Real-time non-invasive quality screening of Yb-doped thin film electrodes using machine learning

Abstract Rapid classification of dopant levels in ultrathin dielectric films is essential for real-time quality control in optoelectronic manufacturing. In this study, we propose a non-invasive, machine learning-based approach to classify Yb doping levels in (Yb: MoO 3 )/Ag/MoO 3 multilayer electrodes using optical transmittance spectra. Transmission data, simulated using effective medium approximation models and validated with experimental measurements, were used to train a multi-layer perceptron classifier. With architectural tuning and the implementation of Broyden–Fletcher–Goldfarb–Shanno (BFGS) optimization, the model achieved near 100% accuracy in classifying discrete doping levels and demonstrated excellent robustness across random seeds and training splits. This method eliminates the need for destructive testing and can be extended to other rare-earth dopants. The results highlight the potential of combining spectral analysis and machine learning for real-time, automated compositional assessment in display technologies and thin-film manufacturing.