An automatic visual system to identify and estimate ionic contamination in printed circuit boards using electrochemical migration patterns

Detection and estimation of ionic contamination in electronics is important, in order to ensure manufacturing quality by detecting the cause and preventing failure mechanisms such as electrochemical migration (ECM). However, such tests require expensive, specialized equipment. This paper proposes a low-cost, automatic, visual-based method, in which the ionic contamination profile of a printed circuit board is determined through the analysis of shape and color features of optical microscope images of ECM failures. Images of copper dendrites were acquired through the water drop test using solutions contaminated with NaCl and Na₂SO₄ from 10-50 ppm, in steps of 10. Thresholding and connected component analysis were used to segment the shorting dendrite. The method used three types of features: global and local shape features, and color features. Two feature selection methods, ReliefF and Correlation-based Feature Selection (CFS) were also tested to measure feature quality and to determine the best feature subsets. The predictive model used was feature-weighted k-nearest neighbor. The study determined that copper dendrites produced with sodium sulfate contaminant were larger and denser compared to sodium chloride. Increasing the amount of contaminant also increased the density of the pattern. At higher sodium sulfate contamination levels, dendrites tended to have reddish tips, while with sodium chloride, branch shapes transitioned from a well-defined appearance to a “stringy” appearance. The system was able to distinguish between the two contaminants at 97.3%, while using only eight descriptors. The system was also able to distinguish between five closely-spaced contaminant levels at 63.38% and 57.14% accuracy for sodium chloride and sodium sulfate respectively. Local shape features, which were not used in previous work, were found to be generally more useful compared to global shape features.