Chaotic route to computation.

Irregular fluctuations in light intensity have long plagued the operation of a wide variety of lasers. During the past decade, researchers have learned how to avoid, control, or exploit such chaotic outbursts by adjusting laser parameters, such as power input (SN: 2/22/92, p. 119). In effect, they can turn a dial to tune a laser's signal so that it flutters randomly, falls into a regular pattern, or remains steady.

Now, researchers have demonstrated in a simulation how a network of chaotic lasers could operate as a computer, adding or multiplying numbers and processing information. William L. Ditto of the Georgia Institute of Technology in Atlanta and Sudeshna Sinha of the Institute of Mathematical Sciences in Madras, India, describe their scheme in the Sept. 7 Physical Review Letters.

The researchers use certain patterns of laser signals to encode different decimal digits, one digit per laser in the network. To perform an operation such as addition, they allow some of the fluctuating light to leak from one laser to another, setting off an avalanche of activity as the linked lasers react to multiple inputs from their neighbors. Ditto and Sinha show that after the system has settled down, the numerical answer is encoded in the network's final behavior pattern.

"There are lots of different ways to do the encoding," Ditto says. For example, "we can change the network connectivity or the chaotic system parameters."

Moreover, the chaotic elements don't have to be lasers. Any network of systems showing chaotic behavior--including electric circuits or nerve cells--will have comparable computational capabilities. "It's an entirely new computing paradigm," Ditto contends. Instead of precisely specifying every tiny step, a programmer simply sets the initial conditions and lets the system work out an answer in its own way.

Ditto and his coworkers plan to test their idea using chaotic ammonia lasers and, ultimately, a network of nerve cells attached to silicon. "This is a glimpse of how we can make common dynamic systems work for us in a way that's more like how we think the brain does computation," says Ditto.