FeRh‐based surface acoustic wave‐driven neuromorphic computing for energy‐efficient AI applications

Abstract Current‐driven spintronic artificial neural networks (ANNs) hold great promise for image recognition but are limited by excessive power consumption. Surface acoustic waves (SAWs) have recently emerged as a disruptive alternative, offering ultralow‐power control over magnetization through the magnetoelastic and acoustothermal effects. In this work, for the first time, we demonstrate a SAW‐driven neuromorphic computing paradigm utilizing FeRh magnetic phase transitions, achieving both ReLU neuron activation and robust synaptic plasticity. Notably, the power density of our neuromorphic devices is reduced by an order of magnitude compared to that of conventional spintronic‐based devices, enabling power‐efficient image recognition with an accuracy exceeding 91%. We also demonstrate that ANNs implemented with our neuromorphic devices can precisely and autonomously assess left ventricular ejection fraction, a key clinical metric for evaluating cardiac function. Our findings establish SAWs as a transformative enabler for next‐generation neuromorphic computing, paving the way for energy‐efficient, high‐precision artificial intelligence in both advanced image processing and medical diagnostics.