Reservoir Computing Based on ZnWO4-x/WO3-x Heterojunction Optoelectronic Memristor for Time-Series Data and Static Image Processing

The nonlinear, dynamic, and memory characteristics, as well as synaptic properties, make memristors particularly suitable for reservoir computing. In this work, a novel optoelectronic memristor of a heterojunction structure is fabricated. Based on that device, reservoir computing systems are further implemented to improve the recognition accuracy of time-series data and optimize the operating complexity of the system, respectively. Specifically, the memristor features an Ag/ZnWO_4-x/WO_3-x/ITO structure, which can exhibit great synaptic properties, as well as obvious nonlinear I–V characteristics and high resistance ratios under electrical modulations. By integrating the memristor with mask technology, a basic reservoir unit can be obtained. With a parallel interconnection of the units, a reservoir system for time-series data processing is constructed, which achieves a high accuracy rate of 96.84% in heart arrhythmia detections. In addition to electrical performance, the memristor also presents good optical responses due to the contribution from the ZnWO_4-x and WO_3-x functional layers. The distinct and stable conductance states under ultraviolet (UV) light excitations help to extend reservoir states and thereby realize a pure optical reservoir computing system. Compared to traditional optoelectronic hybrid systems, the pure optical system can obviously lower the operating complexity since it relies solely on optical modulation. Taking a static image recognition task as an example, the accuracy of the pure optical system reaches as high as 92.73%. This work provides a new type of memristor and instances to achieve high-performance and simple reservoir systems, which may offer innovative insight for advanced reservoir computing.