Parallel Tempered Bayesian Inference for Characterizing Non-Ideal Semiconductors: Carrier Trapping in CdTe Thin Films

We describe an improved Bayesian inference methodology to characterize photovoltaic materials by matching charge carrier simulations to spectroscopy data. A "parallel tempering"scheme is introduced, which efficiently and reliably locates the global maximum in the complex multimodal distributions that are characteristic of cadmium telluride (CdTe) films. Our results show that the standard carrier transport model cannot explain the observed decay of time-resolved photoluminescence (TRPL) data from CdTe films and that there is carrier trapping within low-lying defect states. This inference has been confirmed by temperature-dependent TRPL and time-resolved emission spectroscopy (TRES). Our work shows that Bayesian inference can discriminate between plausible physics models, as well as determine parameter values for a given model. Finally, we have combined TRPL with time-resolved terahertz spectroscopy (TRTS) to describe the dynamics on nanosecond to microsecond time scales. These results show that sample degradation can be detected by its effect on surface recombination.