Robocoater: Automated, Multi-Modal Optical Characterization Platform for Performing Closed-loop Bayesian Optimization of Thin-Film Hybrid Perovskite for PV Application

N. Woodward, B. Guo, M. Chauhan, M. Abolhasani, K. Rayes, A. Amassian
North Carolina State University,
United States

Keywords: automation, machine learning, camera vision, Bayesian optimization, thin film fabrication, perovskite, platform


Hybrid metal-halide perovskites are a promising material for photovoltaics that have made large strides in power conversion efficiency (PCE) in the past decade due to being an affordable, solution-processible, tunable direct bandgap material. Spin coating is a widely adopted technique for the fabrication of perovskite thin films; however, it is a very strenuous, manual process that can vary person-to-person in a given lab. In addition to that, anti-solvent treatment is a crucial step in the fabrication process. Different anti-solvents and drip parameters are required for varying perovskite systems to achieve the optimal thin film for high PCE devices. Optimization of one-step spin coating of a hybrid perovskite system with an anti-solvent drip is a multiparametric problem that requires many human hours and resources. Here we present a fully automated spin coating platform, the RoboCoater, which allows precise control of processing conditions like spin speed, anti-solvent drip timing, and drip volume to achieve accurate and reproducible results that are not feasible by humans. In addition, this platform has integrated in-situ absorbance and photoluminesce measurement capabilities that are synchronized with the spin coating experiment to determine film quality. We have also utilize Bayesian Optimization to reduce the time spent needing to optimize the multiple process parameters of perovskite thin-films in a high-throughput, closed-loop manner to reduce the time and material cost to optimize the perovskite active layer to achieve a higher power conversion efficiency perovskite solar cell. This dramatically reduces the number of person-hours needed to optimize a single anti-solvent for a given perovskite system and allows us to more quickly screen various anti-solvents to find the best performing process parameters. Overall we have built a small, modular 3D printed scientific platform that is much more affordable than the large, commercial optical characterization platforms that can run autonomous experimentation for a wide range of solution processable materials. The RoboCoater defines standardized processing conditions in different research labs to achieve repeatable, peer-executed experimentation to help the community advance together.