Lithography Mask Design Through Training-Free Gradient Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mask generation methods for lithography, particularly for small feature sizes and complex circuit designs, are limited by the need for extensive training data and computational resources, and fail to account for dispersion forces in non-Fraunhofer regimes, making them inefficient and inflexible.
Innovation Solution
A gradient-based optimizer directly generates lithography masks from desired patterns without training, applicable to various lithography systems, including photon, particle-based, and scalar waves, using a computer-implemented method that iteratively refines mask templates through gradient descent to minimize error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing mask generation methods are used, then training data and computational resources are required, but this increases device complexity and generation time
Solution Approach 1:
The patent extracts the essential function of mask generation by removing the training data requirement and extensive computational resources from the process. The gradient-based optimizer directly generates masks from desired patterns without needing large training datasets, thereby reducing device complexity while maintaining manufacturing precision.
Solution Approach 2:
The patent replaces the mechanical system of training-based machine learning with a direct gradient-based optimization approach. This substitution eliminates the need for extensive training data and complex computational models, reducing device complexity while achieving accurate mask generation through iterative refinement guided by gradient descent.
2Manufacturing precision
If existing mask generation methods are used, then training is required, but this increases loss of time
Solution Approach 1:
The patent applies preliminary action by directly generating mask templates through gradient-based optimization without requiring prior training. The method starts with an initial mask template and iteratively refines it using gradient descent, achieving accurate mask patterns faster than traditional training-based methods.
Solution Approach 2:
The patent skips the extensive training phase required by traditional machine learning methods. By using a direct gradient-based optimization approach, the system rushes through the mask generation process, achieving accurate patterns in significantly less time without the need for large training datasets.
3Ease of manufacture
If grid-based binary holography is used, then simple mask structure is achieved, but this limits adaptability to complex circuit designs
Solution Approach 1:
The patent introduces dynamics by using a gradient-based optimizer that can adaptively adjust mask parameters during generation. This allows the system to transition from simple grid-based structures to complex patterns suitable for modern circuit designs, maintaining ease of manufacture while significantly improving adaptability.
Solution Approach 2:
The patent applies parameter changes by iteratively refining mask properties through gradient descent optimization. The method can adjust various parameters such as hole positions, sizes, and distributions to transform simple grid-based masks into complex patterns tailored to specific circuit designs, thereby enhancing adaptability without sacrificing manufacturability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer implemented method and a data processing system for the design and creation of a lithography mask (4) configured to be arranged between a source (2) with a beam (21) and a target (5) wherein the target has a predefined desired pattern (51), wherein the method comprises; - creating an initial mask template (41) with mask template properties (401), - repeatedly until an intermediate error (62) is acceptable (60); - predicting an intermediate pattern (52) produced in the target (5) by the mask template (42), - determining the intermediate error (62) of the intermediate pattern (52) relative to the desired pattern (51), and - improving the mask template properties (401) of the mask template (42) based on the error (60).