BumpPrint: Intrinsic Package Fingerprinting with Scanning Acoustic Microscopy
BumpPrint: Intrinsic Package Fingerprinting with Scanning Acoustic Microscopy
Wednesday, October 7, 2026: 5:00 PM
Summary:
This work presents an intrinsic hardware authentication framework based on Scanning Acoustic Microscopy of solder bump and micro bump arrays in semiconductor packages. The method exploits natural microstructural variability introduced during fabrication to derive a reproducible acoustic fingerprint for each device without embedding dedicated security circuitry. High resolution SAM C scan images are processed through automated bump segmentation, intensity and morphology feature extraction, and unsupervised learning to generate stable bump type maps that serve as package level identifiers. Principal Component Analysis and UMAP are used for dimensionality reduction, followed by clustering based discretization to assign canonical bump classes. Fingerprints are compared using Hamming distance to evaluate uniqueness and repeatability across multiple imaging sessions. Experimental results on several BGA packages show low intra device variation and strong inter device separation. Accelerated thermal cycling further demonstrates that the extracted fingerprints remain stable under environmental stress, with no identity reversals observed. The proposed approach offers a non destructive, physically grounded, and scalable path toward anti counterfeit verification, provenance assurance, and traceability in advanced electronic packaging.
This work presents an intrinsic hardware authentication framework based on Scanning Acoustic Microscopy of solder bump and micro bump arrays in semiconductor packages. The method exploits natural microstructural variability introduced during fabrication to derive a reproducible acoustic fingerprint for each device without embedding dedicated security circuitry. High resolution SAM C scan images are processed through automated bump segmentation, intensity and morphology feature extraction, and unsupervised learning to generate stable bump type maps that serve as package level identifiers. Principal Component Analysis and UMAP are used for dimensionality reduction, followed by clustering based discretization to assign canonical bump classes. Fingerprints are compared using Hamming distance to evaluate uniqueness and repeatability across multiple imaging sessions. Experimental results on several BGA packages show low intra device variation and strong inter device separation. Accelerated thermal cycling further demonstrates that the extracted fingerprints remain stable under environmental stress, with no identity reversals observed. The proposed approach offers a non destructive, physically grounded, and scalable path toward anti counterfeit verification, provenance assurance, and traceability in advanced electronic packaging.