New analog chip acts just like a nerve cell.

It may look like an ordinary computer chip, but the "silicon neuron" is nothing like those that run even the most avantgarde machines.

Two computational neuroscientists have made a new chip that brings researchers a step closer to creating a machine that truly mimics the nervous system. Because the device uses a new type of analog technology, its transistors act like nerve-cell membranes, says Misha Mahowald, a graduate student at the California Institute of Technology in Pasadena. She and Rodney J. Douglas of the University of Oxford in England designed the chip so that its circuits replicate the electrical currents that affect cell membranes and cause nerve cells to fire. They describe the new device in the Dec. 19/26 Nature.

Mahowald and Douglas are now working out the technical details for putting many neurons onto one chip and inter-connecting a series of chips to create "megacircuits" that can model brain function, Douglas says.

In addition, the researchers plan to link the silicon neurons to existing analog chips that mimic sensory nerve cells in the ear (SN: 1/6/90, p.7) or in the retina, Mahowald says.

Scientists currently simulate small groups of neurons by using neural networks, but the computer programs take a long time to run, even on very powerful machines. "[Silicon] neurons could solve these kinds of problems very easily," says Douglas. Even with thousands hooked together, silicon neurons should work at least as fast as the nerve cells they emulate and with little power consumption, he says.

Douglas and Mahowald hope to make an easy-to-use system that neurobiologists can put to work testing ideas about neural circuity. "We'd like to make a cheap, fast tool for the 'neuroscientist in the street,'" Douglas says.

Because even a network of silicon neurons operates in real time and requires little power, Douglas thinks roboticists may find these neurons more useful than artificial intelligence computers for controlling robots that must maneuver on their own.

Computer chips typically use digital signals to operate; thus, a signal is either on or off, but never partly on. Analog devices need less power to run, and they operate very rapidly. Their signals exhibit a continuum of responses, so they lack the precision of their digital counterparts. "But a very important characteristic of the brain," notes Douglas, "is that it doesn't do things very precisely."

For years, Douglas has studied the electrical properties of brain cells involved in vision. His research provided the details needed to design the circuitry in the silicon neuron. "It all rests on the fact that transistors in this subthreshold regime have physical characteristics that are similar to the physical characteristics of a real membrane," Douglas says.

For example, in a nerve cell, the membrane serves a function similar to the gate in a transistor. To conduct a current, a transistor lets electrons through its gate. Likewise, a cell membrane's current results when channels in the membrane open briefly to let ions move into and out of the cell. The types of membrane channels and the rate at which they open result in the different impulses that let nerve cells communicate with each other.

Douglas and Mahowald designed the silicon neuron so that each of its circuits represents a different kind of membrane channel. With these circuits, they can recreate the various types of currents that make a nerve fire.

According to Mahowald, several neuro-biologists have had trouble distinguishing between the patterns generated on oscilloscopes by real neurons and their silicon counterparts.

"It comes closer than anything I could ever imagine," says Oxford neurobiologist Kevan A.C. Martin. "It does in an electronic way what a biological neuron is doing. That's tremendously exciting."