edfas.org 5 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 4 As a result, advanced packaging defects in AI systems are not only more frequent, but also more complex and multi-modal, often involving coupled electrical, thermal, and mechanical failure mechanisms. Typical buried defect locations include microbumps, RDLs, TSVs, hybrid bonding interfaces, underfill layers, and die-to-interposer interfaces, all of which are inaccessible to conventional surface inspection. FAILURE ANALYSIS REQUIREMENTS FOR ADVANCED AI PACKAGING To effectively address these challenges, modern FA must deliver three core capabilities: (1) Early localization of buried defects without destroying high-value AI devices; (2) Micro- to nanoscale root-cause identification across metals, dielectrics, and heterogeneous interfaces; and (3) Data-driven feedback to design, materials, and process teams to accelerate yield learning and reliability improvement. Meeting these requirements demands a systematic and specialized FA workflow, which can be broadly divided into nondestructive FA and destructive FA. NONDESTRUCTIVE FA: LOCALIZING BURIED DEFECTS Nondestructive FA techniques are indispensable for advanced AI hardware due to high device cost, limited sample availability, and the complexity of 2.5D/3D package architectures. These methods are typically applied first to preserve sample integrity while enabling rapid identification of regions of interest for deeper analysis. Scanning acoustic microscopy (SAM) operates in three primary modes, A-scan, B-scan, and C-scan, each providing different types of information about a sample.[2-5] The A-scan captures the raw ultrasonic waveform, recording echo amplitude as a function of transit time, and is typically used to optimize inspection conditions by max- imizing signal amplitude at a specific point of interest. Building on this, the B-scan generates a cross-sectional view of the material, offering depthresolved information that reveals inter- nal structures along a vertical slice. In contrast, the C-scan produces a planar image at a selected depth, effectively mapping features across a specific layer within the sample. For both B-scan and C-scan modes, proper transducer positioning and signal tuning are essen- tial to ensure that reflections from the region of interest are captured with sufficient intensity for accurate imaging. Therefore, the C-scan mode (commonly referred to as C-SAM) is particularly effective for detecting delamination, voids, and cracks at material interfaces, such as underfill-to-die, mold-to-substrate, and interposer layers, and serves as one of the key FA methods in advanced packaging. These defects represent major reliability risks in large-format AI packages, particularly under high thermal and mechanical stress. Figure 2 displays extensive delamination at the die-attach level. A gated planar C-SAM image at the underlayer level shows that the delamination is widespread, with layers in the substrate gradually separating. The rectangular defect pattern at the bottom right indicates a popcorn crack resulting from the sudden expansion of trapped moisture.[5] X-ray computed tomography (XCT: micro-CT and nano-CT).[6-8] A fast overview scan (~30 s) at micrometerscale voxel size is first performed to capture the full structure of the sample, enabling rapid navigation and identification of regions of interest (ROIs), despite some noise and under-sampling artifacts. Based on this, a targeted high-resolution nano-CT scan is conducted on a selected ROI, achieving a voxel size on the order of a few hundred nanometers. This higher-resolution scan provides sufficient data quality for detailed analysis, including the identification of sub-micron voids in bump bonds. Figure 3 demonstrates that XCT provides 3D visualization of internal package structures, including TSVs, microbumps, solder joints, and voids. XCT enables rapid, package-level defect screening and precise localization of buried anomalies, such as voids in microbumps, cracks in RDLs, and TSV defects, without physical cross-sectioning. Fig. 2 C-SAM scanned top layer and underlayer images indicating the extensive delamination between the die surface and the mold compound.[5]
RkJQdWJsaXNoZXIy MTYyMzk3NQ==