Hardware Trojans (HTs) pose a serious threat to the security of Integrated Circuits (ICs). Detecting HTs in an IC is an important but difficult problem due to the wide spectrum of HTs and their stealthy nature. While researchers have been working on enhancing traditional IC tests and developing new methods to try to detect Trojans, there is still a possibility a Trojan will avoid detection during test time and be activated once the chip is in use. A runtime Trojan detection system could monitor an IC during its operational life time and provide a last-line of defense. However, most runtime approaches are infeasible due to the overhead introduced by additional hardware, or computational complexity, or both. In this paper, we propose a hardware-based runtime detection model that overcomes the aforementioned constraints. It applies chaos theory, which has been shown to be effective in several other domains, to characterize dynamic data in a reconstructed phase space, which helps us describe, analyze, and interpret power consumption data (whether chaotic or not). The proposed chaos based approach does not make any assumption on the statistical distribution of power consumption, this makes our model applicable for runtime use given the fact that power consumption is very dynamic as well as heavily application and data dependent. Hardware overhead, which is the main challenge for runtime approaches, is reduced by taking advantage of available thermal sensors present in most modern ICs. For real world implementation, thermal sensor noise cancelation is considered in our proposed model. Our simulation results for detecting Trojans on publicly available Trojan benchmarks demonstrate that the proposed model outperforms the current runtime Trojan detection approaches in terms of detection rate, computational complexity, and implementation feasibility. Approved for Public Release; Distribution Unlimited: 88ABW-2016-4308; Dated 31 AUG 2016.

中文翻译：

---

将混沌理论应用于运行时硬件木马监控和检测

硬件木马 (HT) 对集成电路 (IC) 的安全构成严重威胁。由于 HT 的广谱性及其隐蔽性，检测 IC 中的 HT 是一个重要但困难的问题。虽然研究人员一直致力于增强传统 IC 测试并开发新方法来尝试检测木马，但木马仍有可能在测试期间避免检测并在芯片使用后被激活。运行时特洛伊木马检测系统可以在其运行生命周期内监控 IC 并提供最后一道防线。然而，由于额外硬件或计算复杂性或两者兼有而引入的开销，大多数运行时方法是不可行的。在本文中，我们提出了一种克服上述限制的基于硬件的运行时检测模型。它应用混沌理论，这已被证明在其他几个领域中是有效的，可以在重建的相空间中表征动态数据，这有助于我们描述、分析和解释功耗数据（无论是否混乱）。所提出的基于混沌的方法不对功耗的统计分布做出任何假设，这使得我们的模型适用于运行时使用，因为功耗是非常动态的，并且严重依赖于应用程序和数据。硬件开销是运行时方法的主要挑战，它通过利用大多数现代 IC 中可用的热传感器来减少。对于现实世界的实现，我们提出的模型中考虑了热传感器噪声消除。我们在公开可用的木马基准测试上检测木马的模拟结果表明，所提出的模型在检测率、计算复杂度和实现可行性方面优于当前的运行时木马检测方法。获准公开发行；分发无限：88ABW-2016-4308；日期为 2016 年 8 月 31 日。