The application of machine learning techniques in side-channel analysis has recently received increased attention. Finding the best hyperparameters to achieve optimal performance for machine learning models in side-channel analysis is still a challenging endeavor. In order to solve the problem, we present an efficient ensemble framework designed to support profiled side-channel analysis for attacking cryptographic devices with countermeasures. Our proposed framework can partially mitigate the impact of traditional countermeasures employed in cryptographic devices. Additionally, we introduce a novel voting method called elite voting, which leverages candidate keys with higher probabilities to recover the secret key and adjusts the voting weights for better candidate keys. Experimental results illustrate that our proposed framework can effectively recover the right key from cryptographic devices with countermeasures through multiple experiments. It enhances the signal-to-noise ratio of traces and successfully recovers the right key across various datasets. Furthermore, when compared to traditional methods, our elite voting method further enhances the performance of ensemble learning by reducing the number of traces needed to recover the secret key. It exhibits superior performance compared to other ensemble methods, as it can reduce the minimum required number of traces significantly.

Horizontal analysis is a widely adopted method in side-channel analysis, particularly for public-key cryptosystems, where attackers aim to recover the key from a single trace. Current methods rely on trace segmentation, dimensionality reduction, and classification, but high noise and poor feature preservation hinder accuracy. Noise blurs cryptographic operation segment boundaries, and existing dimensionality reduction techniques fail to maintain the inherent distribution of trace points in a high-dimensional space. As a result, secret information recovery based on clustering remains inaccurate. This paper proposes an automated horizontal analysis framework named UMAP-HC to improve secret information recovery accuracy. The framework employs a sliding segmentation method to locate cryptographic operations in noisy traces with blurred segment boundaries. It leverages uniform manifold approximation and projection (UMAP) for feature preservation and hierarchical clustering for secret information recovery. Experimental results on four open access public-key algorithm power trace datasets, an SM2 power trace collected from a smart card, and an ECC power trace with dummy operation countermeasures demonstrate that UMAP-HC effectively classifies cryptographic operations, accurately locates operation segments, and recovers secret key. It achieves up to 100% recovery accuracy, surpassing previous methods by 40%-70%, with normalized mutual information reaching 1, an improvement of 0.02-0.99 over existing approaches.

As a post-quantum digital signature scheme, Dilithium was specifically designed to withstand known quantum algorithm attacks, and its side-channel resistance has garnered significant research attention. However, current side-channel attacks against Dilithium exhibit several limitations: (1) failure to leverage low-correlation characteristics in power traces, (2) loss functions limited to categorical information extraction from power traces, (3) dependency on specific coefficient recovery conditions while neglecting inter-coefficient statistical dependencies, (4) requirement for separate profiling models per intermediate value, resulting in substantial information loss. To address these limitations, we propose an enhanced template attack framework integrating deep learning with classical template attack methodology. Our approach employs a dual-loss similarity learning mechanism for feature extraction from high-dimensional power traces, enabling the construction of more discriminative templates while preserving weakly correlated features. Through assembly-level analysis of the y polynomial generation routine, we reveal inherent correlations among coefficientsyk0,yk1,yk2,yk3. Building on this discovery, our dual-loss similarity learning framework is designed to capture these inter-coefficient relationships, preserving their intrinsic dependencies while achieving effective inter-class separation and intra-class aggregation properties, which significantly enhances the effectiveness of subsequent template attacks. Experimental results on Cortex-M4 power traces demonstrate our method achieves 32.94% polynomial coefficient recovery accuracy for polynomial coefficients y, outperforming conventional SOD-based (83% improvement), T-Test-based (97%), and PCA-based template attacks (197% enhancement). Furthermore, complete private key recovery is achieved with merely 14 power traces under specific conditions. This DL-enhanced template attack framework demonstrates superior side-channel leakage exploitation, yielding substantial performance enhancements over conventional approaches.

Simple Power Analysis is a commonly used method in Side-Channel Analysis on cryptosystems, which requires a significant amount of labor costs for segmentation. General Pulse Tailor for Simple Power Analysis (SPA-GPT) proposed in CHES 2024 utilizes reinforcement learning to achieve automated segmentation. However, its low efficiency and only targeting public-key algorithms limit its practical applications. In this paper, we propose a practical method, which utilize long short-term memory network and attention mechanism, coupled with a new deep Q-network policy using Simulated Annealing strategy, to solve the contradiction between reinforcement learning and high efficiency in trace segmentation. Moreover, the novel agent proposed in this paper also demonstrates transferability, enabling direct segmentation of a trace under varying lengths and signal-to-noise ratio conditions once the agent has been fully trained. In addition, our new approach is applicable for locating each execution of block ciphers in various encryption modes. Comparative experiments are conducted on 14 datasets, which are collected from software or hardware implementations of RSA, ECC, ML-KEM, AES, PRESENT, and SIMON, running on microcontrollers, FPGAs, or smart cards. Experimental results show that the new method enhances time efficiency by 50.34% to 94.24% while reducing network parameters by 87.84% compared to SPA-GPT.

