物理不可克隆函数的机器学习防御与攻击综述
A Survey of Machine Learning Defense and Attack of Physical Unclonable Function
寇瑜萍 1邓丁 1欧钢 1黄仰博 1牟卫华1
作者信息
- 1. 国防科技大学电子科学学院,湖南长沙 410073
- 折叠
摘要
随着卫星导航技术的发展,芯片、模块和板卡等导航产品被广泛应用到各种导航终端设备上,但这些设备在开放环境中的通信安全问题日益凸显.物理不可克隆函数(Physical Unclonable Function,PUF)是一种新型"硬件指纹"技术,基于PUF的身份认证方式可以对设备进行硬件层面的认证,满足其轻量级和高安全性的认证需求.针对多数PUF易受到机器学习(Machine Learning,ML)建模攻击的问题,对不同的结构改进方法进行介绍,分析了几种常用ML攻击算法的特点,提出了防御和攻击两方面的性能评价方法,从安全性方面讨论了 PUF的未来发展趋势.
Abstract
With the development of satellite navigation technology,navigation products such as chips,modules,and boards are widely used in various navigation terminal devices.However,the communication security issues of these devices in open environments are increasingly prominent.Physical Unclonable Function(PUF)is a new type of"hardware fingerprint"technology.PUF based identity authentication can provide hardware-level authentication for devices,meeting their lightweight and high-security authentication requirements.To solve the problem of the vulnerability of most PUF structures to Machine Learning(ML)modeling attacks,different structure improvement methods are introduced,the characteristics of several commonly used ML attack algorithms are analyzed,the performance evaluation methods for both defense and attack are proposed,and the future development trend from the aspect of security is discussed.
关键词
物理不可克隆函数/导航设备/抗攻击结构/机器学习/可靠性Key words
PUF/navigation equipment/anti-attack structure/ML/reliability引用本文复制引用
基金项目
湖南省自然科学基金(2022JJ30669)
出版年
2024
国家科技期刊平台
NSTL
万方数据