3D Design to SEM Image Prediction for Faster Defect Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional Monte Carlo simulation techniques for predicting defects in semiconductor fabrication are computationally expensive and time-consuming, limiting throughput and unable to render design images from SEM images or determine material properties.
Innovation Solution
A system and method utilizing deep learning predictive models, specifically a conditional generative adversarial network (CGAN), to generate simulated SEM and 3D design images based on training images and 3D design data, enabling faster defect identification and material property prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo simulations are used to generate SEM images from design data, then defect prediction accuracy is improved, but computational time and cost increase significantly
Solution Approach 1:
The system performs preliminary actions by training a deep learning model on Monte Carlo simulation data in advance. Once trained, the model can rapidly generate SEM images without requiring real-time Monte Carlo simulations, thus resolving the contradiction between accuracy and computational time.
Solution Approach 2:
The invention creates a copy of the complex Monte Carlo simulation process through a deep learning model. The trained neural network replicates the simulation outcomes, providing accurate defect predictions without the computational burden of running actual Monte Carlo simulations for each inspection.
2Measurement precision
If Monte Carlo simulations are used for defect prediction, then imaging conditions can be identified, but throughput is reduced due to computational expense
Solution Approach 1:
The system pre-trains the deep learning model with comprehensive Monte Carlo simulation data covering various imaging conditions and defect types. This preliminary training enables the model to rapidly predict defects and identify optimal imaging conditions during actual production, maintaining precision while dramatically improving throughput.
Solution Approach 2:
The invention replaces the computationally intensive Monte Carlo simulation mechanism with a deep learning-based predictive model. This substitution maintains the ability to identify imaging conditions and predict defects while enabling high-speed processing suitable for production throughput.
3Loss of time
If deep learning models are trained on limited data, then training time is reduced, but model accuracy and generalization capability deteriorate
Solution Approach 1:
The system performs preliminary action by pre-training the deep learning model on a comprehensive dataset of Monte Carlo simulations covering diverse defect types, locations, and imaging conditions. This pre-training ensures high model accuracy and generalization capability before deployment, while the trained model then operates rapidly during production without requiring additional training time.
Data Source
AI summary
A system for characterizing a specimen is disclosed. In one embodiment, the system includes a characterization sub-system configured to acquire one or more images a specimen, and a controller communicatively coupled to the characterization sub-system. The controller may be configured to: receive training images of one or more features of a specimen from the characterization sub-system; receive training three-dimensional (3D) design images corresponding to the one or more features of the specimen; generate a deep learning predictive model based on the training images and the training 3D design images; receive product 3D design images of one or more features of a specimen; generate simulated images of the one or more features of the specimen based on the product 3D design images with the deep learning predictive model; and determine one or more characteristics of the specimen based on the one or more simulated images.