Classical cluster-based side-channel analysis (SCA) uses clustering algorithms to analyze power traces and often, principal component analysis to reduce the dimension of data, resulting in that clustering may not deal well with high-dimensional traces, such as cryptographic algorithm implementations with countermeasures. In this article, we propose an intelligent framework for cluster-based SCA, which includes three steps of clustering, classification and correction, for processing large high-dimensional data. By combining unsupervised clustering and supervised deep learning techniques, the framework succeeds in mining the data for additional in-depth information. In addition, unlike traditional cluster-based SCA, our approach focuses on deep learning and deliberately avoids over-reliance on cluster labels during classification. And metrics for correction are adopted to achieve a high level of reliability in key recovery. Experiments on the RSA smart card based on Montgomery ladder implementation and FPGA-based ECC with random delay demonstrate that our framework can significantly improve the success rate with strong robustness.

Timing analysis presents a significant threat to cryptographic modules. However, traditional timing leakage analysis has notable limitations, especially when precise execution times cannot be obtained. In this article, we propose a novel timing leakage analysis method that leverages horizontal leakage in the power/electromagnetic channel by detecting the trace length of encryption processes under varying inputs. To demonstrate the effectiveness of our approach, we conducted systematic experimental evaluations across a range of cryptographic devices. In comparison to timing leakage analysis based on plaintext-ciphertext correlation, our method offers higher accuracy at lower testing costs and exhibits improved resistance to vertical noise.

Side-channel analysis (SCA) based on machine learning, particularly neural networks, has gained considerable attention in recent years. However, previous works predominantly focus on establishing connections between labels and related profiled traces. These approaches primarily capture label-related features and often overlook the connections between traces of the same label, resulting in the loss of some valuable information. Besides, the attack traces also contain valuable information that can be used in the training process to assist model learning. In this paper, we propose a profiled SCA approach based on contrastive learning named CL-SCA to address these issues. This approach extracts features by emphasizing the similarities among traces, thereby improving the effectiveness of key recovery while maintaining the advantages of the original SCA approach. Through experiments of different datasets from different platforms, we demonstrate that CL-SCA significantly outperforms other approaches. Moreover, by incorporating attack traces into the training process using our approach, we can further enhance its performance. This extension can improve the effectiveness of key recovery, which is fully verified through experiments on different datasets.

The module-lattice-based digital signature standard, formerly known as CRYSTALS-DILITHIUM, is a lattice-based post-quantum cryptographic scheme. In August 2024, the National Institute of Standards and Technology officially standardized ML-DSA under FIPS 204. ML-DSA generates one valid signature and multiple rejected signatures during a single signing process. Most side-channel attacks targeting ML-DSA have focused solely on the valid signature, while largely neglecting the hints contained in rejected signatures. Building on prior SASCA frameworks originally proposed for ML-DSA, in this paper we present an efficient and fully practical instantiation of a private-key recovery attack on ML-DSA that jointly exploits side-channel leakages from both valid and rejected signatures within a unified factor graph. This concrete instantiation maximizes the information extracted from a single signing attempt and minimizes the number of required traces for full key recovery. We conducted a proof-of-concept experiment with both reference and ASM-optimized implementations on a Cortex-M4 core chip, where the results demonstrate that incorporating rejected signatures reduces the required number of traces by at least 50.0% for full key recovery. Moreover, we show that using only rejected signatures suffices to recover the key with fewer than 30 traces under our setup. Our findings highlight that protecting rejected signatures is crucial, as their leakage provides valuable side-channel information. We strongly recommend implementing countermeasures for rejected signatures during the signing process to mitigate potential threats.

In side-channel testing, the standard timing analysis works when the vendor can provide a measurement to indicate the execution time of cryptographic algorithms. In this paper, we find that there exists timing leakage in power/electromagnetic channels, which is often ignored in traditional timing analysis. Hence a new method of timing analysis is proposed to deal with the case where execution time is not available. Different execution time leads to different execution intervals, affecting the locations of plaintext and ciphertext transmission. Our method detects timing leakage by studying changes in plaintext-ciphertext correlation when traces are aligned forward and backward. Experiments are then carried out on different cryptographic devices. Furthermore, we propose an improved timing analysis framework which gives appropriate methods for different scenarios.

