VIEWPOINTS : THE GURU OF SILICON CHIPS : Carver Mead Is Shaping an Entire Industry With His Approach to the Gray Matter of Computer Brains

Few people have either the knack or the opportunity to completely redefine a medium, inspire an industry and transform pop culture all at the same time. Guttenberg managed to do it with paper, D. W. Griffith did it with celluloid and William Paley used television.

Carver Mead is doing it with silicon.

Scratch any chip in any computer anywhere in the world and you’ll find silicon--the element that Mead has made his most eloquent medium. Silicon is the raw stuff of computation; etched with the right patterns, it can make bits and bytes sing or dance or turn themselves into pretty pictures. So Mead treats silicon much as the Impressionists treated canvas and Griffith viewed film--as a chance to capture and portray reality in new and provocative ways. The result may not be art, but it’s certainly avant-garde. As is Mead.

From the breadth of his conceptual range down to the tips of his Fu Man Chu mustache, Mead is precisely the sort of high-tech guru you would expect CalTech to fabricate. At 55, he vaguely resembles a silicon Salvador Dali, albeit a little less surreal. He is a clever visionary who also knows how to implement.

“You’ve got to listen to the silicon,” Mead rhapsodizes, “because it’s always trying to tell you what it can do.”

By any measure, Mead has been a good listener. As a consultant to Intel in the early days, he played a crucial role in the development of microminiaturization technology--that is, the technology of making computers smaller. As a professor at CalTech, Mead has been a pioneer in crafting a design approach to the challenge of what are known as Very Large Scale Integrated circuits--considering, for example, how you lay out a million transistors on a dime’s worth of silicon. (Think of it as putting a perfect map of the United States on a surface the size of your thumbnail).

How can you get computer chips to cope with their own growing complexity and design themselves? Mead’s answers to those questions are rewriting the rules of chip design from Silicon Valley to Osaka.

Of course, having helped create one of the most brilliant design methodologies for silicon, Mead promptly proceeded to turn his back on it. Inspired by a course he taught with Nobel laureate physicist Richard Feynman and MacArthur Fellow John Hopfield in the early 1980s on “The Physics of Computation,” Mead had the sort of epiphany that radicalized his view of his medium. He could no longer look at silicon the same way. His new perspective was as profoundly different as photography is from painting.

Instead of using silicon as a medium to design computers, Mead would use silicon as a medium to design nervous systems--the networks of neurons, axons and synapses that shape sight, hearing and touch. Silicon would come to its senses--or, more precisely, the senses would come to silicon.

Traditional computer chips aren’t up to the challenge of replicating the senses--which, if you think about it, only makes sense because we don’t see and hear like computers do. By contrast, Mead’s chips are analogs of the real thing. He has crafted synthetic neurons into a silicon retina that can see not like a movie camera but like an eye; he’s devised a silicon cochlea that hears not like a tape recorder but like the ear--in fact, it is now being considered for cochlea implants for the profoundly deaf.

Other neural network chips emulate memory. What’s more, these chips don’t need to be programmed--they can learn from experience. “We can already do some pretty amazing things,” Mead says. “This gives you a way to deal with the natural world.” Even today, some of these neural network chips can do things that would stump a Cray supercomputer, even if digital computers will always remain the high-speed calculator of choice.

This is nothing less than a rejection of the belief that fancy computer software offers the best pathway to so-called artificial intelligence. Indeed, the future of artificial intelligence may not be in software that emulates how we think people think but in the silicon emulations of neurons, synapses and axons that are the hardware of thought. In other words, instead of artificially creating thought from the top down, Mead wants to build it from the bottom up.

Just as movies and television have become the mirrors of culture and society, this emerging generation of neural netware may become the new mirrors of the senses and thought. “It’s sure going to open up our minds,” says Mead. “Our models of computations are going to be very, very different.”

What will personal computers with neural systems that can see and hear and remember be like? Who needs a computer with a sense of smell? (Actually, the Federal Aviation Administration is exploring the use of a “neural nose” at airports for bomb detection). What does the future of computing look like when the computers are based on biological models of thought instead of computer models? These aren’t speculative questions--these are the questions that this new design metaphor creates.

When the microprocessor was invented at Intel, says Mead, “We all predicted the emergence of personal computers. What we didn’t foresee was that you’d find microprocessors in gas pumps and microwave ovens. None of us foresaw this incredible proliferation.”

Mead has begun to evangelize his new brand of synthetic neurobiology and recently published “Analog VLSI and Neural Systems,” a book not destined to be optioned by literary agent Swifty Lazar but certain to be the cognitive cookbook for the emerging generation of neural net-masters. With Federico Faggin, one of the co-creators of the first microprocessor, Mead has started up Synaptics, one of the first neural network chip companies.

Since mass-producing these chips will be a snap, it’s easy to predict that we’ll start seeing them pouring out of Synaptics and Silicon Valley for industrial uses within four years and out of Japan in consumer electronics within a decade.

My best guess is that they will redefine electronics as surely as television redefined radio. The reality of these times is that silicon neural networks may be as powerful a medium as television networks--but, unlike the television networks, neural networks are destined to get smarter.