3D current reconstruction for Failure Analysis with Quantum Diamond Microscopy
3D current reconstruction for Failure Analysis with Quantum Diamond Microscopy
Wednesday, October 7, 2026: 10:20 AM
Summary:
Understanding electrical current flow within integrated circuits is essential for effective electrical failure analysis (EFA), as it directly reveals the conduction paths associated with defects. Quantum Diamond Microscopy (QDM) has emerged as a powerful, non-destructive technique for magnetic current imaging, enabling quantitative reconstruction of current distributions even in buried structures. In this work, we extend the capabilities of QDM from conventional two-dimensional analysis to three-dimensional (3D) current reconstruction using a data-driven approach. A U-Net–based deep learning model is trained on physically representative current distributions to infer volumetric current density from vector magnetic field measurements acquired at a single observation plane. The method is validated using benchmark datasets with known ground truth, custom-designed test structures, and real device measurements. The results demonstrate accurate depth-resolved reconstruction across a range of stand-off distances and structural complexities with a depth accuracy in the order of 15%. This approach enables enhanced insight into current flow in advanced semiconductor devices and packages, highlighting the potential of combining QDM with machine learning to improve defect localization and root cause analysis in modern EFA workflows.
Understanding electrical current flow within integrated circuits is essential for effective electrical failure analysis (EFA), as it directly reveals the conduction paths associated with defects. Quantum Diamond Microscopy (QDM) has emerged as a powerful, non-destructive technique for magnetic current imaging, enabling quantitative reconstruction of current distributions even in buried structures. In this work, we extend the capabilities of QDM from conventional two-dimensional analysis to three-dimensional (3D) current reconstruction using a data-driven approach. A U-Net–based deep learning model is trained on physically representative current distributions to infer volumetric current density from vector magnetic field measurements acquired at a single observation plane. The method is validated using benchmark datasets with known ground truth, custom-designed test structures, and real device measurements. The results demonstrate accurate depth-resolved reconstruction across a range of stand-off distances and structural complexities with a depth accuracy in the order of 15%. This approach enables enhanced insight into current flow in advanced semiconductor devices and packages, highlighting the potential of combining QDM with machine learning to improve defect localization and root cause analysis in modern EFA workflows.