Defect Image Classification And Detection With Deep Learning

How deep learning is used to improve automatic inspections in semiconductor manufacturing.


Authors: Dan Sebban and Nissim Matatov

Inspection means have increasingly been incorporated into typical manufacturing of boards, substrates and/or systems. A significant number of automatic inspections rely on the analysis of images that are acquired by a multitude of means such as optical, X-ray, infrared, acoustic microscopy. In contrast to automatic inspections, traditional visual inspection is performed manually by humans based on images and can be laborious and inaccurate. Detection of “indeterministic” defect types such as cracks and/or scratches is quite challenging since such defects may have a variety of shapes, locations and severity. Deep learning, a subfield of machine lLearning, has recently advanced the state-of-the-art learning from images and become the standard approach for computer vision tasks. We will present a case study for automating visual inspection of boards by as much as 40%. The application can be extended to identify specific types of defects and to support root cause analysis.

Click here to continue reading.

Leave a Reply

(Note: This name will be displayed publicly)