-
摘要:
基于深度学习的侧信道攻击需要针对密码算法的每一个密钥字节建模并训练，数据采集和模型训练开销大. 针对该问题，提出一种基于多源数据聚合的神经网络侧信道攻击方法. 为筛选具有良好泛化效果的密钥字节泄露数据进行数据聚合，以AES-128算法为例，先基于16个密钥字节的泄露数据训练16个单密钥字节模型，分别实现对16个密钥字节的恢复；其次，设计一种打分机制评估各单密钥字节模型的泛化效果，通过得分排序筛选出对各密钥字节恢复效果最好的单密钥字节模型；最后，以筛选模型所对应的各密钥字节泄露数据集构建多源数据聚合模型进行训练，实现密钥恢复. 实验测试结果表明，多源数据聚合模型具有良好的泛化效果，有效提高了密钥恢复的准确率和效率，降低了恢复密钥所需的能量迹数量，其在采集能量迹较少的情况下依然具有较好的攻击效果.
Abstract:Side channel attack based on deep learning needs to model and train each key byte of the cryptographic algorithm, which costs a lot of data acquisition and model training. To solve this problem, a side channel attack method based on multi-source data aggregation neural network is proposed. In order to screen the leaked data of key byte with good generalization quality for data aggregation, taking AES-128 algorithm as an example, firstly 16 single key byte models are trained based on the leaked data of 16 key bytes, and models are used to recover 16 key bytes respectively. Secondly, a scoring mechanism is designed to evaluate the generalization effect of each single key byte model, and models with the best recovery effect for each key byte are selected according to score sorting. Finally, a multi-source data aggregation model is constructed based on the key byte leaked data sets corresponding to the selected models to realize key recovery. The tested results show that the multi-source data aggregation model has good generalization effect, effectively improves the accuracy and efficiency of key recovery, reduces the number of traces used to recover the key, and the model also has good attack effect in the case of less traces.
-
Keywords:
- side channel attack /
- neural network /
- scoring mechanism /
- data aggregation /
- data labeling
-
终端网络是互联网的重要组成部分，它连接骨干网络和终端网络，对用户体验的影响最为直接. 随着5G/6G、物联网等技术的发展，终端网络的性能需求不断提升，承载着诸如智慧城市和工业互联网等新兴应用，是推动社会数字化转型的重要基础设施，是未来网络演进不可忽视的重要研究对象. 清华大学李振华教授团队通过分析终端网络中存在的用户困惑和技术鸿沟问题，从“可用性、可靠性、可信性”三个关键维度进行研究，提出云原生强化设计的理念，实现终端网络大规模的测量分析与设计优化，并在多个工业系统中取得了良好的应用效果. 文章突出从用户视角出发的设计思想，对提升网络终端的可用性、可靠性与安全性做出了系统性的探索，主要包括以下三个核心点：
1）针对终端网络带给用户的主要困惑，从网速、断连、安全和代际角度全面分析，阐述克服经典设计模式潜在缺陷的研究动力，通过剖析大规模工业终端网络在多样化使用场景下的性能落差问题，总结动机、场景、资源和知识方面的研发鸿沟，为克服现存技术挑战指明解决方向.
2）围绕云原生强化设计的创新模式，综合考量技术和非技术多方面因素，利用服务器无感知基础设施、以微服务形态测量分析大规模终端网络，并针对复杂场景下的异构性能缺陷，跨层跨代协同强化，自适应改进终端网络设计. 最终实现终端网络的整体完善和全面进化，让终端网络服务更加高效、安全和可靠. 这些方法对现实中的网络运营与演进具有重要借鉴意义.
3）实践效果上，该研究团队将理论设计与工业应用相结合，在不同规模和需求的多个工业系统（包括政府运营的专网、大型企业的商业系统以及创业公司的网络应用）中做了调研分析、部署实施和落地改造，有效并高效地解决了其关键问题，提升了服务质量，示范性地推动了大规模复杂终端网络的技术革新.
总体而言，该研究工作系统而全面地分析了终端网络面临的问题，并在理论和实践上进行了有益的探索，形成了一套改善网络性能的方法体系. 这对推动基于云原生的网络技术发展具有较大的参考价值. 后续工作可以在技术普适性和用户感知等方面进行拓展，以建立一个更智能、自主的网络系统，这将对万物互联时代数字社会的进步具有重要意义.
