Analog circuits for electronic neural network

Abstract

A simple neuron for an artificial neural network (ANN) using four analog MOS transistors is described. The synaptic weight (acting as a resistor) is replaced by four MOSFETs operating in the triode region as a linear resistor. The weight of each input can be controlled by a clock signal in order to adapt for a feedforward network, such as a backpropagation chip.< >