In order to investigate the transient behaviour of charge carriers in semiconductors followingphotoexcitation, time-resolved photoluminescence (TRPL) is a powerful approach. Commonly, mono- orbiexponential models are used to analyse the TRPL data. These are simplistic models because they can onlybe used to extract one or two lifetimes from the data, neither of which necessarily have any physical meaning.Following a recent publication, we assess the TRPL data using Bayes' theorem instead.[1,2]
In this methodology, the PL response is calculated from reported physical models using a large number ofparameter combinations.[3,4] The different draws of each parameter combination are then given a probability,depending on how well the calculated overlaps with the experimental data. Depending on the number ofparameters, this can be a time-consuming approach.
In this work, we present a revised approach, expanding upon previous work by incorporating a Markov-ChainMonte-Carlo (MCMC) technique for exploring the n-dimensional parameter space. Computational time is savedby reducing the total number of draws needed to assess each parameter. With some additional optimizations,a typical run of 10'000 parameter combinations can be performed in < 2 hours. As a result, we obtain aprobability distribution for each parameter, which can give many important insights into the underlying physics.For instance, some parameters are non-identifiable based on the given, experimental data set and should bereported as such. 'Regular' fitting approaches would fail to grasp the non-identifiability and may thus lead toconfusing or even contradictory results.
To validate the approach, we use FAPbI
thin films and half-stacks and assess them with TRPL alone at first.We find a good agreement between extracted values for the different parameters and respective valuesreported in literature. In addition, correlations between these parameters can be probed and accuratelydescribed as well and may be useful to understand new physical processes. The extracted parameters arethen used to simulate the response of the sample during an intensity-dependent PLQE measurement and asurprisingly good agreement between the two is found. Based on this, the extracted parameters of the FAPbI
-SpiroOMeTAD half-stack are studied in more detail. Some novel insights into the processes governing PLquenching at the perovskite-HTL interface are gained and open new questions to why some transport layersresult in goo interfaces and other don't. Lastly, we compare the extracted parameters to other techniquesavailable in our group to find good agreements between all of them.
Overall, we show that this method of evaluating data can be extremely powerful, as it won't be limited by'overfitting' and in the case of TRPL allows the extraction of a variety of fundamental parameters from justasimple optical measurement.