西安电子科技大学学报（自然科学版）2025，Vol.52Issue(6)：169-187,19.DOI:10.19665/j.issn1001-2400.20250907
针对轻量级分组密码的矩形攻击深度学习模型
Deep learning-based model for rectangle attack on lightweight block ciphers
摘要
Abstract
In recent years,deep learning based cryptanalysis has gradually become a research hotspot.This study focuses on the security analysis of lightweight cryptographic algorithms SPECK 32/64 and SIMON 32/64.The rectangular attack theory is cleverly embedded into a deep learning framework,and a deep learning cryptanalysis method that integrates the idea of rectangular attacks is proposed.This method introduces neural networks as a probability distribution modeling tool in rectangular attacks,achieving the following three results.First,we search for the optimal differential path between the SPECK 32/64 algorithm and the SIMON 32/64 algorithm in rectangular attacks.Using this path,we construct an 8-round effective discriminator for SPECK 32/64 algorithm with an accuracy of 0.53,and an 11 round rectangular neural network differential discriminator for the SIMON 32/64 algorithm with an accuracy of 0.54.Second,by using multiple ciphertext pairs as inputs to the neural network,more usable feature information is provided for the neural network model.Experimental data show that when the number of ciphertext pairs increases to 16,the 7-round discriminator accuracy of the SPECK 32/64 algorithm reaches 0.94,while the 10 round discriminator accuracy of the SIMON 32/64 algorithm is also as high as 0.84.Finally,we apply the trained rectangular neural network differential differentiator to key recovery attacks and attempt sub key recovery against the 11 round and 12 round SPECK 32/64 algorithms.In 100 independent attack experiments,the success rate of 11 rounds of attacks is 100％,and the success rate of 12 rounds of attacks is 85％.This study provides a new path for the practical application of deep learning in password security assessment.关键词
深度学习/矩形攻击/差分分析/神经网络/密钥恢复攻击Key words
deep learning/rectangle attack/differential cryptanalysis/neural network/key recovery attack分类
信息技术与安全科学引用本文复制引用
孙浩然,栗琳轲,陈杰,刘君..针对轻量级分组密码的矩形攻击深度学习模型[J].西安电子科技大学学报（自然科学版）,2025,52(6):169-187,19.基金项目
国家自然科学基金(62302285) （62302285）
陕西省重点研发计划(2023-YBGY-015) （2023-YBGY-015）
