Physical reservoir computers based on principles of magnonics promise energy efficient data processing and a reduction in the size and weight of the neuromorphic computing devices. The present work is a major step toward all-magnonic implementation of the recently proposed concept of a physical reservoir based on the spin wave active ring. The main component of the ring is a spin wave delay line employing a thin film of yttrium iron garnet (YIG) as the spin wave guiding medium. We propose controlling spin wave propagation in the YIG film electronically to enter input data into the reservoir. To this end, we exploit a physical effect of scattering of backward volume spin waves from a highly localized Oersted field of a dc current flowing through a metallic strip sitting on top of the YIG film. We find experimentally that a very small current (on the order of several milliamps) through the strip is able to control the amplitude of auto-oscillations in the ring. The use of the current control of spin wave propagation as a means to enter input data into the reservoir reduces the number of non-magnetic components of the reservoir to just one (a microwave amplifier). In addition, the proposed current-controlled magnonic reservoir demonstrates a record-high short-term memory capacity of 5.53, as our experiments show. Our findings open up an avenue for reduction of energy consumption by magnonic active-ring-based physical reservoirs, their micro-miniaturization, and all-magnonic implementation.
Read full abstract