Hardware-learning neural network LSI using a highly-functional transistor simulating neuron actions

This paper describes the architecture and the organization of a hardware-learning neural network LSI, in which a newly developed "brain-cell-like" transistor called neuron MOSFET (neuMOS or νMOS) is utilized not only in a neuron cell but also in a synapse cell. In order to implement learning capability on a chip, a new hardware-oriented backpropagation learning algorithm has been developed. The actions for self-learning based on this algorithm are also carried out by νMOS logic circuits.