Overlay Measurement Using Design Data and Deep Learning Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring overlay shift between semiconductor device layers are inefficient due to the complexity of under-layer structures and shrinking design rules, which makes it difficult to measure overlay shift effectively using standard edge detection techniques, which are not effective due to the complexity of under-layer structures and shrinking design rules.
Innovation Solution
Utilizes deep learning segmentation to segment images of semiconductor devices and measure overlay shift between semiconductor devices using design data and deep learning segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If standard edge detection techniques are used for overlay measurement, then the measurement process is simple, but the measurement precision deteriorates due to complexity of under-layer structures and shrinking design rules
Solution Approach 1:
The patent segments the overlay measurement process into distinct stages: acquiring separate images of first and second layers, segmenting these images to extract structural features, and then performing alignment calculations. This segmentation allows complex under-layer structures to be processed systematically, improving measurement precision while maintaining manageable process complexity
Solution Approach 2:
The patent introduces an intermediary alignment calculation process that operates on segmented features from both layers. This intermediary step computes overlay shift based on corresponding features from the first and second layers, bridging the gap between simple image acquisition and precise overlay measurement even in the presence of complex structures
2Device complexity
If traditional overlay measurement methods are used, then the device complexity is low, but the measurement precision deteriorates due to inability to handle complex under-layer structures
Solution Approach 1:
The patent performs preliminary actions by acquiring and segmenting images of individual layers before performing the actual overlay alignment calculation. By pre-processing the images to extract and identify structural features separately, the system prepares clean, organized data that can be accurately aligned, improving measurement precision without requiring overly complex real-time processing
Solution Approach 2:
The patent transitions from direct overlay measurement to a multi-dimensional approach by separately imaging and analyzing the first and second layers in distinct image datasets. This dimensional separation allows the system to handle complex under-layer structures more effectively, as each layer can be processed independently before combining results for final overlay calculation
3Ease of manufacture
If standard edge detection is applied to shrinking design rules, then the method remains simple to implement, but the measurement precision deteriorates due to poor top surface signal
Solution Approach 1:
The patent creates segmented copies of the original images, extracting only the relevant structural features from each layer. These segmented feature copies are then used for alignment calculations, eliminating noise and irrelevant information from the original images. This copying approach maintains implementation simplicity while dramatically improving measurement precision for shrinking design rules
Solution Approach 2:
The patent applies local quality enhancement by segmenting images to extract only the specific structural features relevant to overlay measurement. Rather than processing entire images uniformly, the system identifies and processes local features with higher precision, adapting to varying signal qualities across different regions and improving overall measurement accuracy for advanced design rules
Data Source
AI summary
A method for overlay measuring using design data and deep learning segmentation is disclosed. The method may render selected design layers as rendered design images corresponding to each site of the design layers. A first design layer is rendered as a first rendered design image including a first site and a second design layer is rendered as a second rendered design image including a second site. The method may acquire measured images of a sample including multiple layers. The method may apply a deep learning model to the measured images to segment the measured images into a first segmented layer and a second segmented layer. The method may align a selected rendered design image with a corresponding segmented layer. The method may determine overlay shift between the first layer and the second layer based on alignment of the selected rendered design image and the corresponding segmented layer.