A Response-Feedback-Based Strong PUF with Improved Strict Avalanche Criterion and Reliability

B. Zhu, Xiaowen Jiang, Kai Huang, Miao YuSensors 2023 · DOI

TL;DR

This paper introduces a novel strong Physical Unclonable Function (PUF) design that enhances security against machine learning (ML) and statistical attacks by incorporating response feedback and a Linear Feedback Shift Register (LFSR). It also presents a Two-Level Reliability Improvement (TLRI) method that significantly reduces the overhead for achieving high reliability, striking a better balance between security and practicality for IoT devices.

What Problem Does This Paper Address?

The Internet of Things (IoT) is rapidly expanding, creating a critical need for secure authentication mechanisms for resource-constrained devices. Traditional cryptographic methods are often too power-hungry and require secure key storage, which is vulnerable to physical attacks. Physical Unclonable Functions (PUFs) offer a promising alternative by generating unique device IDs from inherent manufacturing variations, eliminating the need for stored keys. However, existing strong PUFs, particularly Arbiter PUFs (APUFs), are susceptible to sophisticated attacks like ML modeling and statistical attacks due to their limited statistical properties (e.g., poor Strict Avalanche Criterion - SAC). Achieving both high security and reliability in PUFs has been a persistent challenge, with many secure designs suffering from low reliability and vice-versa.

Methods

The core of this work is a novel PUF architecture that combines an Arbiter PUF (APUF) chain with a Linear Feedback Shift Register (LFSR) and a response feedback mechanism.

  • Response-Feedback-Based Strong PUF: The proposed PUF uses an LFSR with a primitive polynomial to generate a challenge for the APUF chain. Crucially, the APUF's response is fed back and XORed with the LFSR output to create a new challenge for the next stage. This feedback loop introduces significant nonlinearity and randomness, making it harder for ML models to learn the PUF's behavior. The LFSR's role is to obfuscate the original challenge, ensuring that even if the LFSR's state is known, the APUF's actual challenge remains hidden.
  • Two-Level Reliability Improvement (TLRI): To address the inherent unreliability of PUF responses (often around 95% or less), the paper proposes TLRI. This method builds upon existing techniques like delay difference testing and majority voting. It uses a Delay Difference Test Circuit (DDTC) to select more reliable responses and then employs a majority voting mechanism. The key innovation is that TLRI significantly reduces the number of voting cycles and overall response generation time required to achieve high reliability (e.g., 95%) compared to traditional majority voting.

Datasets

The paper relies on simulations for evaluation. A total of 1.5 million Challenge-Response Pairs (CRPs) were generated for training ML models, and 10,000 CRPs were used for testing. The simulations were performed using a Python platform. No new benchmark datasets were introduced; rather, the authors generated synthetic data based on their proposed PUF model.

Key Insights and Ideas

  • Feedback is Key for Nonlinearity: The response feedback mechanism is identified as a critical component for introducing the necessary nonlinearity and randomness to resist ML modeling attacks and improve statistical properties like GSAC.
  • LFSR as an Obfuscator: The LFSR doesn't just generate random bits; it actively obfuscates the original challenge, making it difficult for attackers to reverse-engineer the PUF's internal workings.
  • TLRI: Efficiency in Reliability: The TLRI method demonstrates that achieving high reliability doesn't necessarily require a massive increase in computational overhead. By intelligently filtering responses and using optimized voting, significant reductions in voting cycles and generation time are possible.
  • Balancing Act: The paper highlights the trade-off between security (nonlinearity, randomness) and reliability in PUFs and presents a design that effectively balances these two crucial aspects.

Key Results

  • ML Attack Resistance: The proposed PUF demonstrated strong resistance to ML attacks. Even with 1.5 million training CRPs, prediction accuracy for LR, SVM, and DNN models did not exceed 60%.
  • Improved Statistical Properties: The PUF exhibits good Strict Avalanche Criterion (SAC) and Generalized Strict Avalanche Criterion (GSAC), with response flip rates close to 0.5 when challenge bits are flipped.
  • Reliability Enhancement: The TLRI method achieved 95% reliability with less than 35% of the voting times and single-response generation cycles compared to traditional pure majority voting. For a delay of 12, the required MV for 95% reliability was reduced by up to 70% compared to the no-DDTC case.
  • Uniformity and Uniqueness: The simulated PUF instances showed excellent uniformity and uniqueness, close to the ideal values of 50%, indicating good intra-chip consistency and inter-chip distinctiveness.

Limitations and Future Work

  • Simulation-Based: The results are based on simulations, and real-world hardware implementation might reveal different performance characteristics due to fabrication variations and noise.
  • Specific LFSR and APUF: The security and reliability analyses are tied to the specific LFSR and APUF structures used. Further research could explore variations.
  • Future Work: The authors suggest exploring the impact of different LFSR feedback polynomials and APUF configurations. They also implicitly point towards hardware implementation and testing as a natural next step to validate the simulated results. The paper also mentions that the LFSR feedback coefficients are not obfuscated, which could be a potential avenue for future research to further enhance security against reverse attacks.

Summary updated

Abstract

Physical Unclonable Functions (PUFs) are significant in building lightweight Internet of Things (IoT) authentication protocols. However, PUFs are susceptible to attacks such as Machine-Learning(ML) modeling and statistical attacks. Researchers have conducted extensive research on the security of PUFs; however, existing PUFs do not always possess good statistical characteristics and few of them can achieve a balance between security and reliability. This article proposes a strong response-feedback PUF based on the Linear Feedback Shift Register (LFSR) and the Arbiter PUF (APUF). This structure not only resists existing ML modeling attacks but also exhibits good Strict Avalanche Criterion (SAC) and Generalized Strict Avalanche Criterion (GSAC). Additionally, we introduce a Two-Level Reliability Improvement (TLRI) method that achieves 95% reliability with less than 35% of the voting times and single-response generation cycles compared to the traditional pure majority voting method.
BibTeX
@inproceedings{zhu2023response,
  author = {B. Zhu and Xiaowen Jiang and Kai Huang and Miao Yu},
  title = {{A Response-Feedback-Based Strong PUF with Improved Strict Avalanche Criterion and Reliability}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s24010093},
}
Figure