Developing a Machine Vision Inspection System for Electronics Failure Analysis

Abstract

Ball grid array (BGA) package components are widely used in various electronic products, as they shorten the conductive path and greatly reduce the interference during current transmission. The BGA component solder ball has little flexibility for distortion, which may cause stress concentration and solder joint cracking. Red ink dye stain analysis testing can visually analyze the cracking area and the interface of the solder joint. At present, most companies use failure analysis (FA) specialists to observe the components and circuit board end images with a microscope. As component packaging technology becomes more complex and the number of input–output (I/O) signals of the chip increases sharply, FA operations require a long time and may be misjudged due to personnel fatigue and subjective determination. This paper aims to develop a machine vision image recognition system for intelligent decision analysis to judge the severity of solder ball cracking and the cracking interface type. This system uses charge-coupled devices (CCDs) to obtain the image. First, this paper uses affine transformation compensation to perform horizontal correction, which recognizes the actual position coordinates of each solder joint, estimates the solder ball according to the Hough Circle to obtain the center and the radius of the solder ball, and then a 2-D vector map (Gerber) file is used to establish the mask, and the region of interest (ROI) in the image feature is obtained through the iterator of the pixel access program. Second, the color space is converted from red, green, and blue (RGB) to hue, saturation, and value (HSV) to reduce the effects of light variations during image capture. This paper develops the range analysis model (uses the multiple-threshold method to judge whether a particular pixel in an image has the cracking (color) feature, as well as its cracking interface), and uses a morphology filter to filter image noise (filter pixels of overlapping features). The verification results show that the system developed in this paper and FA specialists have a consistency of more than 85% in the judgment of the severity of the cracking, and the false alarm ratio and the escape ratio of the system are about 9% and 16%, respectively. Regarding the judgment result of the solder ball cracking interface type, the consistency reaches 98%, which shows that the automatic failure identification system for solder joints, as developed in this paper, can effectively assist red ink solder joint FA.