Using Machine Learning to Predict Energy Needed to Create Charged Defects in Crystal Structures

A total of 89 hole-dopable oxides were identified using a novel machine-learning framework.

• Accurate predictions of oxygen vacancy formation energies in three charge states were achieved through a single model.
• The approach incorporates data normalization, Fermi level alignment, and treatment of perturbed host states.
• A joint machine-learning model was developed to integrate defect formation energies with band-edge predictions.
• BaGaSbO was highlighted as a potential ambipolar photovoltaic material among the identified oxides.
• The proposed protocol may serve as a standard for future machine-learning studies on point defect formation energies.

Simplified