|
Description
|
The dataset contains synthetic physics simulations and experimental photoluminescence measurements by HEUMANN, RUDOLPH, HUANG, KIRCHARTZ and DREESSEN (Forschungszentrum Jülich). It supports neural-network-accelerated Shockley–Read–Hall (SRH) trap characterisation of semiconductor thin films. The methodology is described in “Heumann et al. (2026): Rapid Parameter Estimation from Photoluminescence Decays of Halide Perovskite Thin Films, arXiv:2609.18438”. The dataset includes an archived snapshot of the Python code. The code extracts SRH trap parameters (N_t, E_t, τn, τp for one or two trap levels) from time-resolved (TRPL) and, optionally, steady-state (SSPL) photoluminescence measurements of semiconductor thin films. The recombination model includes radiative recombination. A neural-network surrogate replaces the rate-equation (ODE) solver for the differential decay time (τ_diff) as a function of the quasi-Fermi level splitting (QFLS). The pipeline works as follows: A physics-informed start-point estimate is followed by a CMA-ES fit on the network, which is then refined with the rate equations. The full Bayesian posterior is sampled with tempered sequential Monte Carlo. This gives credible intervals and the log evidence. Posterior predictive checks are run with both the network and the rate equations. For comparison, the same inference can be run with the rate equations instead of the network. With the surrogate, fit and posterior take seconds instead of hours. The most current version of the code is in the GitHub repository https://github.com/roheumann/rapid-trpl-bayes. At its core, the dataset provides a large synthetic training corpus of τ_diff vs. QFLS curves. They were computed by numerically solving the SRH transient rate equations over a quasi-random Sobol-sampled parameter space: Sampled parameters: excitation density, bandgap, radiative coefficient, and the trap parameters of one or two trap levels. Size: about 1.57 million curves for one trap and 2.10 million curves for two traps. Format: lossless compact HDF5 files, with a script that restores the original training files bit for bit. The two trained neural-network surrogates are included: one trap, and two traps with an additional head that predicts the QFLS window of each curve. Their test errors are σnn = 0.112 and 0.125 in log₁₀ τdiff. The experimental data are TRPL transients, SSPL intensity–QFLS data and an absorptance spectrum from two perovskite thin films: an 85:15 film (450 nm, E_g = 1.625 eV) a PI-treated 77:23 film (1000 nm, E_g = 1.67 eV) These serve as targets for parameter fitting and Bayesian uncertainty quantification. A noise-free synthetic one-trap data set, generated from the one-trap fit of the 85:15 film, is used to validate the method against a known ground truth. A further component is a drift-diffusion model of the TRPL experiment (Beer–Lambert excitation profile; MATLAB reference and Python port). It quantifies the model error σmodel of the 0D rate equations at the best fit of each film, at the measured absorption coefficient: 0.145 dec for the 85:15 film and 0.353 dec for the 77:23 film. Combined with the network error as σtrPL = √(σmodel² + σnn²), this sets the width of the Gaussian log-likelihood in the Bayesian inference (0.191 and 0.374 dec). The five-scenario estimate of the original submission (σ = 0.4936 dec) is kept for comparison. The data cover the relevant variables of the semiconductor photophysical recombination cycle, including trap density (N_t), trap energy level (E_t), electron and hole SRH lifetimes (τn, τp), radiative recombination coefficient (k_rad), bandgap energy (E_g), carrier excitation density, quasi-Fermi level splitting, differential decay time and photoluminescence quantum yield. (2026)
|
|
Keyword
|
perovskite solar cells, Bayesian inference, transient photoluminescence, steady-state photoluminescence, uncertainty quantificaiton, machine learning, Shockley-Read-Hall recombination, charge-carrier lifetime, parameter estimation |
|
Related Publication
|
@misc{heumann2026rapidparameterestimationphotoluminescence, title={Rapid Parameter Estimation from Photoluminescence Decays of Halide Perovskite Thin Films}, author={Robin Heumann and Toby Rudolph and Gaosheng Huang and Thomas Kirchartz and Chris Dreessen}, year={2026}, eprint={2609.18438}, archivePrefix={arXiv}, primaryClass={cond-mat.mtrl-sci}, url={https://arxiv.org/abs/2609.18438}, }
doi: 2609.18438
|