A self-learning neural-network LSI using neuron MOSFETs

Abstract

A functional MOS transistor called a neuron MOSFET (vMOS) which simulates the function of biological neurons is discussed. A method of constructing neural network LSIs that have a self-learning capability using the neuron MOSFET is given. The key is the implementation of a synaptic connection which changes its weight according to various learning algorithms. In addition, the synapse must be free from standby power dissipation and be as small as possible.<>