Hardware cybersecurity has become a key issue, especially for very large integrated circuits. If counterfeit, forged, or defective ICs present a significant threat to system reliability and security. The growing complexity of digital and mixed-signal systems makes it increasingly challenging yet vital to develop robust methods to assess and confirm the reliability and authenticity of ICs. We introduce a new terahertz testing method for non-destructive and unobtrusive identification of counterfeit, damaged, forged or defective ICs by measuring their response to incident terahertz and sub-terahertz radiation at the circuit pins and analyzing the response using artificial intelligence (AI). These responses create unique signatures for ICs. We generated 2D images by measuring the response on a selected pin of a radio frequency IC (RFIC) scanned by a focused terahertz radiation. By applying the data augmentation processes, we created a secure image data set to train the convolutional neural network (CNN) model. An unsecured image data set representing altered or damaged ICs was generated by modifying the original image data. The trained models identified secure devices with a ~94% accuracy.
Read full abstract