An intelligent optical sensor realization

Abstract

An analog VLSI realization of a multi-layer optically coupled neural network is described in this paper. The network has an optical array constructed with 10/spl times/10 photosensitive elements (photoreceptors) as the input devices. The artificial neuron has been implemented by a 2-stage comparator comprised of a differential amplifier and an output inverter to obtain a signum (hard-limiting) nonlinear function. The weighting coefficient values are determined by the size of the transistors used. The network has partial connections between the input layer and the hidden layer, and is fully connected between the hidden layer and output layer. There are 100 input nodes, 16 hidden neurons and 5 output neurons in the network. The physical realization contains approximately 1630 devices that have been fabricated using 3 /spl mu/ single-polysilicon, double-metal, p-well full custom CMOS technology.< >