A Production of the Magnetic Neuron Network
T. Haga, Hideki Tsujimoto, K. Shirae · IEEE Translation Journal on Magnetics in Japan · 1992
An artificial neural network circuit using a magnetic device has been developed. The network consists of magnetic neurons and magnetic synapses. A magnetic neuron starts oscillating when the sum if its input currents exceed the threshold level, and its frequency varies according to the input level. Magnetic synapses can be easily controlled through a bias current. The network fabricated in this work has a three-layer structure, with two input gates and a single output. By adjusting the synapses, 16 kinds of logic functions can be realized. It is, however, very difficult to set correctly all the weights of the synapses in a large-scale network, so we adopted a learning technique. Only the input sets and the ideal output are given; the weights are corrected according to a learning algorithm that eventually attains ideal operation. That is, the network learns.