Semiconductor Defect Detection in Non-Repeating Logic Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current detection methods for programmed die repeating defects in semiconductor manufacturing are inadequate for backgrounds with non-repeating logic features, as they rely on Fourier filtering, which fails to suppress noise from such patterns, limiting the detection of defects in noisy backgrounds.
Innovation Solution
A system and method using single die inspection (SDI) with multiple modes and computer subsystems to select the optimal mode for detecting programmed defects, suppressing noise by determining the mode with the lowest non-defect signal and applying a trained defect detection method, allowing detection in regions with non-repeating patterned features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Fourier filtering is used to detect programmed defects, then defects in repeating patterns can be detected, but defects in non-repeating logic regions cannot be detected due to noise from non-repeating features
Solution Approach 1:
The inspection area is segmented into two distinct regions: a training care area containing only non-defect signals from non-repeating logic features, and a detection care area containing both defect and non-defect signals. This segmentation allows the system to separately characterize background noise and defect signals, resolving the contradiction between detecting defects in repeating patterns versus handling non-repeating regions.
Solution Approach 2:
The system performs preliminary action by training the defect detection method in advance using data from the training care area, which contains only non-defect signals. This preliminary training establishes a baseline understanding of background noise characteristics before actual defect detection is performed in the detection care area, enabling accurate defect identification even in non-repeating logic regions.
2Reliability
If array mode with Fourier filtering is used, then die repeater defects can be detected, but the method requires a substantially quiet background which is not available in non-repeating logic regions
Solution Approach 1:
The system extracts and isolates the non-defect signal component by using a training care area that contains only background noise from non-repeating logic features. By separating the training data from the detection data, the system can extract characteristics of the harmful noise and use this knowledge to suppress it during actual defect detection, thereby reducing the impact of noise from non-repeating patterned features.
Solution Approach 2:
The system implements feedback by using the trained defect detection method, which was trained on noise characteristics from the training care area, to inform and improve the detection process in the detection care area. The training phase provides feedback about noise patterns that are then applied to suppress similar noise during defect detection, enhancing reliability in noisy non-repeating regions.
3Measurement precision
If a training care area is selected that is mutually exclusive of defect detection areas, then the training data is cleaner with less non-defect signal, but the training area cannot overlap with areas where defects are detected
Solution Approach 1:
The system applies local quality by creating different care areas with different purposes and characteristics. The training care area is designed with high quality (clean, non-defect signals only) for training purposes, while the detection care area is designed to cover regions where defects may occur. Each area has optimized properties suited to its specific function, resolving the conflict between training data quality and available area.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate detection of programmed defects in regions with non-repeating logic patterns by reducing noise and improving signal-to-noise ratios, overcoming the limitations of existing methods that require quiet backgrounds or Fourier filtering.
Implementation Method 1
an inspection subsystem configured for generating output for a wafer by scanning the wafer with light and detecting light scattered from the wafer during the scanning
Data Source
AI summary
Systems and methods for detecting programmed defects on a water during inspection of the wafer are provided. One method includes selecting a mode of an inspection subsystem for detecting programmed defects on a wafer that generates output for the wafer having the lowest non-defect signal and at least a minimum signal for the programmed defects. The method also includes selecting a training care area that is mutually exclusive of care area(s) used for detecting the programmed defects during inspection of the wafer. The training care area generates less of the non-defect signal than the care area(s). The method further includes training a programmed defect detection method using the output generated with the selected mode in the training care area and detecting the programmed defects during the inspection of the wafer by applying the trained programmed defect detection method to the output generated in the care area(s) with the selected mode.