Neural Network Based Glitch Physically Unclonable Function
Yusuke Nozaki, Masaya Yoshikawa · 2021
The security issues of artificial intelligence (AI) system are attracting attention. The neural network physically unclonable function (NN PUF) have been proposed to improve the security of AI devices. The NN PUF combines two functions: AI inference and device authentication. In device authentication, a unique ID is generated for each device by using the manufacturing variation of large-scale integration (LSI). The NN PUF uses the delay difference of NN computation time for ID generation. However, the conventional NN PUF is known to have low uniqueness in field programmable gate array (FPGA) implementation. Therefore, the present study proposes a new NN PUF that improves uniqueness. Evaluation experiments with FPGAs showed that the proposed NN PUF significantly improved uniqueness compared to the conventional NN PUF.