loren@tristan.llnl.gov (Loren Petrich) (08/31/90)
This had appeared in a recent NN digest posting: ******************************************************************************* Subject: neural stuff From: ellen_warneke%a1@hp1900.desk.hp.com Date: 23 Aug 90 09:37:00 -0800 [[ Editor's Note: This was forwarded by my friendly local librarian. Does anyone have any more information? -PM ]] DR152.02: The MITSUBISHI ELECTRIC Central Laboratory has developed an optical neuro chip that can recognize characters more accurately than conventional devices. The chip, which is designed to recognize characters, "A," "E," and "J," integrates 32 light emitting devices, 32 photo detectors, and 1,024 spatial modulation devices on an 8mm-x-8mm GaAs (gallium arsenide) chip. Having a quantum-well structure, the chip has 32 neurons. Spatial modulation devices function to connect neurons, and the three character patterns are imbedded on the devices. Since the lab already developed an optical chip for identifying three characters in November 1988, the neuro chip is a second prototype chip. (8/21/90: Nikkei Sangyo [p.5]) ******************************************************************************* Does anyone have any further information on how this chip works? And the characters "A", "E", and "J" are described as having been encoded into the chip. I wonder how this was accomplished, and I wonder how well this chip performs on examples of varying quality that it was not trained on. One of my colleagues has been contemplating an opto-electronic NN design, and he might be interested to learn of a successful implementation. $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ Loren Petrich, the Master Blaster: loren@sunlight.llnl.gov Since this nodename is not widely known, you may have to try: loren%sunlight.llnl.gov@star.stanford.edu