Multimodal Defect-Cause Recommendation for Manufacturing Processes

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Identifying the cause of defects in manufacturing processes, such as semiconductor manufacturing, is time-consuming for human experts and often loses important image information during classification, and existing systems lack effective analysis tools for cross-referencing defect patterns.

Innovation Solution

A neural network model is employed to receive user inquiries about defect phenomena, encode information using text and image modalities, and generate responses by searching for similar cases through a matching module, including a transformer-based encoder and a large language model to recommend defect-causing processes, facilities, and chambers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If human experts manually classify defect patterns and identify defect-causing facilities, then analysis accuracy can be maintained, but the time required for defect analysis becomes significantly large

Engineering Contradiction:

Improvedefect analysis accuracyVSAvoidtime required for defect analysis

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the manual mechanical process of human experts classifying defect patterns with an automated neural network model. The model encodes defect images and text information, searches for similar cases in a database, and recommends defect-causing processes, thereby eliminating the time-consuming manual classification while maintaining analysis accuracy through computational intelligence

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service defect analysis by allowing users to input defect information and receiving automated recommendations from the neural network model without requiring expert intervention. The model independently performs encoding, similarity searching, and defect cause recommendation, making the defect analysis process autonomous and accessible to non-experts

Inventive Principle:

Principle #25Self-service

2Ease of operation

If defect images are codified into classification categories, then defect pattern classification is simplified, but important image information is lost

Engineering Contradiction:

Improvedefect classification simplicityVSAvoidimage information loss

Core Design Contradiction:

Ease of operationVSLoss of information

Solution Approach 1:

The patent segments the defect analysis into multiple modalities: text information encoding and image information encoding are performed separately by different encoders (text encoder and image encoder). This segmentation allows each modality to be processed independently while preserving their unique characteristics, preventing information loss that would occur in traditional single-category codification

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system creates a composite representation by combining text features and image features into a unified query vector through an adapter module. This composite approach integrates the simplicity of text classification with the rich visual information from images, achieving both ease of operation and information preservation

Inventive Principle:

Principle #40Composite materials

3Adaptability or versatility

If a separate analysis system is implemented for cross-analysis of defect-suspected facilities, then comprehensive defect analysis is enabled, but system complexity increases

Engineering Contradiction:

Improvedefect analysis comprehensivenessVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The neural network model serves multiple functions within a single integrated system: it encodes text information, encodes image information, searches for similar cases, and recommends defect-causing processes. This multi-functionality eliminates the need for separate analysis systems while maintaining comprehensive defect analysis capabilities across different facilities and processes

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent merges the text encoding module, image encoding module, similarity searching module, and recommendation module into a single integrated neural network model. This consolidation combines multiple analysis functions that would traditionally require separate systems, reducing overall system complexity while enhancing adaptability and versatility in defect analysis

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20250322001A1Apparatus and method with defect-cause recommending

Publication Date: 2025.10.16 SAMSUNG ELECTRONICS CO LTD

AI summary

An apparatus and method for recommending a defect-causing process are disclosed. The apparatus for recommending a defect-causing process includes a communication interface configured to receive a user's inquiry including identification information related to a defect phenomenon occurring in a target process, and a neural network model configured to search for a similar case related to the defect phenomenon by encoding information related to the defect phenomenon based on the identification information and generate a response to the user's inquiry by using a prompt generated based on the user's inquiry and the similar case.