Dimension reduction of multivariable optical emission spectrometer datasets for industrial plasma processes

Area of Science:

• Analytical Chemistry
• Process Engineering
• Data Science

Background:

• Optical Emission Spectroscopy (OES) generates high-dimensional data from industrial plasma processes.
• Real-time OES data is valuable for inferring critical process parameters like wafer etch rates in semiconductor manufacturing.
• Existing methods struggle with the complexity and high dimensionality of OES data for direct physical process interpretation.

Purpose of the Study:

• To introduce a novel data dimension-reduction method, Internal Information Redundancy Reduction (IIRR).
• To minimize information redundancy in OES datasets while maintaining interpretability for physical processes.
• To enable accurate process monitoring and control through effective data simplification.

Main Methods:

• Developed the Internal Information Redundancy Reduction (IIRR) method operating in the original variable space.
• Identified peak wavelength emissions and their correlative relationships.
• Proposed a new statistic, Mean Determination Ratio (MDR), to quantify information loss.

Main Results:

• The IIRR method effectively reduces the dimensionality of OES datasets with minimal information loss.
• Demonstrated the method's effectiveness using a semiconductor manufacturing dataset.
• Achieved accurate prediction of etch rates from IIRR dimension-reduced spectral data.

Conclusions:

• IIRR is a powerful tool for simplifying complex OES data in industrial plasma processes.
• The method facilitates direct interpretation of physical processes from reduced datasets.
• IIRR enables accurate real-time process monitoring and control, exemplified by etch rate prediction.