Implementation of analog shunting neural networks for optoelectronic detection and processing

A fully parallel and completely analog implementation of a shunting neural network is described. Such a network provides an artificial version of the biological cell connection structure and function commonly found in the more peripheral layers of a sensory organ, such as in the retina of an eye. This implementation is easily integrated with photodetector arrays and can be used to achieve many of the powerful signal processing properties normally associated with biological sensory systems. By using the circuit characteristics of a field-effect transistor biased into the voltage-variable conductance range, the implementation achieves an extremely high packing density with only one transistor per unilateral interconnection path. Dynamic range compression, edge enhancement, and preferential directional sensitivity are discussed in relation to the implementation constraints.<
>