edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 26 NO. 4 18 recipes, using XRM data to visualize the wires in the SEM through the surface of the sample. After optimal parameters were derived, a single wire was cut to demonstrate precision and replication of the milling parameters on other sites. Figure 7 is a 3D XRM image showing one such result, revealing that the open gap created by the fs-laser cut is 20 µm (the spot size of the laser is <15 µm). In this example, the initial fast, low-resolution scan with XRM is achieved in under 30 minutes and provides sufficient data to observe the internal structure of the devices and layout of the interconnects and wires. Mounting the sample on a carbon stub allows for quick transfers for laser milling to generate fiducial markers—the milling itself takes one second. A second 3D XRM scan at higher resolution is performed so both the ROI and surface fiducial markers will be visible in the resulting XRM data. The position of the wires with respect to the laser-milled surface fiducial markers can be determined by overlaying the virtual XRM data with the SEM image and aligning the laser-made fiducials that are visible in both images. Laser milling parameters are optimized with a dose test to make sure the milling only cuts the wire and does not penetrate deeper to damage the die. The subsequent laser milling using three different patterns takes <20 seconds to cut multiple wires, providing several data points to check for repeatability and local variations in the materials due to fillers or additional components. Another scan by the XRM checks the depth of the laser cuts to evaluate which patterns cut the wire reliably without damaging the die. The final steps apply the same workflow targeted at a single wire in another area of the chip, including previous x-ray overview scans that provide low resolution position information to laser mill surface fiducial marks. By following the earlier recipe, laser milling is completed in less than one second. The final x-ray scan verifies that the laser cut precisely targets the wire of interest without damaging the neighboring wire or the die below, indicating the workflow can be employed for precise and targeted sample preparation. These examples show that the recipe is repeatable at different locations within the same sample and for similar packages using the same molding compound materials. However, due to variations in filler materials and other additives, different samples may require further optimization. Although further functional testing of the chip is needed to validate the proposed method, this experiment targeting an interconnect wire for functional testing and fault isolation shows promise as the workflow itself takes about 5 hours and the entire process can be done in 6-8 hours, including data reconstruction and preparation time between steps, and it does so nondestructively. There is opportunity to speed up this workflow by employing AI reconstruction technologies. NOVEL PROCESSES SPEED UP INSPECTION WORKFLOWS These use cases demonstrate that combining 3D XRM and deep learning supports a robust and repeatable nondestructive inspection method for rapid analysis of highly integrated packaging structures with reasonable throughput for process validation and error correction guidance, suited for process development as well as FA quality checks. Fig. 7 3D XRM verification image after execution of optimized process on single wire. The 20 µm gap created by the fs-laser cut is clearly visible and other structures show no signs of damage.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==