AI Mask Image Generation for OPC Rule Compliance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical proximity correction (OPC) methods for generating semiconductor masks rely heavily on user expertise, leading to performance and speed limitations due to the choice of mask generation rules.
Innovation Solution
Employing an AI model to generate mask images by determining a loss function based on differences with corrected mask images, performing operations like opening and closing to adhere to minimum size and spacing rules, and updating the model using gradients to improve accuracy and compliance with fabrication constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional OPC methods are used with user-defined mask generation rules, then masks can be generated following fabrication constraints, but the performance and speed are limited by user expertise and manual rule selection
Solution Approach 1:
The patent replaces the manual mechanical process of rule selection and iterative correction with an AI-based automated system. The neural network model learns optimal mask generation rules from training data and automatically applies them, eliminating the need for manual user intervention in rule selection and significantly accelerating the mask generation process while maintaining high precision
Solution Approach 2:
The AI model performs self-learning from training data consisting of target patterns and corresponding corrected mask images. The system automatically optimizes its own parameters through backpropagation and gradient descent, enabling it to independently generate high-quality masks without continuous manual tuning or user expertise
2Reliability
If traditional OPC methods are used with manual rule selection, then masks can be corrected according to fabrication rules, but the process requires significant user expertise and time
Solution Approach 1:
The system enables self-service by allowing the AI model to automatically learn and internalize fabrication rules from training data. The neural network independently optimizes its parameters through backpropagation, eliminating the need for users to manually select or understand complex mask generation rules while ensuring continuous compliance with fabrication constraints
Solution Approach 2:
The patent implements a feedback mechanism where the loss function calculates the difference between generated masks and corrected reference masks. This error signal is fed back through backpropagation to continuously update and refine the model's parameters, ensuring that the system learns and maintains compliance with fabrication rules autonomously
3Manufacturing precision
If iterative corrections are performed manually in traditional OPC, then mask patterns can be optimized, but the process is time-consuming and dependent on user ability
Solution Approach 1:
The patent replaces the slow manual iterative correction process with automated neural network-based optimization. The AI model performs multiple corrections in parallel during inference, achieving the same or better optimization results instantaneously without the time-consuming sequential manual adjustments required by traditional OPC methods
Solution Approach 2:
The system performs preliminary learning during the training phase, where the model pre-learns optimal correction strategies from numerous examples. This preliminary action enables the model to automatically apply learned corrections to new patterns without requiring time-consuming manual intervention during actual mask generation
Data Source
AI summary
A method for generating a mask image for fabricating a mask includes: generating the mask image from a target pattern using a first artificial intelligence (AI) model; calculating a first loss function based on a difference between the mask image and a mask image corrected according to a mask rule; and updating the first AI model based on the first loss function.