Deep Learning Corner Rounding for OPC Pattern Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical proximity correction (OPC) methods face challenges in ensuring patterning reliability due to the approximation of correction values for various patterns, leading to decreased accuracy and reliability in semiconductor manufacturing.
Innovation Solution
A deep learning-based corner rounding method is introduced, which generates a transform model to output a rounded layout target from a square layout, improving the accuracy and reliability of OPC by using a generative adversarial network (GAN) to correct patterns on semiconductor wafers, and includes an optical rule check to determine defects in the OPCed layout.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing OPC methods use approximation of correction values for various patterns, then the manufacturing process is simplified and faster, but the patterning accuracy and reliability decrease
Solution Approach 1:
The patent replaces traditional rule-based OPC methods with a deep learning-based transform model. The model is trained using generative adversarial networks (GANs) to learn the mapping between square layouts and actual developed patterns, substituting mechanical approximation rules with an intelligent system that can capture complex pattern formation physics and predict accurate correction values for various patterns.
Solution Approach 2:
The patent changes the fundamental parameter representation from simple geometric rules to learned feature representations. The transform model processes input layouts through multiple convolutional layers and transformations, learning optimal parameter adjustments for corner rounding and pattern correction that adapt to different pattern types and process conditions, thereby improving accuracy without manual rule complexity.
2Reliability
If deep learning-based corner rounding is implemented, then patterning reliability is enhanced, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-training the transform model using GANs on extensive simulation data before actual OPC processing. The model learns from synthetic pattern formations and correction examples in advance, so that during actual mask manufacturing, the pre-trained model can quickly predict correction values without requiring complex real-time calculations, thus reducing processing time while maintaining high reliability.
Solution Approach 2:
The patent uses copying by generating synthetic training data through GANs that replicate real pattern formation processes. The transform model is trained on these copied simulations of actual lithography outcomes, allowing it to learn from virtual examples without requiring extensive physical experimentation or lengthy real-world processing, thereby reducing time costs while improving reliability through diverse training scenarios.
Data Source
AI summary
The inventive concept provides a corner rounding method of a deep learning-based optical proximity correction (OPC) pattern by which patterning reliability may be ensured, and an OPC method and a mask manufacturing including the corner rounding method. The corner rounding method of a deep learning-based OPC pattern includes: obtaining a contour of a photoresist (PR) pattern or an etching pattern on a wafer; obtaining a square layout of the PR pattern or the etching pattern corresponding to the contour; generating a transform model through deep learning with the square layout and the contour; and obtaining a rounded layout target with respect to a square layout target by using the transform model.