DRAM Wafer Pixel-to-Design Alignment via Conjunction Region Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target finding algorithms for pixel-to-design alignment (PDA) in DRAM wafers are inadequate due to their reliance on methods designed for logic wafers, which fail to accurately account for the intricate layout of DRAM wafers, leading to inaccurate offsets and compromised reliability.
Innovation Solution
A method and system that apply a bandpass filter to an input frame from an inspection sub-system, calculate one-dimensional projection profiles, determine threshold values, and identify conjunction locations of non-repeating patterns to obtain accurate locations for PDA targets, ensuring reliable alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional target finding algorithms (Harris Corner detector or image contrast metrics) are used for DRAM wafers, then the alignment process can be performed, but the accuracy of PDA alignment deteriorates due to the unique layout of DRAM wafers with repeating patterns
Solution Approach 1:
The patent applies local quality by making the target selection process adaptive to local patterns. The algorithm identifies conjunction regions where non-repeating patterns occur and selectively targets these regions rather than applying a uniform approach across the entire wafer. This local adaptation ensures high-contrast features are captured in the specific context of DRAM wafer layouts with repeating SA and SWD patterns
Solution Approach 2:
The patent changes the parameters of target selection by transitioning from center-based targeting to conjunction region-based targeting. The algorithm modifies the selection criteria to identify regions where multiple non-repeating patterns converge, changing the fundamental parameters of what constitutes a valid target and thereby improving alignment accuracy
2Ease of operation
If targets are selected based on center of SA, SWD or conjunction regions using old algorithm, then the alignment process is simplified, but the accuracy deteriorates because selected targets lack high-contrast features and typically exhibit only one edge
Solution Approach 1:
The patent applies preliminary action by pre-identifying and filtering for conjunction regions before final target selection. The algorithm performs preliminary scanning to locate where non-repeating patterns converge, preparing the groundwork for accurate target identification. This preliminary filtering ensures that only regions with appropriate high-contrast features are considered for alignment
Solution Approach 2:
The patent substitutes the mechanical/algorithmic approach of simple center-based selection with a more sophisticated image analysis mechanism. Instead of relying on geometric center calculations, the system uses projection profiles and pattern recognition to identify conjunction regions, replacing a simple mechanical selection process with an intelligent detection system
Data Source
AI summary
A system and method for pixel-to-design alignment for inspection of a dynamic random-access memory wafer are disclosed. The system includes a controller configured to communicate with an inspection sub-system. One or more processors of the controller are configured to receive an input frame from the inspection sub-system, apply a bandpass filter to the input frame to generate a band-passed image, determine a one-dimensional projection profile along a first direction and a second direction of the band-passed image, determine a local range for a portion of the projection profiles, determine a threshold value for the local range which creates two or more distinct regions, identify one or more conjunction locations of two or more non-repeating patterns based on the two or more distinct regions, and obtain a location for one or more pixel-to-design alignment targets based on the one or more conjunction locations.