Machine learning-driven prediction of band-alignment types in 2D hybrid perovskites

Abstract

Based on intramolecular band alignments between the organic and inorganic units, 2D hybrid perovskites can be of four types (I_a, I_b, II_a and II_b). Specific optoelectronic devices (photovoltaics, light emitting diodes, spintronics, etc.) demand specific charge carrier property that originates due to different types of band alignments. In this study, we have proposed a machine learning technique to classify 2D perovskites based on their band alignment types using molecular and elemental features. Our proposed model can successfully classify type I–II, type I_a–I_b and type II_a–II_b using binary classification and all four types using multiclass classification. We have also formulated an equation for determining the probability of the different band alignment types based on the contribution coefficients of the considered features. We believe such an interpretable glass-box model can open a new paradigm for the study of electronic properties of 2D perovskite materials.