A machine learning quest to design benzodithiophene-based organic solar cell heterojunction layers with promising molecular packing

High-performance organic solar cells (OSCs) based on benzodithiophene (BDT) donors are a current challenge in the field of renewable energy. In this work, we started our machine learning (ML) journey to predict the fill factor (FF) values of 485 organic hole transport layer donors from literature, focusing on the identification of the most influential molecular descriptors. We found that the Random Forest algorithm was the best model for predicting FF values, and that fr_aryl_methyl and SlogP_VSA4 were the most influential descriptors. We then created a database of 11,878 new donors by retrosynthetic approach, and the FF values predicted up to 70 in the initial screening. After a further validity filtering step, a sub-set of 6,166 donors was selected for further analysis, the highest predicted FF in this sub-set being 67. The results presented here show the potential of ML to speed up the design of high-performance OSCs and provide insight into the molecular characteristics that dictate packing and efficiency of benzodithiophene-based donors. The findings of this study could help develop more efficient and stable OSCs, which would help realize the full potential of organic photovoltaics.