Modeling SRAM Start-up Characteristics For Physical Unclonable Functions
Apurva Dargar · Research Repository (Delft University of Technology) · 2011
The security of electronic devices is of crucial importance to companies as well as to users.Moreover, companies that develop Intellectual Property also want to protect them from counterfeiting and overbuilding.Company profits, brand reputations and personal information of the users are at stake if there is a breach in the security of these electronic devices.In the classical approach, a system is secured by storing the cryptographic keys permanently in the non-volatile memories that are present in the security devices.However, this permanent storage of the key makes them easy targets for physical attacks; hence compromising the security of the system.A more secure, cost-effective and elegant solution to this permanent key storage is the use of Physical Unclonable Functions (PUFs).PUF is a method of producing a signature from a physical object, such as an Integrated Circuit, by relying on the non-reproducible physical attributes of a device.These signatures are unique because fabricated circuits exhibit slightly different electrical behavior from one another even if their design, mask and manufacturing process are identical.Various kinds of PUFs exist; examples are Optical PUF, Butterfly PUF and SRAM PUF.However, the start-up value based SRAM PUFs appear to be the most promising ones for usage in ICs.Although the SRAM PUFs are becoming very popular, only a little has been published in the field of modeling and analysis of their start-up behavior.Reproducing the same start-up behavior, every time the chip is powered-on, is very crucial in order to produce the same cryptographic key.This thesis proposes an analytical model for the start-up value based SRAM PUFs; it helps in understanding the impact of both non-technological parameters (such as supply voltage and temperature) as well as technological parameters (such as the geometry of the transistors and threshold voltage) on the behavior of the start-up values of an SRAM.Various experiments have been performed to analyze and quantify their impact.The results obtained indicate a major impact of the non-technology parameters.The reproducibility of start-up values becomes more likely with slower ramp-ups and lower temperatures.For example, the percentage of reproducible bits increase from 93.5% at 1µs ramp-up to 96% at 10ms ramp-up.Amongst the technology parameters, it is observed that a small mismatch of 1.6% in the threshold voltage is enough to flip the start-up value of the cell for 65nm technology.These results have been validated by comparing them with actual silicon data measured at Intrinsic ID.The validation of the results proves the correctness of the analytical model proposed and gives a proof of robustness of the start-up values. Modeling SRAM Start-up Characteristics For Physical Unclonable Functions