Using Design of Experiment and Machine Learning Approaches to Optimize the Effect of Solvent Additives and Processing Parameters on PM6:Y6 Organic Photovoltaics

Bulk heterojunction organic photovoltaics (OPVs) have emerged as a promising alternative to inorganic solar cells over the past two decades due to their properties such as flexibility, semi-transparency, lightweight and low-cost. High-efficiency OPV devices have been obtained with the development of non-fullerene acceptors (NFAs), which show strong absorption, tunable properties, low voltage losses, and high current generation. Over the past years, with a wide bandgap polymer PM6 and a non-fullerene acceptor Y6 blend, PM6: Y6 OPV cells have surpassed 18% power conversion efficiency (PCE). The ideal morphology of the active layer plays an important role in the performance of OPVs, and many approaches such as thermal annealing, solvent vapor annealing, and incorporation of additives are effective ways to control the nanostructure of donor/ acceptor domains in the active layer for efficient exciton dissociation and charge-transport. Due to the large number of components and processing conditions, OPVs are challenging to optimize. The combination of Design of Experiment (DoE) and Machine Learning (ML) is a cost-effective and fast approach to accelerate the development of OPVs. This approach samples the parameter space in a rigorous, systematic way to find the best possible performance for a given range of variables. In this study, the effects of the donor fraction, solution concentration, thermal annealing temperature, spin-speed of spin coating and the amount of different additives, such as ferrocene and acetone, on device performance of PM6:Y6 based OPV cells were optimized simultaneously using a combination of DoE and ML approaches. ML algorithms were performed to analyze the multivariable datasets and, the effect of each processing parameter on PCE was investigated and mapped out to visualize the relationship to PCE. This approach provides a methodology for establishing the effect of processing and composition perturbations on the device performance over a wide range of conditions.