THz response based hardware security and reliability testing powered by deep learning image classification

THz testing has been recently proposed to identify altered or damaged ICs. This method is based on the fact that a modern field-effect transistor (FET) with a sufficiently short channel can serve as a terahertz detector. The response can be recorded while changing the THz radiation parameters and location and compared to a trusted one for classification. We measured the THz response of original and damaged ICs for classification using different Transfer Learning models as a method of deep learning. We have achieved the highest classification accuracy of 98%.