Paths towards high perovskite solar cells stability using machine learning techniques

This work aims to analyze the stability of Perovskite solar cells PSCs using machine learning (ML) techniques. An extremely randomized trees technique, trained with a dataset containing 1050 perovskite device samples with different materials, deposition methods and storage conditions, is used. Pushing by its non linearity and randomity, this approach is an intriguing choice for decreasing the variance of the total model. The effects of data inputs on the stability of the device are investigated by analysing the Decision Trees (DT) constituent of the Extra Trees (ET) while the feature importance technique was used for feature engineering. The two techniques findings are compared with previous experimental results for screening the most optimized manufacturing materials and storage conditions for long-term stability. For regular cells, TiO2/m-TiO2 as electron transport layer (ETL), (2D-3D) perovskite as active layer, P3HT and LiTFSi + TBP as hole transport layer (HTL) and HTL second layer, and Carbon as back contact were found to enhance the device stability with DMF + DMSO as precursor solution and Chlorobenzene as an anti-solvent solution. For inverted cells, BCP and PCBM, MAPBl3-xClx, NiO and DEA, Al back contact were found to improve stability. The obtained results provide evidence of the aptness of the proposed ML strategy in captivating the suitable combination of different layer materials, deposition methods, and storage conditions. Besides, the adopted method unveils the importance of manufacturing techniques in realizing efficient and stable solar cells.