Reservoir Computing with Random Skyrmion Fabrics

The many nanoscale properties of skyrmions have shown promise in applications ranging from non-volatile memory and spintronic logic devices, to enabling the implementation of unconventional computational standards such as Stochastic and Reservoir Computing. Particularly, Reservoir Computing is a type of recursive neural network capable of recognizing and predicting spatio-temporal events. No knowledge of its topology or node weights are required for its basic functioning and, therefore, one can utilize naturally existing networks formed by a wide variety of physical processes. Here we will discuss how random skyrmion clusters can effectively implement a functional reservoir. This is achieved by leveraging the nonlinear resistive response of the individual skyrmions arising from their current dependent AMR. Complex time-varying current signals injected via contacts are modulated nonlinearly by the fabric’s AMR as the current flow distributes along paths of least resistance as it traverses the geometry. By tracking resistances across multiple input and output contacts, we show how the instantaneous current distribution effectively carries temporally correlated information about the injected signal allowing for a numerical demonstration of simple pattern recognition.