Machine learning regression analyses of intensity modulation two-photon spectroscopy (Mlim) in perovskite microcrystals

Abstract: Perovskite thin films hold great promise for optoelectronic applications, such as solar cells and light emitting diodes. A challenge is that defects are unavoidably formed in the material. Thorough understanding of the defect formation and their dynamics has proven challenging based on traditional spectroscopy. Here we integrated the functional intensity modulation two-photon spectroscopy with artificial intelligence - enhance data analyses to obtain a deep understanding of defect-related trap states within perovskite microcrystals. We introduce a novel charge carrier recombination dynamics model that comprehensively includes exciton and electron-hole pair photoluminescence (PL) emissions, as well as the trapping and detrapping equilibrium dynamics. By varying parameters in the dynamic model, a large pool of the temperature dependent intensity modulation PL spectra can be simulated by solving the ordinary differential equations in the charge carrier dynamics model. Then, tree-based supervised machine learning methods and ensemble technique -- regression chain have been used to optimize the Machine learning intensity modulation spectroscopy (Mlim), which helps to determine the parameters of the charge carrier dynamics model based on the temperature dependent intensity modulated PL spectra in perovskite. And the reliability of the Mlim predicted trap property parameters is confirmed by directly comparing the Mlim-retrieved intensity modulation spectra with experimental data. Besides, our approach unravels valuable insights into PL emissions, including those from excitons and free electron-hole pairs, but also provides details of trapping, detrapping, and nonradiative depopulation processes, offering a comprehensive understanding of perovskite material photophysics. This study suggests that Mlim applications hold promise for studying various photoactive devices.