Cryptology ePrint Archive: Report 2020/306

Leakage Assessment in Fault Attacks: A Deep Learning Perspective

Sayandeep Saha and Manaar Alam and Arnab Bag and Debdeep Mukhopadhyay and Pallab Dasgupta

Abstract: Generic vulnerability assessment of cipher implementations against fault attacks (FA) is a research area which is still largely unexplored. The security assessment for FA becomes especially interesting in the presence of countermeasures, as countermeasure structures are not very well-formalized so far, and on several occasions, they fail to fulfil their sole purpose of preventing FAs. In this paper, we propose a general, simulation-based, statistical yes/no test to assess information leakage in the context of FAs. The fascinating feature of the proposed test is that it is oblivious to the structure of the countermeasure/cipher under test, and detects fault-induced leakage solely by observing the ciphertext distributions. Unlike a recently proposed approach, which utilizes t-test and its higher-order variants for detecting leakage at different moments of ciphertext distributions, in this work we present a Deep Learning (DL) based leakage assessment method. Our DL-based method is not specific to moment-based leakages only and thus, can expose leakages in several cases where t-test based technique either fails or demands a prohibitively large number of ciphertexts. Experimental evaluation over a representative set of countermeasures establishes that the DL-based method mostly outperforms the t-test based leakage assessment in terms of the number of ciphertexts required. Further, we present a novel analysis technique to interpret the leakages from the DL models, which is highly desirable for a sound vulnerability assessment. In another vertical of this work, we enhance the leakage assessment test methodology for recently proposed Statistical-Ineffective-Fault-Analysis (SIFA) and establish the efficacy by verifying different countermeasures including a publicly available hardware implementation of a SIFA countermeasure. In the third vertical, we enhance the test for verifying FA-assisted leakages from so-called “non-cryptographic” parts of an implementation. As concrete proof of this, we validate a well-accepted automotive security module called Secure Hardware Extension (SHE) for which the test figured out non-trivial vulnerabilities.

Category / Keywords: implementation / Fault Attack and Block cipher and Information leakage and Deep Learning

Date: received 9 Mar 2020, last revised 13 Mar 2020

Contact author: sayandeep iitkgp at gmail com,alam manaar@gmail com,amiarnabbolchi@gmail com,dmcseiitkgp@gmail com

Available format(s): PDF | BibTeX Citation

Version: 20200313:084452 (All versions of this report)

Short URL: ia.cr/2020/306


[ Cryptology ePrint archive ]