Recognizing the speed limitations of simulating neural networks on single-processor digital computers, this paper
proposes a possible analog circuit approach to the implementation of multiple processing elements for various neural networks. The circuit is interfaced to a standard digital computer for ease of initialization and control, and each artificial neuron features programmable weights and a propagation delay time of less than 500 nanoseconds. The neuron circuit is based on the widely-known McCulloch-Pitts model. The circuit adaptation required for implementation in a Kohonen-type network is also presented. The circuit configurations for the Step and Sigmoid transfer functions are also discussed.