CNN Root Cause Analysis for Image Sensor Defect Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The manufacturing process of image sensor arrays is hindered by defects such as voids, clusters, and excessive leakage due to particles and dust, which require extensive human expertise and resources for root cause analysis, limiting the ability to evaluate all defective pixels and leading to inefficiencies in quality control.

Innovation Solution

A system utilizing Convolutional Neural Networks (CNNs) for automated statistical root cause analysis, combining in-line optical inspection data with end-of-line electrical data to classify defects and generate root cause statistics, reducing the need for human intervention and enabling comprehensive defect evaluation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual root cause analysis is performed by human experts, then analysis accuracy can be maintained, but the analysis time and resources increase significantly

Engineering Contradiction:

Improvedefect analysis accuracyVSAvoidanalysis time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces manual human expert analysis with an automated machine learning system that uses convolutional neural networks to classify defects and identify root causes. This substitution maintains high accuracy while dramatically reducing analysis time from hours/days to minutes, resolving the contradiction between measurement precision and time loss.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system creates digital copies of defect images and data, processing multiple copies simultaneously through the ML model. This allows comprehensive analysis of all defective pixels without the sequential limitations of manual review, improving both speed and completeness while maintaining accuracy through consistent automated classification.

Inventive Principle:

Principle #26Copying

2Reliability

If all defective pixels are evaluated manually, then comprehensive quality assessment is achieved, but the resource requirements and complexity increase

Engineering Contradiction:

Improvequality control completenessVSAvoidanalysis system complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces complex manual analysis processes with an automated machine learning system that handles the complexity internally. The ML model automatically evaluates all defective pixels, correlating optical inspection data with electrical test data to provide comprehensive quality assessment without requiring human experts to manually process each defect, thus maintaining reliability while managing complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs self-service by automatically classifying defects, identifying root causes, and generating quality reports without human intervention. The ML model autonomously processes all defective pixels, correlates data from multiple sources, and provides comprehensive quality assessment, eliminating the need for extensive human resources while maintaining complete evaluation.

Inventive Principle:

Principle #25Self-service

3Productivity

If automated defect classification is implemented, then analysis speed increases, but the precision and expertise level may decrease

Engineering Contradiction:

Improvedefect analysis speedVSAvoiddefect classification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements automated defect classification using convolutional neural networks that have been trained on extensive defect data. This substitution achieves high-speed processing while maintaining or exceeding the precision of manual analysis, as the ML model learns from numerous examples and applies consistent classification criteria across all defects, eliminating human variability while preserving accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs preliminary training and calibration using extensive defect datasets before deployment. This preliminary action prepares the ML model to accurately classify defects at high speed, ensuring that the automated system achieves both productivity and precision by learning from comprehensive training data before actual defect analysis begins.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11803953B2Manufacturing quality improvement through statistical root cause analysis using convolution neural networks

Publication Date: 2023.10.31 DPIX LLC

AI summary

A quality improvement method for an image sensor array includes collecting in-line optical inspection data of the image sensor, collecting end of line electrical data of the image sensor, creating defect maps and obtaining x-y coordinates of the optical inspection data and the electrical data, correlating the defect maps to generate correlated defects, classifying the images of the correlated defects, and generating root cause statistics of the classified correlated defects.