CEST members Jarno Laakso and Patrick Rinke, with collaborators from University of Turku and China, developed new machine learning-based methodology for rapidly predicting perovskite properties. This new approach accelerates computations and can be used to study perovskite alloys. These alloy materials contain many candidates for improved solar cell materials, but studying them has been difficult with conventional computational methods. The researchers demonstrated the effectiveness of the new approach by finding the most stable mixing fractions for an alloy of CsPbCl3 and CsPbBr3 perovskites. Having an efficient method for studying the stability of perovskite alloys is a key step towards engineering solar cells that are more resilient to degradation.
The same methodology that was applied to perovskites in this study can boost the discovery of other new alloy materials. After the initial success with their machine learning approach, Laakso and collaborators are looking into studying more complex perovskite alloys to discover solar cell materials that are highly efficient, nontoxic, and resilient to degradation.
The paper was published in Physical Review Materials doi.org/10.1103/PhysRevMaterials.6.113801.