Navigating high-dimensional processing parameters in organic photovoltaics via a multitier machine learning framework

Area of Science:

• Materials Science
• Renewable Energy
• Computational Chemistry

Background:

• Optimizing organic photovoltaic (OPV) devices involves complex, interdependent processing parameters that dictate bulk heterojunction morphology.
• A significant challenge in OPV research is the high-dimensional nature of fabrication variables and their impact on device performance.

Purpose of the Study:

• To develop a data-driven machine learning framework for rational optimization of OPV photoactive layers.
• To create a standardized database integrating experimental results, fabrication parameters, and device efficiencies.

Main Methods:

• Construction of a comprehensive database of donor/acceptor pairs and nine key fabrication parameters.
• Development of a three-tiered machine learning strategy using gradient boosting regression trees, progressing from baseline to global optimization models.
• Validation of the machine learning model on 78 external systems with previously unseen components.

Main Results:

• The global nine-parameter optimization model achieved a Pearson correlation of >0.9 and >80% success rate in identifying optimal multiparameter configurations.
• The model demonstrated robust generalization, with >75% accuracy in predicting optimal conditions for individual parameters on external systems.
• The framework successfully consolidates over a decade of experimental data for efficient analysis.

Conclusions:

• A practical, data-driven machine learning framework can significantly accelerate the rational optimization of OPV photoactive layers.
• The developed tiered approach effectively captures parameter synergies for improved predictive accuracy.
• This methodology offers a scalable solution for navigating complex parameter spaces in organic electronics research.