Learning-based identification of malicious circuits for trustworthy IoT systems

Trustworthiness is an important aspect for systems for IoT application, especially when it comes to solutions in the domains of security or privacy. Integrated circuits are an essential element of IoT systems, and thus, an attractive target. Therefore, one has to assure over the complete design and fabrication cycle, that no harmful changes have been made to the circuits. This motivated several researchers to focus on the subject of reverse engineering to search for malicious circuits, also known as hardware Trojans, which have been added during design or fabrication processes. Other purposes are the identification of patent violations or the support of existing verification solutions. Having in mind the complexity of this task, this work proposes an approach towards a fully automated solution that focuses on analog circuits. Therefore, an approach for identifying analog circuit structures in netlists extracted from random layouts is presented. Its feasibility is shown at the hand of current mirror circuits with varying architecture. The results indicate that the algorithm has an average accuracy of above 80%, with a maximum detection rate of 100%.