Proceedings of MATSUS23 & Sustainable Technology Forum València (STECH23) (MATSUS23)
DOI: https://doi.org/10.29363/nanoge.matsus.2023.006
Publication date: 22nd December 2022
For the continuous improvement of optoelectronic devices such as organic light-emitting diodes and solar cells in terms of e.g. stability and efficiency, a comprehensive model including all major physical processes is crucial. Further, the combined analysis of measurement and simulation paves the way for the understanding of novel devices such as perovskite solar cells. In this work, an up-scaled R&D solar cell is investigated. Once a model and material & device parameter set describing the cell is found, the optimization of the cell performance can start. The initial challenge, however, is to determine such a set of parameters since the availability of e.g. material parameters is limited and is usually obtained by tailored measurements or taken from literature. The values for certain material parameters can vary depending on the literature source or the sequence of layers in the experimental stack. A way to determine the parameters is fitting. In least-square algorithms the sum of the squared differences between the measurement and simulation is minimized by varying the material parameters until an optimal set is found. This task however can be cumbersome, especially with a large number of unknown parameters. Often the parameters are also correlated which again increases the complexity of the problem. In such situations, domain knowledge is required to facilitate the search for the minimal error.
In this contribution, we discuss traditional and machine-learning assisted approaches [1] to determine the model parameters for perovskite solar cells based on electroluminescence images and current-voltage measurements.