Learning Trajectories from Spin-Wave Dynamics

Abstract

The efficacy of a physical reservoir computer model based on traveling spin waves in a spin-wave delay-line active-ring resonator was demonstrated recently. In the present work, we investigate how this neuromorphic device can be adapted for sensing applications. In this ``reservoir computing for sensing'' framework, we exploit strong coupling of the physical reservoir to its environment to utilize the reservoir as a sensing element. The dynamics of traveling spin waves in delay-line active rings are strongly dependent on the magnetic field and carrier frequency of those spin waves. Treating the spin-wave frequency as an environmental variable, we excite the active ring into different dynamical states by modulating the carrier frequency of a drive signal of microwave pulses injected into the ring. Training a linear regression on the time-multiplexed output from the ring allows the periodic amplitude patterns of the spin waves to be mapped reproducibly onto two-dimensional trajectories, representing periodic ``behavioral'' targets. Our work demonstrates the versatility of a magnonic resonator as a multipurpose computing and sensing device.