Sub-pixel Alignment Using Composite Image Cross-correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing alignment systems face difficulties in accurately positioning material layers with overlapping or differently patterned alignment markers, leading to multiple peaks in cross-correlation values and poor quantization, especially when non-zero offsets are desired, and they struggle with process variations.
Innovation Solution
A system and method that determine an accurate offset vector between material layers using overlapping alignment markers of the same or different patterns by generating composite images, cross-correlating them, and fitting a surface to the correlation values to identify the extremum of maximum correlation, allowing for sub-pixel resolution alignment without separately imaging each layer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated microscopes and complex patterns are used to identify alignment markers, then alignment precision is improved, but manufacturing cost increases prohibitively
Solution Approach 1:
The patent uses simple, inexpensive alignment markers (such as opaque regions or patterned features) that can be easily fabricated alongside the material layers during normal manufacturing processes, replacing the need for expensive specialized microscopes and complex alignment patterns. These markers are integrated into the layers themselves rather than being separate components.
Solution Approach 2:
The patent replaces sophisticated optical microscope systems with a computational image processing approach. Instead of using complex optical hardware to achieve precise alignment, the system uses standard imaging equipment combined with cross-correlation algorithms and surface fitting methods to determine relative positions of material layers with sub-pixel accuracy.
2Area of stationary object
If alignment markers are placed close together to reduce space, then area is reduced, but imaging individual markers becomes difficult or impossible
Solution Approach 1:
The patent merges the imaging of multiple alignment markers into a single combined image capture. By taking one image that contains both alignment markers from different material layers simultaneously, the system eliminates the need to separately image each marker, thereby solving the problem of markers being too close or overlapping to be imaged individually.
Solution Approach 2:
The patent transitions from spatial separation of markers to computational separation through image processing. Even though markers may overlap in the physical domain (same spatial location), the cross-correlation algorithm operates in the correlation domain to uniquely identify and measure each marker's position, effectively adding a computational dimension for differentiation.
3Loss of information
If cross-correlation is performed on overlapping alignment markers, then alignment information is obtained, but multiple peaks appear making location determination impossible
Solution Approach 1:
The patent introduces a surface fitting function (such as a paraboloid or Gaussian surface) as an intermediary between the discrete cross-correlation values and the final alignment measurement. This continuous surface model smooths out the multiple peaks and identifies the true maximum through surface extremum detection, eliminating the ambiguity caused by multiple correlation peaks from overlapping markers.
Solution Approach 2:
The patent changes the parameter space from discrete pixel-based cross-correlation values to continuous surface parameters through fitting. By modeling the correlation values as a continuous surface and finding its extremum, the system achieves sub-pixel resolution and eliminates the discrete peak ambiguity that plagues direct cross-correlation of overlapping markers.
4Adaptability or versatility
If alignment systems compare different patterns across material layers, then versatility is improved, but performance limitations arise due to process variations
Solution Approach 1:
The patent transforms the alignment problem from direct pattern matching to a correlation-based approach where the specific pattern details become less critical. By using cross-correlation to measure relative displacement between layers regardless of their absolute patterns, and then applying surface fitting to extract precise offset values, the system achieves reliable alignment even when comparing different patterns across layers, making it robust against process variations.
Data Source
AI summary
A method for determining an offset vector. The method includes obtaining an image of a first feature. An image of a second feature is also obtained. Also, a combination image of the first feature and the second feature is obtained. A plurality of composite images is utilized to determine an accurate offset vector between the first feature and the second feature in the combination image. The plurality of composite images is based on the image of the first feature and the image of the second feature.