Adaptive Milling Guided by a 3D Digital Twin for Large-Area Deprocessing of Warped Advanced Packages and BEOL Stacks

Thursday, October 8, 2026: 10:00 AM
Renliang Yuan , NVIDIA Corporation, Santa Clara, CA
Elia Halteh , NVIDIA Corporation, Santa Clara, CA
Romha Amha , NVIDIA Corporation, Santa Clara, CA
Dante Paquin , NVIDIA Corporation, Santa Clara, CA
Jonathon Elliott , NVIDIA Corporation, Santa Clara, CA
Chuan Zhang , NVIDIA Corporation, Santa Clara, CA
Jane Li , NVIDIA Corporation, Santa Clara, CA

Summary:

Chip warpage presents a significant challenge to the large-area deprocessing of advanced integrated circuits. Conventional uniform milling techniques often result in the simultaneous exposure of multiple interconnect layers, thereby compromising high-fidelity failure analysis. Here we report a closed-loop deprocessing framework utilizing adaptive milling guided by a 3D digital twin, which is continuously synchronized via high-frequency 2D top-down imaging. This workflow integrates image-based layer/interface identification with spatially customized material removal, employing variable-Z CNC milling for package-level removal and variable-dose focused ion beam milling for back-end-of-line delayering. Validation on warped package structures and a non-planar die region demonstrates that this adaptive approach improves target-layer continuity across large areas, including uniform exposure to the C4 bump level over an approximately 2500 mm2 CoW-scale package area. By enabling warpage-aware material removal, this method facilitates more controlled and scalable physical failure analysis for complex semiconductor architectures.