Deep Generative Models for Fast CAD-Based Semiconductor Image Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating simulated images from design information, such as forward EM simulation and rule-based approximation, are computationally intensive or result in poor predictions, making them infeasible for real production in semiconductor fabrication processes like mask inspection and wafer metrology.
Innovation Solution
A deep generative model is employed to generate simulated images from design information, utilizing a training dataset of aligned CAD and actual images to learn how design information appears on a specimen, allowing for computationally efficient and realistic rendering of semiconductor inspection and metrology images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If forward EM simulation is used to generate simulated images, then the simulation accuracy is improved, but the computational time and complexity increase significantly making it infeasible for real production
Solution Approach 1:
The system pre-generates a comprehensive library of simulated images covering various defect types, locations, and characteristics before actual inspection. This preliminary simulation phase allows the inspection system to quickly reference pre-computed results during production, avoiding real-time computational overhead while maintaining high accuracy in defect detection
Solution Approach 2:
The patent creates a buffer library of simulated images that cushions against the need for real-time simulation during inspection. By having pre-prepared simulated images for various defect scenarios, the system eliminates the time-consuming forward EM simulation step during actual production inspection, resolving the contradiction between accuracy and speed
2Productivity
If rule-based approximation methods are used to generate simulated images, then the computational speed is improved, but the prediction accuracy deteriorates resulting in poor defect detection
Solution Approach 1:
The system creates accurate copies of actual defect images through sophisticated simulation that replicates real defect characteristics, appearances, and variations. These simulated copies are generated with high fidelity to serve as reliable references for defect detection, maintaining both speed and accuracy by using pre-generated realistic defect representations rather than simple rule-based approximations
Data Source
Figure 1
Figure 1a~3
Figure 2
AI summary
Methods and systems for generating simulated images from design information are provided. One system includes one or more computer subsystems and one or more components executed by the computer subsystem(s), which include a generative model. The generative model includes two or more encoder layers configured for determining features of design information for a specimen. The generative model also includes two or more decoder layers configured for generating one or more simulated images from the determined features. The simulated image(s) illustrate how the design information formed on the specimen appears in one or more actual images of the specimen generated by an imaging system.