密码学报（中英文）2025，Vol.12Issue(4)：729-751,23.DOI:10.13868/j.cnki.jcr.000791
嵌入式AI硬件单元的侧信道分析方法概述
Systematic Study on Physical Side-Channel Attack Against Embedded AI Hardware Units
摘要
Abstract
With the development of artificial intelligence(AI)technology,the trained machine learning models are developed on various embedded hardware platforms,such as microcontroller,filed programmable gate array(FPGA),and graphics processing unit(GPU).These platforms are applied to speech recognition,image processing,traffic monitoring,text processing and other industries.Due to broad applications,the embedded AI hardware unit has attracted wide attention across academic and industry.In recent years,side-channel analysis(SCA)has been applied as a new method in the scenario of model extraction,adversarial example attacks,model inversion,and other attack scenarios.The physical security analysis and evaluation of embedded AI hardware units have become the hot topics in academia and industry.This study systematically investigates public works on physical side-channel attacks against embedded AI hardware units.The impacts of physical side-channel analysis on the physical security of embedded AI hardware units are profoundly analyzed as well.This study also summarizes the key problems to be solved in this field,and prospects the physical side-channel analysis of embedded AI hardware units.关键词
侧信道分析/机器学习/模型提取/逆向分析/对抗样本Key words
side-channel analysis/machine learning/model extraction/reverse engineer/adversarial examples分类
信息技术与安全科学引用本文复制引用
金诚斌,高宜文,高锐,尤博,李丹..嵌入式AI硬件单元的侧信道分析方法概述[J].密码学报（中英文）,2025,12(4):729-751,23.基金项目
国家重点研发计划(2024YFF0619202) （2024YFF0619202）
国家自然科学基金(U2336205,62202231) （U2336205,62202231）
云南省新型研发机构培育对象项目(202404BQ040148) （202404BQ040148）
广州市重点研发计划(2023B01J0002)National Key Research and Development Program of China(2024YFF0619202) （2023B01J0002）
National Natural Science Foundation of China(U2336205,62202231) （U2336205,62202231）
Yunnan Provincial New R&D Institution Cultivation Project(202404BQ040148) （202404BQ040148）
Guangzhou Key Research and Development Program(2023B01J0002) （2023B01J0002）
