Hardware-Software Codesign of 2D Neuromorphic Optoelectronic Device for Dynamic Gesture Recognition

In-sensor reservoir computing has recently gained considerable attention for its efficient training process and advanced integration of sensing, storage, and processing functionalities. However, developing a highly efficient in-sensor reservoir computing system remains challenging, mainly due to the lack of suitable devices with appropriate architectures. In this study, a graphene/MoSe₂-based ohmic contact optoelectronic synaptic memory device optimized for in-sensor reservoir computing (RC) is introduced, designed to emulate biological synaptic functions and enable efficient neuromorphic computing. Based on the dynamic characteristics and fading memory of this device, a reservoir computing system for dynamic gesture recognition, including six types of gestures, is stimulated, achieving a recognition rate of 95%. This work provides a potential solution for hardware-software co-design in dynamic gesture recognition.