Bridging First-Principles Materials Predictions and Experimental Solar Cell Performance using an AI-Guided Multiscale Framework

First-principles modeling has become a powerful tool for investigating material properties, placing computational methods at the forefront of modern materials discovery. However, the gap between atomistic predictions and real-world performance limits their ability to drive technological progress, which continues to rely on extensive experimental efforts despite increasing automation of research workflows. Herein, we present an AI-guided multiscale framework that couples density functional theory (DFT) with device-level simulations to connect predicted halide perovskite material properties to solar cell performance, accurately capturing experimentally observed data. First, we establish a DFT-based approach to predict key optoelectronic properties of inorganic perovskites. We then incorporate them into a drift-diffusion model to predict composition-dependent current–voltage behavior, with Bayesian optimization subsequently refining the parameter sets to reproduce experimentally measured trends across two perovskite compositions. Finally, we employ Monte Carlo sampling to construct probability distributions of inferred solar cell parameters, revealing the physical mechanisms by which Sn incorporation enhances the photovoltaic response of CsPbIBr2 solar cells. By translating atomic-scale calculations into device-level performance predictions, this work establishes a proof of concept for targeted, computation-driven materials screening to accelerate the development of next-generation photovoltaic technologies.