Photolithographic Pattern Defect Detection with Pixel Density Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current photolithography techniques struggle to detect nanoscale defects in semiconductor patterns due to the small and subtle nature of defects, lack of training data, and challenges in labeling defects, leading to unreliable detection.
Innovation Solution
A method involving generating pixel density maps with peaks and troughs in SEM images, applying threshold parameters to detect defects based on height, width, depth, and width comparisons, and optimizing these parameters using machine learning for improved defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-based defect detection techniques are used to detect defects from SEM images, then defect detection capability is improved, but the requirement for significant training data increases which makes the process labor-intensive and difficult
Solution Approach 1:
The patent creates synthetic defect patterns by copying and transforming known defect types. It generates artificial defect images by applying transformations (rotation, scaling, translation) to original defect patterns, creating a large training dataset without requiring manual collection and labeling of real defect images.
Solution Approach 2:
The patent performs preliminary defect pattern generation and training data creation before actual defect detection. By pre-generating synthetic defect patterns and creating training datasets in advance, it eliminates the need for labor-intensive manual data collection and labeling during operational phases.
2Measurement precision
If AI-based defect detection techniques are applied to SEM images, then defect detection is attempted, but small and subtle defects are not detected reliably due to noise and lack of training data
Solution Approach 1:
The patent transforms defect detection from direct image analysis to pixel density map analysis. By changing the parameter space from raw pixel values to density distributions, it enhances the visibility of subtle defects and makes them detectable through statistical comparisons of peak and trough characteristics.
Solution Approach 2:
The patent introduces pixel density maps as an intermediary representation between SEM images and defect detection. These density maps serve as a mediator that converts complex image data into simplified statistical features (peak positions, heights, widths), making subtle defects more detectable while reducing noise impact.
3Measurement precision
If manual labeling of defects is performed to create training data, then training data quality is improved, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The patent generates synthetic training data by copying and transforming existing defect patterns. Instead of manually labeling each defect instance, it creates numerous variations through geometric transformations, automatically generating large volumes of labeled training data without human intervention.
Solution Approach 2:
The system performs self-labeling by automatically generating defect patterns with known characteristics. The synthetic defect generation process inherently provides ground truth labels, eliminating the need for manual annotation while maintaining data quality.
4Manufacturing precision
If feature sizes are shrunk to create more intricate patterns, then manufacturing capability is improved, but defect detection becomes more difficult due to nanoscale dimensions
Solution Approach 1:
The patent transitions from two-dimensional image space analysis to a transformed density map space. By converting spatial defect information into density distribution patterns with distinct peaks and troughs, it creates an additional analytical dimension where nanoscale defects manifest as detectable statistical variations rather than sub-visual features.
Solution Approach 2:
The patent changes the detection parameter from direct spatial coordinates to statistical properties of pixel density distributions. This parameter transformation amplifies nanoscale defect signals by representing them as variations in peak height, width, and position, making them detectable despite their small physical dimensions.
Data Source
AI summary
Systems and methods with photolithographic pattern defect detection are provided. A method includes generating a first pixel density map including a first set of peaks and a first set of troughs respectively along a first direction in an image corresponding to a periodic pattern, generating a second pixel density map, for a localized region determined in the first pixel density map dependent on a determined position of a peak and a determined position of a trough, where the second pixel density map includes a second set of peaks and/or a second set of troughs along a second direction of the image that is perpendicular to the first direction, detecting one or more defects in the image by comparing a height/depth and width of a peak or trough of the second set of peaks or troughs with a height/depth threshold parameter and a width threshold parameter, respectively.