Microcrack detection and noise reduction in integrated circuit packages
Loading...
Date
2018-06
Authors
Koh, Ye Sheng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The rise in consumption of electronic products in the recent years have
subsequently led to an increase in manufacturing of integrated circuits (ICs) to meet
consumers’ demands. Thus, it is vital that each IC is inspected for defects that
compromises its quality and usability. This ensures that no defective ICs are used in
conjunction with the manufacturing of electronic products which may severely impact
the end product’s performance and lifespan. One of the common defects is microcrack
on the IC’s package. Image processing is used to detect the presence of microcracks on
the IC and the method currently employed to achieve this is by convolution with multiple
kernels with different configurations. However, this method is time consuming due to
the multiple configurations needed to be tuned and is also susceptible to image noise
which lowers the accuracy of the detected microcracks. Therefore, a better algorithm is
desired to improve the detection performance in terms of time and accuracy. Three
algorithms are tested and evaluated in terms of microcrack detection and noise reduction
which are probability based thresholding, histogram equalization, and modified Perona-Malik’s anisotropic diffusion methods. The first algorithm, probability based
thresholding method consists of two stages, (i) image crack segmentation where the crack
regions are analysed to obtain a suitable thresholding value, and (ii) image denoising
where morphological closing is performed on the image. For the second algorithm,
histogram equalization method has three stages, (i) image contrast enhancement through
histogram equalization, (ii) image crack segmentation which subtracts the histogram
equalized image with the image before histogram equalization process before merging
the images using bitwise operation, and (iii) image denoising using morphological
opening. The third algorithm, modified Perona-Malik’s anisotropic diffusion method
consists of three stages, (i) image crack enhancement which separates the image into its
red, green, and blue channels and enhances the crack features using modified Perona-Malik’s anisotropic diffusion, (ii) image crack segmentation which subtracts the diffused
image with the pre-diffused image before summing the grey values of the images together,
and (iii) image denoising using morphological opening and median filter. Images
processed using modified Perona-Malik’s anisotropic diffusion method produces images
with less noise compared to probability based thresholding method and histogram
equalization method. The method has detected cracks present in three samples out of the
five samples tested. The modified Perona-Malik’s anisotropic diffusion method is thus
proven to produce relatively better performance compared to the other tested methods.