Semiconductor Equipment Diagnosis Using CNN and XAI Defect Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor equipment diagnosis methods face challenges in interpreting Tool To Tool Matching (TTTM) results due to increasing complexity with the number of inputs, making it difficult to correlate these results with product quality and identify defect causes effectively.
Innovation Solution
Employing a Convolutional Neural Network (CNN) and eXplainable AI (XAI) to analyze time series data from substrate treating processes, converting this data into image information, and classifying features to determine substrate defects and identify their causes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If TTTM analysis is applied to measure equipment performance, then equipment compatibility is improved, but interpretation difficulty increases as the number of inputs increases
Solution Approach 1:
The patent replaces traditional mechanical/statistical analysis methods with AI-based image processing. Time series data from multiple equipment inputs is transformed into image format, allowing CNN algorithms to automatically extract features and identify patterns without manual analysis, thus resolving the interpretation difficulty while maintaining TTTM compatibility assessment
Solution Approach 2:
The patent introduces an intermediary transformation step that converts complex time series data into image representation. This intermediary format serves as a bridge between raw multi-parameter equipment data and human-interpretable diagnostic results, enabling both automated analysis and visual interpretation
2Ease of manufacture
If traditional diagnosis methods are used, then implementation simplicity is maintained, but diagnostic accuracy decreases for complex equipment failures
Solution Approach 1:
The patent performs preliminary transformation of time series data into image format before analysis. This preprocessing step prepares the data in an optimal format for pattern recognition, enabling the subsequent CNN-based diagnosis to achieve high accuracy while keeping the overall system implementation straightforward through automated processing
3Measurement precision
If detailed TTTM results are generated for each input, then measurement completeness is improved, but result interpretability worsens
Solution Approach 1:
The patent transforms one-dimensional time series data from multiple inputs into a two-dimensional image representation. This dimensional transformation allows the system to preserve complete measurement information from all inputs while enabling intuitive visual interpretation through image-based pattern recognition and heat map visualization
Data Source
AI summary
The present invention provides an apparatus and method for diagnosing semiconductor equipment for analyzing a determination result and a cause of a defect for a substrate using a deep learning algorithm and explainable AI. The semiconductor equipment diagnosis method comprises acquiring time series data for a parameter related to a substrate treating process of a substrate treating apparatus, generating table information based on the time series data, converting the table information into image information based on data included in the table information, extracting and classifying a feature from the image information, determining whether a substrate treated by the substrate treating apparatus is good or defective based on the feature, and analyzing a cause of a defect of the substrate in response to determining that the substrate is defective.