Image Hashing for Wafer Pattern Synonym Grouping

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current defect detection methods for semiconductor wafers, such as pattern synonym detection, are tedious, manual, time-consuming, and impractical, especially as dimensions shrink, leading to inefficient production schedules and unmanageable groupings that hinder root cause analysis.

Innovation Solution

An image hashing-based unsupervised pattern synonym detection system that groups similar patterns together using fixed length hash strings, reducing the number of design bins and enabling quick, unsupervised pattern synonym detection across wafers, layers, and devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual pattern synonym detection methods are used, then detection accuracy can be achieved, but the process is tedious and time-consuming

Engineering Contradiction:
Improvedetection accuracyVSAvoidanalysis time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces manual mechanical inspection with an automated optical inspection system that captures wafer images and uses image processing algorithms to automatically detect pattern synonyms. This substitution eliminates the time-consuming manual process while maintaining detection accuracy through systematic image analysis.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates hash representations (copies) of wafer images that capture essential pattern features. By working with these compressed hash copies instead of full-resolution images, the system enables rapid comparison and detection of pattern synonyms without analyzing complete image data, significantly reducing processing time.

Inventive Principle:
Principle #26Copying

2Reliability

If OPC rule-based search is used to detect pattern synonyms, then detection capability is improved, but the process becomes slow and cannot be practically implemented for all patterns

Engineering Contradiction:
Improvedetection capabilityVSAvoidprocessing speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent transforms wafer images into fixed-length hash representations, fundamentally changing the data parameter from high-dimensional image data to compact numerical vectors. This parameter transformation enables rapid computation and comparison, allowing the system to process all patterns at high speed while maintaining reliable detection capability through the preserved essential features in the hash representation.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If design-based grouping algorithm is used, then processing speed is improved, but it divides the dataset into an unmanageable number of groups

Engineering Contradiction:
Improveprocessing speedVSAvoidnumber of groups
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent performs preliminary clustering by grouping wafer images based on their hash representations before conducting detailed pattern synonym detection. This preliminary action organizes the dataset into manageable clusters that share similar characteristics, reducing the overall complexity and making subsequent analysis more tractable while maintaining processing speed.

Inventive Principle:
Principle #10Preliminary action

4Loss of time

If exact search algorithm is used, then processing time is reduced, but designs with similar root cause are split into multiple groups

Engineering Contradiction:
Improveturnaround timeVSAvoidroot cause analysis capability
Core Design Contradiction:
Loss of timeVSLoss of information

Solution Approach 1:

The patent applies local quality by focusing detection on specific critical regions and features within wafer images rather than treating all areas uniformly. The hash representation captures essential local features that are most relevant to pattern synonym detection, enabling fast processing while preserving the ability to identify designs with similar root causes through targeted feature analysis.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4200897B1Unsupervised pattern synonym detection using image hashing
Publication Date: 2025.11.05 KLA CORP
  • EP4200897B1 patent drawingFigure 1
  • EP4200897B1 patent drawingFigure 2
  • EP4200897B1 patent drawingFigure 3

AI summary

Images of semiconductor wafers can be hashed to determine a fixed length hash string for each of the images. Pattern synonyms can be determined from the hash strings. The pattern synonyms can be grouped. A degree of similarity between images in the groups is adjustable via a hamming distance. This can be used for various applications, including determination of latent defects.