Science & Technology (Commonwealth Union) – Tandem solar cells utilizing perovskite semiconductors exhibit superior sunlight-to-electricity conversion efficiency compared to traditional silicon solar cells. To propel this technology toward market readiness, enhancements in stability and manufacturing processes are imperative. Researchers from the Karlsruhe Institute of Technology (KIT), in collaboration with two Helmholtz platforms – Helmholtz Imaging at the German Cancer Research Center (DKFZ) and Helmholtz AI – have achieved a breakthrough in predicting the quality of perovskite layers and, consequently, the resulting solar cells. Leveraging Machine Learning and innovative Artificial Intelligence (AI) methods, they can assess the quality based on variations in light emission during the manufacturing process. These findings, instrumental in refining manufacturing processes, are detailed in the publication in Advanced Materials.
Perovskite tandem solar cells combine a perovskite solar cell with a conventional counterpart, often silicon-based. Recognized as a next-generation technology, these cells currently boast an efficiency exceeding 33 percent, a significant improvement over conventional silicon solar cell. Not only do they leverage cost-effective raw materials, but they also offer ease of manufacturing. Achieving this remarkable efficiency entails the production of an extremely thin, high-quality perovskite layer, with a thickness merely a fraction of a human hair.
“Manufacturing these high-grade, multi-crystalline thin layers without any deficiencies or holes using low-cost and scalable methods is one of the biggest challenges,” said tenure-track professor Ulrich W. Paetzold who carried out the study at the Institute of Microstructure Technology and the Light Technology Institute of KIT. Even within what seems like perfect lab conditions, there may be unknown areas that result in variations in semiconductor layer quality: “This drawback eventually prevents a quick start of industrial-scale production of these highly efficient solar cells, which are needed so badly for the energy turnaround,” explains Paetzold.
In the pursuit of understanding the variables influencing coating, an interdisciplinary collaboration has emerged, uniting perovskite solar cell experts from KIT with Machine Learning and Explainable Artificial Intelligence (XAI) specialists from Helmholtz Imaging and Helmholtz AI at DKFZ in Heidelberg. The team has devised AI methods that train and scrutinize neural networks using an extensive dataset comprising video recordings capturing the photoluminescence of thin perovskite layers during the manufacturing process. Photoluminescence denotes the radiant emission from semiconductor layers excited by an external light source. Lukas Klein and Sebastian Ziegler from Helmholtz Imaging at DKFZ elucidate, “Since even experts could not discern specific characteristics on the thin layers, the concept arose to employ an AI system for Machine Learning (Deep Learning) to identify hidden indicators of well or poorly coated layers within the vast dataset of video data.”
In order to sift through and analyze the widely dispersed cues generated by the Deep Learning AI system, the researchers subsequently employed Explainable Artificial Intelligence methods.
The researchers conducted experimental observations revealing that photoluminescence undergoes variations throughout the production process, exerting an influence on coating quality. Lukas Klein and Sebastian Ziegler underscore the pivotal role of employing Explainable Artificial Intelligence (XAI) methods, emphasizing their unconventional approach in systematically identifying factors critical for achieving high-grade solar cells. They assert that this departure from the norm, where XAI is typically utilized as a safeguard in AI model construction, represents a paradigm shift. Klein and Ziegler express their enthusiasm, stating, that gaining highly relevant insights in materials science in such a systematic way is a totally new experience.
The researchers harnessed the conclusions drawn from the photoluminescence variations to propel their investigation forward. Following the targeted training of neural networks, the AI demonstrated the ability to predict the efficiency level of each solar cell, distinguishing between low and high efficiency based on specific variations in light emission at different stages of the manufacturing process.
