Revealing processing stability landscape of organic solar cells with automated research platforms and machine learning

Abstract

Abstract We use an automated research platform combined with machine learning to assess and understand the resilience against air and light during production of organic photovoltaic (OPV) devices from over 40 donor and acceptor combinations. The standardized protocol and high reproducibility of the platform results in a dataset of high variety and veracity to deploy machine learning models to encounter links between stability and chemical, energetic, and morphological structure. We find that the strongest predictor for air/light resilience during production is the effective gap Eg,eff which points to singlet oxygen rather than the superoxide anion being the dominant agent in degradation under processing conditions. A similarly good prediction of air/light resilience can also be achieved by considering only features from chemical structure, that is, information which is available prior to any experimentation.

Keywords

• air stability
• automated screening
• donor/acceptor combinations
• machine learning
• organic solar cells