Since IBM CEO Sam Palmisano visited a team of scientists at IBM Research – Almaden a few weeks ago, he has been joking that they have designed a new computer chip that’s “as large as a worm’s brain.” It’s a good quip. Why would highly-talented researchers spend their time creating a chip like that? It’s also quite an understatement. True, the chip isn’t very smart, in itself. But it signals what could be the beginning of a major new computing architecture that compliments today’s computers. When Palmisano tells the story, he makes sure his audience knows how proud he is of IBM’s researchers.
The chip, a product of IBM’s three-year-old SyNAPSE project, could become a building block for a new generation of computers designed to emulate the animal brain’s abilities for sensing and cognition–all the while consuming many orders of magnitude less power and space than today’s computers. “We believe we have reproduced the core circuit of the brain in silicon,” says Dharmendra Modha, program lead for IBM Research’s cognitive computing department at the Almaden lab. “All mammal brains are built on the same blueprint. We believe that we have found the core design that encapsulates the key architectural principles of the brain.”
Until now, most electronic computers have been based on the thinking of the Hungarian-American scientist John von Neumann. In the mid-1940s, he described a computer architecture based on mathematics that was well-suited to perform speedy calculation and collation of data. But the architecture separates data processing from data storage–or memory. That requires integrated circuits with fast clock speeds, which consume a lot of energy.
The new IBM architecture is quite different. In the brain, memory and processing are packaged together, and, as a result, neurons don’t need to fire often–and very little energy is expended. Ditto with this new computing architecture, which is embodied in neurosynaptic integrated circuits. The idea is to use a large number of these circuits to create a brain-like network of processing power on a chip. The networks can be scaled up massively to add more capabilities. Also, computers made from these components are expected to be able to learn from experiences, find correlations and create hypotheses.
The architecture is particularly well suited for Smarter Planet-type engagements. Increasingly, data about everything from weather to transportation systems to activity on city streets is being gathered from vast sensor networks. If the scientists’ bet plays out, engineers will be able to use neurosynaptic components to build tiny inexpensive sensors. In addition, they’ll be able to build computers using the same components to analyze large amounts of data flowing from different sources. “We can mimic the architecture of the brain to dramatically reduce the cost of gathering information in real-time, sensor-rich, ambiguous, changing environments,” says Modha.
The journal Communications of the ACM published in its August issue an article by Modha and his research colleagues: Cognitive Computing: Unite neuroscience, supercomputing, and nanotechnology to discover, demonstrate, and deliver the brain’s core algorithms.
IBM is conducting its research with the help of a $21 million grant from the US Defense Advanced Research Projects Agency. Collaborators on the project team include scientists from four IBM Research labs, the company’s chip fab in Fishkill, N.Y, two national laboratories, and four universities, including Columbia, Cornell, the University of California, Merced and the University of Wisconsin, Madison.
IBM’s long-term goal is to build a chip system with 10 billion neurons and a hundred trillion synapses, which will consume less than one kilowatt of power and occupy less than two liters of volume–about the size of a shoe box. If IBM’s scientists are successful, they’ll give new meaning to the old phrase, as the worm turns.