Plasma Lithography Imaging Model for Fast 2D Pattern Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating plasma lithography imaging are inefficient and inaccurate, particularly for two-dimensional patterns, due to the high computational demands of numerical simulations and limited applicability of analytical methods to three-dimensional models.
Innovation Solution
A method and apparatus for training a fast imaging model by constructing a structure with periodically repeating training mask patterns, converting these patterns into matrices, and using least square fitting to train a model that directly computes training image patterns, enabling fast and accurate simulation of plasma lithography imaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If numerical simulation methods are used for plasma lithography imaging, then accuracy is improved, but computational efficiency deteriorates
Solution Approach 1:
The patent creates a simplified computational model that copies the essential imaging characteristics of plasma lithography without replicating the full complexity of numerical simulations. By training a machine learning model on numerical simulation data, the system creates a surrogate model that reproduces imaging results with high accuracy but at much lower computational cost, effectively copying the behavior of the complex system in a simplified form.
Solution Approach 2:
The patent performs preliminary numerical simulations to generate training data before deploying the fast imaging model. By pre-computing a comprehensive dataset covering various mask patterns and imaging conditions, the system prepares all necessary information in advance, allowing subsequent imaging tasks to be performed rapidly without repeated heavy numerical computations.
2Productivity
If analytical methods are used for plasma lithography imaging, then computational efficiency is improved, but applicability to three-dimensional models deteriorates
Solution Approach 1:
The patent develops a machine learning-based imaging model that serves multiple functions: it can handle both two-dimensional and three-dimensional mask patterns, accommodate various imaging conditions, and provide results with accuracy comparable to numerical simulations. The universal model is trained on diverse datasets covering different pattern types and dimensions, enabling it to adapt to various imaging scenarios without requiring separate analytical approaches for each case.
3Reliability
If traditional lithography methods are used, then diffraction limits are maintained, but implementation simplicity is preserved
Solution Approach 1:
The patent replaces traditional optical lithography mechanisms with a computational approach using machine learning models. Instead of relying on physical optical systems and their inherent diffraction limits, the system uses trained neural networks to predict imaging results, substituting physical optical processes with computational algorithms that can model evanescent wave effects and near-field imaging without being constrained by diffraction limits.
Data Source
AI summary
A method and an apparatus for training on imaging of plasma lithography. The method comprises: determining a structure for training on an imaging of the plasma lithography is determined, where a training mask pattern repeats periodically along two directions in the structure; constructing a model simulating the structure; obtaining a training image pattern of the plasma lithography is obtained through computation based on the model, where the training image pattern corresponds to the training mask pattern; and training a fast imaging model through the training mask pattern and the training image pattern to obtain a trained imaging model for the training mask pattern.