Molecular crystal memristor-based edge AI platform for energy-efficient and real-time smart grid inspection

Vast power grid infrastructure generates enormous volumes of inspection data from smart meters, unmanned aerial vehicle (UAV) patrols, and high-definition video monitoring. Meeting the demand for real-time analysis places stringent requirements on latency, energy efficiency, and on-device intelligence at the edge. Here, we present a molecular crystal memristor-based edge artificial intelligence (AI) hardware platform that can be directly deployed in inspection devices, enabling real-time grid monitoring with drastically reduced computational and storage overheads. The memristor exhibits highly controllable filamentary switching behavior, stable multi-level conductance states, femtowatt-scale power consumption, and outstanding retention. Leveraging these properties, the platform enables fully hardware-integrated convolution, achieving 97% feature-extraction accuracy and 67.75 TOPS/W energy efficiency, thereby substantially alleviating the computational and storage load of cloud servers. This work establishes a scalable and energy-efficient in-memory computing framework for smart grid inspection and provides a powerful foundation for broader edge AI applications.