Reservoir Computing Seminar Group​

Date: 2025/12/18

Presenter: Prof. Takeshi Yoshimura (Osaka Metropolitan University) 

Title: Physical Reservoir Computing Using Piezoelectric Resonators

Abstract:

In this study, we have demonstrated physical reservoir computing by focusing on the inherent nonlinearity and short-term memory of piezoelectric resonators. The transient response of piezoelectric resonators exhibits memory on the scale of hundreds of milliseconds, making their integration into sensors attractive because it aligns with the temporal characteristics of real-world signals. Furthermore, we introduced a configuration that couples multiple resonators via field-effect transistors. This forms a state space within the reservoir layer where past and present inputs interfere, thereby enhancing computational capability. We confirmed that this simultaneously fulfills key requirements in physical reservoir design: increasing short-term memory length and strengthening nonlinearity[1,2].

Furthermore, this physical reservoir computing using piezoelectric resonators is being explored for deployment in practical intelligent devices. Acoustic sensors using piezoelectric MEMS resonators have demonstrated excellent time-series processing capabilities in the NARMA2 prediction task[3,4]. Additionally, current sensors employing a magnetic excitation tuning fork structure have demonstrated the ability to classify current patterns[5]. These results indicate that piezoelectric resonators can be deployed as edge AI devices integrating sensing and computation.

[1] T. Yoshimura, T. Haga, N. Fujimura, K. Kanda, I Kanno, 22nd IEEE Transducers, 457 (2023)
[2] T. Yoshimura, T. Haga, N. Fujimura, K. Kanda, I Kanno, Jpn. J. Appl. Phys., 62, SM1013 (2023)
[3] S. Yamamoto, M. Kiuchi, T. Yoshimura, 38th IEEE MEMS 550-553 (2025)
[4] S. Yamamoto, M. Kiuchi, T. Yoshimura, 23rd IEEE Transducers, 1592 (2025)
[5] K. Nishimura, N. Fujimura, T. Yoshimura, Jpn. J. Appl. Phys., 63, 09SP23 (2024)