Design for prognostics and security in field programmable gate arrays (FPGAs).

There is an evolutionary progression of Field Programmable Gate Arrays (FPGAs) toward more complex and high power density architectures such as Systems-on- Chip (SoC) and Adaptive Compute Acceleration Platforms (ACAP). Primarily, this is attributable to the continual transistor miniaturisation and more innovative and efficient IC manufacturing processes. Concurrently, degradation mechanism of Bias Temperature Instability (BTI) has become more pronounced with respect to its ageing impact. It could weaken the reliability of VLSI devices, FPGAs in particular due to their run-time reconfigurability. At the same time, vulnerability of FPGAs to device-level attacks in the increasing cyber and hardware threat environment is also quadrupling as the susceptible reliability realm opens door for the rogue elements to intervene. Insertion of highly stealthy and malicious circuitry, called hardware Trojans, in FPGAs is one of such malicious interventions. On the one hand where such attacks/interventions adversely affect the security ambit of these devices, they also undermine their reliability substantially. Hitherto, the security and reliability are treated as two separate entities impacting the FPGA health. This has resulted in fragmented solutions that do not reflect the true state of the FPGA operational and functional readiness, thereby making them even more prone to hardware attacks. The recent episodes of Spectre and Meltdown vulnerabilities are some of the key examples. This research addresses these concerns by adopting an integrated approach and investigating the FPGA security and reliability as two inter-dependent entities with an additional dimension of health estimation/ prognostics. The design and implementation of a small footprint frequency and threshold voltage-shift detection sensor, a novel hardware Trojan, and an online transistor dynamic scaling circuitry present a viable FPGA security scheme that helps build a strong microarchitectural level defence against unscrupulous hardware attacks. Augmented with an efficient Kernel-based learning technique for FPGA health estimation/prognostics, the optimal integrated solution proves to be more dependable and trustworthy than the prevalent disjointed approach.