Two-Metric Helper Data for Highly Robust and Secure Delay PUFs

The Physically Unclonable Function (PUF) has become an inescapable security primitive, as it generates a fingerprint unique to each device, and is natively robust against reverse engineering attacks. However it suffers from many flaws due to a native lack of reliability and sensitivity against modeling and physical attacks. In the state-of-the-art, the enhancement of reliability involves the use of a helper data which takes advantage of Error-Correcting Codes (ECC), situated after the entropy source, thus increasing the overall PUF complexity and creating leakages exploited by Side Channel Attacks.This paper presents a new method to generate Helper Data, situated before the entropy source, which does not use ECC but still improves drastically the reliability of the PUF. It also provides high entropy and proven resistance against the attacks on the helper data. This method applies to "strong" delay PUFs used as master key generators, thus unsensitive to modeling attacks (the PUF responses are never disclosed). The validation has been done on real ASIC devices fabricated in CMOS 45 nm technology and shows that the reliability can be as low as 10−8 of Bit Error Rate.