Enhancing counterfeit detection of integrated circuits through machine learning-assisted THz-TDS analysis

Abstract

Due to the global offshore fabrication of semiconductors, hardware security problems such as counterfeit ICs and Hardware Trojans (HTs) have affected semiconductor device trustworthiness in critical applications. Previous research has proven the encapsulant material difference exists in the counterfeit ICs. What's more by using a pulsed THz signal, Terahertz Time-Domain Spectroscopy (THz-TDS) is able to detect the effective refractive index difference between authentic and counterfeit IC packaging. This research has also successfully observed the reflective index difference between authentic and counterfeit by measuring the layer thickness and THz-TDS time delay. However, the accuracy of calculating the effective refractive index depends on the accuracy of the layer thickness and the time delay measurements. Consequently, reflective index difference may arise from noise encountered during data collection, which can affect the accuracy and repeatability of counterfeit detection tasks. In this paper, we utilize an unsupervised machine learning model to further demonstrate the capabilities of THz-TDS in counterfeit IC detection. Additionally, the potential of using THz-TDS to generate a unique fingerprint is also discussed.