Enhancing prediction accuracy of physical band gaps in semiconductor materials

Accurate band-gap prediction is essential for designing and discov-ering new materials with desired properties. However, current methods for calculating band gaps based on local and semilocal func-tionals lead to significant underestimation, hindering the effectiveness of in silico and high-throughput screening of materials. We present a machine learning model with domain adaptation to rapidly yield accu-rate band-gap prediction of semiconductors (oxides, chalcogenides, nitrides, phosphides, etc.). The approach circumvents the prerequisite for a large amount of physically measured band-gap data, which is notoriously scarce. It instead sources knowledge from a large dataset with underestimated band gaps and subsequently transfers knowl-edge to train a crystal graph convolution neural network (CGCNN) us-ing a small dataset of accurate, physically measured band gaps. The prediction model shows a low mean absolute error (MAE) of 0.23 eV, outperforming those using Perdew-Burke-Ernzerhof (PBE) functionals (MAE = 0.87 eV). Visualization of the learned crystal graph using the t-distributed stochastic neighbor embedding (t-SNE) algorithm re-vealed that the crystal structure and composition have a strong influ-ence on the material band gaps.

Automated summary

Accurate band-gap prediction is crucial for designing and discovering new materials. Current methods often underestimate the band gaps, limiting their effectiveness. This study presents a machine learning model with domain adaptation that can rapidly and accurately predict the band gaps of semiconductors, overcoming the scarcity of measured data.