Accelerating The Development of Thin Film Photovoltaic Technologies: An Artificial Intelligence Assisted Methodology Using Spectroscopic and Optoelectronic Techniques

Thin film photovoltaic (TFPV) materials and devices present a high complexity with multi-scale (from nm to μm), multi-layer and multi-element structures and with complex fabrication procedures. To deal with this complexity, the evaluation of the compositional, structural, morphological, optical, and electrical properties, among others, of TFPV devices is critical to generate a model that allows pushing forward their development and optimization, especially in terms of PV performance. However, the intrinsic complexity of TFPV materials and devices requires approaching their research in a holistic way. This involves the performance of a holistic characterization approach which presents three main challenges: 1) performing a systematic combinatorial characterization (using different characterization techniques) of materials and devices, 2) automating characterization (data acquisition) for generating a statistically relevant amount of information (big data), 3) generating automatized, fast, and multidimensional data processing strategies.<br/><br/>In this work, an approach to solve these three challenges is presented through the creation of a modular and highly customizable automatized characterization platform and methods that provide a fast, holistic, and non-destructive characterization of complex material and devices. The above-mentioned challenges are faced up by the platform in the following way:<br/><br/><b>1) Systematic heterogeneous characterization –</b> integrating dedicated sensors for the evaluation of composition, morphology, crystal structure, thickness, and optoelectronic parameters by different techniques (X-ray fluorescence, Raman, photoluminescence and UV-Vis-NIR reflectance spectroscopies, visual imaging, elastic light scattering, I-V, EQE, electroluminescence).<br/><b>2) Automatize data acquisition –</b> developing dedicated software and protocols for systematic and automatized data acquisition and a data storage structure for efficient data analysis.<br/><b>3) Automatized and fast multidimensional data processing –</b> developing flexible methodologies and algorithms to automatize data conditioning and processing using statistical and AI-assisted data analysis using machine learning algorithms to accelerate the big data analysis and identify correlation patterns and processes to push forward the development of the TFPV technologies.<br/><br/>A real case for the accelerated research of Cu₂ZnSnSe₄-based TPV (CZTSe) devices is then used as practical example for the proposed platform. More than 20 samples (with 5x5 cm² size) are systematically studied as the initial step for large-scale production of CZTSe based PV devices. The devices have Glass/Mo/MoSe2/CZTSe/CdS/i-ZnO/ITO structure and small controlled and non-controlled fabrication processes variations are included in the samples. Each samples is discretized in 3x3 mm² individual solar cells (up to 196 cells per sample), which allows achieving a complete data library that aids obtaining a considerable amount of cell-by-cell information to implement statistically relevant analyses. A systematic characterization combining optical and optoelectronic techniques as described above is performed in each cell. The analysis of the data obtained allows correlating the physicochemical properties of the TFPV materials with variations in device performance. This knowledge provides key information to understand the role of different materials and layers properties, their interrelations and, finally, the current technology limitations. This information is strongly relevant for the future development of the CZTSe technology towards high efficiencies and its future industrialization. Moreover, the proposed platform opens a way for the development of a process monitoring tool for controlling the production of TFPV devices that is easily adaptable to other optoelectronic technologies.