Flash-Based Anti-Aliasing Filter Kernel for Lithography Mask Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing technologies face challenges in converting non-bandlimited pattern layouts into band-limited images for optical lithography processes, as existing methods like DFT-based techniques require costly and computationally intensive sub-pixel sampling to suppress alias noise effectively.
Innovation Solution
A system that uses an anti-aliasing filter (AAF) kernel to convert non-bandlimited pattern layouts into band-limited images by constructing a lookup table with precomputed convolution values, allowing for efficient convolution operations using a flash-based technique, which decomposes polygons into basis functions and interpolates values for grid locations, thereby bypassing the need for sub-pixel sampling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If DFT-based techniques are used to convert non-bandlimited pattern layouts into band-limited images, then alias noise can be suppressed, but computational expense increases significantly due to the need for sub-pixel sampling
Solution Approach 1:
The patent segments the anti-aliasing filter kernel into multiple basis functions (e.g., separable x and y components). Each basis function is precomputed and stored in lookup tables, allowing the convolution operation to be broken down into simpler, reusable components that reduce computational expense while maintaining accuracy in suppressing alias noise.
Solution Approach 2:
The patent performs preliminary computation by precalculating convolution values of the anti-aliasing filter kernel with basis functions and storing them in lookup tables before actual pattern conversion. This preliminary action eliminates the need for repeated expensive computations during runtime, significantly reducing computational expense while preserving alias noise suppression capability.
2Manufacturing precision
If sub-pixel sampling is applied to suppress alias noise in DFT-based techniques, then conversion accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The patent creates copies of precomputed convolution values in lookup tables that can be rapidly accessed during pattern conversion. Instead of performing repeated sub-pixel sampling computations, the system copies and applies precomputed values from lookup tables, maintaining high conversion accuracy while dramatically reducing processing time.
Solution Approach 2:
The patent changes the computational parameters by switching from direct sub-pixel sampling computation to lookup table-based value retrieval. This parameter change maintains the effective resolution and accuracy of sub-pixel sampling while eliminating the computational overhead, thus reducing processing time without sacrificing pattern conversion accuracy.
3Productivity
If flash-based techniques with lookup tables are used for convolution, then computational efficiency improves, but the complexity of constructing and managing lookup tables increases
Solution Approach 1:
The patent segments the anti-aliasing filter kernel into basis functions, which allows the lookup table to be constructed from simpler, reusable components. This segmentation reduces the overall complexity of lookup table construction by breaking down a complex convolution operation into multiple simpler precomputations that can be systematically organized and managed.
Solution Approach 2:
The patent creates basis functions that serve multiple purposes: they are used to construct the lookup tables, they can be applied to different pattern geometries, and they enable the same lookup table to serve various convolution operations. This universality reduces the overall system complexity by eliminating the need for separate computations for different cases.
Data Source
AI summary
Techniques and systems for converting a non-bandlimited pattern layout into a band-limited pattern image are described. During operation, the system receives the non-bandlimited pattern layout which comprises one or more polygons. The system further receives an anti-aliasing filter (AAF) kernel, wherein the AAF kernel is configured to convert a non-bandlimited pattern into a band-limited pattern. The system then constructs an AAF lookup table for the AAF kernel, wherein the AAF lookup table contains precomputed values for a set of convolution functions which are obtained by convolving a set of basis functions with the AAF kernel. Next, the system creates a sampled pattern layout by applying a grid map over the pattern layout. The system then obtains the band-limited pattern image by using the AAF lookup table to convolve the AAF kernel with each grid location in the sampled pattern layout.