Machine Learning Approaches for Predicting Electronic Properties of Oxide Thin Films

Electronic properties of oxide thin films including band gap, carrier concentration, electrical conductivity, mobility, resistivity, and dielectric constant govern their functionality in electronics, spintronics, optoelectronics, and energy devices. However, extracting these properties experimentally or through first-principles simulations is often time-consuming, resource-intensive, and constrained by limited parameter exploration. This study presents a robust machine learning (ML) framework for accurate prediction of electronic properties of oxide thin films using a multi-modal dataset comprising experimental measurements, computational descriptors, and physics-informed engineered features. Multiple ML models were developed, including Random Forests (RF), Gradient Boosting, Artificial Neural Networks (ANN), Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Physics-Informed Neural Networks (PINNs). PINNs achieved superior band-gap prediction (MAE = 0.06 eV, R² = 0.97), while ANN demonstrated the best performance for mobility prediction (RMSE = 1.12 cm²/V·s). Carrier concentration and conductivity predictions showed strong accuracy using ensemble models (R² > 0.93). SHAP-based interpretability revealed that electronic correlations were strongly influenced by strain-induced structural distortions, oxygen partial pressure, cation valence states, surface/interface defect density, and film thickness. Validation against independent DFT calculations and experimental datasets confirmed high generalizability across perovskite, spinel, and binary oxide thin films. The proposed ML framework offers a powerful tool for accelerating electronic property prediction, enabling dynamic optimization of deposition conditions and supporting inverse design of thin films with targeted electronic performance.