chip fetches the instructions to work with the data.
The new IBM chips do not have actual biological components, but are designed as "neurosynaptic cores" with integrated memory designed to emulate synapses, computation to resemble neurons, and communication that is modeled on axons. The overall processing is intended to mimic the distributed and parallel processing in the brain, driven by events.
Radical New Form of Processing
Both of the working prototypes are 45-nanometer SOI-CMOS and have 256 "neurons." One prototype has 262,144 "programmable synapses," and the other 65,536 "learning synapses." The company said the prototypes have already been used to demonstrate simple apps, such as navigation, machine vision, pattern recognition, associative memory, and classification.
The long-range goal: a chip system with 10 billion neurons and 100 trillion synapses, taking up less than two liters of volume and drawing only one kilowatt.
As an example of how this radical new form of processing might be applied, IBM envisions that a cognitive computing system could monitor the world's water with a network of sensors and actuators. Sensor-derived data would include temperature, pressure, wave height, acoustics, and ocean tide, and tsunami warnings would be issued by the system, after it learned what to look for.
Other applications could include an instrumental glove used by a grocer, where the cognitive processing could take data from sights, smells, texture and temperature in a store, and then determine which produce had gone bad or was contaminated.
Al Hilwa, program director for application development at industry research firm IDC, said the new cognitive chips are in keeping with IBM's major effort to make computing more like thinking, as the company demonstrated when its Watson computer bowled down competitors on the Jeopardy game show earlier this year.
But, Hilwa cautioned, don't expect an instrumental glove with learning powers anytime soon. Even after the systems are fully developed, he said, they first will be deployed into very high-performance situations, and, some years later, only then made available for the general business and consumer markets.