Wafer Inspection Pattern Segmentation for Nuisance Defect Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor wafer inspection methods struggle to effectively distinguish between real yield-killing repeater defects and nuisance repeaters, especially in noisy patterns, leading to increased time and cost in identifying and correcting defects during the manufacturing process.
Innovation Solution
A system and method that segment patterns based on noisiness levels, using a standard deviation threshold to differentiate between noisy and quiet regions, allowing for targeted inspection and reducing the detection of nuisance repeaters while maintaining sensitivity to real defects, by dividing the wafer image into segments and applying a threshold multiple times the standard deviation to identify defects-of-interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional inspection methods detect all defects on wafers, then detection sensitivity is improved, but the number of nuisance defects increases and inspection efficiency deteriorates
Solution Approach 1:
The wafer image is divided into multiple segments based on design pattern characteristics. Each segment is then inspected independently with appropriate thresholds, allowing the system to maintain high detection sensitivity for real defects while reducing the impact of nuisance defects in noisy regions through localized analysis.
Solution Approach 2:
Different inspection thresholds and criteria are applied to different segments of the wafer image based on their noise characteristics. Quiet regions use stricter thresholds to detect real defects, while noisy regions use adjusted thresholds to minimize false alarms, thereby improving overall inspection efficiency without sacrificing detection capability.
2Measurement precision
If inspection thresholds are lowered to detect smaller defects, then detection precision is improved, but the number of false positive nuisance defects increases
Solution Approach 1:
The inspection area is segmented into multiple regions based on design pattern noise characteristics. This allows the system to apply different threshold levels to different segments, enabling sensitive detection of small defects in quiet regions while suppressing false positives in noisy regions through localized threshold adjustment.
Solution Approach 2:
The inspection threshold parameter is dynamically adjusted based on the noise level of each segment. By changing the threshold parameter according to local noise characteristics, the system maintains high precision for small defect detection in low-noise areas while reducing false positives in high-noise areas.
3Ease of operation
If the entire wafer is inspected as one region, then inspection simplicity is maintained, but the ability to distinguish noisy from quiet regions is lost
Solution Approach 1:
The wafer image is automatically segmented into multiple regions based on design pattern characteristics and noise levels. This segmentation enables the system to distinguish between noisy and quiet regions, applying appropriate inspection strategies to each while maintaining an automated, user-friendly process.
Solution Approach 2:
Different inspection qualities and thresholds are applied to different regions of the wafer based on their noise characteristics. Quiet regions receive detailed inspection for small defects, while noisy regions receive adjusted inspection parameters, thereby maintaining operational simplicity while achieving precise noise differentiation.
Data Source
AI summary
During semiconductor wafer inspection, an image of a semiconductor wafer is divided into segments. A standard deviation for each of the segments is determined using a difference image. A threshold is then applied to each of the segments. The threshold can be a multiple of the standard deviation. Pixels in the image that include a defect are determined after applying the threshold. The pixels outside the threshold are then labeled as defects-of-interest using the processor.