Performance analysis of perovskite solar cells in 2013–2018 using machine-learning tools

This work presents a review and analysis of a database containing 1921 solar cell device performance data points extracted from 800 publications on the (organo)-lead-halide perovskite solar cell published between 2013 and 2018. The aim is to review the literature to capture the major patterns in the past and analyze the database using machine learning tools to develop heuristics and models to predict the cell performance. The factors related to the fabrication methods of metal-halide perovskite cells like the utilization of mixed cation perovskites, solvent engineering (such as mixed DMF+DMSO solvent), use of chlorobenzene as anti-solvent, employing two or three times spinning as well as factors pertaining the device architecture like utilization of LiTFSI+TBP+FK209 as hole transport layer (HTL) additive and tin oxide (SnO2t) as electron transport layer (ETL) emerge as the significant factors leading to highly efficient device of regular (n-i-p) architecture devices. The use of poly-triarylamine (PTAA) as HTL, mixed cation perovskite and bathocuproine (BCP) as ETL interlayer seems to be important for inverted (p-i-n) cells. It is also clear that the evolution of maximum efficiencies obeys the S-shape logistic curve. Two local S-shaped patterns exist in the data; the first one seems to be completed in 2014–2015, the second curve, which apparently started due to the effective utilization of solvents and anti-solvents, covers the later years, and it is reaching its maturity; the last two records may be indicators of a new trend.