Researchers from Peking University have used artificial intelligence to accelerate the discovery of high-performance halide perovskite materials for photovoltaics.
Their recent study introduces a machine learning (ML) model that accurately predicts bandgap, conduction band minimum (CBM), and valence band maximum (VBM) - critical electronic parameters determining solar cell efficiency and performance. The research directly addresses a key bottleneck in perovskite PV development: the need for faster, cost-effective identification of stable, lead-free, and high-efficiency materials.
Traditionally, such screening has relied on high-throughput experimentation or density functional theory (DFT) simulations - both laborious and energy-intensive. The team’s ML approach achieved strong predictive performance (R² > 0.80, MAE < 0.29 eV) across thousands of halide perovskite candidates.
Of particular value to solar technologists, the model applies to both inorganic and hybrid organic-inorganic perovskites, offering broad utility across research pipelines. Using SHapley Additive exPlanations (SHAP), the team also decoded the key chemical and structural features that influence band alignment - vital for designing high-efficiency tandem and multi-junction architectures.
This work provides practical tools for PV labs aiming to improve spectral absorption, minimize recombination losses, and boost cell voltages.
