講演情報
[8p-N304-13]Dynamic modeling of ultra-tunable Si NEMS resonators for physical reservoir computing applications
〇(B)Nay WinAung1, Takahiro Namazu1,2, Amit Banerjee1 (1.Kyoto Univ. of Advanced Science, 2.SRIS, Tohoku Univ.)
キーワード：
Physical Reservoir Computing、Silicon NEMS Resonator、Duffing Nonlinearity
Physical reservoir computing (PRC) is a hardware-based information processing method promising efficient, low-cost, hardware-based edge-computing for smart-sensor and IoT applications. While PRC models have shown promising results in various classification and time-series prediction tasks, the fundamental underlying mechanisms demand a deeper understanding. In this research, we are exploring an ultra-tunable Si NEMS resonator-based PRC platform for developing a chip-scale, mass-producible PRC system that can provide (a) fundamental insights into PRC mechanisms like memory-nonlinearity tradeoff, and (b) tunability to multitask. In this regard, we exploit the ultrawide electrostatic tunability in resonance frequency, nonlinearity and quality factor of ultrathin NEMS resonators.
The device at the core is a vibrating Si NEMS resonator beam (w ~ 10 nm thick and L ~ 100 µm long), which is electrostatically actuated and capacitively sensed. Such Si NEMS resonators can show ultrawide electrostatic tunability in frequency and nonlinearity. We are particularly interested in the electrostatically tunable cubic nonlinearity of the device, which can be exploited for tuning the nonlinear transformation of an input pattern. In this work, we focus on modelling the complex dynamic response of the NEMS resonator and then explore the performance of the system in PRC implementation vis a vis electrostatic tuning.
The device at the core is a vibrating Si NEMS resonator beam (w ~ 10 nm thick and L ~ 100 µm long), which is electrostatically actuated and capacitively sensed. Such Si NEMS resonators can show ultrawide electrostatic tunability in frequency and nonlinearity. We are particularly interested in the electrostatically tunable cubic nonlinearity of the device, which can be exploited for tuning the nonlinear transformation of an input pattern. In this work, we focus on modelling the complex dynamic response of the NEMS resonator and then explore the performance of the system in PRC implementation vis a vis electrostatic tuning.