Physical reservoir computing is a machine-learning computing method that utilizes device transient and nonlinear responses (response to history) that have not been actively used in electronics. The merits for low-power consumption hardware, such as short machine-learning time and extremely small number of adjustable weights, are attractive for highly-efficiency computing applications near the edges of information and telecommunication network where computing resource is limited. For this computing method, various electronic devices are applicable. Currently, we are conducting research on systems that use memristive device, spintronic device, and ferroelectric device, to efficiently perform information processing of time-series data.

Silicon can have very long electron spin-lifetime owing to its weak spin-orbit interaction and hence it is very attractive for spin-functional device applications. In addition, a representative spin device "silicon-based spin transistor" is promising for next-generation computing technology on a silicon platform, through integration with CMOS circuits. We are mainly conducting on fabtication of spin MOSFETs with an inversion channel, studies on device physics including spin transport phenomena, and the development of device technology to achieve a high spin functionality toward practical use.