Related Experiment Video
Updated: May 5, 2026
11:23
Lensless Fluorescent Microscopy on a Chip
Published on: August 17, 2011
17.6K
LDDP-Net: A Lightweight Neural Network with Dual Decoding Paths for Defect Segmentation of LED Chips
Jie Zhang1, Ning Chen1, Mengyuan Li1
1Mechnical and Vehicle Engineering, Hunan University, Changsha 411082, China.
Sensors (Basel, Switzerland)
|January 25, 2025
Summary
This study introduces LDDP-Net, a lightweight neural network for LED chip defect detection. It achieves high accuracy in chip segmentation, improving semiconductor manufacturing quality.
Area of Science:
- Semiconductor Manufacturing
- Computer Vision
- Artificial Intelligence
Background:
- Chip defect detection is vital for semiconductor production quality and chip performance.
- Existing methods may struggle with information loss and fine-grained segmentation.
Purpose of the Study:
- To propose LDDP-Net, a lightweight neural network for efficient and accurate LED chip segmentation.
- To enhance defect detection capabilities in semiconductor manufacturing.
Main Methods:
- Developed LDDP-Net with a modified MobileNetv3 backbone to reduce information loss.
- Implemented dual decoding paths (coarse and fine-grained) for parallel processing.
- Integrated intermediate-layer features to improve boundary segmentation accuracy.
Main Results:
- LDDP-Net achieved a mean Intersection over Union (mIoU) of 90.29% on the chip dataset.
- The network is lightweight, with 2.98 million parameters and 2.24 Giga Floating Point Operations (FLOPs).
- Demonstrated significant improvements compared to existing advanced methods.
Conclusions:
- LDDP-Net offers an effective and efficient solution for LED chip segmentation.
- The proposed architecture enhances defect detection accuracy and boundary segmentation in semiconductor manufacturing.
