Semiconductor Defect Detection Noise Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection methods in semiconductor manufacturing face challenges due to noise interference, which reduces sensitivity and integrity, especially as design rules shrink and more potential defects are detected, making it difficult to distinguish real defects from noise.
Innovation Solution
A computerized system that captures images of semiconductor specimens, generates difference images based on predefined descriptors for defects of interest and noise, and determines defect candidates using a processing unit with predefined thresholds, effectively filtering out noise and enhancing detection sensitivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reference die image comparison method is used for defect detection, then defect detection capability is provided, but noise from process variations and mechanical variations interferes with detection and decreases sensitivity
Solution Approach 1:
The patent segments the defect detection process into multiple independent analysis components: pattern matching analysis, noise analysis, and defect detection analysis. Each component processes specific aspects of the inspection data separately, allowing noise to be identified and filtered out while maintaining sensitivity to real defects. The segmentation enables parallel processing of different signal characteristics.
Solution Approach 2:
The patent introduces an intermediary noise model that acts as a mediator between the raw inspection signals and the final defect detection. This noise model is trained separately on noise-only data and then used to subtract or suppress noise components from the defect detection process, improving sensitivity without being affected by the same noise that plagues traditional reference comparison methods.
2Quantity of substance
If design rules are shrunk to increase density, then more defects are detected, but real defects become buried within increased noise
Solution Approach 1:
The patent implements dynamic noise modeling where the noise characteristics are not fixed but adapt to the specific inspection conditions and design rules being used. The system dynamically adjusts the noise model parameters based on the current process variations and design complexity, allowing it to maintain effective signal-to-noise separation even as design rules shrink and defect quantities increase.
Solution Approach 2:
The patent changes the parameters of the detection system by using multiple independent analysis components with different sensitivity characteristics. Instead of relying on a single fixed threshold or method, the system adjusts and combines results from pattern matching, noise analysis, and defect detection analyses, each with optimized parameters for their specific function, thereby maintaining precision across varying defect quantities.
3Productivity
If traditional defect detection algorithms are used, then inspection is performed, but noise cannot be effectively filtered and detection integrity is hindered
Solution Approach 1:
The patent performs preliminary noise modeling and characterization before the actual defect detection process. By pre-training noise models on noise-only data and establishing baseline noise characteristics in advance, the system prepares the filtering mechanisms ahead of time, allowing the main inspection process to run efficiently without being bottlenecked by complex real-time noise analysis, thus maintaining both productivity and integrity.
Solution Approach 2:
The patent implements feedback mechanisms where the results from noise analysis and pattern matching are fed back into the defect detection process. The system continuously refines its understanding of what constitutes noise versus real defects based on accumulated inspection data, improving detection integrity over time while maintaining high throughput through automated feedback loops that reduce manual review requirements.
Data Source
AI summary
There are provided system and method of detecting defects on a specimen, the method comprising: capturing a first image from a first die and obtaining one or more second images; receiving: i) a first set of predefined first descriptors each representing a type of DOI, and ii) a second set of predefined second descriptors each representing a type of noise; generating at least one difference image based on difference between pixel values of the first image and pixel values derived from the second images; generating at least one third image, comprising: computing a value for each given pixel of at least part of the at least one difference image based on the first and second sets of predefined descriptors, and surrounding pixels centered around the given pixel; and determining presence of defect candidates based on the at least one third image and a predefined threshold.