Machine Learning Optical Proximity Correction Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lithographic technologies face challenges in accurately predicting and correcting optical proximity errors, particularly in achieving consistent and large process windows for complex design layouts.
Innovation Solution
A machine learning-based method is employed to predict optical proximity corrections by training a model using data from spatially shifted versions of design patterns and their corresponding optical proximity corrections, allowing for the selection of images with optimal process windows and the application of weighting functions to prioritize critical regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning model is trained using spatially shifted versions of design patterns, then prediction accuracy of optical proximity corrections is improved, but training data preparation complexity increases
Solution Approach 1:
The training data is prepared in advance by generating spatially shifted versions of design patterns and their corresponding optical proximity corrections. This preliminary action allows the machine learning model to learn from diverse configurations during training, improving prediction accuracy without adding complexity during actual operation.
Solution Approach 2:
Spatial shifts are applied as parameter changes to the training data, creating variations in the input patterns. This approach enriches the training dataset with different spatial configurations, enabling the model to generalize better and improve prediction accuracy across various design scenarios.
2Manufacturing precision
If weighting functions are applied to prioritize critical regions, then manufacturing precision is improved, but computational complexity increases
Solution Approach 1:
Weighting functions are applied to assign different importance levels to different regions of the design pattern. Critical regions receive higher weights, directing the optimization focus to areas that most impact manufacturing precision, while less critical regions receive lower weights, reducing unnecessary computational effort.
3Stability of the object's composition
If spatially shifted training data is used, then process window consistency is improved, but data processing time increases
Solution Approach 1:
Spatially shifted versions of training patterns are generated in advance during the data preparation phase. This preliminary action ensures that the model is exposed to various spatial configurations during training, improving process window consistency, while the actual production process benefits from the pre-learned knowledge without requiring real-time data generation.
Data Source
AI summary
A method including: obtaining data based an optical proximity correction for a spatially shifted version of a training design pattern; and training a machine learning model configured to predict optical proximity corrections for design patterns using data regarding the training design pattern and the data based on the optical proximity correction for the spatially shifted version of the training design pattern.