Defect formation in CsSnI₃ from DFT and machine learning methods

Sn-based perovskites are promising alternatives to Pb-based perovskites due to their lower toxicity. However, their performance is limited by p-type self-doping. Substitutional doping on the cation site of CsSnI₃ has been proposed to suppress the p-type doping. In this study, we combine density functional theory calculations with machine learning to develop predictive models for the formation energy and charge transition levels of substitutional defects on the Sn site. Using elemental and structural properties as features, we trained models including gaussian process regression, kernel ridge regression, and random forest regression. Our results identify key features influencing defect properties and provide optimized models to predict dopant energetics in CsSnI₃.