N. Akter, J. Suarez, M. Shur, N. Pala
Rensselaer Polytechnic Institute,
United States
Keywords: AI characterization, terahertz, hardware security, reliability testing, VLSI
Summary:
Hardware security is becoming increasingly crucial as IC designs get more complex and sophisticated. To have a strong, reliable, and secure infrastructure, it is critical to distinguish genuine and fault-free ICs from counterfeit, altered, or defective ICs. The analysis of an IC’s response to incident terahertz (THz) waves has emerged as a promising technique to assess their authenticity and reliability. The use of Artificial Intelligence (AI)/Machine Learning (ML) techniques can improve the speed and accuracy of this analysis. We report on the use of a simple convolutional neural network and five distinct transfer learning models to analyze the THz response of radio-frequency ICs (RFICs). We generated a map of 2D images by measuring the response on a selected pin of RFICs scanned by focused terahertz radiation. We increased the number of image datasets by applying the data augmentation processes and then trained the convolutional neural network (CNN) model with and without the transfer learning approach. We obtained the unsecure image dataset representing altered or damaged ICs by damaging the IC with a high voltage and modifying the original image data by generating random noise (~20% to 200%). Our model can successfully distinguish the secure and unsecure IC images with 98% accuracy. We have also developed a software application to automate data processing and AI analysis.