Machine-Learning Prediction of Charged-Defect Formation Energies from Crystal Structures

Area of Science:

• Materials Science
• Computational Materials Science
• Solid State Physics

Background:

• Materials informatics advances enable broader material synthesis.
• Screening semiconductors by defect properties is difficult due to a lack of general prediction frameworks.
• Predicting defect formation energies across multiple charge states from structural data is a key challenge.

Purpose of the Study:

• To develop a general framework for predicting defect formation energies in multiple charge states.
• To introduce a machine learning model integrating defect formation energies and band-edge predictions for virtual screening.
• To identify novel hole-dopable oxides with potential photovoltaic applications.

Main Methods:

• A protocol involving data normalization, Fermi level alignment, and treatment of perturbed host states was developed.
• The protocol was validated by accurately predicting oxygen vacancy formation energies in three charge states using a single model.
• A joint machine-learning model was created to integrate defect formation energies and band-edge predictions.

Main Results:

• The developed protocol accurately predicted oxygen vacancy formation energies.
• The joint machine-learning model successfully integrated defect formation energies and band-edge predictions.
• 89 hole-dopable oxides were identified, including BaGaSbO, a promising ambipolar photovoltaic material.

Conclusions:

• The presented protocol offers a robust method for predicting point defect formation energies.
• The joint machine-learning framework facilitates efficient virtual screening of materials.
• This work is expected to establish a standard approach for machine-learning studies on point defect formation energies.