Despite considerable achievements of deep learning-based side-channel analysis, overfitting represents a significant obstacle in finding optimized neural network models. This issue is not unique to the side-channel domain. Regularization techniques are popular solutions to overfitting and have long been used in various domains. At the same time, the works in the side-channel domain show sporadic utilization of regularization techniques. What is more, no systematic study investigates these techniques’ effectiveness. In this paper, we aim to investigate the regularization effectiveness on a randomly selected model, by applying 4 powerful and easy-to-use regularization techniques to 8 combinations of datasets, leakage models, and deep learning topologies. The investigated techniques are L₁, L₂, dropout, and early stopping. Our results show that while all these techniques can improve performance in many cases, L₁ and L₂ are the most effective. Finally, if training time matters, early stopping is the best technique.

Side-channel attacks represent a realistic and serious threat to the security of embedded devices for already almost three decades. A variety of attacks and targets they can be applied to have been introduced, and while the area of side-channel attacks and their mitigation is very well-researched, it is yet to be consolidated. Deep learning-based side-channel attacks entered the field in recent years with the promise of more competitive performance and enlarged attackers' capabilities compared to other techniques. At the same time, the new attacks bring new challenges and complexities to the domain, making the systematization of knowledge (SoK) even more critical.We first dissect deep learning-based side-channel attacks according to the different phases they can be used in and map those phases to the efforts conducted so far in the domain. For each phase, we identify the weaknesses and challenges that triggered the known open problems. We also connect the attacks to the threat models and evaluate their advantages and drawbacks. Finally, we provide a number of recommendations to be followed in deep learning-based side-channel attacks.

This paper was selected for Top Picks in Hardware and Embedded Security 2020 and it presents a physical side-channel attack aiming at reverse engineering neural networks implemented on an edge device. The attack does not need access to training data and allows for neural network recovery by feeding known random inputs. We successfully reverse engineer information about layers, neurons, activation functions, and weights associated with neurons. This attack opens a new door in the domain of security of neural networks. Follow-up works by other researchers have shown this attack to be applicable for various settings and difficult to protect against.

Deep learning-based side-channel analysis performance heavily depends on the dataset size and the number of instances in each target class. Both small and imbalanced datasets might lead to unsuccessful side-channel attacks. The attack performance can be improved by generating traces synthetically from the obtained data instances instead of collecting them from the target device, but this is a cumbersome and challenging task. We propose a novel data augmentation approach based on conditional Generative Adversarial Networks (cGAN) and Siamese networks, enhancing the attack capability. We also present a quantitative comparative deep learning-based side-channel analysis between a real raw signal leakage dataset and an artificially augmented leakage dataset. The analysis is performed on the leakage datasets for both symmetric and public-key cryptographic implementations. We investigate non-convergent networks’ effect on the generation of fake leakage signals using two cGAN based deep learning models. The analysis shows that the proposed data augmentation model results in a well-converged network that generates realistic leakage traces, which can be used to mount deep learning-based side-channel analysis successfully even when the dataset available from the device is not optimal. Our results show that the datasets enhanced with “faked” leakage traces are breakable (while not without augmentation), which might change how we perform deep learning-based side-channel analysis.

At CRYPTO’19, A. Gohr proposed neural distinguishers for the lightweight block cipher Speck32/64, achieving better results than the state-of-the-art at that point. However, the motivation for using that particular architecture was not very clear; therefore, in this paper, we study the depth-10 and depth-1 neural distinguishers proposed by Gohr [7] with the aim of finding out whether smaller or better-performing distinguishers for Speck32/64 exist. We first evaluate whether we can find smaller neural networks that match the accuracy of the proposed distinguishers. We answer this question in the affirmative with the depth-1 distinguisher successfully pruned, resulting in a network that remained within one percentage point of the unpruned network’s performance. Having found a smaller network that achieves the same performance, we examine whether its performance can be improved as well. We also study whether processing the input before giving it to the pruned depth-1 network would improve its performance. To this end, convolutional autoencoders were found that managed to reconstruct the ciphertext pairs successfully, and their trained encoders were used as a preprocessor before training the pruned depth-1 network. We found that, even though the autoencoders achieved a nearly perfect reconstruction, the pruned network did not have the necessary complexity anymore to extract useful information from the preprocessed input, motivating us to look at the feature importance to get more insights. To achieve this, we used LIME, with results showing that a stronger explainer is needed to assess it correctly.

Profiling attacks, especially those based on machine learning proved as very successful techniques in recent years when considering side-channel analysis of block ciphers implementations. At the same time, the results for implementations of public-key cryptosystems are very sparse. In this paper, we consider several machine learning techniques in order to mount a power analysis attack on EdDSA using the curve Curve25519 as implemented in WolfSSL. The results show all considered techniques to be viable and powerful options. Especially convolutional neural networks (CNNs) are effective as we can break the implementation with only a single measurement in the attack phase while requiring less than 500 measurements in the training phase. Interestingly, that same convolutional neural network was recently shown to perform extremely well for attacking the implementation of the AES cipher. Our results show that some common grounds can be established when using deep learning for profiling attacks on distinct cryptographic algorithms and their corresponding implementations.