edfas.org 7 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 4 image of Fig. 5 contains a scanning electron micrograph of the physical cross-section of the sample. Summation of the thicknesses of the individual layers leads to a defect depth of 665 µm. Despite the higher structural complexity of the sample, both lateral and axial localization were considerably accurate when conducting the additional recursive inversion-based analysis of the heat propagation. The achievable precision will however depend on how close the closest reconstruction plane is located with respect to the actual source depth and thus also depend on the increments of the recursive reconstruction sequence. By combining SAM, XCT, and LIT techniques, FA engineers can identify failure sites with high confidence before any destructive steps are taken. This integrated nondestructive approach significantly reduces analysis cycle time, protects high-value AI devices, and ensures that subsequent high-resolution destructive FA is both targeted and efficient. DESTRUCTIVE FA: ROOT-CAUSE IDENTIFICATION After defect regions are localized through nondestructive analysis, destructive FA techniques provide the spatial resolution and material contrast required for definitive root-cause identification in advanced AI packaging. DualBeam focused ion beam-scanning electron microscopy (FIB-SEM) has become an indispensable tool in both materials science and semiconductor failure analysis due to its unique ability to combine site-specific material removal with high-resolution imaging.[12-16] This capability enables precise cross-sectioning, 3D structural characterization, and direct correlation between physical defects and functional performance, which is critical for understanding increasingly complex device architectures and advanced packaging technologies. Different FIB ion sources are best suited for varying sample sizes and technology requirements. Ga+ FIB excels in high-precision, small-volume applications, such as TEM sample preparation and detailed analysis of logic and memory devices, typically at the 40 µm3 scale. Plasma FIB (PFIB) extends this capability to larger volumes from 200 to 500 µm3, making it well-suited for CMOS, advanced packaging, and power semiconductor analysis, including tasks such as delayering and diagonal milling while maintaining good resolution. For even larger-scale investigations, femtosecond (fs)-laser-assisted PFIB enables above millimeter-scale material removal with high throughput, providing rapid access to deeply buried features in complex 3D structures. Figure 6 demonstrates the use of the Thermo Scientific Helios 5 Laser PFIB triple-beam system to address the complexities of a 4H AMD Radeon VII HBM2 device. The fs laser enables rapid material removal on the order of 1 mm³ (Figs. 6a and 6b), while Xe+ plasma rocking polishing minimizes curtaining effects (Fig. 6c). This polishing step effectively flattens surface artifacts induced by vertically aligned TSV pillars and microbumps following fs-laser ablation. High-resolution SEM images (Fig. 6d) are then acquired and processed using automated iFast software to generate 2D metrology measurements, such as microbump and TSV dimensions, as well as stacking offset (Figs. 6e and 6f). This system is particularly well suited for failure Fig. 5 Axial defect localization in 4H HBM.[10] Left: Electron micrograph of the sample’s physical cross-section. The red marker indicates the location of the I/O diode used for thermal stimulation. The thermal source is located at a depth of 665 µm inside the sample. Right: Cross-sectional view through the sample’s synthetically computed heat source distribution. The thermal wave converged at a depth of 670 µm, only 5 µm lower than the depth estimated from the electron micrograph. The red dashed line is the actual source depth obtained from the electron micrographs. The estimated source depth is indicated by the solid red line.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==