Machine learning and density functional theory simulation of the electronic structural properties for novel quaternary semiconductors

Abstract

In order to accelerate the application of quaternary optoelectronic materials in the field of luminescence, it is crucial to develop new quaternary semiconductor materials with excellent properties. However, faced with vast alternative quaternary semiconductors, traditional trial-and-error methods tend to be laborious and inefficient. Here, we combined machine learning (ML) with density functional theory (DFT) calculation to predict the bandgaps of 2180 quaternary semiconductors, most of which were undeveloped but environmentally friendly. The evaluation coefficient (R²) of the model using a random forest algorithm was up to 0.93 in ML. Four novel quaternary semiconductors with direct bandgaps: Ag₂InGaS₄, AgZn₂InS₄, Ag₂ZnSnS₄, and AgZn₂GaS₄, were selected from the ML model. Then their electronic structures and optical properties were further verified and studied by DFT calculations, which demonstrated that the four quaternary semiconductors had direct bandgaps, a small effective mass, and a large exciton binding energy and Stokes shift. Our calculation could significantly speed up the discovery of novel optoelectronic semiconductors and has a certain reference value for the study of luminescent materials and devices.