Spin-wave coupled spin torque oscillators for artificial neural network

Abstract

The spin-wave coupled spin-torque-oscillator is studied for a processing unit of an artificial neural network by numerically solving the Landau-Lifshitz-Gilbert equation incorporating the Slonczewski spin-torque term. The spin-wave coupling is controlled by tuning the Heisenberg and Dzyaloshinskii-Moriya interactions and also by externally applying the oscillating magnetic field. The activation function of the processing unit is varied by the spin-wave coupling. The artificial neural network computation is also demonstrated by performing the recognition of the handwritten digits in the MNIST database. By using the sigmoid-like activation function, the learning is conducted by the algorithm used for the conventional feedforward neural network. The obtained recognition accuracy is 0.853.