A Hybrid Strong-Weak PUF Against Machine Learning Attacks with High Reliability

Physical Unclonable Function (PUF), as a lightweight hardware security primitive, is widely applied in resource-constrained Internet of Things (IoT) devices. However, the strong PUFs are vulnerable to modeling attacks. In this paper, we propose a configurable hybrid strong-weak PUF (SW PUF) structure that combines the characteristics of strong with weak PUFs, which integrates both strong and weak PUFs into a new PUF design. This proposed SW PUF can be flexibly configured into a strong PUF (an arbiter PUF, a ring oscillator PUF, or a bistable ring PUF) or a weak PUF (SRAM PUF) which is configured by utilizing challenges to select different process biases. Moreover, the presented PUF utilizes unique key streams of weak PUF to bitwise encrypt the raw challenges from the strong PUFs to obfuscate the relationship between the challenge and the response, facilitating an inherent immunity to white-box modeling attacks. Experimental results show that the modeling accuracies of LR, SVM, CMA-ES, and ANN are close to 55% and the SW PUF has high reliability without any post-processing techniques.