An Infrared Near-Sensor Reservoir Computing System Based on Large-Dynamic-Space Memristor with Tens of Thousands of States for Dynamic Gesture Perception

To efficiently process the massive amount of sensor data, it is demanding to develop a new paradigm. Inspired by neurobiological systems, an infrared near-senor reservoir computing (RC) system, consisting of infrared sensors and memristors based on single-crystalline LiTaO3 and LiNbO3 (LN) thin film respectively, is demonstrated. The analog memristor is used as a reservoir in the RC system to process sensor signals with spatiotemporal characteristics. LN crystal structure stacked with oxygen octahedra provides favorable conditions for reliable Mott variable-range hopping conduction, which provides the memristor with tens of thousands of reservoir states within a large dynamic range. With the characteristics, the analog sensor signals with high data fidelity can be directly fed to the memristive reservoir, and the spatiotemporal features can be separated and mapped. The system demonstrated a dynamic gesture perception task, achieving an accuracy of 99.6%, which highlights the great application potential of the memristor in signal sensor processing and will advance the application of artificial intelligence in sensor systems. Crystal ion slicing techniques are used to fabricate a single-crystalline thin film for both the memristor and sensor, which opens up the possibility of realizing monolithic integration of a memristor-based near-sensor computing system. A novel infrared near-sensor reservoir computing (RC) system constructed from an ion-slicing LiTaO3-based infrared array and ion-slicing LiNbO3-based memristor array is demonstrated. Thanks to the excellent capacities of the memristor reservoir, the infrared near-senor RC system successfully and robustly implemented a dynamic gesture perception task with spatiotemporal feature fusion.image