TransNet: Shift Invariant Transformer Network for Side Channel Analysis

Hajra, Suvadeep; Saha, Soumen; Alam, Manaar; Mukhopadhyay, Debdeep

doi:10.1007/978-3-031-17433-9_16

Cited by 6 publications

(17 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…sensitive information) given a trace T. A discriminative model estimates a φ-parametric probability conditional distribution Pr[Y |T, φ] that is as similar as possible to the true unknown joint probability distribution Pr [Y |T]. This approach is beneficial for directly solving a classification problem without modeling unnecessary information and thus, mitigating the impact of some countermeasures such as the desynchronization effect [CDP17a,ZBHV19,Mag19,HSAM22].This reason leads the side-channel community to investigate the DL approaches to improve the profiled SCA [CDP17a, KPH + 19, ZBHV19, BPS + 20] relying on discriminative approach. While [WPP22] combines a DL dimensionality reduction method with template attacks as an alternative to the Principal Component Analysis [APSQ06], the Linear Discriminant Analysis [BGH + 15] or the Kernel Discriminant Analysis [CDP17b], all the end-to-end DLSCA models proposed in the state-of-the-art are based on the discriminative approach (e.g.…”

Section: Related Work and Limitations Of Discriminative Modelsmentioning

confidence: 99%

“…While [WPP22] combines a DL dimensionality reduction method with template attacks as an alternative to the Principal Component Analysis [APSQ06], the Linear Discriminant Analysis [BGH + 15] or the Kernel Discriminant Analysis [CDP17b], all the end-to-end DLSCA models proposed in the state-of-the-art are based on the discriminative approach (e.g. fully-connected neural networks [MZ13, MHM14, Wei20], ResNets [ZS19, JZHY20, GJS20, MS21], RNNs [LLY + 20], transformer neural network [HSAM22], attention mechanisms [LZC + 21]). However, due to the lack of theoretical results, the discriminative models can be seen as black-box tools, and the design of models can be a real challenge even against unprotected cryptographic implementations.…”

Section: Related Work and Limitations Of Discriminative Modelsmentioning

confidence: 99%

“…Indeed, even if a wide-range of architectures have been studied in DLSCA context (e.g. fully-connected neural networks [PCP20, PHJ + 18], ResNets [JZHY20], transformer neural network [HSAM22], Support Vector Machine (SVM) [PHJ + 18], Random Forest [PHJ + 18]), there are no practical rules to construct them due to a lack of understanding between DL and SCA paradigms. Consequently, the discriminative models force the evaluator to consider the DLSCA approach as black-box tools which is problematic from an evaluation perspective.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Conditional Variational AutoEncoder based on Stochastic Attacks

Zaid

Bossuet²,

Carbone³

et al. 2023

TCHES

View full text Add to dashboard Cite

Over the recent years, the cryptanalysis community leveraged the potential of research on Deep Learning to enhance attacks. In particular, several studies have recently highlighted the benefits of Deep Learning based Side-Channel Attacks (DLSCA) to target real-world cryptographic implementations. While this new research area on applied cryptography provides impressive result to recover a secret key even when countermeasures are implemented (e.g. desynchronization, masking schemes), the lack of theoretical results make the construction of appropriate and powerful models a notoriously hard problem. This can be problematic during an evaluation process where a security bound is required. In this work, we propose the first solution that bridges DL and SCA in order to get this security bound. Based on theoretical results, we develop the first Machine Learning generative model, called Conditional Variational AutoEncoder based on Stochastic Attacks (cVAE-SA), designed from the well-known Stochastic Attacks, that have been introduced by Schindler et al. in 2005. This model reduces the black-box property of DL and eases the architecture design for every real-world crypto-system as we define theoretical complexity bounds which only depend on the dimension of the (reduced) trace and the targeting variable over F2n . We validate our theoretical proposition through simulations and public datasets on a wide range of use cases, including multi-task learning, curse of dimensionality and masking scheme.

