Machine Learning Model for Lithography Mask Pattern Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lithographic processes face challenges in accurately reproducing patterns with dimensions smaller than the classical resolution limit, particularly due to proximity effects and the need for sophisticated fine-tuning steps like optical proximity correction (OPC) to achieve desired electrical functionality and performance in semiconductor manufacturing.
Innovation Solution
The development of machine learning models, such as convolutional neural networks (CNNs), to predict mask patterns and optical proximity corrections based on cost functions that determine differences between predicted and target patterns, enabling more accurate and efficient pattern transfer in lithographic processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional lithographic processes are used to print patterns smaller than the classical resolution limit, then manufacturing precision deteriorates due to proximity effects, but switching to sophisticated fine-tuning steps like optical proximity correction increases device complexity and computational requirements
Solution Approach 1:
The patent changes the fundamental parameter of pattern representation from discrete geometric shapes to continuous level set functions defined by implicit equations. This allows smooth deformation and optimization of patterns while maintaining manufacturing feasibility, resolving the contradiction between pattern accuracy and process complexity by enabling continuous parameter optimization rather than discrete rule-based corrections
Solution Approach 2:
The patent performs preliminary optimization of patterns at the mask design stage using level set methods and gradient-based optimization. By pre-correcting proximity effects and optimizing pattern geometry before fabrication, the method eliminates the need for complex post-processing and fine-tuning steps, thereby reducing overall device complexity while maintaining high manufacturing precision
2Manufacturing precision
If sophisticated fine-tuning steps like optical proximity correction are applied to achieve desired electrical functionality, then manufacturing precision improves, but productivity decreases due to increased processing time and computational complexity
Solution Approach 1:
The patent replaces traditional iterative optical simulation and manual OPC procedures with an automated level set optimization framework that uses gradient descent and implicit function theory. This substitution of the optimization mechanism dramatically reduces computational time while maintaining pattern accuracy, thereby improving manufacturing throughput without sacrificing precision
Solution Approach 2:
By performing all necessary optimizations at the mask design stage using efficient level set methods, the patent eliminates the need for time-consuming iterative adjustments during fabrication. The patterns are pre-optimized to account for proximity effects and manufacturing variations, enabling direct fabrication without additional fine-tuning steps and thus improving productivity
3Manufacturing precision
If machine learning models are trained to predict mask patterns, then device complexity increases due to model training requirements, but manufacturing precision improves for printing features smaller than 20 nm
Solution Approach 1:
The patent trains machine learning models to copy and generalize from a relatively small set of rigorously optimized reference patterns. Once trained, the models can rapidly predict optimal mask patterns for new designs without requiring full optimization computations. This copying approach transfers the computational burden from production to a one-time training phase, improving manufacturing precision for small features while reducing ongoing computational complexity
Data Source
AI summary
Methods of training machine learning models related to a patterning process, including a method for training a machine learning model configured to predict a mask pattern. The method including obtaining (i) a process model of a patterning process configured to predict a pattern on a substrate, wherein the process model comprises one or more trained machine learning models, and (ii) a target pattern, and training the machine learning model configured to predict a mask pattern based on the process model and a cost function that determines a difference between the predicted pattern and the target pattern.