Powering AI at the edge: A robust, memristor-based binarized neural network with near-memory computing and miniaturized solar cell

Memristor-based neural networks provide an energy-efficient platform for AI and can enable self-powered operation when paired with energy harvesters. However, most existing designs use analog in-memory computing that requires a stable power supply, incompatible with unstable harvesters. This work fabricates a robust binarized neural network with 32,768 hafnium-oxide memristors, powered by a miniature wide-bandgap solar cell for edge applications. The circuit uses a resilient digital near-memory computing approach with complementarily programmed memristors and logic-in-sense-amplifier XNOR operation, eliminating compensation or calibration and operating across diverse conditions. Under high illumination, inference matches lab power supply performance; under low illumination, it remains functional with slightly reduced accuracy, naturally entering an approximate computing mode. Neural network simulations for image classification show misclassifications under low illumination are mainly difficult-to-classify cases. This approach advances self-powered AI and intelligent sensors for health, safety, and environmental monitoring.

Fadi Jebali, Atreya Majumdar, Clément Turck, Kamel-Eddine Harabi, Mathieu-Coumba Faye, Eloi Muhr, Jean-Pierre Walder, Oleksandr Bilousov, Amadéo Michaud, Elisa Vianello, Tifenn Hirtzlin, François Andrieu, Marc Bocquet, Stéphane Collin, Damien Querlioz, Jean-Michel Portal