免于参考芯片的指纹自认证硬件木马检测方法研究

The hardware always controls electronic information systems with extremely high value. However, for a long time, the security of hardware is often ignored. With the emergence of hardware Trojans, hardware security and information system security are becoming increasingly vulnerable. Hardware Trojan can leak confidential information, or destroy the critical equipment. Due to a large number of devices rely on imports or fabricated overseas, the hardware security situation is particularly serious in our country. Most of the existing hardware Trojan detection works rely on golden chips. However, the golden chips are extremely difficult to obtain. This project explores golden chips-free hardware Trojan detection methods. We will construct hardware Trojan detection methods based on the automatic recognition or self-reference of chips’ generalized fingerprints. First, we explore from two aspects of fingerprints, logic structure fingerprints and side-channel parameters fingerprints. Furthermore, the generalized fingerprints of the chips are studied. The main contents are as follows: 1) the hardware Trojan detection method based on self-authentication of the logic structure fingerprints; 2) the hardware Trojan detection method based on self-authentication of the side-channel parameters fingerprints; 3) from the aspect of generalized characteristics, we explore hardware Trojan detection methods based on machine learning algorithms, in which, we will also study the adaptive optimization of chip’s various fingerprints to generate the most suitable fingerprints or indirect features for the hardware Trojan detection; 4) the comparison and the combination of the above works will be studied. This project will lay the theoretical foundation for the hardware security detection of information systems without golden chips.

硬件往往控制着具有极高价值的电子信息系统。然而长期以来，硬件的安全性却被忽视。随着硬件木马的出现，硬件安全和信息系统安全面临着严重的威胁。硬件木马能够泄露机密信息或者损毁关键设备。由于大量设备依赖进口或由国外流片生产，我国的硬件安全形势尤为严峻。已有的硬件木马检测工作大多需要参考芯片，而参考芯片难以获得。本课题旨在探索免于参考芯片的硬件木马检测方法，将构建基于芯片广义指纹自动识别或自参考的硬件木马检测方法。首先从逻辑结构指纹和旁道参数指纹两个角度进行探索，然后递进为广义指纹。内容包括：1）基于结构指纹自认证的硬件木马检测方法；2）基于旁道参数指纹自认证的硬件木马检测方法；3）从广义指纹层面，探索基于机器学习的硬件木马检测方法，并研究对芯片各类指纹进行自适应优化，生成最适于木马检测的指纹或间接特征；4）对上述工作的比较与结合研究。本课题将为无参考样本情形下信息系统的硬件安全检测奠定理论基础。

硬件木马的出现严重威胁着硬件的安全性和上层信息系统的安全性，这些安全威胁已经引起了工业界、军方和政府的高度重视。已有的硬件木马检测方法大多依赖于参考芯片，而参考芯片难以获得，甚至不存在。本课题探索免于参考芯片的硬件木马检测方法，面向硬件木马检测的四类实际检测场景，研究提出了成体系的硬件木马检测方法框架。研究内容包括：（1）提出自适应优化的二元分类型硬件木马检测方法。将硬件木马检测问题建模为分类问题，采用电路设计流程中的仿真信息对算法进行训练。经过训练的分类器能够自动识别不含木马的和含木马的电路。（2）提出基于协同训练的硬件木马检测方法。利用无标签IC和不精确的仿真模型构建精确的检测方法。其思想是两个算法能够在待测IC中识别出不同的模式，因而能够标记出一些IC给另一个算法进一步训练。（3）提出基于无监督聚类分析的免参考模型硬件木马检测方法。将硬件木马检测建模为两个模型：基于划分的检测模型和基于密度的检测模型。（4）提出基于聚类集成对抗不可信测试方的免参考模型硬件木马检测框架。本方法不需要参考模型，能够揭露不可信测试方对木马检测结果的恶意修改，并且检测精度较高。