Inline Defect Analysis for Semiconductor Yield Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect review processes in semiconductor manufacturing are inefficient due to the inclusion of nuisance defects, which do not impact yield, leading to inaccurate yield prediction and unnecessary resource allocation.
Innovation Solution
An inline defect analysis system that characterizes and prioritizes defects as nuisance, regular, or critical based on their impact on yield, using a nuisance defect detector to identify and exclude nuisance defects from the review process, focusing on defects with significant yield impact.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all defects are reviewed regardless of importance, then defect review completeness is improved, but review efficiency and yield prediction accuracy deteriorate due to inclusion of nuisance defects
Solution Approach 1:
The patent extracts and removes nuisance defects from the defect review process by classifying defects into categories (nuisance, random, systematic) and excluding those with low yield impact. This allows the review process to focus only on defects that truly affect yield, improving both efficiency and accuracy.
Solution Approach 2:
The patent applies different review standards to different defect types based on their local characteristics and impact. By characterizing each defect's severity and yield impact, the system applies selective review intensity - thorough review for critical defects, minimal or no review for nuisance defects.
2Measurement precision
If optical inspection sensitivity is increased to detect sub-micron defects, then defect detection capability is improved, but false alarm rate increases due to nuisance defects
Solution Approach 1:
The patent converts the harmful effect of increased sensitivity (which causes false alarms) into a benefit by using the same high sensitivity to detect both true defects and nuisance defects, then applying classification algorithms to distinguish between them. The high sensitivity is maintained for detection capability while false alarms are managed through intelligent sorting.
Solution Approach 2:
The patent introduces an intermediary classification system that sits between the inspection tool and the review process. This intermediary layer (the defect characterization system) processes all detected defects and filters out nuisance defects before they reach the review stage, reducing false alarms while maintaining detection sensitivity.
3Productivity
If defect review focuses only on critical defects, then review efficiency is improved, but review completeness deteriorates
Solution Approach 1:
The patent segments defects into distinct categories (nuisance, random, systematic) based on their characteristics and yield impact. This segmentation allows the review process to handle different defect types differently - efficiently reviewing critical defects while appropriately managing less critical ones, thus balancing efficiency and completeness.
Data Source
AI summary
In one embodiment, an inline defect analysis method includes receiving geometric characteristics of individual defects and design data corresponding to the individual defects, determining which of the individual defects are likely to be nuisance defects using the geometric characteristics and the corresponding design data, and refraining from sampling the defects that are likely to be nuisance defects.