Cell Counting Algorithm for CMP Spectrum Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical monitoring techniques for chemical mechanical polishing (CMP) are unreliable due to sensitivity to peak shifts in spectra caused by variations in underlying layers and noise, leading to non-uniformity in wafer processing.
Innovation Solution
Implementing a cell counting technique to identify the best matching reference spectrum by dividing the spectral space into cells, counting crossing and separating cells, and calculating a dissimilarity ratio to determine the most accurate matching spectrum, reducing sensitivity to peak positions and noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sum of squared differences algorithm is used to identify best matching reference spectrum, then matching speed is improved, but reliability deteriorates due to sensitivity to peak shifts and noise
Solution Approach 1:
The patent divides the spectral space into multiple cells (bins) along the wavelength axis. Instead of comparing entire spectra directly, the algorithm segments the spectrum into discrete regions and counts the number of peaks in each cell. This segmentation reduces the impact of peak position shifts and noise by aggregating information across multiple wavelength points, thereby improving reliability while maintaining computational efficiency.
2Ease of manufacture
If sum of squared differences algorithm is used to identify best matching reference spectrum, then computational simplicity is improved, but measurement precision deteriorates due to noise sensitivity
Solution Approach 1:
The spectrum is divided into multiple cells along the wavelength axis, and the algorithm counts peaks in each cell rather than comparing individual wavelength points. This segmentation approach maintains computational simplicity while improving precision by reducing sensitivity to noise and peak position variations.
Solution Approach 2:
The patent creates a simplified representation of the spectrum by counting peaks in each cell and storing these counts as a characteristic signature. This copied representation (peak counts per cell) is then used for comparison with reference spectra, reducing the impact of noise while preserving the essential spectral features needed for accurate matching.
3Difficulty of detecting and measuring
If traditional optical monitoring is used during polishing, then process monitoring capability is improved, but manufacturing precision deteriorates due to non-uniformity from material removal rate variations
Solution Approach 1:
The patent implements real-time optical monitoring during the polishing process, measuring spectra at multiple wavelength points and comparing them against a library of reference spectra. The system uses the cell counting algorithm to identify the best matching reference spectrum and determines the polishing endpoint based on this comparison. This feedback mechanism allows for precise control of the polishing process, compensating for variations in material removal rate and achieving uniform results across wafers.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the reliability of endpoint detection and reduces within-wafer and wafer-to-wafer non-uniformity, providing a more precise control over the polishing process.
Implementation Method 1
a spectrum is measured from a substrate during polishing
Implementation Method 2
through a window in the polishing pad
Data Source
AI summary
A method of controlling polishing of a substrate is described. A controller stores a library having a plurality of reference spectra. The controller polishes a substrate and measures a sequence of spectra of light from the substrate during polishing. For each measured spectrum of the sequence of spectra, the controller finds a best matching reference spectrum from the plurality of reference spectra and generates a sequence of best matching reference spectra. The controller uses a cell counting technique for finding the best matching reference spectrum. The controller determines at least one of a polishing endpoint or an adjustment for a polishing rate based on the sequence of best matching reference spectra.