Design of experiments with the support of machine learning for process parameter optimization of all‐small‐molecule organic solar cells

Abstract Traditionally, squaraine dyes have been studied and employed in biomedical research due to their excellent optical properties, and the molecules are being adopted in different research fields such as organic solar cells. In this study, we investigate correlations between solar cell performance and processing parameters of all‐small‐molecule bulk heterojunction solar cells comprising squaraine (SQ) as electron donor (D) and non‐fullerene small molecules (e.g., ITIC) as electron acceptor (A) with the help of machine learning (ML) and design of experiment (DoE) methods. Among the five predictive ML models tested with the selected parameters, the eXtreme gradient boosting model shows the satisfactory results with quite high coefficient of determination of 0.999 and 0.997 in training and testing sets, respectively. By measuring the contribution of each input variable to solar cell efficiency, four process parameters, that is, the total concentration, the ratio of D/A, the rotational speed of spin coating, and the annealing temperature, are found to be the key features strongly correlated to solar cell efficiency. From contour plots in DoE, the highest solar cell efficiency of approximately 5% can be predicted under the conditions of 15 mg mL −1 in solution concentration, a 1:2 mix ratio of D and A, rotational speeds ranging from 800 to 900 rpm, and annealing temperatures within 100–110°C. Using the suggested parameter conditions, we fabricated solar cells, achieving a quite high efficiency of approximately 4%. Besides the global optimization conditions, we also employ the solvent vapor annealing combination to the thermal annealing to facilitate further mobilization of molecules and more optimized microstructure of bulk heterojunction films, resulting in a further enhancement in solar cell efficiency of more than 20%.