@

초록

This study proposes a fully hardware-based reservoir computing (RC) system, which utilizes a single ferroelectric thin-film transistor (FeTFT) based on a Si3N4/Hf0.5Zr0.5O2 (HZO)/indium gallium zinc oxide (IGZO) tri-layer stack. The proposed FeTFT operates in three independent memory modes according to two different input signals, electrical stimulation and optical stimulation: electric long-term (LT), electric short-term (ST), and optical ST. Through this, non-volatile memory characteristics and volatile memory characteristics were successfully integrated within a single device. The non-volatile electric LT mode is based on the ferroelectric polarization switching mechanism of the HZO layer and was utilized as the readout layer for the RC, demonstrating excellent endurance and retention characteristics. The volatile electric ST mode is based on the charge trapping mechanism of the Si3N4 trap layer, while the optical ST mode is based on the optical ionization mechanism of the IGZO channel. These two ST modes perform the function of a 4-bit reservoir layer with short-term characteristics such as paired-pulse facilitation. Finally, the fully hardware-based RC system, which organically integrates these three modes and applies the hybrid mapping model, achieved a high recognition accuracy of 92.43% in the Modified National Institute of Standards and Technology dataset recognition task.

키워드