Machine learning of kesterite materials using cost-effective hybrid density functional theory

Kesterite materials have received much attention for earth-abundant photovoltaic applications, however, their record cell efficiency is still lower than other thin film solar cells. We performed high-throughput calculations using hybrid density functional theory to find an alternative absorber material. The atomic structures were cost-effectively optimized by employing a downsampled k-point grid for the Fock exchange potential, and therefore the electronic band gap ([Formula presented]) was obtained in good agreement with the conventional hybrid calculation. Based on hybrid DFT calculations, we develop machine learning (ML) models to predict the E_g and suggest the empirical equation for E_g with the accuracy of the root-mean-square-error of 0.137 eV and R² of 0.921. We found that the distance between group-IV and group-VI elements is important to estimate E_g. Using the developed ML models, we searched materials with predicted E_g within the optimum range of 1.0−1.5 eV and also compared their E_g with hybrid functional calculations. Based on our ML and DFT calculations, we suggest nine materials whose ML predicted and DFT calculated E_g in the optimum range for photovoltaic applications.