Proceedings of MATSUS Spring 2025 Conference (MATSUSSpring25)
DOI: https://doi.org/10.29363/nanoge.matsusspring.2025.071
Publication date: 16th December 2024
Dye-sensitized solar cells (DSSCs) are promising for glazing applications due to their potential for semi-transparency. However, their photovoltaic performance and light transmittance are largely determined by the dye and electrolyte used, both of which are fixed during the manufacturing process. Electrolytes are critical to energy technologies, yet their optimization is challenging due to the complexity of their formulations and the multitude of interacting chemical components. Typically, optimization requires numerous experiments, as the effects of these components are often correlated and difficult to analyze independently.
In this study, we employed a design of experiments (DoE) methodology combined with machine learning (ML) to design electrolytes that effectively balance two typically conflicting properties: visible transparency and power conversion efficiency (PCE). The model required only a limited number of experiments for training and exhibited excellent predictive agreement with experimental results.
First, we optimized iodine-based electrolytes to fabricate solar cells with a visible transparency range of 34% and a maximum PCE of 2.94%. We then extended this approach to electrolytes based on alternative redox systems. Using our data-driven modeling approach, we optimized a TEMPO-based electrolyte, achieving photochromic semi-transparent cells with a 42% transmittance variation and a PCE of 2.16%. For opaque cells, this novel electrolyte delivered a PCE of 3.46% with a photochromic dye and an impressive 7.64% PCE when paired with a non-photochromic dye.
This work was funded under the European Union's Horizon 2020 research and innovation programme (grant agrement number 832606; project PISCO).