InSe Dynamic Memristor-Based Reservoir Computing for Temporal and Spatial Signal Recognition
Jing Chen, Junqiang Zhu, Ping Li, Jianguo Hu, Qiuliang Li, Zhenhua Wang, Zheng Zhang, Hengji Li, Siqi Lin, Xiaofei Yue, Tian‐Ling Ren, Hong Liu, Min Jin, Lin Han
Abstract
Multimodal recognition techniques are pivotal for advancing contemporary artificial intelligence, particularly in enhancing visual perception. However, research on electronic devices capable of robust multimodal recognition remains limited. In this study, we employ an InSe/Al 2 O 3 / Pb(Zr 0 · 2 Ti 0 · 8 )O 3 (PZT) heterostructure as a dynamic memristor. Moreover, the InSe dynamic-memristor-based optoelectronic reservoir computing (RC) system is developed for multimodal recognition of temporal and spatial signals. Under electrical modulation, the InSe dynamic-memristor-based parallel RC system has been employed to efficiently process temporal signal tasks, such as a waveform classification task with a normalized root-mean-square error (NRMSE) of 0.0873, and a spoken-digit recognition task achieving a recognition accuracy of 99%. Under optical modulation, the InSe dynamic-memristor-based RC system has been used for processing a spatial signal task for number-image recognition with 100% accuracy. The InSe dynamic-memristor-based optoelectronic RC system sets the stage for future interactive AI vision applications based on 2D electronic devices.