Assessment of Reproducibility, Hysteresis, and Stability Relations in Perovskite Solar Cells Using Machine Learning

Abstract

Herein, the hysteresis and reproducibility of perovskite solar cells and their relations with power conversion efficiency and long‐term stability are analyzed using machine learning tools. A hysteresis dataset containing 387 cells from 194 articles in the literature is constructed and analyzed using association rule mining, while the reproducibility is analyzed using the pooled variance of 24 142 cells from 438 articles. It is found that mixed cation perovskites, two‐step spin coating or multiple spin coating in one step, dimethylformamide + dimethyl sulfoxide as precursor solution, poly[bis(4‐phenyl)(2,4,6‐trimethylphenyl)amine as hole transport layer (HTL), lithium bis(trifluoromethanesulfonyl)imide + 4‐tert‐butylpyridine + tris(2‐(1H‐pyrazol‐1‐yl)‐4‐tert‐butylpyridine) cobalt(III) as HTL dopant, and carbon as back contact are found to be beneficial for both low hysteresis and high reproducibility in regular (n–i–p) cells. In addition to the perovskite material and deposition techniques mentioned earlier, the toluene as antisolvent, bathocuproine as electron transport layer interlayer, and Ag as back contact are found to have positive impacts in inverted (p–i–n) cells. It is also found that those factors are also highly relevant for power conversion efficiency and stability, clearly relating these four most commonly discussed performance measures for perovskite cells.