Utilizing rate-independent hysteresis for analog computing

Abstract Physical systems exhibiting hysteresis are increasingly being used in neuromorphic and in-memory computing research. Generally, the resistance switching of devices with rate-independent hysteresis are being investigated for their use as trainable weights in neural networks, whereas the dynamics of devices showing rate-dependent hysteresis are being investigate for their potential as nodes, for example in reservoir computing systems. In our work we instead investigate the computing potential of a simple rate-independent hysteresis system. We show that by driving a system of only two linear branches with time-multiplexed inputs it is possible to generate nonlinear transforms and perform timeseries prediction tasks.