Quantum Materials - 22

Abstract

Physical reservoir computing (PRC) is an artificial neural network in which reservoir computing is performed on physical devices with nonlinear mapping capability to high-dimensional space, and the PRC is attracting attention as edge AI operating on terminal device due to its low learning cost and high processing speed. However, some fatal issues are remained on the road to the implementation as follows; low computational power, high electrical power consumption, and large device volume. In recent years, among such physical devices for the PRC (i.e., physical reservoir), we tracked down that chaotic interfered spin waves in a ferromagnetic material show excellent computing capability as physical reservoir.[1-3] Herein, we newly fabricated an iono-magnonic reservoir to improve the high dimensionality of the physical reservoir through in situ manipulation of the chaotic interfered spin waves in all-solid state redox device.[4,5]
Figure (a) shows a schematic illustration of the fabricated iono-magnonic reservoir using two excitation and two detection antennas on a Y₃Fe₅O₁₂ (YIG) single crystal. Proton (H+) in a Nafion is inserted into the YIG by gate voltage (V_G) application, and electrons are doped into tetrahedral Fe site, leading to decrease in magnetization. Furthermore, spin wave characteristics (i.e., amplitude and frequency) are significantly modulated by V_G application as shown in Fig. (b). Using the reservoir state consisting of various chaotic interfered spin waves at various V_G state, a chaotic time series prediction task described by a Mackey-Glass equation was performed (Fig. (c)). As shown in Fig. (d), our result was overwhelmingly superior to any other physics reservoirs and was comparable to the performance of a high-performance simulated neural network constructed precisely. We will discuss the effect of proton insertion in the YIG and the complexity of the chaotic spin wave interference that led to the high performance.[5]
This work was partially supported by Innovative Science and Technology Initiative for Security Grant Number JPJ004596, ATLA, Japan, and JSPS KAKENHI Grant Numbers JP22H04625 and JP19H05814 (Grant-in-Aid for Scientific Research on Innovative Areas “Interface Ionics”).