摘要
2D lead-halide perovskites have garnered considerable attention owing to their superior environmental stability and tunable optoelectronic properties, which can be precisely controlled through varying quantum well (QW) width (denoted by the integer n). However, the commonly observed phenomenon of mixed QW width distributions poses a major obstacle to achieving optimal device performance, necessitating an in-depth understanding of how QW width distributions depend on chemical composition and thermodynamic stability. In this work, a robust machine learning (ML)-based energy model is developed, rigorously benchmarked against first-principles calculations, enabling extensive molecular-level simulations of 2D perovskites with butylammonium (BA) and phenethylammonium (PEA) spacer cations. Through hybrid Monte Carlo simulations capable of modeling significantly larger systems than first-principles methods, a universal and rapid evolution is demonstrated from initially homogeneous single-phase QW structures toward energetically favored mixed-phase distributions. Remarkably, the formation of these mixed phases arises primarily due to the enhanced thermodynamic stability of low-n layers, driven by the strong affinity of self-assembled spacer cations to the inorganic
framework compared with methylammonium cations. These findings highlight how ML-powered multiscale modeling provides unprecedented insights into complex 2D perovskite microstructures, thus offering valuable guidelines for the rational design and molecular engineering of next-generation perovskite-based optoelectronic devices.