Intel Corp. has developed a prototype microchip that is stunningly quick at recognizing handwriting, identifying military targets and performing other tasks that are difficult or impossible for conventional chips.
The new NI 1000 chip, developed with financial support from a Pentagon research agency, is one of an unusual breed called "neural networks."
These chips and their associated software work more like the human brain than the Intel microprocessors used in millions of personal computers.
Artificial neural networks are modeled on the human brain's biological network of hundreds of billions of special cells, or neurons, that transmit information back and forth, making possible behavior like learning and recognition.
Because they can recognize visual or sound patterns at high speed, neural nets are being applied to tricky tasks such as distinguishing voices, fingerprints and ZIP codes.
The Pentagon's Defense Advanced Research Projects Agency wants such chips to, for example, identify submarines and other targets.
The first samples of the NI 1000 went to Sunnyvale-based Lockheed Missiles & Space, which works on anti-missile devices, among other defense products.
Intel's new chip is expected to be particularly useful in handwriting recognition, a rapidly growing market. The NI 1000 is said to be a hundred times faster than other technologies used for that purpose.
Nestor Inc., a Rhode Island-based company, developed a version of its handwriting-recognition software for use with the NI 1000. While a scanner based on a fast version of Intel's 486 microchip can recognize about 30 handwritten characters per second, the NI 1000 is expected to recognize 5,000 to 10,000.
"We don't think anything else is even close," said Michael Glier, a Nestor vice president.
Although the chip requires too much electrical current for use in small computers, Intel is looking ahead to improved versions that could be used in hand-held machines activated with pen devices.
Intel has lagged other companies in selling chips for such machines, Mr. Glier said, but the company could eventually outpace the competition with neural-net chips that work with its popular microprocessors.