Compact Hybrid Strong SOIPUF Exploiting Secret Key Generation and Dual-Edge Response Extraction for Enhanced Modeling Attack Resistance

With a huge number of device-specific challenge response pairs (CRPs), strong physical unclonable function (PUF) is a promising technology for low-cost loT device authorization. However, they are typically vulnerable to machine learning (ML) attacks due to the correlation among CRPs. Even though the existing hybrid strong PUFs (weak PUF assisted strong PUF) such as PRNG [1], AESSbox [2] and SPN [3] can ensure a high security through utilizing a huge amount of weak PUFs and extensive nonlinear computations, their complex implementation can result in an extremely large area overhead (e.g., 16.7 $\mathrm{MF}^{2}$ in [1], 23.8 $\mathrm{MF}^{2}$ in [2], and $25 \mathrm{MF}^{2}$ in [3], respectively). By taking advantage of the highly compact property of arbiter-based PUFs and addressing its weakness in ML resilience, this work presents a hybrid strong PUF using symmetrical obfuscated interconnection (SOI) [4], with the secret key generator (SKG) array for challenge obfuscation to combat against both black box and white box ML attacks while requiring an area of only $3.8 \mathrm{MF}^{2}$. Moreover, unlike the conventional arbiter-based PUFs that only employs one single edge as the trigger signal, we further propose the dual-edge (DE) response extraction technique to fully utilize the entropy sources from the SOI block for enhancing both the security and randomness with minimal hardware overhead. We further introduce the meta-stable detection (MD) technique to enhance the PUF reliability. This work can attain a measured ML prediction accuracy of 50.04% with up to 20 M training CRPs and a worse-case bit error rate (BER) of $4 \times 10^{-6}$, while achieving an extremely large number of stable CRP space of $\sim 4 \times 10^{31}$.