Fig. 2: Non-destructive detection of defects in TSVs with high resolution utilizing acoustic interferometry.

a SAM C-scan image showing a TSV array with a subgroup of array marked as G1. The letters a, b, c and d denote the position of TSVs within the subgroup G1. b Automated classification and localization of the TSVs within the exemplary ROI utilizing the End-to-End convolutional neural network (E2E-CNN). TSVs with homogeneous fringes are predicted as Class 1 (green) TSVs. Class 2 (blue) indicates a single inhomogeneity. Class 3 (brown) shows a blossom-like shaped interference pattern. The predictions of E2E-CNN are color coded in green, blue and brown for Class 1, 2 and 3, respectively. c Magnified SAM C-scan images of the TSVs associated with Class 1, Class 2 and Class 3, highlighted with green, blue and brown frames, respectively. d Intensity plot (low intensity corresponds to black, yellow to high intensity) in the x-y-plane for Class 3 at the defocused position Z₇. The distorted fringes along the circumference can be seen leading to a blossom-shaped pattern. White dashed line indicates the position of the TSV. e SEM images illustrating ‘TSV(C1a)’ and indicating a crack, with an opening of about 200 nm, within the metallized wall of the TSV. The crack is better resolved in the magnified image (yellow box). Two red arrows are used to highlight the crack position. f Cross sectional SEM image (after ion slicing) of the crack in ‘TSV(C1a)’ with a cone crack. The crack is indicated within the red dotted circle and pointed using a red arrow marker.