Machine Learning Assisted Inverse Design of Low Resistivity In–Ga–Sn–Zn Oxide Sputtering Targets

ABSTRACT

This study introduces an expedited design methodology that synergizes machine learning with particle swarm optimization (PSO) for the inverse design of low‐resistivity In–Ga–Sn–Zn quaternary sputtering targets. By integrating eight physical descriptors and two processing variables within a leakage‐controlled ensemble‐learning workflow, a high‐performing calibrated surrogate was developed. The finalized five‐fold out‐of‐fold ensemble, followed by the monotone isotonic mapping specified in the computational pipeline, achieved a calibration‐adjusted coefficient of determination ( R ² = 0.895) on the pooled cross‐validated predictions. Leveraging the calibrated surrogate, three optimal compositions were identified through PSO, with predicted resistivity values ranging from 0.483 to 0.495 mΩ·cm. Experimental validation confirmed measured resistivities of 0.46, 0.47, and 0.49 mΩ·cm, respectively, with all relative errors below 5%. Furthermore, all verified targets exhibited a relative density exceeding 99%. These prospective experiments support the practical value of the proposed framework for efficiently developing high‐performance, low‐resistivity oxide sputtering targets.