ML Model for High-Resolution Image Reconstruction from Inspection Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor inspection and metrology processes face limitations due to low resolution images from optical and electron beam inspection tools, which hinder accurate defect classification and pattern feature analysis, leading to increased time and costs for higher resolution image generation and defect review.
Innovation Solution
A system utilizing a trained machine learning model that transforms low-resolution images from inspection tools into higher resolution images by pre-training and re-training with simulated and actual data, leveraging design information to enhance image reconstruction and reduce the need for physical specimen re-examination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If higher resolution images are generated using high magnification optical systems or SEM, then image resolution is improved, but processing time and operational complexity increase
Solution Approach 1:
The system performs preliminary defect detection using inspection images at lower resolution, then selectively generates higher resolution images only for locations where defects are detected. This preliminary action avoids the time-consuming process of generating high resolution images for the entire specimen, resolving the contradiction between measurement precision and processing time.
Solution Approach 2:
Instead of uniformly processing the entire specimen at high resolution, the system applies high resolution image generation locally only to specific regions where defects are detected. This local quality approach maintains measurement precision for critical areas while significantly reducing overall processing time by avoiding unnecessary high resolution processing elsewhere.
2Measurement precision
If higher resolution images are generated for defect review, then defect classification accuracy is improved, but device complexity and operational requirements increase
Solution Approach 1:
The system performs preliminary defect detection using inspection images before initiating high resolution image generation. This preliminary action identifies which specific regions require detailed analysis, simplifying the operational process by avoiding unnecessary high resolution processing and reducing device complexity while maintaining defect classification accuracy.
Solution Approach 2:
The system integrates multiple functions into a unified workflow: defect detection, high resolution image generation, and defect review are combined in a single automated process. This multi-functionality reduces operational complexity by eliminating the need for separate manual operations while maintaining high defect classification accuracy.
3Measurement precision
If physical specimen re-examination is performed at discrete locations, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The system performs preliminary defect detection using inspection images to identify locations requiring high resolution analysis. This preliminary action enables selective re-examination only at necessary locations, maintaining defect detection accuracy while improving productivity by avoiding unnecessary re-examination of defect-free areas.
Solution Approach 2:
The system applies high resolution analysis locally only to regions where defects are detected, rather than uniformly processing the entire specimen. This local quality approach maintains measurement precision for critical defect areas while significantly improving overall inspection throughput by reducing unnecessary processing elsewhere.
Data Source
AI summary
Methods and systems for determining information for a specimen are provided. The embodiments described herein are configured for training a machine learning (ML) model for generating higher resolution images of a specimen from images of the specimen generated by an inspection subsystem. The training includes a pre-training step that is performed using only simulated images and a re-training step that is performed using actual images of a test specimen. The higher resolution images generated by the trained ML model from lower resolution inspection images can be used for applications to including nuisance filtering and defect classification.