Semiconductor Image Reconstruction Verification Using Defect Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ML-based image reconstruction techniques for semiconductor examination introduce uncertainties and inconsistencies, leading to unreliable synthetic images that can compromise defect detection and other examination processes, due to unpredictable network behavior and the introduction of artificial artifacts.
Innovation Solution
A robust framework using two machine learning models is employed to generate and verify the quality of synthetic images, ensuring accuracy by comparing defect maps with ground truth information, thereby ensuring the reliability of the reconstruction process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML-based image reconstruction is used to generate synthetic images, then image quality and examination effectiveness are improved, but reliability deteriorates due to unpredictable network behavior and artificial artifacts
Solution Approach 1:
The patent implements a feedback mechanism where a second ML model processes both the input image and synthetic image to generate a defect map, which is then compared with ground truth defect information. This feedback loop enables verification of synthetic image quality and identification of artifacts, thereby improving reliability while maintaining image quality enhancements.
Solution Approach 2:
The patent introduces an intermediary verification system consisting of a second ML model that acts as a mediator between the image reconstruction process and the final examination results. This intermediary generates defect maps and compares them with ground truth data, serving as a quality control mechanism that ensures the reliability of synthetic images without compromising their enhanced quality.
2Device complexity
If a single ML model is used for image reconstruction, then the process is simple, but quality verification and artifact detection are insufficient
Solution Approach 1:
The patent segments the image reconstruction and verification process into two distinct ML models: a first model dedicated to generating synthetic images from input images, and a second model dedicated to processing images and generating defect maps. This segmentation allows each model to specialize in its specific function, improving overall quality verification accuracy while maintaining reasonable process simplicity through clear division of responsibilities.
Data Source
AI summary
There is provided a system and method of examining a semiconductor specimen. The method includes obtaining an input image of the semiconductor specimen; processing the input image using a first machine learning (ML) model, to obtain a synthetic image corresponding to the input image, where the synthetic image is reconstructed to resemble a target image pertaining to the specific application; processing, by a second ML model, the synthetic image and one of the input image or the target image of the synthetic image, to obtain a defect map indicative of defect distribution in the input image or the target image with respect to the synthetic image; and verifying quality of the synthetic image based on the defect map. The first ML model is previously trained for image reconstruction for a specific application, and the second ML model is previously trained for defect detection.