Neuromorphic computation with a single magnetic domain wall

Abstract

Classifying or predicting complex time-dependent signals is a challenging computational task. Reservoir computing (RC) is an efficient neuromorphic computing approach that is ideally suited to such tasks, and is typically implemented in software using a recurrent neural network (RNN) with fixed synaptic weights (the reservoir) connected to a single, trainable readout layer. However, more efficient implementations of RC are possible if the software RNN reservoir is substituted with a physical system with the correct properties, such as non-linear response to input signals and inherent memory [1].In this work, we exploit the chaotic and non-linear dynamics of a single magnetic domain wall (DW) trapped between two anti-notches in a Ni nanostrip as a hardware-based reservoir. We have modelled the structure (inset of Fig. 1) using both a simple 1D model and micromagnetic simulations with an oscillating field. As shown in Fig. 1, complex oscillatory dynamics over a range of field amplitudes are observed. A chaotic and multi-period regime occurs between 0.4 kA/m and 1.2 kA/m and a non-linear regime at higher fields. Thus, the DW oscillator can be used to transform an input magnetic field to the DW position output suitable for RC.A time-multiplexed input signal is injected into the DW oscillator by modulating the applied field magnitude, and we show how this approach allows the device to perform classification tasks as a reservoir.First, we demonstrate how the scaling to convert input data into applied fields affects the classification accuracy of the sine or square wave task, showing that the best recognition rate is obtained at the edge of a chaotic regime of oscillation (Fig.2). We also demonstrate that the DW oscillator can perform complex tasks, such as spoken digit recognition, and that coupling multiple DWs together enhances the computational capabilities of the reservoir. Our work opens new avenues for research into neuromorphic computing using nanomagnetic hardware.  Bifurcation plot showing the DW position once per cycle for a given applied field using the 1D model. Inset: Schematic of the system.  Classification accuracy of the Sine-Square task over a range of field centre (H0) and field width (ΔH) values.