Surface Acoustic Wave Induced Nano-oscillator Based Reservoir Computing

Abstract

Reservoir computing (RC) is a computational framework where a fixed, non-linear reservoir maps the inputs into higher-dimensional space and the readout is trained with linear regression and classification [1]. Spintronic nanomagnetic devices are well suited for physical reservoir computing due to their interactive non-linear dynamics, recurrence characteristic, enduring lifetime, and low energy consumption [2-3]. A frustrated nanomagnet layer can also implement physical RC [4]. In this work, we demonstrate via micromagnetic simulation that a nanomagnet array excited by surface acoustic waves (SAW) can be used as a reservoir to classify sine and square waves with high accuracy. The simulated nanomagnet array has an input, 7 outputs (Fig. 1) and a 4 GHz SAW is applied to the input nanomagnet. The SAW amplitude is varied in such a way that its envelope forms a random sequence of sine and square waves of 100 MHz frequency (Fig. 2 (a)). The non-linear responses of the output nanomagnets due to this input are processed by reading the reservoir state every 1 ns while the period of the input is 10 ns. The envelopes of the output magnets’ magnetization are shown in Fig. 2 (b) which are used to train the output weights obtained by regression via Moore-Penrose pseudoinverse. For the classification, a random sequence of 100 square and sine wave samples are used. About 80 % of the total samples are trained, and the rest of the samples are used for testing. Various combination of the output nanomagnets exhibit 100 % training accuracy and 100 % testing accuracy which is indicative of the nanomagnet oscillator array being well suited for physical reservoir computing applications. Thus, this proposed strategy to implement reservoir computing could lead to energy efficient neuromorphic devices.Acknowledgment: M.F.F.C, W.A, D.B and J.A are supported in part by the National Science Foundation grant CCF-1815033.  Figure 1: A micro-magnetic snapshot of the input, the reservoir and output nanomagnets.  Figure 2: (a) SAW and input response (b) The envelopes of the responses vs. time (ns) of several output nanomagnets.