SeMIA: Self-Similarity based IC Integrity Analysis

Counterfeit chips in the supply chain as well as hardware Trojan attacks pose serious threats to the semiconductor industry. If undetected before deployment, they can lead to serious consequences including system performance/reliability issues during field operation and potential revenue/reputation loss for a trusted manufacturer. Currently, no unified detection method is available that can simultaneously address these integrity violations in integrated circuits (ICs). In addition, most existing detection approaches require a set of golden chips as a reference, which significantly increases the test cost and complexity. Furthermore, in some scenarios, it may be extremely difficult to obtain golden chips. In this paper, we present a novel unified IC integrity analysis approach that can effectively detect both recycled counterfeit ICs (the most dominant form of counterfeiting) as well as Trojan attacks in ICs without the need of golden chips. The proposed approach, referred to as SeMIA, exploits intrinsic structural self-similarity in a design (e.g., multiple cores, multiple functional units of the same type, different parts of an adder) to isolate recycled chips and hardware Trojan attacks under large inter- and intra-die process variations. It compares dynamic current (IDDT ) signatures between two adjacent similar circuit structures using an appropriate isolation metric to detect such attacks with high degree of confidence. SeMIA does not rely on any embedded structure for authentication, thus it comes at virtually zero hardware overhead and can be applied to chips already produced. Through extensive simulations, we show that for 15% inter- and 10% intra-die variations in threshold voltage for a 45nm CMOS process, over 98% of recycled chips can be reliably identified. Finally, experimental measurements on Field Programmable Gate Array (FPGA) chips demonstrate the effectiveness of SeMIA for protection against both attacks.