Research Paper:Non-Invasive Composition Identification in Organic Solar Cells via Deep Learning
Summary
This study introduces a novel, non-invasive framework for identifying active-layer compositions in organic photovoltaic (OPV) devices using deep learning. The main goal was to overcome the limitations of invasive characterization techniques that risk device damage. Researchers systematically generated a diverse dataset of simulated full-device absorption spectra (350–750 nm) for six common donor–acceptor material pairs (P3HT:PCBM, PTB7-Th:PCBM, PBDB-T:IT4F, PBDB-T:ITIC, PM6:Y6, and PM6:IT4F). Crucially, fabrication-related variability, specifically active-layer thickness variations of ±15% (±20 nm), was incorporated to ensure a realistic training environment. A multilayer perceptron (MLP) neural network was then optimized, achieving classification accuracies exceeding 99% on both training and testing sets. The model demonstrated remarkable robustness and minimal sensitivity to factors like random initialization or data partitioning.
Publication Details
Chang, Y.-H.; Zhang, Y.-L.; Cheng, C.-H.; Wu, S.-H.; Li, C.-H.; Liao, S.-Y.; Tseng, Z.-C.; Lin, M.-Y.; Huang, C.-Y. (2025), Non-Invasive Composition Identification in Organic Solar Cells via Deep Learning. Nanomaterials, 15, 1112. https://doi.org/10.3390/nano15141112.
Fluxim Tools Used
The commercial optical simulation software Setfos was employed to generate the extensive spectral dataset. Setfos utilizes the transfer matrix method (TMM) to accurately model wavelength-dependent reflection, transmission, and absorption in multilayer thin-film optoelectronic devices. Its key role was in simulating the absorption spectra of six distinct OPV donor–acceptor systems within a standard device architecture, while systematically varying the active-layer thickness to account for realistic fabrication tolerances and capture thickness-induced interference effects.
Why it Matters
This research provides a fast, reliable, and non-destructive solution for identifying active-layer compositions in OPV devices, addressing a critical need for quality control in manufacturing and research. The findings pave the way for seamless integration into automated manufacturing diagnostics and quality control workflows, thereby accelerating the development and commercialisation of next-generation OPV technologies. Fluxim Setfos® was instrumental by enabling the accurate and efficient generation of a diverse, high-quality synthetic dataset. This allowed for robust training of the deep learning model, overcoming the cost, time, and practical limitations of purely experimental data collection for such a large and varied dataset.