A data-driven method for enhancing the image-based automatic inspection of IC wire bonding defects

Visually inspecting integrated circuit (IC) wire bonding defects is important to ensuring the product quality after the packaging process. The availability of IC X-ray images offers an unprecedented opportunity of studying the image-based automatic IC wire inspection. In this paper, a data-driven me...

Full description

Saved in:
Bibliographic Details
Published inInternational journal of production research Vol. 59; no. 16; pp. 4779 - 4793
Main Authors Chen, Junlong, Zhang, Zijun, Wu, Feng
Format Journal Article
LanguageEnglish
Published London Taylor & Francis 18.08.2021
Taylor & Francis LLC
Subjects
Algorithms
Artificial neural networks
automatic inspection
Bonding
Classification
Data mining
Defects
Feature extraction
Image enhancement
Image segmentation
Inspection
Integrated circuits
intelligent manufacturing
Semiconductors
Wire
wire bonding quality
X-ray images
Online AccessGet full text

Cover

More Information
Summary:Visually inspecting integrated circuit (IC) wire bonding defects is important to ensuring the product quality after the packaging process. The availability of IC X-ray images offers an unprecedented opportunity of studying the image-based automatic IC wire inspection. In this paper, a data-driven method consists of data pre-processing, feature engineering, and classification is developed to address such problem. The data pre-processing is composed of a chip identification algorithm for locating and separating IC chip image patches from the raw images as well as a wire segmentation algorithm for obtaining the wire region. Next, geometric features extracted from the segmented wires are fed into classification models for identifying defects. Five data mining methods are utilised to develop classification models. The vision detection system (VDS) and convolutional neural networks (CNN) are considered as benchmarks. In computational studies, the effectiveness of the developed method is validated by using X-ray images collected from a semiconductor back-end factory in Mainland China. A comparative analysis is conducted to determine the most suitable classifier for the developed method in the chip classification and the SVM model is finally selected. Advantages of the developed method are verified by benchmarking against the VDS and CNN.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0020-7543
1366-588X
DOI:10.1080/00207543.2020.1821928