Non-Invasive Composition Identification in Organic Solar Cells via Deep Learning

Area of Science:

• Materials Science
• Organic Electronics
• Computational Chemistry

Background:

• Current organic photovoltaic (OPV) device analysis often requires destructive methods.
• Accurate identification of active-layer composition is crucial for OPV performance and reliability.

Purpose of the Study:

• To develop a non-invasive technique for classifying OPV active-layer compositions.
• To leverage simulated absorption spectra and deep learning for accurate material identification.

Main Methods:

• Simulated full-device absorption spectra were generated with ±15% thickness variation for a diverse dataset.
• A multilayer perceptron (MLP) neural network was trained and optimized using various configurations.
• The model's robustness was tested against random initialization and data partitioning.

Main Results:

• The optimized MLP model achieved classification accuracies exceeding 99% on both training and testing datasets.
• The classification accuracy demonstrated minimal sensitivity to random initialization and data splitting.
• The developed method proves effective for non-destructive OPV composition analysis.

Conclusions:

• Deep learning applied to spectral data offers a reliable, non-invasive method for OPV composition classification.
• This approach has the potential to be integrated into automated manufacturing for diagnostics and quality control.
• The findings pave the way for advanced, non-destructive characterization techniques in organic electronics.