Band Alignment of Oxides by Learnable Structural-Descriptor-Aided Neural Network and Transfer Learning

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Band alignment of semiconductors, insulators, and dielectrics is crucial for device performance.
• Ionization potential and electron affinity determine surface-dependent band-edge positions.
• Accurate determination requires complex experiments or simulations.

Purpose of the Study:

• To develop a machine learning model for predicting band alignment in nonmetallic oxides.
• To enable rapid and systematic prediction of band positions for various oxide surfaces.

Main Methods:

• Utilized a high-throughput first-principles calculation dataset of ~3000 oxide surfaces.
• Developed a neural network model trained on bulk structure and surface termination information.
• Extended the model to incorporate multiple-cation effects and apply to ternary oxides.

Main Results:

• The neural network accurately predicts band positions for relaxed binary oxide surfaces.
• The model effectively handles multiple-cation effects and transfers to ternary oxides.
• Achieved accurate predictions using only bulk structure and surface termination data.

Conclusions:

• Machine learning offers an efficient method for determining band alignment in nonmetallic oxides.
• This approach facilitates systematic understanding and materials screening for electronic applications.
• Enables prediction of band alignment for a vast range of solid surfaces.