评述专家
亮点论文
李振华, 王泓懿, 李洋, 林灏, 杨昕磊. 大规模复杂终端网络的云原生强化设计[J]. 计算机研究与发展，2024，61（1）:2−19. DOI: 10.7544/issn1000-1239.202330726
-
![]()
图 2 不同模型恢复各密钥字节的能量迹数
Figure 2. Number of traces for each key byte recovered by different models
表 1 16个模型恢复对应密钥字节的最少能量迹数
Table 1 Minimum Number of Traces for Sixteen Models Recovering the Corresponding Key Byte
模型 能量迹数 模型 能量迹数 模型 能量迹数 模型 能量迹数 M1 50 M5 31 M9 18 M13 28 M2 40 M6 15 M10 50 M14 33 M3 49 M7 27 M11 68 M15 41 M4 30 M8 39 M12 35 M16 33 
下载: 导出CSV
表 2 16 个模型恢复16 个密钥字节的能量迹数
Table 2 Number of Traces for Sixteen Models Recovering Sixteen Key Bytes
模型 密钥字节 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 M1 50 138 78 95 138 141 68 137 81 420 71 384 121 287 177 172 M2 113 40 137 83 160 71 67 170 256 192 171 329 466 208 352 155 M3 96 182 49 300 332 183 87 442 275 299 71 870 738 482 149 374 M4 113 97 245 30 61 68 71 109 125 115 183 142 58 287 181 80 M5 57 129 409 49 21 165 51 191 63 518 150 111 70 415 355 49 M6 125 24 89 70 182 15 39 284 402 191 148 1622 154 164 97 367 M7 66 72 63 128 157 87 27 332 146 359 39 675 134 314 104 328 M8 112 108 165 100 127 125 148 39 40 45 66 74 42 68 117 125 M9 81 306 174 173 93 192 108 49 18 68 66 103 27 122 120 195 M10 308 125 274 168 219 102 211 76 48 50 107 92 107 47 249 161 M11 194 273 183 215 389 151 148 130 103 157 68 404 145 152 108 711 M12 550 274 540 90 135 575 333 79 39 49 140 35 54 74 265 181 M13 147 178 501 177 105 146 131 41 46 72 56 92 28 58 70 150 M14 211 183 508 147 205 116 197 116 72 46 89 175 63 33 184 162 M15 86 123 56 168 188 123 49 86 150 120 42 356 66 274 41 417 M16 99 145 329 54 52 191 118 103 82 142 252 101 92 421 378 33
下载: 导出CSV
表 3 各个模型的得分
Table 3 Scores of Each Model
模型 得分 模型 得分 模型 得分 模型 得分 M1 13.2 M5 17.7 M9 21.9 M13 22.5 M2 13.5 M6 15.9 M10 15.9 M14 12.9 M3 7.2 M7 18.6 M11 6.0 M15 19.2 M4 19.5 M8 28.5 M12 16.8 M16 14.7
下载: 导出CSV
表 4 由得分得到的各个模型的排序
Table 4 Ranking of Each Model According to Scores
排名 模型 排名 模型 排名 模型 排名 模型 1 M8 5 M15 9 M10 13 M1 2 M13 6 M7 10 M6 14 M14 3 M9 7 M5 11 M16 15 M3 4 M4 8 M12 12 M2 16 M11
下载: 导出CSV
表 5 6 个模型对16个密钥字节的恢复结果
Table 5 Recovery Results of Six Models for 16 Key Bytes
模型 密钥字节 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 文献[13] 550 550 550 550 550 550 550 550 550 550 550 550 550 550 550 550 Mi 1156 1410 913 1032 1022 2000 700 1165 1987 1844 1401 468 567 563 1532 2000 Mix_8 41 55 116 36 66 73 38 172 81 106 62 129 88 277 91 50 MO_8 42 73 41 77 120 65 40 190 93 109 93 172 137 108 100 155 Mix_10 27 39 71 32 37 41 25 115 60 68 65 124 84 117 91 33 MO_10 42 37 33 56 67 34 53 160 84 110 72 116 204 110 125 136
下载: 导出CSV
表 6 7个模型对16个密钥字节的恢复结果
Table 6 Recovery Results of Seven Models for 16 Key Bytes
模型 密钥字节数据 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 文献[9] 4 Mix_8 32 16 45 4 4 40 5 4 5 16 9 5 5 37 7 9 MO_8 4 5 5 4 5 6 4 6 9 23 15 29 17 47 34 12 Mix_10 30 6 35 4 5 4 6 4 4 7 10 6 5 28 8 8 MO_10 4 4 5 5 5 4 5 7 6 9 15 18 11 28 34 11 Mix_12 13 6 21 4 5 4 5 4 4 6 9 5 6 20 6 6 MO_12 3 4 5 4 5 4 5 6 7 7 10 6 7 22 25 11
