Smart Neural Network Array for Pattern Recognition

It is proposed to develop a 64 x 64 smart pixel array that implements the pulse-coupled neural network (PCNN) when it is integrated in a compact optical system. The PCNN has been shown to be a powerful image analysis and recognition tool. Each pixel will contain an adaptive threshold circuit with adjustable dynamics, a liquid crystal modulator that outputs neural pulses and implements the linking field, and a photodetector that senses the optically computed internal activity. A scanner provides a means of reading out the outputs of the entire array in succession and displaying them on an oscilloscope, for alignment and adjustment purposes. In Phase 1, the PCNN smart pixel will be designed and simulated. Test pixel layouts will be submitted for fabrication by a silicon foundry. The layout of the complete smart pixel array will be performed. In Phase 2, the silicon array will be fabricated and liquid crystal modulators will be added. Electrical tests will be conducted, and the device will be integrated in a compact optoelectronic system. The performance of the system will be integrated with an optical correlator. There are numerous important commercial applications for cost-effective , highly integrated vision and target recognition systems, including robotic vision, vehicle navigation, security systems, and character recognition. The PCNN is very attractive because it can be implemented as a compact optoelectronic system that lends itself well to volume production.