All physical reservoir computing system with tunable temporal dynamics for multi‐timescale information processing

Abstract Physical reservoir computing (PRC) offers an effective computing paradigm for spatiotemporal information processing with low training costs. Achieving controllable regulation over the temporal dynamics of devices to meet the computational demands of each physical layer is a key challenge for realizing high‐performance PRC chips. Here, we proposed a homogeneously integrated all‐PRC with tunable temporal dynamics. Utilizing the modulation effect of oxygen vacancies on the energy barrier of the pentacene/ZnO interface, short‐term memory, and long‐term memory switching characteristics have been achieved within the same device structure. Furthermore, by altering the gate voltage, the reservoir exhibited a broad range ratio of temporal characteristics (>10²), which provides the potential to map information with different temporal characteristics. Inspired by the process of encoding and reconstructing spatiotemporal information in the human visual system, a biomimetic obstacle recognition system has been constructed to assist visually impaired individuals in walking, demonstrating excellent accuracy in obstacle types (100%) and distances (97.2%) recognition. This work offers a promising avenue for the development of an integrated PRC system with multi‐timescale information processing capability. image