AI-Driven Failure Analysis for the Next Era of Semiconductors
AI-summarised brief · reviewed before publication
ZEISS Microscopy experts Heiko Stegmann and Allen Gu outline how AI-driven deep learning automates semiconductor failure analysis workflows previously reliant on human expertise. The study highlights three applications: optical inspection of bond pads, 3D X-ray nanotomography, and FIB-SEM tomography. An AI segmentation model achieved over 90% agreement with human inspectors and zero false positives in bond pad areas. A neural-network method called “DeepRecon” reduced X-ray nanoCT scan times from 60 hours to 15 hours while maintaining visualization reliability for hybrid bond voids. Additionally, an automated FIB-SEM workflow developed with GlobalFoundries identified open defects in M1 interconnects by comparing images against layout data, potentially reducing scan times to three hours. The authors emphasize that AI augments rather than replaces engineers, handling repetitive detection tasks. This automation is critical as the industry advances toward smaller nodes and complex 3D integration, preventing failure analysis capacity from becoming a production bottleneck.
💡 Why It Matters
- · As chip architectures grow increasingly complex, manual inspection methods physically cannot keep pace with production demands.
- · Automating these tedious detection tasks prevents quality control from stalling the entire manufacturing pipeline, ensuring that advanced semiconductor yields remain viable despite shrinking node sizes.
