AI robot with rapid imaging improves perovskite solar cell fabrication

Researchers at the University of Osaka have developed an automated platform that combines robotics, machine learning and rapid imaging to improve the fabrication of perovskite solar cells. The system addresses a persistent challenge in thin-film photovoltaics: the quality of a perovskite layer is highly sensitive to small fluctuations in temperature, humidity, deposition conditions and drying behavior. Because the liquid coating transforms quickly into a crystalline film, traditional trial-and-error methods often fail to reveal which process variables matter most. The platform, described in ACS Nano, coordinates automated film deposition with optical imaging and environmental measurements. It captures film color changes on a subsecond timescale and resolves the motion and impact of antisolvent droplets within milliseconds. By observing these rapid transformations in real time, the team linked specific dynamics during film formation to the quality of the resulting perovskite material. Lead author Ryosuke Kado said the goal is to measure what happens to the film while it is forming, rather than relying solely on post-fabrication testing. Senior author Akinori Saeki added that automation enables data collection beyond what is practical manually, while machine learning helps identify relationships among many variables. The researchers applied the platform to study solution-to-film crystallization, with particular attention to antisolvent droplets, which are commonly used to promote film formation. Their analysis identified an optimal range of droplet behavior. Using these insights to refine the coating process, the team produced a lead-based perovskite solar cell with a power conversion efficiency of 24.4 percent. The work demonstrates that real-time observation and automated analysis can connect processing conditions directly to device performance. More broadly, the system offers a systematic route toward better-perovskite fabrication by converting fast, hard-to-observe manufacturing events into actionable data. The team said the approach could help photovoltaic research move beyond empirical optimization and support the development of more reliable and practical solar cells.