Explainable Artificial Intelligence Driven Methodology for Accelerated Research of Complex Systems: Case Study of Thin‐Film Photovoltaic Kesterite‐Based Technology

Abstract Materials research increasingly targets systems of higher structural and compositional complexity, which hinders their understanding and slows down technological development. To address this, an explainable artificial intelligence (XAI)‐driven methodology is presented, which is designed to accelerate research in complex systems based on novel materials by identifying the origins of performance limitations and reproducibility issues. Thin‐film photovoltaic devices based on kesterites (Cu₂ZnSnSe₄) are used as a case study due to their multilayer architecture and strong sensitivity to process parameters. More than 1750 solar cells are fabricated under nominally identical conditions, with intrinsic fluctuations leading to performance dispersion. Each device is characterized through automated, multimodal techniques including Raman spectroscopy, photoluminescence, X‐ray fluorescence, spectral reflectance, and hyperspectral imaging. The resulting dataset is used to train Principle Component‐Linead Discriminant Analysis (PC‐LDA) algorithm for two different types of target values: discrete and continuous. In‐depth analysis of the obtained AI models by different approaches identified structural properties of the i‐ZnO and CdS layers as the dominant factors affecting both efficiency and process reproducibility. This feedback is experimentally validated through modified fabrication processes, resulting in improved device performance. The proposed methodology enables a closed‐loop research approach, offering explainable physical insights and paving the way for the implementation of self‐driving laboratories.