Machine learning for enhanced semiconductor characterization from time-resolved photoluminescence

To enhance the accuracy and effectiveness of optoelectronic characterization, Bayesian inference has been applied to the statistical analysis of time-resolved photoluminescence (TRPL) data with large-scale graphics-processing unit (GPU)-based simulations of electron dynamics. A simulated TRPL dataset, derived from a CH3NH3PbI3-xClx perovskite absorber, was used as a case study. From a power scan, Bayesian inference extracts values of the (ambipolar) carrier mobility, free-carrier density, radiative recombination rate, and the overall lifetime of the nonradiative recombination processes. Analysis of the experimental TRPL data yields similar parameter values. The independent contributions of bulk and surface recombination can be distinguished via the introduction of an additional sample thickness. Further experiments separate the front and back surface recombination velocities and can distinguish the electron and hole mobilities. Ultimately, Bayesian inference enables a significant increase in the information yield from TRPL measurements.