Intelligent diagnosis of flip chip solder bumps using high-frequency ultrasound and a naive Bayes classifier.

Machine learning is widely used in industrial diagnosis and can be used to obtain accurate results automatically and online. The flip chip is a useful technology for microelectronic packaging. As flip chip technology trends towards high density and fine pitch, with the introduction of materials of low dielectric constant and lead-free materials, intelligent non-destructive inspection of solder bumps for defects becomes more difficult. Machine learning is a method for realising online inspection and overcoming the effects resulting from human factors such as eye fatigue. In this paper, the authors propose an intelligent diagnosis method for the inspection of flip chip solder bumps for defects. A scanning acoustic microscope with a high-frequency transducer obtains ultrasonic images of the flip chips. A vertical projection method is proposed to segment the flip chip ultrasonic images into solder bump images. Eight features are extracted from each solder bump image, including five geometric features and three frequency domain features. A machine learning method using a naive Bayes classifier is introduced for classification and recognition. The results demonstrate a high rate of accurate classification. Thus, this approach has strong potential for the intelligent diagnosis of flip chip solder bumps online. [ABSTRACT FROM AUTHOR]