Novel design of an M-valued digital system using M-zero neural network

An M-zero neural network (MZNN) is derived from the study of the asymptotic convergence properties of a system of N coupled nonlinear differential equations. The network is a one-layer, feedback neural system consisting of N artificial neurons. Each neuron has M odd zeros in its input-output response function. These neurons can be realized by simple Zener-diode circuits. The MZNN will converge any N-b analog input to an N-b M-ary (or M-valued) digital output, and it can memorize this output even when the input is removed. Because of these properties, it can be used in the design of a fast, M-valued computing system. The origin of the MZNN system and its application to a practical design problem-the design of a four-valued digital multiplier using MZNNs and analog arithmetic units-are studied