Resolving carrier kinetics in perovskite through deep-learning-assisted photoluminescence analysis

Understanding charge carrier recombination kinetics is vital for optimizing perovskite materials and devices. Time-resolved photoluminescence (TrPL) is widely used to investigate carrier kinetics, but kinetic parameters inferred from TrPL alone often suffer from intrinsic non-uniqueness. Therefore, quantitatively resolving complex multichannel recombination processes remains challenging. Here, we demonstrate the non-uniqueness problem in TrPL. Through incorporating TrPL with excitation-dependent photoluminescence quantum yield (PLQY), the non-uniqueness problem is substantially alleviated. Based on this insight, we develop a deep learning (DL) framework to extract kinetic parameters from joint TrPL–PLQY profiles. This framework avoids repeated iterative parameter optimization for each new dataset within the predefined kinetic model. The DL model is trained on a large dataset generated from physically grounded simulations. Six physically relevant key kinetic parameters spanning several orders of magnitude, namely, effective external band-to-band electron-hole bimolecular recombination coefficient (γeh), electron capture coefficient (γec), hole capture coefficient (γhc), trap density (NT), electron emission coefficient (γee), and Auger recombination coefficient (γAug), can be rapidly predicted. High accuracy is achieved, with coefficients of determination of 0.996, 0.938, 0.969, 0.988, 0.905, and 0.973 for the respective parameters. The framework is further validated experimentally. The DL-predicted kinetic parameters are physically consistent with device-level characterization results. Moreover, when working with PLQY, we identify a meaningful TrPL acquisition window (0–2 μs) and decay cutoff (at 0.01), providing guidance toward standardized TrPL characterization protocols. The proposed framework provides a fast and quantitative approach for disentangling complex recombination pathways. Further, this work represents a significant step toward data-driven analysis of perovskite semiconductor kinetics. Charge carrier recombination in perovskites is difficult to quantify because photoluminescence decay alone gives non-unique results. Chen et al. combined decay and quantum-yield measurements with deep learning to predict six kinetic parameters and standardize measurements.