Memristor-Based Sensing–Memory–Computing Integrated Sensors: Mechanisms, Materials, Architectures, and Applications

Abstract

Conventional sensing systems based on physically separated sensing, memory, and computing units suffer from substantial data-transfer overhead, latency, and energy consumption, limiting their deployment in edge intelligence and the Internet of Things. Memristive in-sensor computing has emerged as a promising hardware paradigm by integrating stimulus perception, state retention, and information processing within a single device or array. This review discusses memristor-based sensing-memory-computing integrated sensors from four perspectives: response modality, material system, fabrication and integration strategy, and device architecture. Working mechanisms and sensing behaviors are classified into digital memristors for event-driven functions and binary or low-bit synapses, and analog memristors for gradual weight modulation, feature extraction, and high-precision neuromorphic computing. Representative material platforms, including oxides, two-dimensional materials, organic materials, halide perovskites, electrochemical metallization systems, ferroelectric systems, and phase-change and correlated materials, are compared in terms of operating mechanisms, stimulus-coupling pathways, advantages, and limitations. The effects of thin-film deposition, post-treatment, patterning, and array integration on device uniformity, manufacturability, and scalability are further discussed. Finally, recent progress in visual, tactile, and biomimetic perception is reviewed, and key challenges and future directions are outlined for next-generation systems for edge intelligence.