3D NAND Defect Binning Using 1D Projection and Trench Centers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor wafer inspection systems lack the resolution to accurately detect defects in 3D NAND structures, particularly in vertical stacks, due to limitations in handling local and global gray level variations, leading to separation errors and noise-affected results.
Innovation Solution
A method and system utilizing a broad-band plasma inspection system to generate a one-dimensional projection of a 3D semiconductor wafer image, creating a mask through auto-correlation and auto-convolution to determine trench centers, and performing location-based binning of defects by calculating the distance to neighboring trench centers, thereby enhancing defect detection sensitivity and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Image-based SuperCell (IBS) method is used to find defects on 3D NAND structures, then defect detection capability is provided, but local and global gray level variations cause separation errors and noise affects results
Solution Approach 1:
The patent segments the 3D NAND structure image into multiple one-dimensional projections along different directions. By dividing the complex two-dimensional image analysis into multiple one-dimensional projections, the method reduces the impact of local and global gray level variations on defect detection accuracy.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. By analyzing one-dimensional projections from multiple directions and combining results, the method adapts to local variations in gray levels while maintaining overall detection reliability.
2Measurement precision
If inspection resolution is increased to detect smaller defects, then defect detection sensitivity improves, but system complexity and manufacturing costs increase
Solution Approach 1:
The patent replaces complex high-resolution imaging systems with a simpler approach using multiple lower-resolution one-dimensional projections. By substituting mechanical/optical complexity with computational processing of multiple projections, the method achieves enhanced defect detection sensitivity without proportionally increasing system complexity.
Solution Approach 2:
The patent transforms the two-dimensional image analysis problem into multiple one-dimensional projection analyses. By adding the dimension of multiple projection directions, the method achieves better defect detection sensitivity while keeping individual projection systems simpler.
3Ease of manufacture
If traditional inspection methods are used, then manufacturing process is simple, but defects in vertical stacks cannot be accurately located
Solution Approach 1:
The patent segments the defect detection process into multiple one-dimensional projections and then combines them to reconstruct two-dimensional defect locations. This segmentation approach maintains relative manufacturing simplicity while recovering the lost spatial information about defect locations in vertical stacks.
Solution Approach 2:
The patent uses one-dimensional projections from multiple directions to recover two-dimensional location information. By analyzing defects along different one-dimensional axes and combining the results, the method retrieves the vertical stack location information that would be lost in single-direction inspection.
Data Source
AI summary
Location-based binning can separate defects on different rows of channel holes in a 3D NAND structure to corresponding bins. A one-dimensional projection of an image is generated and a one-dimensional curve is formed. A mask is generated from the one-dimensional curve. Defects in the image are detected using the mask and location-based binning is performed.