ML Classification Engine for Semiconductor Defect Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In integrated circuit fabrication, the shrinking size of critical dimensions and increasing circuit density pose challenges in accurately reproducing circuit patterns using conventional optical lithography, leading to difficulties in defect detection and classification on photomasks, with existing methods being inefficient in distinguishing between similar errors and requiring extensive manual inspection.

Innovation Solution

The use of machine learning techniques, specifically autoencoding and classification engines, to compress and classify 2D shape features in semiconductor or flat panel display designs, allowing for faster and more accurate identification of defects by grouping similar errors and prioritizing features for further inspection, thereby enhancing the efficiency of defect detection and classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If conventional optical lithography is used to reproduce circuit patterns, then manufacturing process is simple, but manufacturing precision deteriorates as critical dimensions shrink

Engineering Contradiction:

Improvepattern reproduction accuracyVSAvoidlithography process complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary classification system that acts as a mediator between the lithography process and defect detection. By classifying defects into categories (process defects, pattern defects, contamination defects) and further subdividing them, the system enables more precise analysis and correction of pattern reproduction issues without directly complicating the lithography process itself.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces manual defect inspection and classification with an automated machine learning-based classification engine. This substitution of mechanical/manual processes with computational systems enables faster, more consistent, and more accurate defect categorization, thereby improving manufacturing precision without proportionally increasing process complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If manual inspection methods are used to detect defects, then inspection accuracy can be maintained, but inspection time increases significantly

Engineering Contradiction:

Improvedefect inspection speedVSAvoiddefect classification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent replaces manual defect inspection with an automated machine learning classification system that processes defect data rapidly and consistently. The classification engine uses trained models to categorize defects automatically, maintaining high accuracy while dramatically increasing inspection speed and throughput.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The classification system is designed to be self-learning and self-improving through machine learning algorithms. The system automatically categorizes defects, identifies patterns, and refines its classification accuracy over time without requiring constant manual intervention or recalibration, thereby maintaining both speed and precision.

Inventive Principle:

Principle #25Self-service

3Measurement precision

If all pattern instances are inspected individually for design rule violations, then detection completeness is achieved, but inspection time increases due to repetition

Engineering Contradiction:

Improvedefect detection completenessVSAvoidinspection time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent merges the inspection process by classifying defects at a pattern level rather than individually inspecting each pattern instance. The classification engine groups similar defects and applies learned patterns to identify violations across multiple instances efficiently, maintaining detection completeness while reducing redundant inspection time.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system performs preliminary classification and pattern recognition before detailed individual inspection. By pre-processing defect data through the machine learning classifier and identifying likely violation patterns, the system prepares the inspection process in advance, enabling faster subsequent verification while maintaining thoroughness.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11263496B2Methods and systems to classify features in electronic designs

Publication Date: 2022.03.01 D2S INC

AI summary

Methods for matching features in patterns for electronic designs include inputting a set of pattern data for semiconductor or flat panel displays, where the set of pattern data comprises a plurality of features. Each feature in the plurality of features is classified, where the classifying is based on a geometrical context defined by shapes in a region. The classifying uses machine learning techniques.