[19p-A41-1~6] Towards social applications of physical reservoir computing using new materials and new principles

2:10 PM - 2:50 PM JST(5:10 AM - 5:50 AM UTC)

[19p-A41-2] Theoretical analysis of physical reservoir computing using spin waves

〇Natsuhiko Yoshinaga^1,3,5, Satoshi Iihama², Yuya Koike⁴, Shigemi Mizukami³ (1.Future Univ. Hakodate, 2.Nagoya Univ., 3.Tohoku Univ. AIMR, 4.Tohoku Univ., 5.AIST)

Keywords:

spin wave,reservoir computing

Recent advances in neural networks demonstrate extraordinary performances in various tasks. However, their current technologies are software-based. Therefore, the computational speed, size, and energy consumption of the neural networks are limited by the properties of conventional electric computers. Neuromorphic computing, more specifically physical reservoir computing, is a promising direction because it can go to the realm of nanoscale, GHz speed, and low energy consumption if we choose the physical system properly. The challenge to achieving nanoscale physical reservoir computing is to clarify the mechanism of its high computational performance. In this study, we found the universal scaling property for high performance of physical reservoir computing in nanoscales. We demonstrate it by using reservoir computing of spin waves for various tasks. The system exhibits state-of-art performance in nanoscales with GHz speed.