Physical Reservoir Computing using HZO-based FeFETs for Edge-AI Applications
Shinichi Takagi, Kasidit Toprasertpong, Eishin Nako, Ryō Suzuki, Shin-Yi Min, Mitsuru Takenaka, Ryosho Nakane · 2023
we have proposed and demonstrated reservoir computing (RC) by using HZO/Si FeFETs. In our FeFET reservoir, the time response of currents of Si FeFETs is utilized as the virtual nodes. We experimentally present the fundamental properties of RC performance and address several methods to enhance RC performance. This scheme is applied to two typical AI tasks, prediction of nonlinear time- series data and speech recognition. We experimentally show a classification accuracy of 98.1 % in a speech recognition task to classify the audio waveforms of ‘0’ to ‘9’ spoken digits.