edfas.org 9 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 1 segmentation using more training data, it should be possible to exploit all differences between experiment and reference by suitable algorithmic comparison of distributions. Yet another reference-free approach to be investigated in the future is directly classifying open and intact interconnects in the experimental dataset, like the approach used in reference 3. CONCLUSION The results achieved with automated slice and view 3D tomography dataset using the latest generation of FIB-SEM systems in combination with ZEISS Atlas 3D software are quite impressive. Virtual slicing in all directions, especially the top-down view, significantly helps to interpret defect type and find root cause. In addition, auto slice and view can save resources and maximize the tool utilization, e.g., overnight or over the weekend. An automated defect detection using the recorded tomography dataset is much more challenging than expected. The five approaches based on comparison with CAD data: A, B, C (conventional gray-value based) and D, E (deep learning-based methods) so far are not yet capable of reliably identifying defects without extensive data preprocessing or manual data exploration. This can be attributed to the inherent structural disparities between the CAD and FIB-SEM data sets. In future work, more training data is needed to improve deep learning segmentation and refine the models. Nevertheless, the necessary hardware and software are available, and sufficient computing power is in place for model training. It is only a matter of time before automation in failure analysis becomes a reality. The authors agree with a statement from David Albert: “Use computer power versus people power to solve problems. Save the people power for the difficult signatures (problems).[9]” ACKNOWLEDGMENTS This article was also presented at ISTFA 2025. REFERENCES 1. D. Mello, et al.: “New Approach in Physical Failure Analysis Based on 3D Reconstruction,” Proc. Int. Symp. Test. Fail. Anal., 2022, p. 201-205, doi.org/10.31399/asm.cp.istfa2022p0201. 2. H. Stegmann and A. Laquerre: “FIB-SEM Tomography Acquisition and Data Processing Optimization for Logic and Memory Structures,” Proc. Int. Symp. Test. Fail. Anal., 2023, p. 387-392, doi.org/10.31399/ asm.cp.istfa2023p0387. 3. H. Stegmann and F. Cognigni: “Few-Shot AI Segmentation of Semiconductor Device FIB-SEM Tomography Data,” Journal of Failure Analysis and Prevention, 25, 2025, doi.org/10.1007/ s11668-025-02203-w. 4. M. Tan and Q.V. Le: “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” Proc. Int. Conf. Machine Learning, 2019, proceedings.mlr.press/v97/tan19a.html. 5. A.P. Aitken, et al.: “Checkerboard Artifact Free Sub-pixel Convolution: A Note on Sub-pixel Convolution, Resize Convolution, and Convolution Resize,” 2017, doi.org/10.48550/arXiv.1707.02937. 6. E. Dehaerne, et al.: “Electron Microscopy-based Automatic Defect Inspection for Semiconductor Manufacturing: A Systematic Review,” Journal of Micro/Nanopatterning, Materials, and Metrology, 24(2), p. 020901. 7. S. Choi, et al.: “Machine Learning Based SEM Image Analysis for Automatic Detection and Classification of Wafer Defects,” SEMI Advanced Semiconductor Manufacturing Conference (ASMC), 2024, p. 1-4, doi.org/10.1109/ASMC61125.2024.10545512. 8. P. Isola, et al.: “Image-to-Image Translation with Conditional Adversarial Networks,” Nov 2018, doi.org/10.48550/arXiv.1611.07004. 9. D. Albert, et al.: “Yield Basics for Failure Analysts,” Tutorial presented at ISTFA 2024, doi.org/10.31399/asm.cp.istfa2024tpb1. ABOUT THE AUTHORS Pascal Limbecker received his diploma in electrical engineering at the University of Applied Sciences in Zwickau, Germany in 2005. Immediately afterwards, he began his career as failure analysis engineer at AMD Saxony, Germany. With the transition to GlobalFoundries in 2009, he has more than 20 years of experience in defect localization using OBIRCH, lock-in thermography, photon emission, nanoprobing, circuit editing, and chip modification using FIB-SEM tools. He is currently a principal member of technical staff and the technical lead engineer in the FA Lab at GlobalFoundries in Dresden, Germany. Rong Wu received her Ph.D. in physical chemistry from Freie Universität Berlin, Germany, in 2024. Following her doctoral study, she pursued postdoctoral research at Helmholtz Zentrum Berlin, working on memristive devices for neuromorphic applications. In 2025, she joined GlobalFoundries, Dresden, Germany, as a failure analysis engineer. Her current work focuses on FIB-SEM failure analysis of memory and logic devices. Heiko Stegmann studied physics and worked on his doctorate at the University of Heidelberg, Germany, 1990-1998, specializing in analytical TEM for biophysics applications. He did his post-doc in 3D TEM of motor proteins at the Max Planck Institute for Medical Research, Heidelberg, Germany, 1998-2000. Stegmann was a senior materials analyst at AMD
RkJQdWJsaXNoZXIy MTYyMzk3NQ==