Deep Learning Device-Pattern Alignment for Precision Digital Overlay
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lithography systems face challenges in accurately aligning subsequent patterned layers due to the decreasing size of alignment marks, leading to inaccuracies in image projection systems and the use of standard image processing algorithms.
Innovation Solution
A deep learning model is employed to recognize unique device patterns, replacing alignment marks, and align subsequent patterned layers by capturing images with cameras, identifying real objects, and fitting shift and rotation models to locate template boxes within the field of view.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If alignment marks are used to align subsequent layers, then alignment function is provided, but measurement precision deteriorates as marks decrease in size
Solution Approach 1:
The patent replaces the mechanical/optical alignment mark detection system with a deep learning-based image recognition system. Instead of relying on physical alignment marks that become increasingly difficult to detect, the system uses neural networks to identify and locate device patterns directly from captured images, substituting the traditional mark-based mechanical alignment approach with an intelligent software-based solution that maintains high precision even as feature sizes shrink.
2Manufacturing precision
If standard image processing algorithms are applied, then processing simplicity is maintained, but manufacturing precision deteriorates in aligning subsequent layers
Solution Approach 1:
The patent substitutes traditional standard image processing algorithms with a deep learning-based recognition system. The neural network model automatically learns and extracts features from captured images, performing complex pattern recognition and alignment calculations that go beyond the capabilities of standard algorithms, thereby achieving superior manufacturing precision despite the increased computational complexity.
3Reliability
If alignment marks are used, then alignment reference is provided, but reliability deteriorates due to inaccuracy in locating marks
Solution Approach 1:
The patent replaces the unreliable alignment mark location system with a deep learning-based pattern recognition system. The neural network directly identifies device patterns and calculates their positions with high precision, eliminating the reliability issues associated with shrinking alignment marks. The system uses the device patterns themselves as alignment references, removing the need for separate alignment marks that become increasingly difficult to locate accurately.
Data Source
AI summary
Embodiments described herein relate to a system, methods, and non-transitory computer-readable mediums that accurately align subsequent patterned layers in a photoresist utilizing a deep learning model and utilizing device patterns to replace alignment marks in lithography processes. The deep learning model is trained to recognize unique device patterns called alignment patterns in the FOV of the camera. Cameras in the lithography system capture images of the alignment patterns. The deep learning model finds the alignment patterns in the field of view of the cameras. An ideal image generated from a design file is matched with the camera with respect to the center of the field of view of the camera. A shift model and a rotation model are output from the deep learning model to create an alignment model. The alignment model is applied to the currently printing layer.