SEM Edge Detection via Inverse Linescan Noise Subtraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge detection methods in scanning electron microscopes (SEMs) face challenges in accurately measuring feature roughness due to noise contamination, leading to biased measurements that are dependent on tool settings and struggle with high noise levels, especially at smaller feature sizes.
Innovation Solution
The implementation of an inverse linescan model that calibrates and fits measured linescan information to determine feature geometry, allowing for unbiased roughness measurements by averaging along axes of symmetry to separate noise from actual roughness, and subtracting noise from power spectral density (PSD) to obtain accurate roughness parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional edge detection methods are used in SEMs, then edge positions can be detected, but measurement precision deteriorates due to noise contamination and bias dependence on tool settings
Solution Approach 1:
The patent extracts and removes the noise component from the power spectral density (PSD) measurement. By separating the noise PSD from the total measured PSD, the method isolates the true roughness signal, eliminating the bias that would otherwise be present in conventional measurements that cannot distinguish noise from actual feature roughness.
Solution Approach 2:
The patent changes the parameter being measured from the raw PSD to the noise-subtracted PSD. By transforming the measurement parameter to represent only the true roughness component (after removing noise contributions), the method achieves unbiased roughness measurements that are independent of SEM tool settings and noise levels.
2Object-affected harmful factors
If noise filtering is applied to improve edge detection, then noise is reduced, but measurement precision deteriorates due to loss of actual roughness information
Solution Approach 1:
Instead of filtering out noise, the patent extracts and separately measures the noise component through PSD analysis. By identifying the noise PSD in frequency domain and subtracting it from the total PSD, the method removes only the noise contribution while preserving all actual roughness information across all frequency ranges.
Solution Approach 2:
The patent introduces the power spectral density (PSD) as an intermediary tool to separate noise from roughness. The PSD transformation allows noise and signal to be distinguished in the frequency domain, serving as a mediator that enables selective removal of noise without affecting the true roughness characteristics.
3Device complexity
If conventional edge detection is used, then processing is simpler, but measurement precision deteriorates due to inability to separate noise from roughness
Solution Approach 1:
The patent introduces power spectral density (PSD) analysis as an intermediary computational step that enables noise separation. While this adds complexity compared to simple threshold-based edge detection, the PSD intermediary provides the mathematical framework needed to accurately distinguish noise from true roughness, achieving superior measurement precision.
Solution Approach 2:
The patent replaces physical noise filtering mechanisms with a computational signal processing approach. Instead of using hardware filters that physically remove frequency components, the method uses mathematical operations on the PSD to subtract noise contributions, achieving noise removal through computation rather than physical filtering.
Data Source
AI summary
An edge detection system is provided that generates a scanning electron microscope (SEM) linescan image of a pattern structure including a feature with edges that require detection. The edge detection system includes an inverse linescan model tool that receives measured linescan information for the feature from the SEM. In response, the inverse linescan model tool provides feature geometry information that includes the position of the detected edges of the feature.