Seeing
inside the
chip.
I develop multimodal, AI-assisted non-destructive failure analysis for 3D integrated circuits and advanced semiconductor packaging — fusing X-ray, acoustic, terahertz, and optical inspection to localize defects buried in heterogeneous material stacks.
Multimodal fusion for 3D ICs
Feature- and decision-level AI fusion of X-ray, acoustic, terahertz, and optical inspection to reveal bonding defects invisible to any single modality.
X-ray CT & laminography
TGV/RDL and HBM3 characterization from 5 µm to 300 nm voxel resolution — void formation, copper pullout, delamination, and material-aware reconstruction.
Terahertz fingerprinting
Near-field THz time-domain spectroscopy as a distinguishing modality for IC identification, provenance, and tracking-and-tracing applications.
PCB assurance & netlist extraction
Semi-supervised deep learning that cuts the labeled-data burden for PCB component detection and interconnect inspection under X-ray and optical imaging.
Multimodal X-ray characterization of through-glass vias and RDL structures for heterogeneous integration.
Material-aware X-ray characterization and spectral reconstruction for semiconductor assurance.
A large-scale multimodal characterization dataset for NVIDIA H100 HBM3 memory systems.
Open to collaboration
Interested in multimodal inspection, failure analysis, or semiconductor assurance datasets? Get in touch or browse the full record.
From single-modality PCB inspection to multimodal AI failure analysis
Over the past five years my work has followed a trajectory from single-modality printed-circuit-board inspection toward multimodal, AI-assisted failure analysis of heterogeneously integrated semiconductor packages — carried out primarily at the University of Florida Security and Assurance Laboratory (SCAN Lab) under Dr. Navid Asadi, with collaborations at NIST, DOE, DHS, and industry partners.
Multimodal characterization of heterogeneously integrated devices
The conceptual core of my dissertation: heterogeneous material stacks in advanced packages exceed the sensitivity of any single non-destructive modality, motivating a fused-inspection framework combining X-ray, acoustic, terahertz, and optical measurements with AI-based fusion. I have built out this workflow using the SCAN Lab's Sigray Apex hybrid X-ray CT/laminography, Nordson Sonoscan SAM, Protemics TeraCube near-field THz-TDS, and optical microscopy — investigating the relationships between imaging physics, material properties, and defect observability.
Multimodal X-ray characterization of through-glass vias
In collaboration with NIST, I designed and executed X-ray CT and laminography workflows across wafer- and die-scale samples at voxel resolutions from 5 µm to 300 nm, characterizing high-aspect-ratio TGV structures with via pitches from 100 to 400 µm. The work investigated void formation, copper pullout, delamination, interface defects, and process-induced variability across bonded and freestanding wafer configurations, with region-of-interest and montage acquisition strategies to optimize throughput and defect detectability.
Multimodal dataset for HBM3 memory systems
A large-scale multimodal characterization dataset for NVIDIA H100 HBM3 memory systems, integrating X-ray CT, X-ray laminography, and scanning acoustic microscopy. The work performs non-destructive imaging of HBM interconnects, microbumps, TSVs, substrate interfaces, and multilayer package structures, curating thousands of annotated defect and interconnect features to support agentic AI workflows, automated failure analysis, and advanced packaging metrology — and to benchmark the multimodal fusion methods central to my dissertation.
Material-aware X-ray for semiconductor assurance
A Department of Energy-funded initiative on semiconductor assurance and material-aware X-ray characterization. My contributions include multimodal workflows integrating multi-energy X-ray CT and laminography with spectral reconstruction, methods for reconstructing effective atomic number and electron density within 3D X-ray volumes, and software pipelines for material-aware 3D PCB reconstruction — GERBER-to-material representations, spectral CT integration, and layer-alignment visualization.
PCB assurance & netlist extraction
My earliest doctoral work — the methodological foundation on which the current multimodal work rests. I demonstrated that semi-supervised deep learning can substantially reduce the labeled-data burden for PCB component detection under X-ray (RAPID 2024), extended it to a multimodal X-ray-and-optical setting (SPIE 2024), and developed an optical PCB-interconnect inspection pipeline (ISTFA 2024) later extended into a journal article (JFAP 2025), with a high-throughput X-ray extension reported at RAPID 2025.
