2D Vanadium Carbide/Oxide Heterostructure‐Based Artificial Sensory Neuron for Multi‐Color Near‐Infrared Object Recognition

Abstract

Near‐infrared (NIR) photon detection and object recognition are crucial technologies for all‐weather target identification in autonomous navigation, nighttime surveillance, and tactical reconnaissance. However, conventional NIR detection systems, which rely on photodetectors and von Neumann computing algorithms, are plagued by energy inefficiency and signal transmission bottlenecks. Herein, a vanadium carbide/oxide (V2C/V2O5‐x) heterostructure is designed and synthesized by a topochemical conversion method. The V2C/V2O5‐x heterostructure‐based memristor exhibits stable threshold‐type resistance switching (RS) behavior with low coefficient of variation in transition voltages (1.62% and 1.7%) over thousands of cycles, and maintains stable performance even after storage for 90 days. Benefiting from the NIR responsivity of V2C and the volatile RS enabled by vacancy‐enriched V2O5‐x, devices exhibit a linear variation in threshold voltage in response to NIR light power density and wavelength. Based on the multi‐color NIR modulable RS characteristics and the YOLOv7 algorithm model, an artificial neural network (ANN) architecture achieves average recognition accuracies of 89.6% for cars and 85.9% for persons on the FLIR dataset. This work reveals a heterostructure with versatile functionalities for neuromorphic devices and establishes a memristor‐based ANN platform for multi‐color object detection and recognition in complex real‐world scenarios.