Polymer dendrites: making tiny connections.

As an information processor, the human brain derives its power not from the brute force of fast transmission, but from its ability to process in parallel. It owes this skill to its vast array of interconnections. Brain cells sprout branches, or dendrites, that enables them to communicate with many neighboring brain cells, creating complex trees of associations.

The effort to make computers that can "thinks," or at least simulate human thought, remains hamstrung by flat silicon chips that process information in sequence rather than in parallel. But that may begin to change.

In the Dec. 24 Science, Michael J. Sailor, a chemist at the University of California, San Diego, and his colleagues report a new method for connecting minute, hairlike wires that could someday lead to computer chips capable of storing information in three dimensions. Such interconnections could, in theory, make possible computers with quick, parallel processing.

Specically, the chemists describe a way to cause electrically conducting polymers to form dendritic branches that can slectively link with other branches in a solution. Using the polymer poly (3-methylthiophene), which only grows when conducting current, they caused particular polymer dendrites to link with each other electrically by alternating a current between them.

"The polymer dendrites alternate between conducting and nonconducting states until they come into contact with each other," the team says. "When an actively polymerizing strand electrically contacts a nonconductive strand, the nonconductive strand switches into its conductive state in the region close to the connection."

This technique, while in an early phase of research, could potentially enable scientists to build information webs with "nodes," or processors, connecting to many other nodes three-dimensionally. "There are really no good tools to construct 00croelectronic devices in three dimensions right now," says Sailor. "But we're moving in that direction."

"We've been able to show that we can take an arbitrary number of wires and hook them up in an arbitrary number of ways," Sailor adds. "The wires don't have 0o connect one to one. One wire can connect to five others, which is essential to parallel processors."

"Your brain can recognize a dollar bill in a split second, a task that takes a serial computer a fairly long time," says Sailor, because it processes information in parallel. "When you pack a logic system into three dimensions, a large amount of information fits into a small volume and it works much faster."

But before Sailor can think about such ideas as neural networks, he must overcome some basic hurdles. To be useful for computing, the wires must conduct current only in one direction, not two. He also wants to simplify and automate this procedure, shrinking it down, shortening the distance between connections, and building more complex arrays than the simple test nodes he has fabricated. Ideally, he wants to grow the wires, make the connections, and insulate them "in one pot," without having to move the wires between solutions.

"The dream is to make a thinking machines analogous to a living brain," Sailor says. "This procedure is one of many tools that might be needed to make one."