ENHANCING THE SECURITY SCALABILITY OF ARBITER PUFS USING MEMORY-BASED WEAK PUFS
The vulnerability of conventional Arbiter Physical Unclonable Functions (PUFs) to machine learning-based modeling attacks significantly constrains their applicability in secure, resource-constrained environments. This thesis proposes an enhanced architectural solution, the Feed-Forward Memory-Based Arbiter PUF (FF-MB-APUF), which integrates volatile memory-based Weak PUFs with nonlinear feed-forward logic to improve entropy and strengthen resistance against modeling attacks. To evaluate the proposed design, a comprehensive experimental framework was developed, utilizing up to 50 million Challenge-Response Pairs (CRPs). The FF-MB-APUF variants were rigorously assessed for their statistical characteristics, including Bit Error Rate (BER), Strict Avalanche Criterion (SAC), randomness, and inter-device uniqueness. Notably, this work presents the first large-scale, systematic analysis of feed-forward loop placements across a wide configuration space. Experimental findings demonstrate that both the quantity and precise positioning of feed-forward loops critically affect modeling resilience. For adversarial benchmarking, state-of-the-art modeling strategies, including Deep Neural Networks (DNNs) and the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), were implemented and tuned. While prior studies report a 92% prediction accuracy using 5 XORs and 8 feed-forward loops with 3 million CRPs, the proposed FF-MB-APUF achieved significantly improved robustness, with only 84% prediction accuracy even under 10 million CRPs using 63 feed-forward loops and no XOR obfuscation. This highlights the intrinsic non-linearity and increased complexity introduced by the memory-based and feed-forward mechanisms. The optimized FF-MB-APUF configuration, incorporating 63 feed-forward loops across 64 stages, demonstrated superior resistance to modeling attacks, enhanced randomness (49.23%), and improved inter-device uniqueness (49.20%), resulting in a balanced output distribution and high entropy. These results establish the FF-MB-APUF as a scalable, hardware-efficient, and secure primitive ideal for next-generation embedded systems and low-power IoT deployments.
History
Degree Type
- Master of Science
Department
- Computer Science
Campus location
- Fort Wayne