Pattern Shape Evaluation Device for LER Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating line-edge roughness (LER) in semiconductor manufacturing are affected by noise, leading to inaccurate measurements, especially when the signal-to-noise ratio is low, and are prone to errors due to random noise and electron beam irradiation effects.
Innovation Solution
A pattern shape evaluation device that uses a moving average of signal profiles to extract edge positions and calculates noise floor height based on power spectral density data, employing a rectangular window function to improve noise rejection and edge position accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If edge position is extracted directly from CD-SEM image, then measurement process is simple, but measurement precision deteriorates due to random noise
Solution Approach 1:
The patent applies preliminary action by performing moving average processing on signal profiles before extracting edge positions. This preprocessing step reduces random noise in the image data, ensuring that the subsequent edge extraction produces more accurate results without complicating the overall measurement workflow.
Solution Approach 2:
The patent introduces an intermediary element - the moving average processed signal profile - that mediates between the raw noisy image data and the final edge position measurement. This intermediary processing layer filters out random noise while preserving the essential edge information, thereby improving measurement precision.
2Measurement precision
If multiple image processing methods are applied to reduce noise, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent performs moving average processing as a preliminary step before edge extraction and LER calculation. This single preprocessing operation reduces random noise effectively, improving measurement precision without requiring multiple complex image processing methods or sophisticated equipment.
Solution Approach 2:
The patent changes the parameter of signal processing by applying moving average to the signal profiles. This parameter change - processing signals in groups and averaging them - reduces the impact of random noise on edge position extraction, thereby improving LER measurement accuracy with simple computational steps.
3Measurement precision
If Fourier transform is used to calculate PSD for noise removal, then measurement precision improves, but calculation time increases
Solution Approach 1:
The patent extracts and removes random noise from LER measurements by applying moving average processing to signal profiles before edge extraction. This approach eliminates the need for computationally intensive Fourier transforms and power spectral density calculations, achieving noise removal with simpler and faster operations.
Solution Approach 2:
The patent uses a simple, computationally inexpensive method - moving average processing - instead of complex Fourier transform-based noise removal. This disposable-like simplicity in the processing approach achieves effective noise reduction without the time-consuming calculations of spectral analysis methods.
Data Source
AI summary
Line-edge roughness or line width roughness is evaluated while preventing influence of noise caused by a device or an environment. Therefore, an averaged signal profile 405 in which a moving average of S pixels (S is an integer greater than 1) is taken in a Y direction is obtained from a signal profile showing a secondary electron signal amount distribution in an X direction with respect to a predetermined Y coordinate obtained from a top-down image, an edge position 406 of a line pattern is extracted based on the averaged signal profile, and a noise floor height is calculated based on a first power spectral density 407 of LER data or LWR data based on the extracted edge position and a second power spectral density 409 of a rectangular window function corresponding to the moving average of the S pixels.