The transition to a sustainable society must be carried out minimizing the environmental impact of energy production. Photovoltaic (PV) energy is a clean source with great potential that can be implemented from large industrial plants to self-consumption systems in homes. The current generation of PV systems focuses on integrating PV in diverse environments where mature technologies fall short, solving various integration needs such as tandem devices, semitransparency, flexibility, appearance, stability, and efficiency under variable lighting conditions. This increases the complexity of system design by involving multiple material properties that require precise control of the fabrication process, a challenge for both research development and industrial up-scaling. Accelerating the research and development of materials and devices requires a fluent, fast, and statistically relevant flow of data between the synthesis and characterization. However, the advanced laboratory characterization methods and data analysis are often expensive, slow and difficult to integrate between the production process steps.
This work presents the results of the Platform Zero project (HE 101058459), whose objective is to robotize, automate, and hybridize material characterization techniques through massive data acquisition systems and artificial intelligence (AI) analysis. Different thin film PV technologies are studied at different maturity levels: industrial (perovskite, CIGS), pre-industrial (new CIGS-based), and research (kesterite). Modules up to 30x30 cm2 are analyzed on both flexible and rigid substrates, evaluating how small disturbances in the fabrication process modify material properties such as thickness, composition, crystalline quality, structural defects, and doping, among others. A multimodal and multiscale characterization is performed combining data from different spectroscopic techniques (Raman, reflectance, transmittance, PL, EL, LIBS), optical methods (hyperspectral vision, TR-PL, thermography), and electrical measurements (IV, EQE). The statistically relevant results from the different characterization techniques help to implement strategies for the automation of data acquisition and fusion, and to generate a big database to develop a supervised method based on AI algorithms for the rapid, self-explainable, and massive data analysis. As a result, analysis rates of up to 0.2 m2/min for sheet-to-sheet and roll-to-roll processes were achieved.
The proposed methods based on multimodal characterization and AI provide fast and automate data acquisition and analysis, which accelerates the research of thin film PV materials and represents a further step in the development of self-driving laboratories. In addition, applying these methods to the in-line monitoring of the fabrication process allows to identify deviations from the standard process, optimizing the production efficiency and reducing the environmental impact of thin film PV research and industry.