“…Wankerl et al [ 24 ] used fully convolutional networks to automatically detect solder voids in x-ray images, and more recently Kong et al [ 23 ] have proposed an algorithm that combines deep learning-based model to increase the void indexing accuracy. Regardless the ease in void detection when using the automated methods, many studies still rely on detected 2D voids or a cross-sectional view of solder joints depending on the test condition and failure mechanism [ 13 , [25] , [26] , [27] , [28] , [29] ]. Jiang et al [ 13 ] reported that the void size and position are critical variables affecting both shear strength and thermal resistance, based on both FE simulation and experiment.…”