Implementing a magnonic time-delay reservoir computer model

Abstract

Recently we demonstrated experimentally that microwave oscillators based on the time delay feedback provided by traveling spin waves could operate as reservoir computers. In the present paper, we extend this concept by adding the feature of time multiplexing made available by the large propagation times/distances of traveling spin waves. The system utilizes the nonlinear behavior of propagating magnetostatic surface spin waves in a yttrium-iron garnet thin film and the time delay inherent in the active ring configuration to process time dependent data streams. Higher reservoir dimensionality is obtained through the time-multiplexing method, emulating "virtual" neurons as temporally separated spin-wave pulses circulating in the active ring below the auto-oscillation threshold. To demonstrate the efficacy of the concept, the active ring reservoir computer is evaluated on the short-term memory and parity check benchmark tasks, and the physical system parameters are tuned to optimize performance. By incorporating a reference line to mix the input signal directly onto the active ring output, both the amplitude and phase nonlinearity of the spin waves can be exploited, resulting in significant improvement on the nonlinear parity check task. We also find that the fading memory capacity of the system can be easily tuned by controlling the active ring gain. Finally, we show that the addition of a second spin-wave delay line configured to transmit backward volume spin waves can partly compensate dispersive pulse broadening and enhance the fading memory capacity of the active ring.