edfas.org 31 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 1 approach can substitute for laser voltage probing in backside-powered devices, enable measurements on live digital circuits, and support detection of resistive gate defects. Joshua Ballard of Tiptek LLC detailed the requirements for UHF AC nanoprobing, emphasizing the need for sharp, highly conductive probes with coaxial shielding to minimize loss and cross talk. After showing a TEM cross-section of a shielded UHF probe, he introduced a system incorporating a local amplifier to overcome system capacitance. Comparative transmission measurements demonstrated the significant gain achieved by using coaxially shielded probes over bare ones. Hanaul Noh of Park Systems presented scanning microwave impedance microscopy (sMIM), an AFM-based technique that maps local material properties using a shielded probe guiding a 3 GHz microwave signal. He demonstrated applications for semiconductors, dielectrics, embedded structures, and 2D materials, concluding with a quantitative method for determining local semiconductor doping levels. David Albert from IBM joined the panel discussion, which centered on the major challenges in UHF nanoprobing and SPM. Audience members raised issues including probe wear, intermittent-failure characterization, backside power-delivery methodologies, and the relationship to e-beam probing. Discussion also addressed quantification challenges in SPM, bandwidth limits of high-frequency techniques, and the impact of capacitive loading from the probe itself. The session concluded with speculation on a future tool that could merge high-frequency nanoprobing with e-beam detection to mitigate such loading effects. ISTFA 2025 ARTIFICIAL INTELLIGENCE USER GROUP Chair/Co-Chairs: Christoph Maier, representing Florian Felux, Peter Hoffrogge, Thomas Rodgers, and Konstantin Schekotihin christoph.maier@infineon.com, peter.hoffrogge@pvatepla.com, thomas.rodgers@zeiss.com, konstantin.schekotihin@aau.at The ISTFA 2025 Artificial Intelligence User Group session, including panelist Navid Asadi from the University of Florida, explored practical ways to leverage AI in failure analysis while protecting proprietary data. The most discussed point was data protection, with participants cautioning against sharing raw images, which conflicts with the need for FA-related images/data to improve AI tools for FA. Alternative approaches, such as exchanging model weights and using pre-trained networks, were proposed. Collaboration requires clear legal frameworks and explicit scoping of what can be shared. The group emphasized the importance of pragmatic data-minimizing exchanges, governance, and legal clarity to enable cooperation. Human-in-the-loop reinforcement learning can keep experts in control while improving models. Sourcing data through academic partners and public facilities, as well as using synthetic data, were also discussed as potential solutions. Additionally, the group highlighted the value of using already approved Artificial Intelligence User Group presenters and co-chairs. “COMPARATIVE TRANSMISSION MEASUREMENTS DEMONSTRATED THE SIGNIFICANT GAIN ACHIEVED BY USING COAXIALLY SHIELDED PROBES OVER BARE ONES.” publications to seed training and building a shared literature-derived database. Live polling indicated that many labs already manage substantial repositories of image and text data as well as measurements, providing a foundation for AI applications. AI adoption levels of the participants vary widely,
RkJQdWJsaXNoZXIy MTYyMzk3NQ==