Lithography Patterning Model Training for Low-k1 Pattern Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lithographic technologies face challenges in accurately reproducing patterns with dimensions smaller than the classical resolution limit, particularly due to the complexity of low-k1 lithography, where sophisticated fine-tuning steps are required to achieve desired electrical functionality and performance, and there is a need for improved methods to measure and predict pattern characteristics on substrates.
Innovation Solution
Training a patterning process model using a deep learning convolutional neural network (CNN) to cooperatively predict patterns on substrates, combining physical models with machine learning to enhance accuracy and reduce overfitting, and employing machine learning models to determine optical proximity corrections and etch biases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional lithographic methods are used to print patterns with dimensions smaller than the classical resolution limit, then manufacturing precision deteriorates, but using low-k1 lithography with sophisticated fine-tuning steps improves manufacturing precision while increasing device complexity and process time
Solution Approach 1:
The patent replaces complex physical lithographic processes with computational models. A trained patterning process model (using machine learning and physical models) predicts pattern characteristics, substituting the need for sophisticated fine-tuning steps and multiple lithographic iterations. This computational approach maintains manufacturing precision while reducing process complexity.
Solution Approach 2:
The patent performs preliminary computational analysis before actual lithography. The trained model predicts pattern outcomes in advance, allowing optimization of lithographic parameters beforehand. This preliminary action prevents the need for complex real-time adjustments and fine-tuning steps during the manufacturing process.
2Manufacturing precision
If sophisticated fine-tuning steps are applied to achieve desired pattern accuracy, then manufacturing precision improves, but loss of time increases due to multiple processing steps
Solution Approach 1:
The patent substitutes time-consuming physical trial-and-error fine-tuning with rapid computational modeling. The trained patterning process model instantly predicts pattern outcomes, eliminating the need for multiple iterative fine-tuning steps and associated wait times for measurement and adjustment.
Solution Approach 2:
The patent creates a virtual copy of the lithographic process through computational modeling. The trained model simulates pattern formation and predicts outcomes without requiring actual physical iterations, allowing rapid virtual experimentation and optimization that eliminates real-time processing delays.
3Measurement precision
If physical models alone are used to predict patterning processes, then device complexity remains low, but measurement precision deteriorates due to model inaccuracies
Solution Approach 1:
The patent creates a composite modeling approach combining physical models with machine learning models. The physical model provides theoretical foundation while the machine learning model captures complex empirical relationships. This composite model achieves high measurement precision by leveraging the strengths of both approaches without requiring overly complex individual components.
Solution Approach 2:
The patent merges physical modeling with machine learning in a unified patterning process model. The physical model handles fundamental processes while the machine learning model handles empirical variations and complex interactions. This merging achieves high prediction accuracy while keeping individual model components at manageable complexity levels.
Data Source
AI summary
A method for training a patterning process model, the patterning process model configured to predict a pattern that will be formed by a patterning process. The method involves obtaining an image data associated with a desired pattern, a measured pattern of the substrate, a first model including a first set of parameters, and a machine learning model including a second set of parameters; and iteratively determining values of the first set of parameters and the second set of parameters to train the patterning process model. An iteration involves executing, using the image data, the first model and the machine learning model to cooperatively predict a printed pattern of the substrate; and modifying the values of the first set of parameters and the second set of parameters such that a difference between the measured pattern and the predicted pattern is reduced.