During the past two decades, Physical Unclonable Functions (PUF) remained under discussion as strong hardware security primitives. As compared to conventional technology, memristor-based PUFs got tremendous attraction due to their simple and easy architecture, high endurance, low energy consumption, small size, and low cost. The inherent high variability of memristive devices makes it an ideal candidate for the generation of unique fingerprints for individual devices. By exploiting this variation, we propose a novel architecture dual memristive crossbar (DuMXbar) that integrates two memristive crossbar PUFs implemented with two different memristor devices. Each PUF generates highly uncorrelated response pairs. The simulation results of DuMXbar PUF prove that it is comparatively more robust against machine learning (ML) based classification algorithms like logistic regression and support vector machine. We have also proposed a secure authentication protocol for DuMXbar PUF to restrict the prediction accuracy of ML algorithms close down to 50% (ideal).

在过去的二十年中，物理不可克隆功能 (PUF) 作为强大的硬件安全原语仍在讨论中。与传统技术相比，基于忆阻器的PUF由于其结构简单易行、耐用性高、能耗低、体积小、成本低而受到极大的关注。忆阻器件固有的高可变性使其成为为单个器件生成独特指纹的理想选择。通过利用这种变化，我们提出了一种新颖的架构双忆阻交叉开关（DuMXbar），它集成了使用两个不同忆阻器器件实现的两个忆阻交叉开关 PUF。每个 PUF 生成高度不相关的响应对。DuMXbar PUF 的仿真结果证明它相对于基于机器学习 (ML) 的分类算法（如逻辑回归和支持向量机）具有更强的鲁棒性。我们还为 DuMXbar PUF 提出了一种安全认证协议，以将 ML 算法的预测精度限制在 50%（理想）。