Transistor-Structured Artificial Dendrites for Spatiotemporally Correlated Reservoir Computing

作者：Yao Ni, Yumo Zhang, Jingjie Lin, Xingji Liu, Yue Yu, Lu Liu, Wei Zhong, Yayi Chen, Rongsheng Chen, Hoi Sing Kwok, Yuan Liu · 发表于：IEEE Electron Device Letters · 年份：2025 · DOI：10.1109/led.2025.3598823 · 被引用次数：1 · 研究领域：Neural Networks and Reservoir Computing、Advanced Memory and Neural Computing、Neural Networks and Applications

Neuromorphic electronics, which integrate sensing, storage, and computing to boost efficiency and performance, offer a low-cost, energy-efficient alternative for temporal data processing with edge computing potential when applied to reservoir computing (RC). However, current neuromorphic RC systems struggle with diverse spatial information inputs due to device design limits. Here, we present the first artificial dendrite horizontally cascaded by Ga-Sn-O (GTO)-based synaptic transistors with efficient electron-ion coupled films as the gate dielectric. This design allows unlimited lateral gate expansion, source/drain and gate interchangeability, and sustains a microampere-level output current across a record centimeter-scale gate-channel distance. The artificial dendrite maintains stable weight modulation through$10^{\mathbf {{5}}}$electrical cycles and$10^{\mathbf {{7}}}$bending cycles, with performance restorable via dielectric film replacement. This work pioneers the demonstration of spatiotemporally correlated reservoir computing using artificial dendrites-based RC systems, achieving highly accurate recognition and introducing a new paradigm to the field.