Combinatorial and Machine Learning Techniques for Complex Thin Film Photovoltaics: Accelerated Research and Process Monitoring Methodologies

Abstract

[eng] Materials and thin film photovoltaic (TFPV) devices exhibit a high level of complexity, characterized by their multi scale, multi layer, and multi element structures, along with complex manufacturing processes. Addressing this complexity requires a comprehensive assessment of their physicochemical properties, which is crucial to establish an integrated model that outlines strategies for their development and improvement. However, this evaluation process is notably time intensive and demands a high level of expertise. In this context, the integration of Combinatorial Analysis (CA) and Artificial Intelligence (AI) emerges as a powerful tool, significantly accelerating the advancement of these materials and devices. Unfortunately, s everal obstacles stand between researchers and these tools, including the absence of automated measurement systems, ambiguity in CA and ML workflows, a deficiency in programming skills, and skepticism regarding the reliability of ML results. This thesis , presented as article explores these challenges, focusing on TFPV, by proposing a methodology and making tools available that overcomes these obstacles by harmonizing CA and ML within a user friendly framework. This is achieved by first establish ing a clear and structured approach to apply CA and ML, followed by providing accessible, open source, and freely available tools. These tools are designed not only to facilitate the implementation of the proposed methodology but also to seamlessly integra te into existing research workflows. The thesis commences with the motivation, where it accentuates the need of rapidly transitioning to renewable and sustainable sources of energy, where solar photovoltaic (PV) technology is a major contributor. It emphas izes that TFPV can help with faster integration of PV due to their versatility, able to be used in scenarios where traditional technologies cannot. Then, the text makes an introduction to the current state of the PV market and projecting its application, to then go in more details about the fundamentals of TFPV technology. Then, an overview of AI and its role in energy research is presented, also introducing its fundamental working principles and advanced concepts . After , research problems and gaps in the field are identified and the thesis’ objectives are listed, aim ing to solve those problems and fill those knowledge gaps. Then, the methodology used in this research is laid out, specifying procedures , equipment, materials, software, and tech niques . After the methodology, the four published articles are incorporated . The first article describes and demonstrates a novel characterization methodology based on normal reflectance measurements and ML algorithms for precise, low cost, and scalable assessment of the thickness of AlO x nanometric layers . The second article presents a combinatorial approach for the analysis of CZGS (Cu 2 ZnGeSe 4 ) solar cells using the methodology in the first article , providing a deep understanding of the stoichiometric limits and point defects formation in the CZGS compound and the influence of parameters on the performance of the devices. The third article introduces spectrapepper , a Python library for streamlin ing the analysis of complex materials and devices, such as multi layered TFPV solar cells, using spectroscopy. And finally , the fourth article introduces pudu, a Python library designed to enhance the interpretability of ML models in spectroscopic data analysis, aiming to increase the transparency and scientific impact of ML results with sensitivity analysis After presen ti ng the articles, further exploratory experiments with promising promising preliminary results are preliminary results are presented. Thispresented. This follow the follow the methodology in the methodology in the published articlespublished articles, being a natural follow up and extension of the methodology presented. , being a natural follow up and extension of the methodology presented. ThThuus,s, servingserving as the next steps to be followed as the next steps to be followed toto further advance in the development of further advance in the development of TFPVTFPV devices with the aid of devices with the aid of MLML.. In a final instance,In a final instance, the conclusions show how the the conclusions show how the objectives of the thesis are fulfilled with the objectives of the thesis are fulfilled with the presented work, and how the use of CA and presented work, and how the use of CA and ML can help the advancement of the TFPV field.ML can help the advancement of the TFPV field.

[spa] Los materiales y dispositivos fotovoltaicos de capa fina (TFPV) presentan un alto nivel de complejidad, caracterizados por sus estructuras multi-escala, multi-capa y multi-elemento, junto con procesos de fabricación complejos. Abordar esta complejidad requiere una evaluación exhaustiva de sus propiedades fisicoquímicas, la cual es crucial para lograr un modelo integral que delineé estrategias para su desarrollo y mejora. Sin embargo, este proceso de evaluación es notablemente intensivo en tiempo y exige un alto nivel de experiencia. En este contexto, la integración del Análisis Combinatorio (CA) y la Inteligencia Artificial (AI) surge como una herramienta poderosa, acelerando significativamente el avance de estos materiales y dispositivos. Desafortunadamente, numerosos obstáculos se interponen entre los investigadores y estas herramientas, incluyendo la ausencia de sistemas de medición automatizados, la ambigüedad en los flujos de trabajo para CA y ML, una deficiencia en habilidades de programación y el escepticismo respecto a la fiabilidad de los resultados de ML. Esta tesis, en forma de compendio de artículos, explora estos desafíos y desarrolla los objetivos en base a estos con enfoque en TFPV. Propone una metodología que supera estos obstáculos armonizando CA y ML dentro de un marco amigable para el usuario. Esto se logra estableciendo primero un enfoque claro y estructurado para aplicar CA y ML, seguido de la provisión de herramientas accesibles, de código abierto y de acceso libre. Estas herramientas están diseñadas no solo para facilitar la aplicación de la metodología propuesta, sino también para integrarse sin problemas en los flujos de trabajo de investigación existentes. El texto concluye examinando el cumplimiento de los objetivos, haciendo énfasis en como CA y ML ayudarán en el desarrollo de tecnología TFPV.