Publication date: 17th February 2025
On the value of device characterization for the optimization of solar cells
The biggest advantage of organic photovoltaic, namely the flexibility of organic synthetic chemistry is also the biggest challenge for material selection and process optimization. To find the best process in a highly multidimensional process parameter and material space, traditional Edisonian optimization methods are insufficient.
Sophisticated optimization algorithms that help to explore the high-dimensional process parameter space have recently been applied. Thanks to increasing computing power and modern approaches in the field of machine learning, these methods are continuously improving [1 - 3].
Many of the optimization approaches are black-box optimization methods based on purely mathematical models and do not, or only partially, consider the physical domain knowledge underlying organic photovoltaics. In this case, it is sufficient to measure only the efficiency of the respective devices.
In this contribution, we want to quantify how far, domain knowledge about the material parameters and the device physics of the organic solar cells is beneficial to the actual optimization process. Therefore, we have implemented a virtual laboratory to compare different process optimization strategies based on different amounts of device characterization. The virtual laboratory is essentially a benchmark function for the optimization that is based on actual data obtained from several hundred organic solar cells and associated characterization data.
Within the virtual laboratory, we compare black-box optimization with an optimization algorithm that is virtually fed characterization data and that can make predictions in the process parameter space using drift-diffusion simulation as well as optical modeling. We show that prior knowledge of optical constants combined with optical modeling is particularly advantageous for identifying local maxima for larger layer thicknesses.