Hybrid Spectral Data Compression for Efficient and Interpretable Bosch Process Monitoring

Virtual Metrology (VM) is used to estimate waferlevel process outcome using in-situ data, which helps in reducing metrology cost and time. In cyclical plasma etching processes such as Bosch etching, Optical Emission Spectroscopy (OES) provides high-dimensional, temporally resolved data that is rich in process information but computationally burdensome. We propose a hybrid data compression framework for OES data (3648 Channels, 25 Hz) using discrete wavelet transform (DWT) filtering, principal component analysis (PCA), and functional PCA (fPCA). The resulting compressed embedding is interpretable, compact, and predictive, enabling accurate VM while preserving plasma physics insights. Experiments on 100 wafers etched through 100 Bosch cycles demonstrate that the embedding achieves strong data compression (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;0.001 \%$</tex> of the total volume) and supports accurate prediction of etch depth, outperforming baselines.