Virtual EDS System for Real-Time Defect Composition Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting the material composition of defects in semiconductor manufacturing, such as energy dispersive x-ray spectroscopy (EDS), are time-consuming and potentially destructive, and struggle with accurately predicting composition from scanning electron microscope (SEM) images due to defects' random shapes, sizes, and textures, as well as diverse appearances based on source and composition.
Innovation Solution
A computer-vision based system that analyzes SEM images to virtually execute EDS operations in real-time, using a processor to extract image features, classify them, predict the associated elements and their shapes, and grade defects based on predetermined criteria, thereby eliminating the need for destructive methods and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional EDS method is used to detect defect composition, then measurement precision is improved, but loss of time increases and the wafer may be damaged
Solution Approach 1:
The patent creates a virtual copy of the EDS system using machine learning models trained on SEM images and corresponding EDS spectral data. The virtual EDS system reproduces the material composition analysis functionality of physical EDS equipment, enabling rapid defect characterization without requiring actual X-ray spectroscopy measurements, thus reducing inspection time while maintaining composition detection accuracy
Solution Approach 2:
The patent performs preliminary training of machine learning models using datasets containing SEM images and corresponding EDS spectral information before actual inspection. This preliminary action creates a pre-trained virtual EDS system that can rapidly analyze defects without requiring time-consuming real-time EDS measurements during production, thereby reducing loss of time while preserving measurement precision
2Measurement precision
If conventional EDS method is used to detect defect composition, then measurement precision is improved, but object-affected harmful factors increase due to potential wafer damage
Solution Approach 1:
The patent introduces a virtual EDS system based on machine learning as an intermediary between SEM imaging and material composition analysis. Instead of directly using physical EDS equipment that requires X-ray excitation and may damage the wafer, the virtual system processes SEM images through trained models to predict composition, eliminating the harmful effects of physical EDS while maintaining measurement precision
Solution Approach 2:
The patent creates a virtual replica of EDS functionality that mimics the analytical capabilities of physical EDS equipment without requiring actual X-ray spectroscopy measurements. This copy enables defect composition detection while avoiding the potential wafer damage associated with conventional EDS methods
3Productivity
If virtual EDS system is implemented, then productivity is improved through real-time analysis, but device complexity increases due to machine learning infrastructure
Solution Approach 1:
The patent develops a universal virtual EDS system using machine learning models that can analyze multiple types of defects across different semiconductor devices through a single platform. This multi-functional approach enables real-time inspection throughput improvement without requiring separate complex systems for each defect type, thereby managing device complexity while enhancing productivity
Data Source
AI summary
Provided is a method for virtually executing an operation of an energy dispersive x-ray spectrometry (EDS) system in real time production line by analyzing a defect included in a material undergoing inspection based on computer vision, the method including receiving a scanning electron microscope (SEM) image of the material including the defect, extracting an image-feature from the SEM image of the material, classifying the extracted image-feature under a predetermined label, predicting, based on the classified image-feature, an element associated with the defect included in the material and a shape of the predicted element, and grading the defect included in the material based on comparing the predicted element with a predetermined criteria.