show abstract

Section: Related Work and Limitations Of Discriminative Modelsmentioning

confidence: 99%

Section: Related Work and Limitations Of Discriminative Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Conditional Variational AutoEncoder based on Stochastic Attacks

Zaid

Bossuet²,

Carbone³

et al. 2023

TCHES

View full text Add to dashboard Cite

show abstract

“…Commonly used network architectures including Multilayer Perceptron (MLP) [4], [6], [7], [28]- [30], Convolutional Neural Network (CNN) [6], [7], [10], [11], [20], [31]- [33], and the recently employed Transformer [34]. CNNs used in DL-PSCA typically consist of two components: a feature extractor comprising convolutional layers, and a classifier consisting of several fully-connected layers that integrate the extracted features and perform classification predictions.…”

Section: B Deep Learning-based Profiled Side-channel Analysismentioning

confidence: 99%

ACNN: Arbitrary Traces Attack based on LeakageArea Detection

Xiao

Tang

2023

Preprint

View full text Add to dashboard Cite

Deep Learning-based Profiled Side-Channel Analysis (DL-PSCA) has become an important research method in the field of Side-Channel Analysis (SCA). The current research of DL-PSCA is based on publicly available datasets. Both the training set and attack set are derived from the same batch of collected data and are well-organized, resulting in the disregard of differences arising from the condition of trace measurement that exists in actual attack scenarios. Specifically, the timing of measurement is inconsistent due to the fact that attackers have different levels of control in terms of profiling and attacking devices. To bridge the gap, we assert that it is necessary for a network to explicitly identify areas of leakage and subsequently predict the leaked values. By decomposing conventional DL-PSCA into tasks of finer granularity, we have been able to restructure DL-PSCA as a regression task that operates in asliding window manner. Therefore, we present Arbitrary Convolutional Neural Network (ACNN) that is composed of a feature extractor and a regressor, which scans the input trace of arbitrary length for leakage localization and leakage value prediction in a sliding window way. We use the public datasets DPAv4.2 and ASCAD respectively to simulate variable traces and verify the effectiveness of our approach on unmasked and masked devices. The experiments show that no matter if the implementation is protected with masking scheme, the network can accurately identify the leakage areas and predict the leaked values. In terms of key recovery performance, our architecture is on par with state-of-the-art.

show abstract

“…In 2017, the attention [8] mechanism was used in machine translation tasks and achieved good results, becoming a new generation of deep learning architecture. In 2022, TransNet [9] introduced the long-term attention mechanism [10] to side-channel attacks. TransNet can reduce the guessing entropy to below 1 using only 400 records in multiple desynchronized datasets and has achieved significant results on several datasets.…”

Section: Introductionmentioning

confidence: 99%

Side-Channel Attacks Against the FESH Algorithm

Fan,

2024

FCIS

View full text Add to dashboard Cite

The FESH algorithm is a block cipher algorithm based on finite field operations. Currently, no research has been conducted on its side-channel attack security. Therefore, this study proposes two methods to address this issue: a correlation power analysis attack method targeting the FESH algorithm, and a template attack method based on an improved TransNet model. The first method theoretically analyzed the vulnerabilities of the FESH algorithm and successfully obtained valid leaked information through a correlation power attack; The second method introduced BlurPool blurring and downsampling techniques, as well as normalization operations, which reduced the training parameters of the improved model by approximately 50%. Additionally, the validation was performed on both the FESH dataset and the desynchronized ASCAD public dataset, which provided evidence that the entropy estimates were significantly better than those of the original TransNet model. The experimental results highlight the importance of considering side-channel security when implementing the FESH algorithm.

show abstract

TransNet: Shift Invariant Transformer Network for Side Channel Analysis

Cited by 6 publications

References 25 publications

Conditional Variational AutoEncoder based on Stochastic Attacks

Conditional Variational AutoEncoder based on Stochastic Attacks

ACNN: Arbitrary Traces Attack based on LeakageArea Detection

Side-Channel Attacks Against the FESH Algorithm

Contact Info

Product

Resources

About