DESIGN OF NEURAL NETWORKS BASED ON WAVE-PARALLEL COMPUTING TECHNIQUE

A Wave-Parallel Computing (WPC) technique is proposed to address the i nterconnection problem in massively-parallel analog hardware required for implementing neural networks. The fundamental concepts are the fre quency multiplexing of signals on a single line, and their wave-parall el processing without decomposition. This paper discusses the realizat ion of a Hopfield-type massively-connected neural network, and shows t hat the WPC-based network exhibits much lower topological complexity c ompared with the original network. We also investigate the possible im plementation of WPC using the present MOS technology, and discuss the performance evaluation in terms of the degree of multiplexing and the processing speed.