Terahertz TDS for IC identification
Broadening the modality set beyond X-ray and optical imaging, I first-authored work establishing near-field THz time-domain spectroscopy as a distinguishing modality for IC identification and provenance (SPIE Terahertz Emitters, Receivers, and Applications XV, 2024). This introduced the Protemics TeraCube capability into my multimodal workflow and provides the terahertz contribution to the fusion framework.
Program leadership & broader contributions
I serve as technical lead for a multi-million-dollar NSF Engines-funded semiconductor R&D initiative, leading development of multimodal datasets and AI-assisted failure-analysis workflows. Across two program years I contributed to the DHS HS-POWER program on automated microelectronics assurance under Dr. Pauline Paki, co-authoring work on advanced-packaging raw-materials assurance (SPIE 2025) and mentoring three student interns to publishable outputs. I have co-authored numerous SCAN Lab papers spanning explainable-AI metrics, physics-informed neural networks, SAM adaptation, counterfeit-IC classification, and VLM/LLM inspection assistants.
Peer-reviewed & refereed work
Blue marks first authorship. Full profile: Google Scholar ↗
US patent applications
Two 2026 US patent applications, co-invented at the University of Florida SCAN Lab, formalize outputs of my multimodal inspection and physics-informed reconstruction work.
Explainable AI metrics for IC packaging inspection
Methods for quantifying and explaining AI-driven inspection decisions for integrated-circuit packaging, providing interpretable metrics that support root-cause failure attribution in non-destructive analysis.
Image reconstruction with multimodal fusion & physics-informed neural network
A reconstruction approach that combines multimodal measurement data with a physics-informed neural network, improving image quality and defect sensitivity for advanced-packaging inspection beyond any single modality.
Patrick Craig
I am a PhD candidate in Electrical and Computer Engineering at the University of Florida, researching at the Security and Assurance Laboratory (SCAN Lab) under Dr. Navid Asadi. My work develops multimodal, AI-assisted non-destructive inspection and failure-analysis methods for 3D integrated circuits and advanced semiconductor packaging.
Over five years I have built inspection workflows spanning optical microscopy, X-ray CT and laminography, scanning acoustic microscopy, and near-field terahertz spectroscopy — and serve as technical lead for a $1M+ NSF Engines-funded semiconductor R&D initiative, in collaboration with NIST, Department of Energy, and DHS.
PhD, Electrical & Computer Engineering
BS, Electrical Engineering
Graduate Research Assistant
- Technical lead for a $1M+ NSF Engines-funded semiconductor R&D initiative, developing multimodal datasets and AI-assisted failure-analysis workflows.
- Built multimodal characterization workflows integrating optical microscopy, X-ray CT/laminography (Sigray Apex Hybrid), SAM (Nordson Sonoscan), and near-field THz-TDS (Protemics TeraCube).
- Contributed to 15+ conference presentations, journal publications, and technical demonstrations across federally funded and industry-sponsored programs.
Hardware & Logistics Intern
- Performed system-level performance analysis of SATCOM and RADAR subsystems, evaluating communication efficiency, robustness, and architectural constraints.
- Supported architectural trade-off studies assessing design choices impacting performance, reliability, and integration.
- Authored hardware and system documentation and validated electrical system requirements against program-level specifications.
Programming & analysis
Python, C++, MATLAB · NumPy, Pandas, SciPy, Matplotlib · KLayout
Non-destructive evaluation
Optical microscopy · X-ray CT / laminography · Scanning acoustic microscopy (SAM) · THz-TDS
Test & validation
Device characterization · failure analysis · lab instrumentation · measurement data acquisition
- Technical lead — NSF Engines semiconductor R&D initiative ($1M+)
- DHS HS-POWER program — automated microelectronics assurance (two program years)
- NIST collaboration — multimodal X-ray characterization of TGV/RDL
- Mentored three bachelor's- and master's-level interns to publishable outputs on assurance and counterfeit detection (DHS HS-POWER, two summers).
- Frequent co-author and lead across SCAN Lab student projects (SPIE, ECTC, RAPID, PAINE, ISTFA).
Get in touch
Open to collaboration on multimodal inspection, failure analysis, and semiconductor assurance. The full CV is available below.
Download CV (PDF)