Simulated Image Generation for Semiconductor Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor fabrication processes face challenges in achieving high precision and uniformity due to defects and process variations, which are difficult to detect and correct using traditional examination methods.
Innovation Solution
A system and method utilizing a machine learning module with a displacement matrix to generate simulated images of semiconductor specimens from design images, optimizing parameters to correct pixel positions and reduce defects and process variations, enabling efficient and precise examination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional examination methods are used to detect defects on semiconductor specimens, then the examination process can be performed with simple equipment, but the detection precision and ability to identify process variations are insufficient
Solution Approach 1:
A simulated image is introduced as an intermediary between the design image and the actual specimen image. The simulated image, generated by a machine learning module, serves as a reference that bridges the gap between ideal design specifications and real-world manufacturing variations, enabling more precise defect detection without requiring complex examination equipment
Solution Approach 2:
Instead of directly analyzing the complex actual specimen image, the system creates a copied representation (simulated image) that reproduces the essential features and process variations. This simulated copy can be precisely compared against the design image to identify defects, achieving high measurement precision while keeping the examination system relatively simple
2Productivity
If automated examination processes are implemented to increase examination effectiveness, then defect detection capability is improved, but the complexity of the examination system increases
Solution Approach 1:
The machine learning module is trained using actual specimen images and design images, enabling it to automatically learn and adapt to process variations. Once trained, the module autonomously generates simulated images and identifies defects without requiring complex manual configuration or intervention, thereby increasing examination effectiveness while maintaining system simplicity
Solution Approach 2:
The machine learning module is trained in advance using a dataset of images before actual defect detection begins. This preliminary training action enables the module to automatically perform defect detection and classification during production, significantly improving examination productivity without requiring complex real-time analysis systems
3Measurement precision
If the machine learning module is trained with more comprehensive training samples, then the accuracy of simulated image generation is improved, but the training time and computational resources increase
Solution Approach 1:
The system uses a displacement matrix that captures the essential geometric transformations and process variations needed for accurate simulation. Rather than requiring exhaustive training on all possible variations, the displacement matrix provides a compact representation that achieves high simulated image accuracy with reduced training time and computational resources
Data Source
AI summary
There is provided a system to examine a semiconductor specimen, the system comprising a processor and memory circuitry configured to obtain a training sample comprising an image of a semiconductor specimen and a design image based on design data, train a machine learning module, wherein the training includes minimizing a function representative of a difference between a simulated image generated by the machine learning module based on a given design image, and a corrected image corresponding to a given image after correction of pixel position of the given image in accordance with a given displacement matrix, wherein the minimizing includes optimizing parameters of the machine learning module and of the given displacement matrix, wherein the trained machine learning module is usable to generate a simulated image of a specimen based on a design image of the specimen.