Cryptology ePrint Archive: Report 2020/1031

Profiled Deep Learning Side-Channel Attack on a Protected Arbiter PUF Combined with Bitstream Modification

Yang Yu and Michail Moraitis and Elena Dubrova

Abstract: In this paper we show that deep learning can be used to identify the shape of power traces corresponding to the responses of a protected arbiter PUF implemented in FPGAs. To achieve that, we combine power analysis with bitstream modification. We train a CNN classifier on two 28nm XC7 FPGAs implementing 128-stage arbiter PUFs and then classify the responses of PUFs from two other FPGAs. We demonstrate that it is possible to reduce the number of traces required for a successful attack to a single trace by modifying the bitstream to replicate PUF responses.

Category / Keywords: applications / Profiled attack, deep learning, side-channel analysis, bitstream modification and arbiter PUF

Date: received 26 Aug 2020

Contact author: yang11 at kth se

Available format(s): PDF | BibTeX Citation

Version: 20200827:031408 (All versions of this report)

Short URL: ia.cr/2020/1031


[ Cryptology ePrint archive ]