CASCADABLE OPTICAL NEURON ON A CHIP

OPTICAL NEURAL NETWORKS HAVE DEVELOPED TO THE POINT WHERE VIABLE ARCHITECTURES SUCH AS OPTICAL ASSOCIATIVE PROCESSORS FOR VISUAL PERCEPTION HAVE BEEN DESCRIBED IN DETAIL. HUNDREDS OF PAPERS ARE PUBLISHED AND SEVERAL TOPICAL CONFERENCES ARE HELD ANNUALLY ON THE SUBJECT. HOWEVER, FEW DEMONSTRATIONS HAVE BEEN REPORTED OFACTUAL OPERATING SYSTEMS DUE TO THE UNAVAILABILITY OF AN "IDEAL" OPTICAL NEURON HAVING OPTIC GAINCASCADABILITY, HIGH ON/OFF CONTRAST, APPROPRIATE TWO-DIMENSIONAL GEOMETRY AND NXN PIXEL DENSITIES, ADEQUATE SWITCHING SPEEDS, AND TOLERABLE SWITCHING POWER REQUIREMENTS. CLEARLY, ARCHITECTURE DESIGN HAS OUTPACED DEVICE DEVELOPMENT. IN THIS PROPOSAL WE DESCRIBE A REVOLUTIONARY NEWDEVICE WHICH WILL MEET THESE REQUIREMENTS. THE DEVICE WHICH IS APTLY CALLED AN "OPTICAL NEURON" CONSISTS OF A VERTICAL-CAVITY SURFACE-EMITTING LASER MONOLITHICALLY INTEGRATED WITH A PHOTOTRANSISTOR. THESE MONOLITHICALLY INTEGRATED DEVICES WILL BE CASCADABLE AND WILL HAVE HIGH OPTICAL GAIN (THUS THE ABILITY TO BE MASSIVELY INTERCONNECTED). THEY WILL BE FABRICATED USING SIMPLE PLANAR VLSI TECHNOLOGY AND EMIT LIGHT PERPENDICULAR TO THE SURFACE. CONSEQUENTLY THEY ARE IDEALLY SUITED TO OPTICAL FOURIER TRANSFORMATIONS USING ONLY LENSES. THE PHASE I PROJECT WILL (1) DEMONSTRATE AN OPTICAL NEURON ON ACHIP, (2) DESIGN A OPTIMIZED SECOND GENERATION DEVICE, (3) IDENTIFY A NEURAL NETWORK ARCHITECTURE BASED ON THE OPTICAL NEURONS, AND (4) OUTLINE PHASE-II. PHASE II SHOULD INCLUDE THE DEVELOPMENT OF, FOR EXAMPLE, A NEURAL OPTICAL COMPUTER OR A NEURAL NETWORK FOR PERFORMING PATTERN RECOGNITION.