The efficiency and cost of silicon PUF-based applications, and in particular key generators, are heavily impacted by the level of reproducibility of the bare PUF responses under varying operational circumstances. Error-correcting codes can be used to achieve near-perfect reliability, but come at a high implementation cost especially when the underlying PUF is very noisy. When designing a PUF-based key generator, a more reliable PUF will result in a less complex ECC decoder and a smaller PUF footprint, hence an overall more efficient implementation. This paper proposes a novel insight and resulting technique for reducing noise on memory-based PUF responses, based on adapting supply voltage ramp-up time to ambient temperature. Circuit simulations on 45nm Low-Power CMOS, as well as actual silicon measurements are presented to validate the proposed methods. Our results demonstrate that choosing an appropriate voltage ramp-up for enrollment and adapting it according to the ambient temperature at key-reconstruction is a powerful method which makes memory-based PUF response noise up to three times smaller.
展开▼
