Machine Learning Model for Optical Image Resolution Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for optically imaging and analyzing small features in semiconductor fabrication are limited by optical resolution, requiring time-consuming and expensive follow-up imaging with SEM, and are challenging to create training data for machine learning models.
Innovation Solution
A characterization system and method that uses machine learning techniques to enhance image quality by correlating training images with design images, allowing for the generation of enhanced images without the need for expensive SEM imaging, using optical images for training and enhancement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If optical imaging is used to image small features, then imaging speed and efficiency are improved, but image resolution and measurement precision deteriorate
Solution Approach 1:
The patent creates virtual SEM images by copying and transforming optical image data through machine learning models. The system generates synthetic high-resolution images that mimic SEM quality without actually performing SEM imaging, thus maintaining optical imaging speed while achieving SEM-level resolution through computational copying of the visual information
Solution Approach 2:
The patent replaces the mechanical SEM imaging system with a computational approach using machine learning models trained on optical and SEM image pairs. Instead of physically using an SEM to capture high-resolution images, the system substitutes the mechanical imaging process with an algorithmic transformation that converts optical images into enhanced representations
2Measurement precision
If SEM imaging is used to obtain detailed images, then image resolution is improved, but time consumption and cost increase
Solution Approach 1:
The patent performs preliminary training by creating a machine learning model using a dataset of optical-SEM image pairs before actual inspection. This preliminary action captures the transformation rules between optical and SEM images, so that during production inspection, only the trained model needs to be applied to optical images to generate SEM-quality outputs without performing actual SEM imaging
Solution Approach 2:
The system creates virtual copies of SEM images through computational processing of optical images using the trained machine learning model. These synthetic images replicate the detailed information normally requiring SEM capture, eliminating the need for time-consuming physical SEM imaging while maintaining diagnostic quality
3Measurement precision
If machine learning training data is created using SEM images, then model accuracy is improved, but cost and complexity increase
Solution Approach 1:
The patent makes the optical imaging system multi-functional by enabling it to serve both as the primary inspection tool and as the source for generating training data. The same optical microscope used for rapid imaging also provides the training dataset when paired with corresponding SEM images, eliminating the need for separate dedicated training data acquisition infrastructure
Solution Approach 2:
The system creates training data by copying and pairing existing optical images with their corresponding SEM images. This copying approach reuses available data without requiring additional physical specimens or complex data generation procedures, simplifying the training data creation process while maintaining model accuracy
Data Source
AI summary
A system and method for enhancing image quality. The system and method acquire a machine learning model trained for correlating one or more training images and one or more training design images. The system and method receive one or more sample specimen images corresponding to one or more features of a sample specimen. The system and method enhance the one or more sample specimen images by generating one or more enhanced images with the machine learning model based on at least the one or more sample specimen images.