下载: 导出CSV
表 7 5个模型对16个密钥字节的恢复结果
Table 7 Recovery Results of Five Models for 16 Key Bytes
模型 密钥字节 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 Mi 95 44 40 Mix_8 5 7 6 4 5 4 5 5 4 6 6 5 6 8 6 15 MO_8 4 6 4 4 4 5 6 7 7 6 8 7 8 8 8 20 Mix_10 5 6 7 5 4 4 4 5 4 4 8 6 7 6 8 12 MO_10 4 5 4 4 4 4 7 7 7 8 13 9 7 11 9 16
下载: 导出CSV
表 8 针对Present算法16 个模型恢复16 个密钥字节的能量迹数
Table 8 Number of Traces for Sixteen Models Recovering Sixteen Key Nibbles for Present Algorithms
模型 密钥字节 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 M1 4 169 357 M2 6 317 394 319 181 55 73 82 83 167 M3 55 25 217 M4 5 14 78 489 130 M5 19 438 379 M6 337 15 11 159 79 184 341 240 421 M7 270 236 4 75 257 97 352 M8 169 8 473 108 478 230 M9 5 M10 72 114 54 442 198 9 35 139 65 463 M11 340 6 150 M12 35 143 86 398 31 5 51 51 186 M13 119 332 73 5 M14 19 145 165 166 45 402 6 189 163 M15 130 403 176 209 118 172 219 11 M16 480 477 220 390 229 235 5
下载: 导出CSV
表 9 16 个模型的得分
Table 9 Scores of Sixteen Models
模型 得分 模型 得分 模型 得分 模型 得分 M1 8.1 M5 7.8 M9 3 M13 10.5 M2 21 M6 15.6 M10 20.7 M14 19.8 M3 8.1 M7 11.7 M11 8.1 M15 13.5 M4 10.2 M8 9.6 M12 21.6 M16 9.3
下载: 导出CSV
表 10 不同模型对各密钥半字节的恢复结果
Table 10 Recovery Results of Different Models for Each Key Nibble
模型 密钥字节数据 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 Mi 4 6 25 5 19 11 4 8 5 9 6 5 5 6 11 5 Mix_8 272 3 171 11 3 2 8 2 54 3 2 2 3 129 Mix_10 250 2 397 4 3 3 3 2 104 1 4 2 5 254 Mix_12 2 2 13 2 474 1 2 4 3 38 2 2 6 2 3 Mix_16 4 3 5 5 4 3 2 4 4 5 4 3 4 3 3 6
下载: 导出CSV
-
[1] 王安,葛婧,商宁,等. 侧信道分析实用案例概述[J]. 密码学报,2018,5(4):383−398 doi: 10.13868/j.cnki.jcr.000249 Wang An, Ge Jing, Shang Ning, et al. Practical cases of side-channel analysis[J]. Journal of Cryptologic Research, 2018, 5(4): 383−398 (in Chinese) doi: 10.13868/j.cnki.jcr.000249
[2] Zhang Libang, Xing Xinpeng, Fan Junfeng, et al. Multi-label deep learning based side channel attack[C/OL] //Proc of the 2019 Asian Hardware Oriented Security and Trust Symp. Piscataway, NJ: IEEE, 2019[2022-01-10]. https://ieeexplore.ieee.org/document/9006657
[3] Ghandali S, Ghandali S, Tehranipoor S. Profiled power-analysis attacks by an efficient architectural extension of a CNN implementation[C] //Proc of the 22nd Int Symp on Quality Electronic Design. Piscataway, NJ: IEEE, 2021: 395−400
[4] Kim J, Picek S, Heuser A, et al. Make some noise: Unleashing the power of convolutional neural networks for profiled side-channel analysis[J]. IACR Transactions on Cryptographic Hardware and Embedded Systems, 2019, 2019(3): 148−179
[5] Maghrebi H, Portigliatti T, Prouff E. Breaking cryptographic implementations using deep learning techniques[C] //Proc of the 5th Int Conf on Security, Privacy, and Applied Cryptography Engineering. Berlin: Springer, 2016: 3−26
