Data and Code for: Mutual Information-Based Wavelength Selection for Micro-Leak Detection in N2 Semiconductor Process Environments
Description
This dataset contains the Self-Plasma Optical Emission Spectroscopy (SPOES) data and MATLAB code supporting the article "Mutual Information-Based Wavelength Selection for Micro-Leak Detection in N2 Semiconductor Process Environments" (Vacuum, 2026). Simulated air leaks (79% N2 / 21% O2) were introduced into a chamber operated with N2 process gas (2500 sccm, 2.5 Torr). The exhaust gas was monitored with a SPOES sensor covering 2048 wavelength channels (200–850 nm, 0.95 nm optical resolution). The data come from two independent experiments. Wavelength selection experiment (folder "Wavelength Selection"): leaks of 20–60 ppm injected at intervals of at least 10 minutes. It contains the raw SPOES spectra (Wavelength_Select_N2_1.csv, Wavelength_Select_N2_2.csv) and the measured leak input signal (Leak_timing _Labeling.xlsx). The script Wavelength_Selection_Algorithm.m ranks all wavelengths by their similarity to the leak input using Pearson, Spearman and Mutual Information (MI) and compares them with a peak-based PCA baseline. MI selects 245.2 nm, which corresponds to the NO γ-band emission. The script reproduces Figs. 5–7 and Table 4. Leak detection experiment (folder "Leak_detection Algorithm"): leaks of 10–40 ppm injected at intervals of at least 15 minutes. The file "Experiment conditions for leak detection.mat" holds three measurement sessions (AData, BData, CData; samples × 2048 wavelengths) with sample labels (labels_A/B/C; 0 = normal, 1 = leak). Machine_optimization.m compares logistic regression, KNN and random forest under stratified group-based 6-fold cross-validation, searches 128 wavelet denoising configurations, and evaluates detection accuracy per concentration (Table 5, Table 6). The final system reaches 98.42% accuracy with a 0.99% false alarm rate. Wavelet_filter.m reproduces Fig. 9. confusion_Matrix.m reproduces Fig. 8. Run the scripts with the MATLAB current folder set to the folder that contains them. Run order and the mapping between scripts and the paper's figures and tables are described in README.md. Requirements: MATLAB R2022a or later with the Statistics and Machine Learning, Wavelet, Signal Processing and Parallel Computing Toolboxes.
Files
Institutions
- Yonsei UniversitySeoul, Seoul