This paper contributes to the study of PUFs vulnerability against modeling attacks by evaluating the security of XOR BR PUFs, XOR TBR PUFs, and obfuscated architectures of XOR BR PUF using a simplified mathematical model and deep learning (DL) techniques. Obtained results show that DL modeling attacks could easily break the security of 4-input XOR BR PUFs and 4-input XOR TBR PUFs with modeling accuracy $\sim$ 99%. Similar attacks were executed using single-layer neural networks (NN) and support vector machines (SVM) with polynomial kernel and the obtained results showed that single NNs failed to break the PUF security. Furthermore, SVM results confirmed the same modeling accuracy reported in previous research ($\sim$ 50%). For the first time, this research empirically shows that DL networks can be used as powerful modeling techniques against these complex PUF architectures for which previous conventional machine learning techniques had failed. Furthermore, a detailed scalability analysis is conducted on the DL networks with respect to PUFs' stage size and complexity. The analysis shows that the number of layers and hidden neurons inside every layer has a linear relationship with PUFs' stage size, which agrees with the theoretical findings in deep learning. Consequently, A new obfuscated architecture is introduced as a first step to counter DL modeling attacks and it showed significant resistance against such attacks (16% - 40% less accuracy). This research provides an important step towards prioritizing the efforts to introduce new PUF architectures that are more secure and invulnerable to modeling attacks. Moreover, it triggers future discussions on the removal of influential bits and the level of obfuscation needed to confirm that a specific PUF architecture is resistant against powerful DL modeling attacks.

中文翻译：

---

深入研究：使用带有简化数学模型的深度学习技术来对抗 XOR BR 和 TBR PUF（攻击和对策）

本文通过使用简化的数学模型和深度学习 (DL) 技术评估 XOR BR PUF、XOR TBR PUF 和 XOR BR PUF 的混淆架构的安全性，为研究 PUF 抵御建模攻击的漏洞做出了贡献。获得的结果表明，DL 建模攻击可以轻松破坏 4 输入 XOR BR PUF 和 4 输入 XOR TBR PUF 的安全性，建模精度为 $\sim$ 99%。使用具有多项式内核的单层神经网络 (NN) 和支持向量机 (SVM) 执行了类似的攻击，获得的结果表明单个 NN 未能破坏 PUF 安全性。此外，SVM 结果证实了先前研究中报告的相同建模精度（$\sim$ 50%）。首次，这项研究凭经验表明，DL 网络可以用作强大的建模技术，以应对这些复杂的 PUF 架构，而之前的传统机器学习技术在这些架构上已经失败。此外，针对 PUF 的阶段大小和复杂性，对 DL 网络进行了详细的可扩展性分析。分析表明，层数和每层内的隐藏神经元与 PUF 的阶段大小呈线性关系，这与深度学习中的理论发现一致。因此，引入了一种新的混淆架构作为对抗 DL 建模攻击的第一步，它对此类攻击表现出了显着的抵抗力（准确率降低了 16% - 40%）。这项研究为优先考虑引入更安全且不易受到建模攻击的新 PUF 架构的努力迈出了重要一步。此外，它引发了未来关于移除有影响的位以及确认特定 PUF 架构能够抵抗强大的 DL 建模攻击所需的混淆级别的讨论。