High-rate capability of III–V anodes enabled by machine-learning-guided compositional design

High-performance anodes are critical for advancing lithium-ion batteries beyond the limits of conventional graphite. Here, we present a machine-learning-guided framework for discovering multicomponent group III–V semiconductor anodes. By integrating Bayesian optimization with combinatorial thin-film synthesis, we efficiently navigated the complex compositional space of quaternary InGaAsSb thin films. Remarkably, the optimized composition delivers a capacity of approximately 600 mAh g −1 even at an ultrahigh current density of 10 A g −1 , representing one of the highest rate performances reported for a binder-free single-layer anode. Structural analyses revealed columnar grains with Ga-rich nanodots and compositional fluctuations that mitigate mechanical stress and suppress degradation. These findings highlight the potential of compositional complexity to enhance durability and rate capability. More broadly, our results demonstrate that coupling machine learning with physical synthesis provides a versatile paradigm for intelligent materials discovery in energy storage systems.