Power Module Die Defect Detection With Two-Stage AI Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting defects in silicon carbide (SiC) based power devices in power modules are inefficient and inaccurate, particularly when dealing with a large number of power semiconductor dies, leading to anomalous behavior and potential failure.
Innovation Solution
A system and method utilizing a die detection model and a defect detection model, both based on convolutional neural networks, to automatically identify the positions and defects of power semiconductor dies in images, allowing for simultaneous prediction of bounding boxes and class probabilities, with the ability to stop manufacturing processes if defects are detected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis is used to detect defects in power semiconductor dies, then the detection process is simple to implement, but the detection speed and accuracy are insufficient leading to anomalous behavior
Solution Approach 1:
The patent replaces manual visual inspection with an automated optical inspection system that uses machine learning models (YOLOv5, RetinaNet) to detect defects in power semiconductor dies. The system captures images of the dies and uses trained detection models to automatically identify defects, eliminating the need for manual analysis and significantly improving both detection speed and accuracy.
2Reliability
If optical microscopy is used to detect defects in SiC based power devices, then the detection method is non-destructive, but the detection efficiency is low when dealing with a large number of power semiconductor dies
Solution Approach 1:
The patent implements preliminary action by training multiple detection models (YOLOv5, RetinaNet) in advance with labeled defect data before actual inspection. The pre-trained models can rapidly analyze images of power semiconductor dies during production, eliminating the need for real-time model training and enabling fast, efficient detection of defects across large numbers of dies.
Solution Approach 2:
The patent changes the parameter of detection methodology from manual optical inspection to automated machine learning-based optical inspection. By introducing AI models with different architectures and training them on specific defect types, the system maintains the non-destructive nature of optical microscopy while dramatically improving detection efficiency for large batches of power semiconductor dies.
3Device complexity
If a single detection model is used for both die positioning and defect detection, then the system complexity is reduced, but the detection precision for both tasks deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the detection task into two separate specialized models: one for die positioning and another for defect detection. Each model is trained specifically for its designated task, allowing both to achieve high precision. The system first uses the positioning model to locate dies, then uses the defect detection model to identify anomalies within the positioned dies.
Solution Approach 2:
The patent implements a multi-functional inspection system that handles both die positioning and defect detection through coordinated specialized models. Rather than requiring a single universal model, the system uses multiple models that work together to provide comprehensive inspection capabilities, improving overall detection precision while maintaining manageable system complexity through modular architecture.
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
The system enables faster and more accurate defect detection compared to manual analysis, reducing manufacturing costs by identifying defects early and preventing anomalous behavior, especially when using a dedicated circuit like an ASIC or FPGA.
Implementation Method 1
the camera includes an optical microscope configured to sense ultraviolet light rays
Implementation Method 2
the passivation layer being transparent to the ultraviolet light rays
Data Source
AI summary
In an embodiment, a method includes: capturing a first image of a power module, the power module including a power electronics circuit, the power electronics circuit including power semiconductor dies; identifying positions of the power semiconductor dies in the first image with a die detection model; extracting second images of the power semiconductor dies from the first image according to the positions of the power semiconductor dies in the first image; and identifying defects of the power semiconductor dies in the second images with a defect detection model, the defect detection model being different from the die detection model.