[16p-D419-1~18] 10.2 Fundamental and exploratory device technologies for spin

▲ [16p-D419-17] Nano-scale reservoir computing based on propagating spin-waves in a ferromagnetic thin film

〇Satoshi Iihama^1,2, Yuya Koike^3,2,5, Shigemi Mizukami^2,4, Natsuhiko Yoshinaga^2,5 (1.FRIS, Tohoku Univ., 2.AIMR, Tohoku Univ., 3.Tohoku Univ., 4.CSIS, Tohoku Univ., 5.MathAM-OIL, AIST)

Keywords:spintronics, physical reservoir computing, spin-wave

Recently, neuromorphic computing based on spintronics technology has attracted great attention for development of future energy-efficient artificial intelligence. Reservoir computing is one of promising scheme for physical implementation of neuromorphic computing. The physical reservoir computing based on spintronics technology such as spin-torque oscillators and spin-waves has been reported experimentally, however, the performance such as memory capacity (MC) has been poor compared with other reservoirs such as photonics reservoir. Therefore, it is demanded to develop experimental schemes for realization of high-performance reservoir computing. Here, we propose nano-scale reservoir computing based on propagating spin-waves in a ferromagnetic thin film. Schematic illustration of reservoir computing based on propagating spin-waves is shown in Fig. 1(a). Multiple physical nodes are placed on a ferromagnetic thin film where spin-waves are excited by spin-transfer torque through injecting current and magnetization can be detected by using the magnetoresistance effect at each physical node. Simulations are performed by micromagnetic simulator mumax³. Benchmark tasks such as MC, information processing capacity (IPC), and normalized root mean square error (NRMSE) for NARMA10 task using spin-wave reservoir were evaluated. Figure 1(b) shows one of result for NARMA prediction task. NRMSE ~ 0.2, high-performance comparable to previous reports, were evaluated. MC, IPC, and NRMSE of NARMA10 tasks using spin-wave reservoir will be presented [1].
This work is supported by JST PRESTO (Grant Number: JPMJPR22B2), JST FOREST Program (Grant Number: JPMJFR2140), JSPS KAKENHI (Grant Number: 21H04648, 21H05000), and X-NICS of MEXT (Grant Number: JPJ011438). S.M. thanks to CSRN of CSIS at Tohoku Univ.
[1] S. Iihama, Y. Koike, S. Mizukami, N. Yoshinaga, arXiv:2301.02193