Statistical Model on CRAFT

作     者：WANG Caibing GUO Hao YE Dingfeng WANG Ping WANG Caibing;GUO Hao;YE Dingfeng;WANG Ping

作者机构：State Key Laboratory of Information Security Institute of Information EngineeringChinese Academy of Sciences School of Cyber Security University of Chinese Academy of Sciences Tianjin Aerospace Zhongwei Data System Technology Co. Ltd. 

出 版 物：《Chinese Journal of Electronics》 (电子学报(英文))

年 卷 期：2022年第31卷第4期

页      面：698-712页

核心收录：

学科分类：11[军事学] 1105[军事学-军队指挥学] 0839[工学-网络空间安全] 08[工学] 110505[军事学-密码学] 110503[军事学-军事通信学] 

基　　金：supported by the National Key R&D Program of China(2018YFA0704704) Natural Science Foundation of China (NSFC)(61772519) the Chinese Major Program of National Cryptography Development Foundation (MMJJ20180102) 

主　　题：Differential cryptanalysis Statistical models Success probability Data complexity Bimodal behavior CRAFT Differential fault analysis (DFA) attacks 

摘      要：Many cryptanalytic techniques for symmetric-key primitives rely on specific statistical analysis to extract some secrete key information from a large number of known or chosen plaintext-ciphertext pairs. For example, there is a standard statistical model for differential cryptanalysis that determines the success probability and complexity of the attack given some predefined configurations of the attack. In this work, we investigate the differential attack proposed by Guo et al. at Fast Software Encryption Conference 2020 and find that in this attack, the statistical behavior of the counters for key candidates deviate from standard scenarios, where both the correct key and the correct key xor specific difference are expected to receive the largest number of votes. Based on this bimodal behavior, we give three different statistical models for truncated differential distinguisher on CRAFT(a cryptographic algorithm name) for bimodal phenomena. Then, we provide the formulas about the success probability and data complexity for different models under the condition of a fixed threshold value. Also, we verify the validity of our models for bimodal phenomena by experiments on round-reduced of the versions distinguishers on CRAFT. We find that the success probability of theory and experiment are close when we fix the data complexity and threshold value. Finally, we compare the three models using the mathematical tool Matlab and conclude that Model 3 has better performance.