Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (9): 43-48.DOI: 10.3778/j.issn.1002-8331.1807-0007

Previous Articles     Next Articles

Improved Incremental SVDD Learning Algorithm for Hardware Trojan Detection

LI Xiongwei1, WEI Yanhai1, WANG Xiaohan1, XU Lu2, SUN Ping3   

  1. 1.Shijiazhuang Campus, The Army Engineering University of PLA, Shijiazhuang 050003, China
    2.Unit 31432 of Strategic Support, China
    3.Unit 61785 of PLA, China
  • Online:2019-05-01 Published:2019-04-28


李雄伟1,魏延海1,王晓晗1,徐  璐2,孙  萍3   

  1. 1.陆军工程大学 石家庄校区,石家庄 050003
    3.中国人民解放军 61785部队

Abstract: Hardware Trojan detection method based on power side-channel analysis and Support Vector Data Description(SVDD) algorithm, it is necessary to incrementally learn new signal samples to optimize the detection model. For the underfitting problem caused by the unconstrained learning range of the new sample of Incremental SVDD learning(ISVDD), an SVDD incremental learning algorithm for hardware Trojan detection is proposed. The algorithm uses the variance, mean and median relationship between the new sample and the original sample to construct the adaptive parameter, selects more effective new model training samples to improve model detection accuracy while reducing learning time. A multi-chip FPGA side-channel signals acquisition platform is used to collect the signals of three chips with different process variations, and the same-sized hardware Trojans implemented in each chip are detected. Experimental results show that the proposed algorithm has higher detection accuracy than ISVDD, which verifies its effectiveness.

Key words: hardware Trojan, side-channel analysis, Support Vector Data Description(SVDD), incremental learning

摘要: 在基于功耗旁路分析与支持向量数据描述(Support Vector Data Description,SVDD)算法的硬件木马检测方法中,为优化完善检测模型需要对新增信号样本进行增量学习,针对经典SVDD增量学习(Incremental SVDD learning,ISVDD)对新增样本学习范围无约束而导致的欠拟合问题,提出了一种适用于硬件木马检测的SVDD增量学习算法。该算法利用新增样本与原始样本之间方差、均值和中位数的关系构建自适应参数,选取更为有效的新模型训练样本,在减少学习时间的同时提高模型检测精度。采用多芯片FPGA旁路信号采集平台分别对3片受工艺扰动不同的芯片进行信号采集,并对各芯片中所植入的相同规模硬件木马进行检测,实验结果表明,该算法较ISVDD相比有更高的检测精度,验证了其有效性。

关键词: 硬件木马, 旁路分析, 支持向量数据描述, 增量学习