Machine Learning Insights into Band Alignments of van der Waals Heterostructures

The integration of two-dimensional (2D) materials into van der Waals heterostructures (vdWHs) enables the stacking of atomically thin layers through weak vdW interactions, offering tunable properties for next-generation optoelectronic and catalytic applications. In this study, we systematically predicted the band alignment types of over 32,000 vdWHs, constructed by pairing any two of the 256 semiconductor monolayers from the 2D Semiconductor Computational Database (2DSdb) [J. Phys. Chem. Lett. 2022 13, 11581], based on Anderson’s rule. Nearly 100 features were extracted from the physical properties of these vdWHs to establish a descriptor database for the triclassification of vdWHs using machine learning models, including extreme gradient boosting, random forest, and gradient boosting classifier, achieving accuracy rates between 85 and 87%. SHapley Additive exPlanations (SHAP) analysis identified electronegativity, oxygen fraction, valence electrons in the p-orbital, and lattice constants as the most influential features. The resulting database is expected to provide valuable guidance for experimentalists in designing nanodevices and photocatalysts.