High-Frequency Transmission Spectroscopy of Integrated Circuits

High-frequency transmission spectroscopy (HFTS) is a novel technique for the classification of a wide range of materials in biomedical, environmental, security, and manufacturing domains. HFTS is based on the fusion of scattering parameter measurements and machine learning classification techniques to identify materials of interest in novel environments. This work seeks to demonstrate the efficacy of HFTS in the domain of integrated circuit classification. Scattering parameter measurements are obtained through a custom interface between an electric network analyzer and the device under test. These scattering parameters are used as features to perform classification along three levels of discrimination: Device Type (timer vs. microcontroller), Part Number (MSP430G2210 vs MSP430G2230), and Lot ID. These measurements are used to perform both supervised classification and unsupervised anomaly detection. Results reported in this letter show that 100% classification and anomaly detection accuracies are achievable for discrimination at the device type and part number level. At the Lot ID level, 97% accuracy for supervised classification was obtained. This technique has wide application to the fields of semiconductor manufacturing and hardware security by providing a low-cost, low-complexity technique for identification of defective or counterfeit devices.