Abstract:Over the past few years, deep learning has been getting progressively more popular for the exploitation of side-channel vulnerabilities in embedded cryptographic applications, as it offers advantages in terms of the amount of attack traces required for effective key recovery. A number of effective attacks using neural networks have already been published, but reducing their cost in terms of the amount of computing resources and data required is an ever-present goal, which we pursue in this work. We focus on the ANSSI Side-Channel Attack Database (ASCAD), and produce a JAX-based framework for deep-learning-based SCA, with which we reproduce a selection of previous results and build upon them in an attempt to improve their performance. We also investigate the effectiveness of various Transformer-based models.

What problem does this paper attempt to address?

The paper attempts to address the problem of how to use deep learning techniques to improve attack efficiency and reduce the required computational resources and data volume in side-channel attacks on the AES encryption algorithm. Specifically: 1. **Research Background**: Side-channel attacks (SCA) are a type of network attack that exploits weaknesses in system implementations to recover information, such as by analyzing power consumption, electromagnetic radiation, etc. In recent years, deep learning has become increasingly popular in side-channel attacks because it can achieve effective key recovery with fewer attack traces. 2. **Objective**: The main objective of this paper is to study deep learning architectures that can effectively reduce resource consumption (such as computational resources and required data volume). The authors particularly focus on the application of Transformer models and their variants in side-channel attacks and attempt to improve existing models to enhance performance. 3. **Dataset**: The research is primarily based on the ASCAD database (ANSSI SCA Database), which is a standard dataset widely used for evaluating deep learning-based side-channel attack methods. ASCAD includes two versions: ASCADv1 and ASCADv2, targeting different microcontrollers and AES implementations. 4. **Methodology**: The authors reproduced previous research results and made improvements based on them. They explored several deep learning architectures, including Convolutional Neural Networks (CNN) and Transformer models. Specifically, the authors attempted to replace the existing encoder structure with Transformer blocks to achieve linear complexity and adapt to the ASCADv2 dataset. 5. **Challenges**: A major issue encountered during the research was the difficulty in model convergence. The authors found that even simple networks struggled to converge when processing the ASCADv1 variable key dataset, which could be due to insufficient network complexity or differences in optimizer implementation details. 6. **Contributions**: The paper proposes new training strategies, such as adaptive learning rate schedules and regularization techniques, to improve model performance. Additionally, it explores whether the categorical cross-entropy loss function is suitable for this type of task and considers loss functions specifically designed for side-channel attack scenarios. Overall, this paper aims to enhance the effectiveness of side-channel attacks by improving deep learning architectures, particularly achieving efficient information recovery under resource-constrained conditions.

Investigating Efficient Deep Learning Architectures For Side-Channel Attacks on AES

Deep Learning Side-Channel Collision Attack

Deep learning side-channel attack against hardware implementations of AES

Effective Deep Learning-based Side-Channel Analyses Against ASCAD.

Exploiting the Incomplete Diffusion Feature: A Specialized Analytical Side-Channel Attack Against the Aes and Its Application to Microcontroller Implementations

On the performance of non‐profiled side channel attacks based on deep learning techniques

Research on Side-Channel Analysis Based on Deep Learning with Different Sample Data

A Domain Generalization Methodology for Side-channel Attack

A Review and Comparison of AI Enhanced Side Channel Analysis

MDASCA: an enhanced algebraic side-channel attack for error tolerance and new leakage model exploitation

Design of Convolutional Neural Networks Architecture for Non-Profiled Side-Channel Attack Detection

Deep learning‐assisted and combined attack: a novel side‐channel attack

Practical aspects on non-profiled deep-learning side-channel attacks against AES software implementation with two types of masking countermeasures including RSM

Adversarial Attack Based Countermeasures against Deep Learning Side-Channel Attacks

A Transfer Learning Approach for Electromagnetic Side-channel Attack and Evaluation

DeepTheft: Stealing DNN Model Architectures through Power Side Channel

SAT based error tolerant algebraic side channel attacks

Physical Side-Channel Attacks on Embedded Neural Networks: A Survey

Overview of Side Channel Cipher Analysis Based on Deep Learning

Enhancing Deep-Learning Based Side-Channel Analysis Through Simultaneously Multi-Byte Training.

Side-channel Attack Using Word Embedding and Long Short Term Memories.