Abstract

The development of optical computers of any type is based on the notion that semiconductor technology imposes limitations in the performance of current computers which prevent them from being effectively used for the solution of a class of interesting computational problems. If optics is used instead, these limitations will be lifted and we will therefore be able to now solve these interesting problems. Global connectivity is perhaps the most distinctive feature of optics vis-a-vis semiconductor technology, and the development of optical neural computers can be viewed as an attempt to exploit this feature. In a neural network each elementary computational unit, the neuron, directly communicates to thousands of others, while in electronic computers each gate is typically connected to only two or three gates. With optics it is feasible to realize the dense connectivity that is evident in neural networks. This provides the impetus for examining neural network models of computation to get ideas about how to build optical computers whose performance is clearly better than their electronic counterparts.

© 1987 Optical Society of America