[6] Cagli E, Dumas C, Prouff E. Convolutional neural networks with data augmentation against jitter-based countermeasures[C] //Proc of the 19th Int Conf on Cryptographic Hardware and Embedded Systems. Berlin: Springer, 2017: 45−68
[7] Benadjila, R, Prouff, E, Strullu, R. et al. Deep learning for side-channel analysis and introduction to ASCAD database[J]. Journal of Cryptographic Engineering, 2020, 10(2): 163−188 doi: 10.1007/s13389-019-00220-8
[8] Wang Huanyu, Dubrova E. Federated learning in side-channel analysis [C] //Proc of the 16th Int Conf on Information Security and Cryptology. Berlin: Springer, 2020: 257−272
[9] Perin G, Chmielewski L, Picek S. Strength in numbers: Improving generalization with ensembles in machine learning-based profiled side-channel analysis[J]. IACR Transactions on Cryptographic Hardware and Embedded Systems, 2020, 2020(4): 337−364
[10] Won Y S, Han D G, Jap D, et al. Non-profiled side-channel attack based on deep learning using picture trace[J]. IEEE Access, 2021, 9: 22480−22492 doi: 10.1109/ACCESS.2021.3055833
[11] Zaid G, Bossuet L, François D, et al. Ranking loss: Maximizing the success rate in deep learning side-channel analysis[J]. IACR Transactions on Cryptographic Hardware and Embedded Systems, 2020, 2021(1): 25−55
[12] 王恺,严迎建,郭朋飞,等. 基于改进残差网络和数据增强技术的能量分析攻击研究[J]. 密码学报,2020,7(4):551−564 doi: 10.13868/j.cnki.jcr.000389 Wang Kai, Yan Yingjian, Guo Pengfei, et al. Research on power analysis attack based on improved residual network and data augmentation technology[J]. Journal of Cryptologic Research, 2020, 7(4): 551−564 (in Chinese) doi: 10.13868/j.cnki.jcr.000389
[13] Wang Ping, Chen Ping, Luo Zhimin, et al. Enhancing the performance of practical profiling side-channel attacks using conditional generative adversarial networks[EB/OL]. 2020[2022-01-10]. https://eprint.iacr.org/2020/867
[14] Luo Zhimin, Zheng Mengce, Wang Ping, et al. Towards strengthening deep learning-based side channel attacks with mixup[EB/OL]. 2021[2022-01-10]. https://eprint.iacr.org/2021/312
[15] Abdellatif K M. Mixup data augmentation for deep learning side-channel attacks[EB/OL]. 2021[2022-01-10]. https://eprint.iacr.org/2021/328
[16] Zhang Hongyi, Cisse M, Dauphin Y N, et al. Mixup: Beyond empirical risk minimization[C/OL] //Proc of the 6th Int Conf on Learning Representations. Amherst, MA: OpenReview. net, 2018 [2022-01-10]. https://openreview.net/pdf?id=r1Ddp1-Rb
[17] Nassar M, Souissi Y, Guilley S, et al. RSM: A small and fast countermeasure for AES, secure against 1st and 2nd-order zero-offset SCAs[C] //Proc of the 15th Design, Automation and Test in Europe Conf and Exhibition. Piscataway, NJ: IEEE, 2012: 1173−1178
[18] Gilmore R, Hanley N, O’Neill M. Neural network based attack on a masked implementation of AES[C] //Proc of the 2015 IEEE Int Symp on Hardware Oriented Security and Trust. Piscataway, NJ: IEEE, 2015: 106−111
-
期刊类型引用(1)
1. 王星宇. 浅析新时代背景下计算机科学技术发展的新方向. 数字通信世界. 2024(03): 164-166 . 
百度学术
其他类型引用(0)
计量
- 文章访问数: 184
- HTML全文浏览量: 79
- PDF下载量: 109
- 被引次数: 1
