SEM Roughness Measurement Using Inverse Linescan Noise Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring roughness in pattern structures, particularly in noisy SEM images, suffer from bias due to measurement noise, leading to inaccurate and unreliable results, which affect the quality of manufacturing processes and device performance.
Innovation Solution
A system and method that utilize an inverse linescan model to detect edges in SEM images without filtering, generating a biased power spectral density dataset, evaluating high-frequency noise, and subtracting predicted noise to obtain an unbiased roughness measure, thereby filtering out artifacts and revealing true feature roughness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If noise filtering is applied to SEM images before roughness measurement, then measurement reliability is improved, but measurement precision deteriorates due to loss of true feature information
Solution Approach 1:
The patent segments the power spectral density into distinct frequency components: low-frequency region representing true feature roughness, mid-frequency region containing transition information, and high-frequency region representing noise. By analyzing and separating these segments, the method selectively processes different frequency components to remove noise while preserving true feature information.
Solution Approach 2:
The patent converts the harmful effect of noise into a beneficial tool by using the noise characteristics in the high-frequency region to identify and remove noise from the entire spectrum. The noise present in the measurement is transformed into a reference signal that enables selective noise subtraction, thereby improving measurement accuracy without losing feature information.
2Ease of operation
If conventional roughness measurement methods are used on noisy SEM images, then ease of operation is maintained, but measurement precision deteriorates due to noise-induced bias
Solution Approach 1:
The patent implements a feedback mechanism where the power spectral density analysis provides information about noise characteristics that is fed back into the measurement process. The high-frequency noise components are used to generate a noise model that is then subtracted from the total PSD, creating a corrected roughness measurement that compensates for noise-induced bias.
Solution Approach 2:
The patent changes the measurement parameter from direct spatial domain roughness calculation to frequency domain analysis using power spectral density. This parameter transformation enables selective processing of different frequency components, allowing noise removal while preserving true feature roughness information that would be lost in conventional spatial domain filtering.
3Measurement precision
If noise is removed from roughness measurements, then measurement precision is improved, but device complexity increases due to additional processing steps
Solution Approach 1:
The patent introduces the power spectral density as an intermediary representation that mediates between the raw noisy measurement data and the final corrected roughness value. This intermediary frequency domain representation enables systematic noise identification and removal through mathematical operations, providing a structured approach to noise reduction without requiring complex hardware modifications.
Data Source
AI summary
In one embodiment, a method includes receiving measured linescan information describing a pattern structure of a feature, applying the received measured linescan information to an inverse linescan model that relates measured linescan information to feature geometry information, and identifying, based at least in part on the applying the received measured linescan model to the inverse linescan model, feature geometry information that describes a feature that would produce a linescan corresponding to the received measured linescan information. The method also includes determining, at least in part using the inverse linescan model, feature edge positions of the identified feature, analyzing the feature edge positions to determine errors in the manufacture of the pattern structure, and controlling a lithography tool based on the analysis of the feature edge positions.