Publication date: 5th November 2025
Deeper understanding of material systems is essential to accelerate optimization of solar cell devices. This is particularly important for emerging materials such as organic semiconductors and halide perovskites, which are less investigated compared to crystalline silicon. However, accurate estimation of material properties from optoelectronic measurements is often challenging due to complexity and interdependency.
In the past, numerical solvers such as drift-diffusion simulation have enabled researchers to reproduce characterization data using known material parameters and compare the simulation results with experimental data to analyze solar cells fabricated in a lab. [1-3] Nevertheless, the traditional fitting routine is time-consuming and relies on a deterministic approach that neglects other possible parameter combinations.
To address these limitations, we have implemented machine learning techniques to the parameter inference problem. First, we reduce the parameter space of a material based on our understanding of physics to maintain an acceptable level of computation work. Next, we develop a neural network model that ensures robust computation, outperforming classical numerical simulations. [4, 5] Finally, we integrate this model with existing Bayesian Parameter Estimation [6-8] methods. By analyzing four current-voltage (JV) curves at different light intensities, we were able to infer electronic properties such as mobilities, recombination coefficients, and defect densities. Moreover, estimation of posterior probability distribution upon every additional input of JV data allows us to quantify the information content in JV curves. Lastly, we performed device performance prediction over light intensities using the inferred parameter set and validated the usefulness of the suggested data analysis workflow.
