Unbiased Roughness Measurement via Noise Subtraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring roughness of pattern structures in noisy images, such as those from scanning electron microscopes, suffer from bias due to measurement noise, leading to inaccurate and unreliable results, especially for small feature sizes and high noise levels, which affects semiconductor manufacturing and other fields.
Innovation Solution
A physics-based inverse linescan model is employed to separate noise from actual roughness, allowing for unbiased measurements by subtracting edge detection noise from biased parameters and predicting unbiased power spectral density data to accurately characterize feature roughness without relying on image filtering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional roughness measurement methods are used on noisy images, then measurement process is simple, but measurement precision deteriorates due to noise bias
Solution Approach 1:
The measurement process is segmented into distinct stages: acquiring multiple images at different focus settings, separating in-focus images from out-of-focus images, and processing them through different computational paths. This segmentation allows noise characterization to be performed independently on out-of-focus images while maintaining accurate roughness measurement on in-focus images, thereby improving measurement precision without excessive complexity
Solution Approach 2:
Out-of-focus images serve as an intermediary medium to characterize measurement noise. By using these images as a proxy for noise sources, the system can separate noise components from actual roughness signals in the in-focus images, improving measurement accuracy while using computational rather than hardware complexity
2Reliability
If image filtering is applied to reduce noise, then noise impact is reduced, but measurement precision deteriorates due to loss of roughness detail
Solution Approach 1:
The methodology segments the image dataset into in-focus and out-of-focus categories, applying different processing strategies to each. Out-of-focus images are used exclusively for noise characterization without affecting the actual roughness measurement, while in-focus images retain full detail for accurate roughness quantification. This segmentation eliminates the need to filter in-focus images, preserving measurement precision while improving noise robustness
Solution Approach 2:
The system creates a computational copy of the noise characteristics using out-of-focus images as proxies. By copying noise properties from out-of-focus images and subtracting them from in-focus images, the method reduces noise impact without applying destructive filtering operations that would lose roughness detail, thereby maintaining measurement precision while improving reliability
3Reliability
If multiple measurement points are used to reduce noise impact, then measurement reliability improves, but loss of time increases due to extended measurement duration
Solution Approach 1:
The measurement process is segmented such that multiple images are acquired efficiently at different focus settings in a single measurement cycle. Out-of-focus images are captured quickly for noise characterization while in-focus images provide the actual roughness data. This segmentation allows parallel processing of multiple measurement points without sequential overhead, improving reliability while minimizing time loss
Solution Approach 2:
Out-of-focus images are captured and processed in advance to establish noise characteristics before the actual roughness measurement is performed. By performing this preliminary noise characterization, the system eliminates the need for repeated noise assessment during the main measurement, reducing total measurement time while maintaining high reliability through multiple measurement points
Data Source
AI summary
In one embodiment, a method includes determining, by a processor, a measurement of edge detection noise; receiving a measurement of a biased parameter including measurement noise; based on the measurement of edge detection noise and a number of measurement points, determining a contribution of edge detection noise to the biased parameter; determining an unbiased parameter by subtracting the contribution of noise from the biased parameter including the measurement noise; and outputting the unbiased parameter.