Advanced Characterization and Degradation Analysis of Perovskite Solar Cells using Machine Learning and Bayesian Optimization

In the race to design the next generation of solar cells, perovskites are widely seen as the likely candidate to push the performance of conventional silicon solar cells. Despite recent breakthroughs, this technology still has to overcome several challenges before it can be commercially deployed. With a vast array of material combinations and fabrication processes to choose from, researchers often spend years trying to optimize their solar cell design and understand performance bottlenecks. Herein, we combine physics modeling, machine learning, and experimentation to better understand the complex relationship between device performance and the underlying material parameters. Specifically, we use Bayesian inference and traditional machine learning techniques to extract material properties from simulated and experimental current-voltage curves of perovskite solar cells measured under indoor and outdoor conditions. This approach, which can be generalized to other semiconductor devices, allows us to gain valuable insight into photovoltaic performance and degradation without resorting to time-consuming and laborious characterization techniques, thus accelerating the pace of experimental research.