Brains Can Teach Chips A Thing Or Two

Apparently, these are the kinds of conversations a person has when hanging around brain surgeons.

Kerry Bernstein, a senior technologist in IBM's high-speed server group, was chatting with his friend, neurosurgeon Marty Bednar, about their work. They struck a deal: Bednar, who works for the drugmaker Pfizer, would teach Bernstein how the brain works, and Bernstein would explain what makes microchips tick.

"What we found is that it's not just an analogy, but in many cases, it's the same fundamental physics that govern the operation of the neuron and the transistor," Bernstein says. That conversation led to Bernstein's developing a seminar for an annual neuroscience symposium, plunging him into the world of science and linking the fields of the brain and chip, a connection that could help shape the future of computing.

It's already having an impact on IBM computer development. For example, the brain runs at low speed and power compared with microchips but makes up for it with massive parallelism--a neuron can fan out to connect with thousands of other neurons, versus at best 10 other circuits with which a typical chip talks. The idea of massive parallelism is central to IBM's $100 million effort to build the world's fastest computer, nicknamed Blue Gene.

Researchers like Bernstein are desperate. Today's chip technology will last only a few decades before it's physically impossible to get more computing power out of it. They need radically new ideas. That's why Bernstein spends about a quarter of his time tracking experiments such as the one last summer in Germany in which scientists transplanted snail nerve cells onto an electronic circuit. That kind of "wet" chip, which combines biochemical and electronic signals, is probably many years away. But Bernstein predicts that once these theories take hold, the effect will be dramatic. "When things change," he says, "they're going to change very quickly."