Adaptive Defect Detection Thresholds for Semiconductor Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor wafer inspection methods face challenges in efficiently detecting defects at varying pixel intensities, leading to increased recipe setup time and difficulty in distinguishing between yield-relevant and nuisance defects, especially as design rules shrink.
Innovation Solution
A system and method that use a processor to determine detection thresholds based on probability density functions and estimated shape parameters, adapting these thresholds to the distribution of difference images, allowing for automated defect detection and optimized laser power without significant user intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed threshold is used for defect detection at each pixel intensity, then the detection method is simple, but the probability of outlier detection varies across different intensity levels leading to inaccurate defect identification
Solution Approach 1:
The patent implements dynamic threshold adjustment by estimating the probability distribution of background pixel values at each intensity level and setting thresholds based on statistical parameters (mean and standard deviation) specific to each intensity segment. This transforms the static fixed-threshold method into a dynamic adaptive method that responds to local intensity characteristics, resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The patent changes the detection parameter (threshold value) based on the intensity level of pixels. By segmenting the image into different intensity ranges and applying intensity-dependent thresholds calculated from the probability distribution of background values in each segment, the system adapts the detection parameters to match local conditions, thereby improving detection accuracy without excessive complexity.
2Measurement precision
If intensity-based segmentation with different threshold values is used for defect detection at different segments, then defect detection accuracy improves, but recipe setup time increases due to extended manual tuning
Solution Approach 1:
The patent enables the inspection system to automatically determine optimal threshold values for each intensity segment by estimating the probability distribution of background pixel values from the image data itself. This self-calibration capability eliminates the need for manual recipe tuning, allowing the system to adapt to different wafer types and inspection conditions autonomously, thus improving detection accuracy without increasing setup time.
Solution Approach 2:
The patent performs preliminary estimation of the background probability distribution and calculation of optimal threshold values before actual defect detection begins. By pre-computing the statistical parameters (mean and standard deviation) for each intensity segment from training data or initial scans, the system prepares the detection parameters in advance, eliminating the need for time-consuming manual tuning during production setup.
3Measurement precision
If more sensitive inspections are applied to detect smaller defects, then detection capability improves, but the population of nuisance defects detected increases dramatically
Solution Approach 1:
The patent applies local quality by using intensity-dependent thresholds that adapt to local background characteristics. By calculating the probability distribution of background values specifically for each intensity segment and setting thresholds based on local statistical properties rather than global fixed values, the system achieves high sensitivity for small defects while maintaining appropriate discrimination against nuisance defects that have different intensity characteristics.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables efficient defect detection across different pixel intensities and layers, reducing manual setup time and improving the accuracy of defect identification, while maintaining sensitivity to strong defects, thus enhancing the yield and profitability of semiconductor manufacturing.
Implementation Method 1
a light source that generates a beam of light
Implementation Method 2
a detector that receives the beam of light reflected from the wafer
Data Source
AI summary
For semiconductor inspection images, detection thresholds can be determined based on probability density functions at a pixel intensity. The detection thresholds can then be applied to an image. This can find outliers at a fixed probability at all pixel intensity levels by estimating the probability distribution of underlying data and adapting the detection threshold values. Laser power can be optimized based on the detection thresholds.