OPTICAL HOLOGRAPHIC NEURAL NETWORK FOR TARGET RECOGNITION APPLICATIONS

IN ORDER TO MAKE HIGHLY-PARALLEL LARGE-SCALE OPTICAL NEURAL NETWORKS OR ASSOCIATIVE MEMORY FOR TARGET RECOGNITION APPLICATIONS, PHYSICAL OPTICS CORPORATION (POC) PROPOSES A PRACTICAL HIGHLY-PARALLEL FULLY-INTERCONNECTED OPTICAL HOLOGRAPHIC NEURAL NETWORK ARCHITECTURE BASED ON POC'S PROPRIETARY MATERIALS. THE PROPOSED ARCHITECTURE PROMISES A PRACTICAL ENGINEERING REALIZATION WITH POTENTIAL TO STORE UP TO 10(12) WEIGHTED CONNECTIONS FOR 10(6) NEURONS. THE MATERIALS ARE BASED ON POC'S DICHROMATED GELATIN POLYMER GRAFT (DCGPG), A STATIC MATERIAL AND REAL-TIME PHOTOPOLYMER, A DYNAMIC MATERIAL. THESE HAVE THE FOLLOWING UNIQUE CHARACTERISTICS, NOT ACHIEVABLE SIMULTANEOUSLY BY ANY OTHER STATE-OF-THE-ART TECHNOLOGIES (I.E., REFRACTIVE CRYSTAL AND BINARY SPATIAL LIGHT MODULATORS): 1) LARGE SIZE (>10X10 QUARE INCHES) AND LOW COST, 2) HIGH SENSITIVITY AND HIGH RESOLUTION, 3) HIGH STORAGE CAPACITY, 4) REAL-TIME READ/WRITE/ ERASE CAPABILITY (FOR REAL-TIME PHOTOPOLYMER). A PHOTOTYPE NEURAL NETWORK WITH A MODERATE NUMBER OF NEURONS (?1000) FOR TARGET RECOGNITION APPLICATIONS WILL BE BUILT AND DEMONSTRATED IN PHASE I OF THIS PROGRAM. A DETAILED ENGINEERING DESIGN OF A ROBUST ADAPTABLE NONLINEAR IMAGING NEURAL NETWORK WILL BE PRESENTED BASED ON THE EXPERIENCE OF FABRICATING THE PROTOTYPE NEURAL NETWORK.