The concepts between Internet of Things (IoT) and edge computing are increasingly intertwined, as an edge-computing architecture generally comprises a (large) number of diverse IoT devices. This, however, increases the potential attack vectors since any one of these connected IoT devices can be targeted to facilitate other malicious cyber activities. Physical attacks are generally harder to mitigate and less studied, in comparison to their cyber counterparts. Thus, in this article we present an attack framework targeting true random number generators (TRNGs), which are a key component in cryptosystems for edge devices. We then demonstrate how such a framework can guide our investigation of a commercial ASIC chip that runs ring-oscillator-based TRNG. Specifically, we show that our template power attack, low voltage fault attack, and voltage glitch fault attack do not require prior knowledge of the TRNG implementation.

Various Artificial Intelligence (AI) techniques are combined with classic side-channel methods to improve the efficiency of attacks. Among them, Genetic-Algorithms-based Correlation Power Analysis (GA-CPA) is proposed to launch attacks on hardware cryptosystems to extract the secret key efficiently. However, the convergence efficiency of GA-CPA is unsatisfactory due to two problems: the randomly generated initial population generally have low fitness, and the mutation operation in each iteration hardly produces high-quality individuals because of the confusion and diffusion characteristics of S-boxes. In this paper, we propose an analysis framework of GA-CPA which focuses on solving these two problems. First, we explore the list of candidate key bytes which is the result of Correlation Power Analysis (CPA) on a limited number of power traces, so that the population can be initialized with high quality candidates. Second, we improve the mutation operation by guiding the candidate key to mutate in a higher-fitness direction instead of randomly. Third, we make full use of the fitness calculation method and combine it with key enumeration algorithms to further improve the efficiency of key recovery. Simulation experimental results show that our method reduces the number of traces by 33.3% and 43.9% compared to CPA with key enumeration and GA-CPA respectively when the success rate is fixed to 90%. Real experiments performed on SAKURA-G confirm that the number of traces required in our method is much less than the numbers of traces required in CPA and GA-CPA. Besides, we adjust our method to deal with DPA contest v1 dataset, and achieve a better result of 40.76 traces than the winning proposal of 42.42 traces. The computation cost of our proposal is nearly 16.7% of the winner.

Side-channel analysis achieves key recovery by analyzing physical signals generated during the operation of cryptographic devices. Power consumption is one kind of these signals and can be regarded as a multimedia form. In recent years, many artificial intelligence technologies have been combined with classical side-channel analysis methods to improve the efficiency and accuracy. A simple genetic algorithm was employed in Correlation Power Analysis (CPA) when apply to cryptographic algorithms implemented in parallel. However, premature convergence caused failure in recovering the whole key, especially when plenty of large S-boxes were employed in the target primitive, such as in the case of AES. In this article, we investigate the reason of premature convergence and propose a Multiple Sieve Method (MS-CPA), which overcomes this problem and reduces the number of traces required in correlation power analysis. Our method can be adjusted to combine with key enumeration algorithms and further improves the efficiency. Simulation experimental results depict that our method reduces the required number of traces by and , compared to classic CPA and the Simple-Genetic-Algorithm-based CPA (SGA-CPA), respectively, when the success rate is fixed to . Real experiments performed on SAKURA-G confirm that the number of traces required for recovering the correct key in our method is almost equal to the minimum number that makes the correlation coefficients of correct keys stand out from the wrong ones and is much less than the numbers of traces required in CPA and SGA-CPA. When combining with key enumeration algorithms, our method has better performance. For the traces number being 200 (noise standard deviation ), the attacks success rate of our method is , which is much higher than the classic CPA with key enumeration ( success rate). Moreover, we adjust our method to work on that DPA contest v1 dataset and achieve a better result (40.04 traces) than the winning proposal (42.42 traces).

The integral attack, exploits the balanced property of the output in the distinguisher. Usually, adversaries append some rounds after the distinguisher, guess the corresponding key bits and check whether the target bits are balanced. Few works add rounds before the distinguisher to make the key recovery attack. In the first full‐round attack on MISTY1, Todo adds one FL layer (key‐dependent linear function) before the distinguisher. In this study, the authors extend his method and give a general method, which they can use to extend some rounds (non‐linear) before the distinguisher to attack more rounds with data complexity smaller than the whole space and little extra time consumption. The basic idea is that for different subkeys guessed in the forward rounds, they set different constant values for the input of the distinguisher. Finally, the selected data space is not full. For substitution permutation network (SPN) (Feistel with SPN round function) structures with 4 bit S‐box and bit permutation, they estimate the data complexity when adding one round before the distinguishers for all 4 bit S‐boxes. Using the method, they improve the integral attacks on PRESENT, RECTANGLE, TWINE and LBlock, and their results could cover one more round.