Understanding Scalable Perovskite Solar Cell Manufacturing with Explainable AI

Published: 27 Oct 2023, Last Modified: 15 Nov 2023NeurIPS XAIA 2023EveryoneRevisionsBibTeX
TL;DR: Combining computer vision and XAI, our study deciphers complex perovskite thin-film data, linking thin-film formation processes to solar cell performance and advancing energy materials science.
Abstract: Large-area processing of perovskite semiconductor thin-films is complex and evokes unexplained variance in quality, posing a major hurdle for the commercialization of perovskite photovoltaics. Advances in scalable fabrication processes are currently limited to gradual and arbitrary trial-and-error procedures. While the in-situ acquisition of photoluminescence videos has the potential to reveal important variations in the thin-film formation process, the high dimensionality of the data quickly surpasses the limits of human analysis. In response, this study leverages deep learning and explainable artificial intelligence (XAI) to discover relationships between sensor information acquired during the perovskite thin-film formation process and the resulting solar cell performance indicators, while rendering these relationships humanly understandable. Through a diverse set of XAI methods, we explain not only *what* characteristics are important but also *why*, allowing material scientists to translate findings into actionable conclusions. Our study demonstrates that XAI methods will play a critical role in accelerating energy materials science.
Submission Track: Full Paper Track
Application Domain: Natural Science
Survey Question 1: Currently, advances in material science fabrication processes are limited by gradual trial-and-error optimization. Our interdisciplinary study opens up a new path for tackling this long-standing issue by first leveraging deep learning to learn the relationships between video data acquired during fabrication and resulting performance indicators and subsequently employing explainable artificial intelligence (XAI) methods to render these relationships humanly understandable. As a result, the study demonstrates that XAI methods streamline the optimization of fabrication processes, reducing trial-and-error approaches, and offer versatile applications across fields like electronics and energy, as evidenced by our insights from solar cell manufacturing data.
Survey Question 2: In our study, we aim to optimize a solar cell production process that has nominally the exact same production parameters such as material composition, temperature, etc. but still yields solar cells of varying quality due to non-measurable small human and technical errors. Classical statistical analyses cannot be applied in this case due to the lack of varying independent variables. Only the video data recording the actual production contains information about the actual parameters which due to its high dimensionality and complexity can only be analyzed using our combination of computer vision and XAI methods.
Survey Question 3: For feature importance/saliency maps, we utilized Expected Gradients, Integrated Gradients, Guided Backpropagation, and Guided GradCAM, for counterfactual examples Genetic Counterfactuals (GeCo) and for concept testing the Testing of Concept Activation Vectors (TCAV).
Submission Number: 13