Soft Matter and Complex Systems Seminar
sala B0.14, ul. Pasteura 5
2023-05-15 (09:30) 
Tony Ladd (University of Florida)
Bayesian Monte-Carlo for photovoltaic materials characterization
Quantifying rates of charge carrier recombination is an important step in developing solar cells with high power conversion efficiencies (PCE). However, recovering characteristic parameters such as carrier mobility, dopant concentration, and recombination rate constants is hindered by the interplay between carrier dynamics and the various recombination mechanisms. Interpretation of optoelectronic measurements, such as time-resolved photoluminescence (TRPL), usually relies on analytically tractable simplifications of the underlying physics models for carrier mobility and recombination, which can sacrifice much of the information content of the measurement.
Bayesian inference is a means of efficiently solving inverse problems. Rather than trying to locate a single optimum solution, it creates a probability distribution in the parameter space, based on the differences between simulated and measured data. In this case the forward problem is straightforward – given a set of material parameters, what does the TRPL curve look like. The inverse problem is to determine the material parameters from a given set of experimental data. I will present results to show that a small number of TRPL scans are sufficient to determine accurate values for most of the key optoelectronic parameters: carrier mobility, doping level, Auger and radiative recombination rates, and the bulk and surface contributions to the non-radiative decay. Minimizing non-radiative decay is a key to designing solar cells with high PCE.
Bayesian inference is quite computationally expensive, requiring a numerical solution of the classical drift-diffusion equations (in 1D) for each parameter set. To reduce the computational demands to manageable proportions, we have implemented a Metropolis Monte Carlo sampling, which can efficiently locate and sample the high probability region of the distribution.
Please note that, since it is outside of the Soft Matter and Complex Systems seminar schedule, attendance is not compulsory (although warmly encouraged!) for students taking the seminar.
Bayesian inference is a means of efficiently solving inverse problems. Rather than trying to locate a single optimum solution, it creates a probability distribution in the parameter space, based on the differences between simulated and measured data. In this case the forward problem is straightforward – given a set of material parameters, what does the TRPL curve look like. The inverse problem is to determine the material parameters from a given set of experimental data. I will present results to show that a small number of TRPL scans are sufficient to determine accurate values for most of the key optoelectronic parameters: carrier mobility, doping level, Auger and radiative recombination rates, and the bulk and surface contributions to the non-radiative decay. Minimizing non-radiative decay is a key to designing solar cells with high PCE.
Bayesian inference is quite computationally expensive, requiring a numerical solution of the classical drift-diffusion equations (in 1D) for each parameter set. To reduce the computational demands to manageable proportions, we have implemented a Metropolis Monte Carlo sampling, which can efficiently locate and sample the high probability region of the distribution.
Please note that, since it is outside of the Soft Matter and Complex Systems seminar schedule, attendance is not compulsory (although warmly encouraged!) for students taking the seminar.