Determining Parameters of Perovskite Half-Devices Using Time-Resolved Photoluminescence and Bayesian Inference

There is an urgent need for the rapid analysis of losses in the field of perovskite-based photovoltaics. A common method to assess losses is the analysis of half-devices, which are an absorber material interfaced with a charge transport layer, using time-resolved photoluminescence. Although the process of data acquisition is relatively straightforward, the interpretation is not as simple due to the numerous convoluted physical processes that can affect the PL decay. In this project, we developed a novel methodology that combines a Markov-Chain Monte-Carlo (MCMC) sampler with Bayesian inference to extract 8 device-relevant parameters from the PL decays. Instead of identifying the parameter set that most accurately describes the data, we evaluate all potential parameter combinations that are consistent with the data. Therefore, it is possible to differentiate between the recombination at each of the interfaces and the bulk of the perovskite material.

Values extracted from our data are consistent with those reported in the literature. This method has the advantage of requiring only a single sample, as opposed to using a thin film on glass as a comparison. This is superior because it avoids secondary effects, such as the substrate-dependent crystallisation of the perovskite film. Additionally, it makes the loss analysis of half-devices much faster. Some novel insights into the processes governing PL quenching at the perovskite-HTL interface are gained and open new questions to why some transport layers result in good interfaces and others don't. We demonstrate that this approach to data evaluation is extremely powerful, as it is not constrained by "overfitting". In the case of TRPL, it enables the extraction of an extensive range of fundamental parameters from optical measurements that are relatively simple to acquire.