INSIGHT-X: Multimodal VLM Assistant for Artifact Localization, and Mitigation Guidance in X-ray Inspection
Year: 2026
Type: article
Abstract: Advanced semiconductor packages rely on X-ray CT and laminography for non-destructive inspection of buried interconnects and internal structures. However, dense metal features and heterogeneous materials often introduce artifacts, including ring artifacts, beam-hardening streaks, geometric distortions, and missing-cone artifacts. These artifacts obscure critical features during inspection, reduce reconstruction fidelity, and complicate defect interpretation, requiring expert-driven and iterative troubleshooting. To address these challenges, we propose INSIGHT-X, a vision–language model framework for artifact diagnosis, localization, root-cause and mitigation in X-ray inspection workflows. INSIGHT-X formulates artifact analysis as a multimodal question-answering task by integrating reconstructed X-ray images, artifact annotations, and expert knowledge into an artifact-centric dataset. Through parameter-efficient fine-tuning, the model learns to identify artifact types, localize affected regions, and generate interpretable explanations. The system also provides artifact-aware cause and mitigation recommendations, including general acquisition and reconstruction adjustments, and calibration guidance. By combining visual understanding, spatial localization with explainable answers, INSIGHT-X transforms expert knowledge into a scalable AI assistant for improving inspection consistency, accelerating troubleshooting, and supporting decision-making in advanced semiconductor manufacturing.
Source: IMAPSource Proceedings
DOI: 10.4071/001c.173631
Institution: University of Florida
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7225331430