Semiconductor Pattern Defect Removal via Grouped Image Parameter Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As semiconductor device integration increases, the challenge of compensating for differences between designed and manufactured patterns becomes more significant, leading to pattern defects that existing techniques struggle to effectively address.
Innovation Solution
A method is developed to classify sample pattern data into standard normal and weak groups based on image parameters, extract determination functions, and perform optical rule checking to identify and correct pattern defects, thereby improving defect prediction and removal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing pattern compensation techniques are used, then manufacturing process is simple, but pattern defect removal accuracy is insufficient
Solution Approach 1:
The patent segments pattern data into multiple groups (normal group, weak group, and suspicious weak group) based on defect probability. This segmentation allows the system to focus computational resources on patterns with higher defect likelihood, improving detection accuracy without requiring complete analysis of all patterns. The segmentation is achieved through clustering algorithms that divide the pattern space into distinct regions with different defect characteristics.
Solution Approach 2:
The patent employs parameter changes by using multiple image parameters (focus parameter, exposure parameter, pattern width) to characterize patterns at different stages. By adjusting and analyzing multiple parameters simultaneously, the system can accurately distinguish between normal and defective patterns. The determination functions dynamically adjust parameter thresholds based on learned characteristics from training data, enabling adaptive defect detection.
2Reliability
If comprehensive pattern analysis is performed on all sample patterns, then defect detection coverage is high, but processing time increases
Solution Approach 1:
The patent applies partial action by performing comprehensive analysis only on patterns classified as suspicious weak patterns, rather than analyzing all patterns equally. The system first performs rapid initial classification to identify a small subset of patterns with high defect probability, then applies detailed analysis only to this subset. This approach maintains high defect detection coverage while significantly reducing overall processing time by avoiding exhaustive analysis of clearly normal patterns.
Solution Approach 2:
The patent implements preliminary action through the initial classification step that divides patterns into normal, weak, and suspicious weak groups before detailed defect analysis. By performing this preliminary sorting based on image parameters and determination functions, the system prepares the data in advance to enable focused analysis on only those patterns that require detailed inspection, thereby reducing total processing time while maintaining comprehensive defect detection.
3Measurement precision
If multiple determination functions are extracted and compared, then classification accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the determination function extraction process by creating separate determination functions for different pattern groups (normal group determination function, weak group determination function). Each function is optimized for its specific group characteristics, allowing accurate classification without requiring a single complex universal function. This segmentation reduces computational complexity by dividing the problem into smaller, more manageable sub-problems that can be solved independently.
Solution Approach 2:
The patent applies local quality by extracting determination functions that are specific to local pattern characteristics rather than using a single global function. The normal group determination function and weak group determination function are each tailored to the specific features of their respective groups. This local optimization allows each function to achieve high accuracy for its target group without the computational overhead of a comprehensive universal function, as each function only needs to handle the specific variations within its group.
Data Source
AI summary
The method includes classifying sample pattern data into a standard normal group and a standard weak group based on a first criterion. The method further includes extracting a normal group determination function by calculating an image parameter with respect to each piece of sample pattern data included in the standard normal group, and extracting a weak group determination function by calculating the image parameter with respect to each piece of sample pattern data included in the standard weak group. The method also includes classifying the object pattern data into a normal group and a weak group by calculating the image parameter with respect to object pattern data based on a first proximity between the normal group determination function and the object pattern data and a second proximity between the weak group determination function and the object pattern data.