Machine Learning Sub-Resolution Assist Feature Generation for Lithography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical proximity correction techniques, including conventional OPC and inverse lithography technology, face challenges in addressing variations in focus conditions and iso-dense bias problems, especially when using off-axis illumination, which can lead to inaccuracies in feature reproduction during photolithography, particularly for complex layouts.
Innovation Solution
The implementation of machine learning-based sub-resolution assist feature (SRAF) generation, which classifies layout features, determines the necessity and characteristics of SRAFs using specific machine learning models, and combines them with modified layout features to improve pattern fidelity and reduce optical proximity effects, leveraging training samples from inverse lithography technology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional OPC or model-based OPC is used to correct optical proximity effects, then pattern fidelity is improved, but computational complexity and processing time increase significantly for full-chip applications
Solution Approach 1:
The patent segments the full-chip layout into multiple tiles or regions, applying optical proximity correction independently to each tile. This divides the computationally intensive full-chip problem into smaller, manageable sub-problems that can be processed in parallel, reducing overall computational complexity while maintaining pattern fidelity through localized correction
Solution Approach 2:
The patent uses machine learning models trained on representative layout patterns to generate correction rules that are then copied and applied to similar patterns across the entire chip. Instead of performing complex simulations for every feature, the system learns from a subset of examples and replicates the corrections, significantly reducing computational burden while preserving accuracy
2Manufacturing precision
If inverse lithography technology is used to generate optimized mask patterns, then pattern fidelity and process window are improved, but runtime computation cost becomes prohibitive for full-chip applications
Solution Approach 1:
The patent performs preliminary inverse lithography technology computations on representative training samples to establish a machine learning model before full-chip processing. This preliminary action captures the essential correction patterns, allowing the subsequent full-chip application to use the pre-trained model for rapid prediction without repeating the computationally intensive ILT optimization for every feature
Solution Approach 2:
The patent replaces expensive, time-consuming full ILT computations with a lightweight machine learning inference process for full-chip applications. The heavy computational work is confined to the one-time model training phase, while the actual production uses the inexpensive, fast ML model that can be applied repeatedly without significant additional computational cost
3Manufacturing precision
If sub-resolution assist features are added to address iso-dense bias problem, then manufacturing precision is improved, but layout complexity and processing difficulty increase
Solution Approach 1:
The patent implements a self-service system where the machine learning model automatically identifies regions requiring sub-resolution assist features and determines their optimal placement and dimensions. The system autonomously analyzes layout patterns, predicts iso-dense bias issues, and generates appropriate SRAF corrections without requiring manual intervention or complex rule-based processing, thereby managing layout complexity through automation
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the accuracy and quality of mask synthesis designs by automatically generating curvilinear main features and SRAFs, improving pattern fidelity and process window, while reducing computational costs associated with full-chip ILT applications.
Implementation Method 1
When light illuminates the photomask, the transmitted light diffracts. The higher spatial frequencies the regions of the mask have, the higher angles the light diffracts at.
Implementation Method 2
This can lead to optical proximity effects such as a pull-back of line-ends from their desired position, corner rounding and a bias between isolated and dense structures
Data Source
AI summary
Layout features in a layout design are classified into groups of layout features. A machine learning-based SRAF generation process is then performed to generate sub-resolution assist features for layout features in each of the groups of layout features. Each of the groups of layout features has a specific machine learning model. The machine learning-based SRAF generation process comprising: dividing regions where sub-resolution assist features are likely to be placed into areas of interest, extracting a feature vector for each of the areas of interest based on a layout area centered at the each of the areas of interest, determining whether the each of the areas of interest should be part of a sub-resolution assist feature by using the feature vector as an input of the specific machine learning model, and generating the sub-resolution assist features based on results of the determining.