Accelerated design of stable halide perovskite heterostructure film in hostile condition via surface modifier

The bulk halide perovskite heterostructure has been adopted for perovskite-based optoelectronic devices owing to their interesting interfacial phenomena; however, most of them are unstable in hostile condition and the extremely large virtual design space causes problems for traditional trial-and-error methods to optimize them. In this study, we combine photoelectrochemical experiments, machine learning and first-principles calculations (Exp + ML + DFT) to accelerate the design of halide perovskite heterostructures in aqueous solution, in an effort to improve stability of halide perovskite materials in hostile condition (prototypical CH ₃NH ₃PbI ₃ quickly deactivates in water within seconds). The optimization of halide perovskite heterojunction with surface modifier is speeded up via machine learning using extra tree algorithm, and a large virtual design space consisting of halide perovskite heterostructure composites is predicted. Five new water-stable perovskite heterostructure candidates predicted via machine learning are experimentally validated, and the first-principles calculations reveal the detailed atomic and electronic structures of an exemplar stable heterostructure candidate. This study demonstrates the effectiveness of machine learning to accelerate the design of stable halide perovskite heterostructure materials in hostile condition.