Brain-Inspired Computing with Spintronics Devices

Abstract

Neural networks and artificial intelligence utilizing artificial neurons and synapses are attracting much attention. Spintronic devices are considered to be suitable for mimicking artificial synapses and artificial neurons because of nonvolatility of information and rich nonlinearity of spin dynamics. We focused on the nonlinearity of spin dynamics and formed a virtual artificial neural network by using the time multiplexing method. By using reservoir computing for learning rules, we succeeded in speech recognition with a high recognition rate of 99.6%. These results pave the way for hardware implementation of artificial intelligence.