Bandgap prediction of two-dimensional materials using machine learning.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• The bandgap of two-dimensional (2D) materials is critical for semiconductor device applications.
• Conventional first-principles calculations for bandgap prediction are computationally expensive.
• Machine learning (ML) offers a potential alternative for rapid property prediction.

Purpose of the Study:

• To investigate the efficacy of ML algorithms in predicting the bandgaps of 2D materials.
• To compare the performance of different ML models for this task.
• To assess the impact of including spin-orbit coupling (SOC) data on prediction accuracy.

Main Methods:

• Utilized four ML algorithms: gradient boosted decision trees, random forests, support vector regression, and multi-layer perceptron.
• Trained models using data from the Computational 2D Materials Database (C2DB).
• Evaluated model performance using R-squared (R2) and root-mean-square error (RMSE), both with and without spin-orbit coupling (SOC) as a feature.

Main Results:

• Gradient boosted decision trees and random forests achieved R2 >90% and RMSE ~0.24-0.27 eV without SOC.
• Support vector regression and multi-layer perceptron showed R2 >70% and RMSE ~0.41-0.43 eV without SOC.
• Including SOC data significantly improved all models, reducing RMSE to ~0.09-0.17 eV and achieving R2 >94%.

Conclusions:

• ML models, particularly gradient boosted decision trees and random forests, are highly effective for predicting 2D material bandgaps.
• Incorporating spin-orbit coupling (SOC) data substantially enhances prediction accuracy.
• ML provides a precise and efficient method for determining 2D material properties, accelerating semiconductor device design.