A Physical Unclonable Function Based on Inter-Metal Layer Resistance Variations and an Evaluation of its Temperature and Voltage Stability

Keying material for encryption is stored as digital bistrings in non-volatile memory (NVM) on FPGAs and ASICs in current technologies. However, secrets stored this way are not secure against a determined adversary, who can use probing attacks to steal the secret. Physical Unclonable functions (PUFs) have emerged as an alternative. PUFs leverage random manufacturing variations as the source of entropy for generating random bitstrings, and incorporate an on-chip infrastructure for measuring and digitizing the corresponding variations in key electrical parameters, such as delay or voltage. PUFs are designed to reproduce a bitstring on demand and therefore eliminate the need for on-chip storage. In this dissertation, I propose a kind of PUF that measures resistance variations in inter-metal layers that define the power grid of the chip and evaluate its temperature and voltage stability. First, I introduce two implementations of a power grid-based PUF (PG-PUF). Then, I analyze the quality of bit strings generated without considering environmental variations from the PG-PUFs that leverage resistance variations in: 1) the power grid metal wires in 60 copies of a 90 nm chip and 2) in the power grid metal wires of 58 copies of a 65 nm chip. Next, I carry out a series of experiments in a set of 63 chips in IBM's 90 nm technology at 9 TV corners, i.e., over all combination of 3 temperatures: -40oC, 25oC and 85oC and 3 voltages: nominal and +/-10% of the nominal supply voltage. The randomness, uniqueness and stability characteristics of bitstrings generated from PG-PUFs are evaluated. The stability of the PG-PUF and an on-chip voltage-to-digital (VDC) are also evaluated at 9 temperature-voltage corners. I introduce several techniques that have not been previously described, including a mechanism to eliminate voltage trends or 'bias' in the power grid voltage measurements, as well as a voltage threshold, Triple-Module-Redundancy (TMR) and majority voting scheme to identify and exclude unstable bits.Electrical EngineeringDoctoralUniversity of New Mexico. Dept. of Electrical and Computer EngineeringPlusquellic, JamesZarkesh-Ha, PaymanHelinski, RyanPerez-Gonzalez, Fernand