Multi-Wavelength Defect Classification Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for inspecting and classifying defects on semiconductor substrates are inaccurate and lack objectivity due to reliance on single-wavelength light, leading to poor detection and classification of defects, especially with increasing numbers of defects on substrates, which affects semiconductor manufacturing productivity.
Innovation Solution
A method using multi-wavelength light to irradiate and split reflected light, forming a characteristic matrix from image information that includes contrasts, polarities, sizes, and statistical values to accurately classify defects by comparing with reference matrices using algorithms like thumb print, nearest neighbor, or decision tree algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single-wavelength light is used for defect inspection, then the inspection process is simple, but detection accuracy and classification reliability deteriorate
Solution Approach 1:
The patent segments the inspection process by separating different wavelength inspections into distinct measurement systems. Each wavelength (e.g., blue, green, red) provides specific characteristic information about defect types, allowing the system to handle complex defect classification by dividing the inspection task into multiple specialized measurements rather than relying on a single complex multi-wavelength system.
Solution Approach 2:
The patent introduces wavelength as an additional dimension to the inspection data. Instead of inspecting defects at a single wavelength, the system collects image information across multiple wavelength dimensions (blue, green, red channels), transforming the inspection from two-dimensional (spatial) to three-dimensional (spatial+wavelength) data collection, which enables more accurate defect characterization and classification.
2Extent of automation
If manual review process is used for defect classification, then classification can be performed, but productivity and objectivity deteriorate
Solution Approach 1:
The patent implements self-service through automatic defect classification using algorithms that process the multi-wavelength image data independently. The system automatically identifies defect types (particles, scratches, bridges, collapses) by analyzing the characteristic matrix derived from multi-wavelength images, eliminating the need for manual inspector intervention and enabling high-throughput automated classification that maintains both accuracy and productivity.
Solution Approach 2:
The patent replaces the mechanical human review process with an automated computational system. Instead of inspectors manually examining and classifying defects, the system uses image processing algorithms that automatically analyze the characteristic matrices generated from multi-wavelength images, substituting human cognitive processing with automated computational analysis that is both faster and more consistent.
3Productivity
If reference number of defects is used to determine review process, then inspection can be streamlined, but reliability deteriorates when critical defects are missed
Solution Approach 1:
The patent changes the parameter used for defect evaluation from a simple count (reference number) to a multi-dimensional characteristic matrix that captures wavelength-specific information. By analyzing the pattern of reflectance across multiple wavelengths rather than just the total number of defects, the system can reliably identify critical defect types even when the overall defect count is low, maintaining both efficiency and reliability.
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
This approach enables rapid and accurate detection and classification of defects, improving the reliability and objectivity of defect inspection, reducing the need for manual review and increasing semiconductor manufacturing productivity by effectively handling varying defect shapes and properties.
Implementation Method 1
irradiating a multi-wavelength light beam onto the object, splitting the light beam reflected from the object by wavelengths
Implementation Method 2
splitting the light beam reflected from the object by wavelengths
Data Source
AI summary
A method of classifying defects of an object includes irradiating multi-wavelength light onto the object, splitting light reflected from the object into light beams, each of the light beams having different wavelengths, obtaining image information of the object based on each of the light beams, forming a characteristic matrix that represent the wavelengths and the image information, and analyzing the characteristic matrix to determine types of the defects on the object. Thus, the defects may be accurately classified using a difference between reactivity of each of the defects in accordance with variations of the wavelengths and inspection conditions.