Histogram Scaling for Wafer Defect Detection Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor wafer inspection, process variation due to thickness differences in wafer layers causes color noise, making it difficult to detect defects and leading to false positives, as the same layer can have varying thicknesses affecting image gray levels and reflectivity, thus requiring an improved image processing technique.
Innovation Solution
A system and method that involve generating a test image and a reference image, calculating and adjusting their gray level histograms through histogram scaling, applying these adjustments to produce a difference image with reduced noise, and identifying defects within this image, which includes subtracting the mean, multiplying by a gain factor, and adding a constant intensity offset, and masking pixels exceeding certain thresholds to enhance defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If histogram scaling is applied to adjust gray level histograms, then process variation is reduced and defect detection sensitivity is enhanced, but device complexity increases due to additional image processing steps
Solution Approach 1:
The patent applies histogram scaling as a preliminary processing step before defect detection. By pre-adjusting the gray level histograms of reference and test images to match each other, the system prepares the images for more accurate defect detection. This preliminary normalization reduces process variation and enhances defect visibility, making the subsequent detection process more effective.
Solution Approach 2:
The patent changes the parameters of the gray level histograms through scaling operations. Specifically, it adjusts the mean and standard deviation of the histograms to match between reference and test images. This parameter transformation equalizes the background gray levels caused by process variation, thereby enhancing defect detection sensitivity without requiring complex hardware modifications.
2Reliability
If tolerance in histogram is increased to account for process variation, then false positives are reduced, but some defects are lost due to loss of detection sensitivity
Solution Approach 1:
The patent transforms the histogram parameters (mean and standard deviation) to match between reference and test images. This parameter alignment allows the system to maintain tight detection thresholds while accounting for process variation. By normalizing the gray level distributions beforehand, the system can use stricter tolerance criteria without losing defects, thus improving reliability without sacrificing detection sensitivity.
Solution Approach 2:
The patent extracts and corrects the process variation component from the images through histogram scaling. By separating the background gray level variation (caused by process variation) from the actual defect signals, the system can set more appropriate detection thresholds. This extraction of the noise component allows for better discrimination between true defects and false positives.
3Manufacturing precision
If color variation correction is attempted, then process variation is addressed, but it becomes difficult to identify the best mode for defect inspection due to increased complexity
Solution Approach 1:
The patent applies parameter changes to the gray level histograms through scaling operations that match the mean and standard deviation between reference and test images. This straightforward parameter transformation addresses color variation and process variation without introducing complex multi-mode inspection requirements. The method maintains a single, consistent inspection approach while effectively correcting for manufacturing variations.
Data Source
AI summary
Gray level histograms for a test image and a reference image are adjusted by histogram scaling. Parameters from the histogram scaling are applied to the test image and the reference image. After the parameters are applied, the reference image and the test image are compared to produce a difference image, such as by subtracting the reference image from the test image. Noise in the difference image can be reduced, which improves defect identification in the difference image. In addition, noisy structures in the difference image which are elongated in vertical or horizontal direction can be found. If the noise exceeds a certain threshold, the structures may not be inspected.