Ryo Iguchi received his Ph.D. in Engineering at Keio University in 2013. After working at Tohoku University, in 2017, he joined the Research Center for Magnetic and Spintronic Materials at NIMS and engaged in research on spin-dynamics based thermal physics. Since 2025, he has been working in Neuromorphic Devices group at the Research Center for Materials Nanoarchitectonics. His research interests include spintronics, thermal measurements, and neuromorphic devices.

Neuromorphic computing is an important concept for the future of information technology, offering promising solutions to enable efficient information processing and thereby reduce energy consumption. As the volume of information continues to increase rapidly, new approaches are required to advance society. Physical reservoir computing (PRC) is one such approach, leveraging the nonlinear dynamics of physical materials, such as spin waves, for information processing.

Spin waves, collective excitations of electron spins in magnetic materials, exhibit oscillatory behavior and inherent nonlinearity due to their quantum characteristics, making them highly attractive for PRC applications. Recent progress [1-4] has demonstrated the potential of spin waves for PRC. However, practical implementation remains challenging, primarily due to the lack of feasible architectures for efficient information extraction and input delivery without significant hardware resources. Thus, a central challenge lies in maximizing the utilization of rich material dynamics and designing architectures capable of efficiently reading out and extracting information.

In this work, we tackled these challenges by developing a hardware-implementable, high-performance read-out architecture and by exploring physics-based input engineering to exploit spin wave dynamics. By thoroughly designing our system to capture the oscillatory behavior of spin waves, our framework achieves record performance in benchmark tests, while significantly reducing hardware demands. Furthermore, our physics-inspired input encoding opens new avenues for optimizing PRC performance by leveraging material dynamics. We hope these developments will pave the way toward highly efficient edge AI systems, capable of intelligent information processing at the edge of trillion-sensor networks.

1. R. Nakane, G. Tanaka, and A. Hirose, IEEE Access, 6, 4462 (2018) DOI: 10.1109/ACCESS.2018.2794584
2. L. Körber et al., Nat. Commun. 14, 3954 (2023) DOI: 10.1038/s41467-023-39452-y
3. W. Namiki et al., Adv. Intell. Syst. 5, 2300228(2023) DOI: 10.1002/aisy.202300228
4. S. Hikasa et al., Adv. Electron. Mater. e00780 (2026) DOI: 10.1002/aelm.202500780
5. J. Chen, R. Iguchi et al., arXiv:2603.04901 (2026) DOI: 10.48550/arXiv.2603.04901