Voltage-mode reservoir computing with ferroelectric CMOS inverters

作者：Rikuo Suzuki, Kasidit Toprasertpong, Ryosho Nakane, Eishin Nako, Mitsuru Takenaka, Shinichi Takagi · 发表于：Applied Physics Express · 年份：2025 · DOI：10.35848/1882-0786/ade199 · 被引用次数：2 · 研究领域：Neural Networks and Reservoir Computing、Advanced Memory and Neural Computing、Neural Networks and Applications

Abstract We propose reservoir computing (RC) utilizing a CMOS inverter composed of ferroelectric FETs (FeCMOS) to enhance energy efficiency and computational capability. We confirmed that the output voltage of an FeCMOS exhibits hysteresis characteristics originating from ferroelectric polarization dynamics and the FeCMOS RC has the capability to solve fundamental nonlinear problems. Furthermore, we introduced a technique to enhance the computational capability of FeCMOS RC by adjusting the center of the operating voltage according to the threshold voltage. This approach facilitates transient dynamics with a wider range of intermediate output voltage and enhances the performance of the RC system. We also demonstrate that FeCMOS RC can solve nonlinear time-series prediction tasks with higher energy efficiency than conventional FeFET RC systems.