Weak Pattern Quantification in Semiconductor Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor inspection methods, such as optical and SEM, face challenges in distinguishing defects from noise, especially for systematic defects related to subtle pattern variations, leading to incomplete quantification of features and subjective classification.
Innovation Solution
A method and system for weak pattern quantification in semiconductor fabrication, involving the acquisition of inspection data, identification of failing pattern types, grouping of similar patterns, and comparison of modulated instances to identify local differences and metrology sites, utilizing an inspection tool and review tool with processor-driven image analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If optical and SEM inspection methods are used to identify systematic defects, then defect detection capability is improved, but the ability to distinguish defects from noise deteriorates
Solution Approach 1:
The patent segments the inspection process into multiple stages: initial optical inspection to identify hot spots, SEM review to localize failures, and automated classification to categorize defects. This multi-stage segmentation allows each method to focus on specific tasks, improving overall defect detection while maintaining distinction accuracy through specialized analysis at each stage.
Solution Approach 2:
The patent introduces an automated classification system as an intermediary between optical inspection and SEM review. This intermediary uses design information and image analysis to pre-process and prioritize defects, reducing the burden on subsequent inspection stages and improving the ability to distinguish true defects from noise by filtering out false positives before detailed analysis.
2Adaptability or versatility
If manual classification is employed to group defects into good, bad and marginal classes, then classification flexibility is improved, but objectivity and consistency deteriorate
Solution Approach 1:
The patent implements an automated classification system that performs defect categorization without human intervention. The system uses design information, inspection data, and analysis algorithms to automatically assign defects to classes (good, bad, marginal), eliminating subjective bias while maintaining consistent classification criteria across all defects.
Solution Approach 2:
The patent incorporates feedback mechanisms where classification results are continuously refined based on inspection data patterns and design rule compliance. The system learns from accumulated data to improve classification accuracy while maintaining objectivity, allowing flexible adaptation to different defect types while ensuring consistent and reliable categorization.
3Productivity
If design-assisted automatic classification is used, then classification speed is improved, but information loss due to inability to quantify features deteriorates
Solution Approach 1:
The patent performs preliminary quantification of pattern features before classification by measuring critical dimensions, pattern fidelities, and geometric properties from inspection images. This preliminary action captures detailed feature information that would otherwise be lost, enabling both fast automated classification and retention of quantitative data for subsequent analysis.
Solution Approach 2:
The patent nests multiple levels of analysis within the classification system: basic defect detection, detailed feature quantification, design rule checking, and pattern fidelity analysis. Each nested level builds upon the previous one, allowing the system to maintain high classification speed at the outer levels while preserving detailed quantitative information in inner levels for comprehensive defect characterization.
Data Source
AI summary
A weak pattern identification method includes acquiring inspection data from a set of patterns on a wafer, identifying failing pattern types on the wafer, and grouping like pattern types of the failing pattern types into a set of pattern groups. The weak pattern identification method also includes acquiring image data from multiple varied instances of a first pattern type grouped in a first group, wherein the multiple varied instances of the first pattern type are formed under different conditions. The weak pattern identification method also includes comparing images obtained from common structures of the instances of the first pattern type to identify local differences within a portion of the first pattern type. Further, the weak pattern identification method includes identifying metrology sites within the portion of the first pattern type proximate to a location of the local differences within the portion of the first pattern type.