edfas.org 15 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 26 NO. 4 larger die as well as chip stacking, because localized reflow allows better control of chip gap height and tilt throughout the bonding process. High accuracy and repeatability in all dimensions are required to provide a reliable systemin-package.[5] To assess the qualitative and quantitative ability of 3D XRM to determine both the alignment accuracy of bump/ via stacks and the quality of TCB interconnects, a 15-die stack made by Interuniversity Microelectronics Centre (imec) was selected as the test vehicle. It consisted of a patterned test chip type-O (PTCO) die sequentially stacked vertically and bonded onto a patterned test chip type-P (PTCP) substrate chip with 20 µm solder bump pitch (Fig. 2). Each PTCO die contains a 5 μm high Cu bump 12.5 μm in diameter on the top side, connected to an 8.5 μm Cu/Sn bump 7.5 μm in diameter, connected by a 5 μm diameter Cu TSV. Alignment and bond validation involve analysis of 3828 peripheral bump/via stacks.[6] The highest-resolution scans performed in this study used voxels of 0.7 µm to scan narrow-pitch bumps located in each corner of the array. These are key for die-to-die alignment in the TCB tool and are identical in pattern, aside from any expected rotation due to their location. To speed up the workflow, an AI model was trained on a single corner of narrow-pitch bumps by doing a two-hour scan, followed by a 15-minute scan of another corner that applied a deep-learning algorithm. After all volumetric data was collected, the software rendered it into 3D virtual slices to achieve a large data set in a noticeably short amount of time while ensuring the analysis of the Cu pillar alignment and solder bump width and height would be statistically relevant. Fig. 2 TSV shift from die 1-15 can be noted by positional shift relative to the white vertical line in this colorized virtual XRM cross-section. The 0.7 μm-voxel-resolution 3D data volume from which the slice was extracted is shown in the colorized cross-sectional 3D rendering in the right-hand image. Fig. 3 Statistical data comparison for mean and standard deviation of TSV shift between all iterations of scan methodology. Bump diameter and height were also assessed (not shown).
RkJQdWJsaXNoZXIy MTYyMzk3NQ==