Crystalline Wafer Defect Characterization Without Destructive Etching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional defect characterization methods for silicon carbide (SiC) wafers, such as etching, are destructive, expensive, and limited in sample size, hindering efficient defect detection and characterization, which is crucial for improving crystal growth and device yield.
Innovation Solution
Nondestructive defect characterization using deep convolutional neural networks (DCNNs) for SiC wafers, utilizing photoluminescence microscopy and machine learning to detect and identify defects like dislocations and stacking faults without etching, enabling faster and more accurate defect counting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If destructive etching is used for defect characterization, then defect detection accuracy is improved, but wafer cost increases and sampling is limited
Solution Approach 1:
The patent replaces the mechanical/chemical etching process with a non-contact optical measurement system. A scanner captures images of the wafer surface, and machine learning algorithms automatically identify and characterize defects such as dislocations, micropipes, and stacking faults. This substitution eliminates the need for destructive chemical etching while maintaining defect detection capability.
Solution Approach 2:
The patent creates a digital copy or representation of the wafer surface through imaging and processing. Instead of physically altering the wafer through etching, the system generates a digital model that contains all necessary defect information. This digital copy can be analyzed repeatedly without affecting the actual wafer, enabling both accurate measurement and full wafer utilization.
2Measurement precision
If conventional etching methods are used, then defect identification is achieved, but processing time and cost increase
Solution Approach 1:
The patent performs defect characterization earlier in the manufacturing process using non-destructive imaging, before wafers undergo subsequent processing steps. The scanning and machine learning analysis can be conducted on wafers immediately after growth or basic preparation, providing early feedback without requiring time-consuming etching procedures later in the process.
Solution Approach 2:
The patent changes the measurement parameters from chemical etching conditions to optical imaging parameters. Instead of controlling etching chemistry, temperature, and time, the system controls imaging parameters such as scan speed, light source intensity, and image processing algorithms. This parameter change enables faster, non-contact measurement while achieving equivalent or superior defect characterization.
3Productivity
If only a few wafers are sampled for destructive testing, then processing cost is reduced, but information availability for process improvement is limited
Solution Approach 1:
The patent makes the characterization system universal by enabling inspection of all wafers rather than requiring separate destructive testing for sampling. The same non-destructive scanning system that provides quality control also generates comprehensive process improvement data. This multi-functionality eliminates the trade-off between sampling volume and information availability, as all wafers contribute to both quality assurance and process optimization.
Solution Approach 2:
The patent implements a feedback mechanism where defect data from comprehensive wafer inspection is fed back to crystal growth and manufacturing processes. The machine learning system identifies defect patterns and trends across all inspected wafers, providing actionable feedback for process improvement. This continuous feedback loop enables systematic reduction of defects in subsequent production batches, directly improving manufacturing efficiency and yield.
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 cost-effective, high-fidelity defect characterization of SiC wafers, allowing increased sampling and accelerated feedback loops for crystal growth improvement, while preserving wafers for device fabrication.
Implementation Method 1
utilizing photoluminescence microscopy and machine learning to detect and identify defects
Data Source
AI summary
A method of analyzing semiconductor wafers includes capturing a first image of a first crystalline material, etching a first surface of the first crystalline material to delineate etch defects in the first crystalline material, and capturing a second image of first crystalline material after etching the first surface of the first crystalline material. Based on the second image, labels of etch defects delineated in the first surface of the first crystalline material are generated. The first image and the labels of etch defects are spatially coordinated to form a defect map identifying one or more defects in the first image based on the delineated etch defects, and based on the defect map and nondestructive data obtained from a second crystalline material, defects in the second crystalline material are identified.