RDL-PUF: A Reconfigurable Double-Latch PUF with Multi-Nonlinear Obfuscation against Modeling-Attack

Abstract

Strong physical unclonable functions (PUFs) are widely used for lightweight device authentication, but their challenge-response mappings are vulnerable to machine-learning modeling attacks. Existing PUFs based on double-latch cells can improve modeling resistance, yet their fixed physical paths limit entropy extraction; meanwhile, cryptographic post-processing such as SHA-256 introduces considerable hardware overhead. To address these issues, this paper proposes a reconfigurable double-latch PUF (RDL-PUF). The proposed design uses six-input lookup table (LUT6) primitives to implement reconfigurable NAND gates with reset, and introduces path selection without additional LUT resources, thereby improving physical path diversity and enhancing exploitable entropy in responses. To further improve modeling resistance, a lightweight obfuscation mechanism is developed by combining modular addition, rotation, XOR, and AND/OR Boolean nonlinear operations, which strengthens structural nonlinearity with low hardware cost. The design is implemented on Xilinx Artix-7 FPGAs and occupies 464 LUTs and 261 FFs. Experimental results show that the prediction accuracies of DNN and MLP attacks are reduced to 50.07% and 50.04%, respectively.