Benchmarking a magnon-scattering reservoir with modal and temporal multiplexing

Abstract

Physical reservoir computing has emerged as a powerful framework for exploiting the inherent nonlinear dynamics of physical systems to perform computational tasks. Recently, we presented the magnon-scattering reservoir, whose internal nodes are given by the fundamental wave-like excitations of ferromagnets called magnons. These excitations can be geometrically-quantized and, in response to an external stimulus, show transient nonlinear scattering dynamics that can be harnessed to perform memory and nonlinear transformation tasks. Here, we test a magnon-scattering reservoir in a single magnetic disk in the vortex state towards two key performance indicators for physical reservoir computing, the short-term memory and parity-check tasks. Using time-resolved Brillouin light scattering microscopy, we measure the evolution of the reservoir's spectral response to an input sequence consisting of random binary inputs encoded in microwave pulses with two distinct frequencies. Two different output spaces of the reservoir are defined, one based on the time-averaged frequency spectra and another based on temporal multiplexing. Our results demonstrate that the memory and nonlinear transformation capability do not depend on the chosen read-out scheme as long as the dimension of the output space is large enough to capture all nonlinear features provided by the magnon-magnon interactions. This further shows that solely the nonlinear magnons in the physical system, not the read-out, determine the reservoir's capacity.