Semiconductor Image Binarization for Noisy Line Width Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of accurately measuring line widths in semiconductor devices is exacerbated by noise in images obtained during the manufacturing process, leading to reduced measurement accuracy, particularly in channel holes, due to unclear edges.
Innovation Solution
An image processing system and method that utilizes edge detection, image binarization, and machine learning to enhance the correlation between line widths measured from different directional views, optimizing the binarization condition to improve measurement accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image binarization is used for line width measurement, then the process is simple and fast, but measurement accuracy deteriorates in high-noise images
Solution Approach 1:
The patent introduces a learning device as an intermediary component that bridges the traditional image processing system and the measurement outcome. This learning device trains a machine learning model using multiple images with different noise levels and binarization conditions, then applies the trained model to automatically determine optimal binarization conditions for new images, thereby improving measurement accuracy without significantly increasing system complexity
Solution Approach 2:
The patent changes the binarization condition parameters (threshold values, processing methods) based on learned patterns from training data. Instead of using fixed binarization parameters, the system dynamically adjusts these parameters by comparing multiple candidate binarized images and selecting the one that produces the most consistent line width measurements across different noise levels, thus improving measurement precision
2Measurement precision
If machine learning is applied to optimize binarization conditions, then line width measurement accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs the computationally intensive machine learning training process in advance, before actual production measurement. The learning device trains the model using a large dataset of images with various noise levels and binarization conditions during a preliminary training phase. Once trained, the model can quickly determine optimal binarization conditions for new images during production, reducing real-time processing time while maintaining high measurement accuracy
3Reliability
If multiple binarization conditions are tested to find the optimal one, then measurement reliability is improved, but the number of processing steps increases
Solution Approach 1:
The patent implements a feedback mechanism where the system generates multiple binarized images with different conditions, compares their line width measurements, and uses this comparison feedback to select the optimal binarization condition. The learning device learns from this feedback during training and automatically applies the learned selection criteria to new images, improving measurement reliability while maintaining processing efficiency through automated decision-making
Data Source
AI summary
An image processing system, including an input interface configured to receive a first direction image corresponding to a view of a semiconductor device in a first direction, and a second direction image corresponding to a view of the semiconductor device in a second direction which intersects the first direction at a first height at which the first direction image is generated; a processor configured to perform an edge detection operation for detecting an edge based on the first direction image, and to perform an image binarization operation on the first direction image; and a learning device configured to compare a first line width obtained based on the image binarization operation, and a second line width obtained based on the second direction image through machine learning, and to learn a condition of the image binarization operation which maximizes a correlation between the first line width and the second line width.