Design, Fabrication, and Run-time Strategies for Hardware-Assisted Security

Thumbnail Image


Publication or External Link






Today, electronic computing devices are critically involved in our daily lives, basic infrastructure, and national defense systems. With the growing number of threats against them, hardware-based security features offer the best chance for building secure and trustworthy cyber systems. In this dissertation, we investigate ways of making hardware-based security into a reality with primary focus on two areas: Hardware Trojan Detection and Physically Unclonable Functions (PUFs). Hardware Trojans are malicious modifications made to original IC designs or layouts that can jeopardize the integrity of hardware and software platforms. Since most modern systems critically depend on ICs, detection of hardware Trojans has garnered significant interest in academia, industry, as well as governmental agencies. The majority of existing detection schemes focus on test-time because of the limited hardware resources available at run-time. In this dissertation, we explore innovative run-time solutions that utilize on-chip thermal sensor measurements and fundamental estimation/detection theory to expose changes in IC power/thermal profile caused by Trojan activation. The proposed solutions are low overhead and also generalizable to many other sensing modalities and problem instances. Simulation results using state-of-the-art tools on publicly available Trojan benchmarks verify that our approaches can detect Trojans quickly and with few false positives. Physically Unclonable Functions (PUFs) are circuits that rely on IC fabrication variations to generate unique signatures for various security applications such as IC authentication, anti-counterfeiting, cryptographic key generation, and tamper resistance. While the existence of variations has been well exploited in PUF design, knowledge of exactly how variations come into existence has largely been ignored. Yet, for several decades the Design-for-Manufacturability (DFM) community has actually investigated the fundamental sources of these variations. Furthermore, since manufacturing variations are often harmful to IC yield, the existing DFM tools have been geared towards suppressing them (counter-intuitive for PUFs). In this dissertation, we make several improvements over current state-of-the-art work in PUFs. First, our approaches exploit existing DFM models to improve PUFs at physical layout and mask generation levels. Second, our proposed algorithms reverse the role of standard DFM tools and extend them towards improving PUF quality without harming non-PUF portions of the IC. Finally, since our approaches occur after design and before fabrication, they are applicable to all types of PUFs and have little overhead in terms of area, power, etc. The innovative and unconventional techniques presented in this dissertation should act as important building blocks for future work in cyber security.