Machine-Learning-Assisted Screening of Interface Passivation Materials for Perovskite Solar Cells

Interface passivation using an ammonium salt can effectively improve the power conversion efficiency (PCE) of perovskite solar cells (PSCs). Despite significant PCE improve-ment achieved in previous studies, the selection criteria for ammonium salts are not fully understood. Here we apply a machine-learning (ML) method to investigate the relationship between the molecular features of ammonium salts and the PCE improvement of PSCs. We establish an ML model using an experimental data set of 19 salts to predict the PCE improve-ment after passivation. Three molecular features (hydrogen bond donor, hydrogen atom, and octane-water partition coefficient) are identified as the most important features of selecting an ammonium salt for passivation. The ML model is further used to screen ammonium salts from a pool of 112 salts in the PubChem database. FAMACs and FAMA-based PSCs fabricated with a model-recommended salt (2-phenylpropane-1-aminium iodide) achieve PCEs of 22.36% and 24.47%, respectively.

Automated summary

Interface passivation using an ammonium salt improves the power conversion efficiency of perovskite solar cells. A machine-learning method is used to investigate the relationship between the molecular features of ammonium salts and the efficiency improvement. Hydrogen bond donor, hydrogen atom, and octane-water partition coefficient are identified as important features for selecting an ammonium salt. The ML model is used to screen salts and achieves high PCEs for FAMACs and FAMA-based PSCs.