中国科学院大学学报2021，Vol.38Issue(4)：494-502,9.DOI:10.7523/j.issn.2095-6134.2021.04.008
指令级功耗特征的硬件木马检测高效机器学习
Efficient machine learning methods for hardware Trojan detection using instruction-level power character
摘要
Abstract
Integrated circuits (IC) are vulnerable to hardware Trojans (HTs) due to the globalization of semiconductor design and outsourcing fabrication.Stealthy HTs which activate malicious aging operations are ususlly hide in normal behaviors.Therefore,it is a challenge to detect those HTs by general test and verification approaches.In this paper,we build an efficient machine learning (ML) framework to classify the genuine and Trojan-insert chips using instruction-level side-channel power characters.Different instructions and HTs are used as feature sets to construct the algorithm models.In order to evaluate the performance of the method,we implemented five HTs benchmarks of MC8051 micro-controller in Altera Stratix Ⅱ FPGA,and presented analysis on five formulated ML models in both supervised and unsupervised modes.The test results showed that the detection accuracy of supervised Naive Bayes is 95％ in average,which is the highest among the ML models.The supervised SVM consumed the shortest running time,with an average of 0.04 s.We also verified that one-class SVM can be a valuable method without golden reference,which has accuracy in the range from 17％ to 72％ even in Harsh learning condition.关键词
硬件木马/机器学习/旁路功耗/指令级/检测Key words
hardware Trojans/machine learning/side-channel power/instruction-level/detection分类
信息技术与安全科学引用本文复制引用
李莹,陈岚,佟鑫..指令级功耗特征的硬件木马检测高效机器学习[J].中国科学院大学学报,2021,38(4):494-502,9.基金项目
Supported by Beijing Natural Science Foundation (4184106),National Internet of Things and Smart City Key Project Docking(Z181100003518002),Beijing Science and Technology Project (Z171100001117147) （4184106）
