Concurrent Optimization of Device Architecture, Transport Layers, and Active Layer for Organic Photovoltaics by Machine Learning

Optimizing organic photovoltaic (OPV) devices requires a holistic understanding of all functional layers, hindered by multiscale, multiphysics interactions difficult to fully model traditionally. While machine learning (ML) excels in active layer development, a framework simultaneously addressing device architecture and transport layers remains underexplored. This work introduces a multicomponent ML framework for systematic analysis of all key subsystems. Three dedicated databases encompassing device architecture, transport layers, and active layer processing (subdivided into thermal and solvent vapor annealing) were constructed. For each database, tailored ML models were developed to decode relationships between subsystem-specific parameters, such as active area, transport layer thickness, and annealing conditions, and the final device efficiency. The ensemble method demonstrated strong predictive accuracy and reliable identification of optimal configurations. This concurrent, component-wise optimization establishes a new paradigm connecting material selection, interfacial engineering, and macroscopic design for rational development of high-performance OPV.