Defect formation in CsSnI₃ from density functional theory and machine learning

Sn-based perovskites as low-toxicity materials are actively studied for optoelectronic applications. However, their performance is limited by p-type self-doping, which can be suppressed by substitutional doping on the cation sites. In this study, we combine density functional theory (DFT) calculations with machine learning (ML) to develop a predictive model and identify the key descriptors affecting formation energy and charge transition levels of the substitutional dopants in CsSnI₃. Our DFT calculations create a dataset of formation energies and charge transition levels and show that Y, Sc, Al, Zr, Nb, Ba, and Sr are effective dopants that pin the Fermi level higher in the band gap, suppressing the p-type self-doping. We explore ML algorithms and propose training a random forest regression model to predict the defect formation properties. This work shows the predictive capability of combining DFT with machine learning and provides insights into the important features that determine the defect formation energetics.