Selected XOR PUF: 抵抗機器學習攻擊與擴展挑戰-響應對空間的設計策略

Abstract

物理不可克隆函數（PUF）是一種技術，利用硬件製造過程中引入的微小隨機性來生成獨特且無法復制的響應特徵。PUF廣泛應用於硬件安全領域，如身份驗證和密鑰生成。然而，傳統的PUF結構，如XOR PUF，面臨來自機器學習攻擊的風險，這些攻擊可以通過分析挑戰-響應對（CRP）推斷出PUF的內部結構，從而破壞其安全性。 本研究旨在改進XOR PUF結構，提出一種新設計：Selected XOR PUF，以增強其抵抗機器學習攻擊的能力，同時提高CRP的使用效率。通過引入新的設計和算法，我們希望顯著提升PUF的安全性，使其在實際應用中更加可靠和高效。實驗結果表明，改進後的PUF在抵抗機器學習攻擊方面表現出更高的效果，並顯著提高了CRP的使用效率，為未來的PUF應用提供了更強的保障。 此外，我們還開發了另一種設計：Twice Selected XOR PUF，其在抵抗機器學習攻擊方面表現出更顯著的效果。我們還研究了不穩定的CRP空間對PUF的影響，並總結了結論與未來可能的延伸方向。

A Physical Unclonable Function (PUF) is a technology that leverages the tiny randomness introduced during the hardware manufacturing process to generate unique and unreplicable response characteristics. PUFs are widely used in the field of hardware security, such as authentication and key generation. However, traditional PUF structures, like XOR PUF, face risks from machine learning attacks, which can infer the internal structure of the PUF by analyzing Challenge-Response Pairs (CRPs), thereby compromising its security. This study aims to improve the XOR PUF structure by introducing a new design: Selected XOR PUF, to enhance its resistance to machine learning attacks while increasing the efficiency of CRP usage. By incorporating new designs and algorithms, we aim to significantly boost the security of PUFs, making them more reliable and efficient in practical applications. Experimental results show that the improved PUF exhibits higher effectiveness in resisting machine learning attacks and significantly enhances CRP usage efficiency, providing stronger assurance for future PUF applications. Furthermore, we have developed an additional design: Twice Selected XOR PUF, which demonstrates even more significant resistance to machine learning attacks. We also investigated the impact of unstable CRP Space on PUFs, concluding with potential future directions.