一种基于深度神经网络的多阶段PUF抗建模能力评估方法OA

To address the limitation of current evaluation methods in assessing the modeling resistance of Physical Unclonable Function(PUF),a three-level modeling threat model that PUFs encounter is de-fined.This model details the attacking target,the adversary's knowledge and capabilities,the attack-ing strategy,and the attacking mode.A deep neural network-based multi-stage evaluation method is proposed to assess the resistance of PUFs to machine learning modeling attacks,reliability-based at-tacks,and power/electromagnetic side-channel attacks.The method is carried out in three sequential phases,using feed-forward neural network modeling attacks and side-channel modeling attacks as the assessment tools,ensuring a comprehensive PUF modeling resistance assessment.The evaluation re-sults indicate that only a small number of the tested PUFs display resistance to machine learning and reliability modeling attacks.However,none of them demonstrate resistance to power side-channel mod-eling attacks.