The differential deep learning analysis proposed by Timon is the first non-profiled side-channel attack technique that uses deep learning. The technique recovers the secret key using the phenomenon of deep learning metrics. However, the proposed technique made it difficult to observe the results from the intermediate process, while the neural network had to be retrained repeatedly, as the cost of learning increased with key size. In this paper, we propose three methods to solve the aforementioned problems and any challenges resulting from solving these problems. First, we propose a modified algorithm that allows the monitoring of the metrics in the intermediate process. Second, we propose a parallel neural network architecture and algorithm that works by training a single network, with no need to re-train the same model repeatedly. Attacks were performed faster with the proposed algorithm than with the previous algorithm. Finally, we propose a novel architecture that uses shared layers to solve memory problems in parallel architecture and also helps achieve better performance. We validated the performance of our methods by presenting the results of non-profiled attacks on the benchmark database ASCAD and for a custom dataset on power consumption collected from ChipWhisperer-Lite. On the ASCAD database, our shared layers method was up to 134 times more efficient than the previous method.

In recent years, deep learning-based side-channel attacks have established their position as mainstream. However, most deep learning techniques for cryptanalysis mainly focused on classifying side-channel information in a profiled scenario where attackers can obtain a label of training data. In this paper, we introduce a novel approach with deep learning for improving side-channel attacks, especially in a non-profiling scenario. We also propose a new principle of training that trains an autoencoder through the noise from real data using noise-reduced labels. It notably diminishes the noise in measurements by modifying the autoencoder framework to the signal preprocessing. We present convincing comparisons on our custom dataset, captured from ChipWhisperer-Lite board, that demonstrate our approach outperforms conventional preprocessing methods such as principal component analysis and linear discriminant analysis. Furthermore, we apply the proposed methodology to realign de-synchronized traces that applied hiding countermeasures, and we experimentally validate the performance of the proposal. Finally, we experimentally show that we can improve the performance of higher-order side-channel attacks by using the proposed technique with domain knowledge for masking countermeasures.

As side‐channel analysis and machine learning algorithms share the same objective of classifying data, numerous studies have been proposed for adapting machine learning to side‐channel analysis. Howe...

Recently power attacks on RSA cryptosystems have been widely investigated, and various countermeasures have been proposed. One of the most efficient and secure countermeasures is the message blinding method, which includes the RSA derivative of the binary-with-random-initial-point algorithm on elliptical curve cryptosystems. It is known to be secure against first-order differential power analysis (DPA); however, it is susceptible to second-order DPA. Although second-order DPA gives some solutions for defeating message blinding methods, this kind of attack still has the practical difficulty of how to find the points of interest, that is, the exact moments when intermediate values are being manipulated. In this paper, we propose a practical second-order correlation power analysis (SOCPA). Our attack can easily find points of interest in a power trace and find the private key with a small number of power traces. We also propose an efficient countermeasure which is secure against the proposed SOCPA as well as existing power attacks.