Template Attacks vs. Machine Learning Revisited (and the Curse of Dimensionality in Side-Channel Analysis)

Lerman, Liran; Poussier, Romain; Bontempi, Gianluca; Markowitch, Olivier; Standaert, François‐Xavier

doi:10.1007/978-3-319-21476-4_2

Cited by 86 publications

(53 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For these experiments, we can see that the result of machine learning approach is better compared to TA, especially with SVM and RF. This is more or less similar to the observations made in [110] about RF (requiring fewer number of training data with ability to cope with some irrelevant features),…”

Section: Methodssupporting

confidence: 83%

Physical attacks on block ciphers

Jap¹

View full text Add to dashboard Cite

The security of a cryptosystem is often compromised, not from a theoretical point of view, but by the leakage caused by the physical implementation of the cryptographic algorithm. A new class of attacks, called physical attacks, has shown the capability to exploit the unintentional physical behaviors from the cryptographic device, which usually provide enough information to recover the secret keys. Dierent methods have been proposed for conducting the attacks. Two of the main focus of physical attacks are side-channel attacks and fault attacks. For side-channel attacks, the strongest cryptanalysis can be carried when the attacker can prole the targeted device. In proling based side-channel attacks, a model is constructed to characterize the leakage behavior from the device. Recently, machine learning algorithms have been proposed as alternatives for the classical proling based attacks. Machine learning and side-channel analysis are two dierent elds of study, however they are similar, in a sense that both are mostly dealing with the same problem (i.e., classication). For fault attacks, the aim of the attacker is to disrupt the execution of cryptographic algorithms. Based on the erroneous results, it is possible to gain some additional information regarding the secret key. Many methods can be used to force a fault to the device, however, laser fault injection is still considered as the preferred tools for injecting faults, due to its high precision and repeatability.

show abstract

Section: Methodssupporting

confidence: 83%

Physical attacks on block ciphers

Jap¹

View full text Add to dashboard Cite

show abstract

“…When working with the typical assumption for profiled SCA that the profiling phase is not bounded, the situation actually becomes rather simple if neglecting computational costs. If the attacker is able to acquire an unlimited (or, in real-world very large) amount of traces, the template attack (TA) is proven to be optimal from an information theoretic point of view (see e.g., [1,2]). In that context of unbounded and unrestricted profiling phase, ML techniques seem not needed.…”

Section: Introductionmentioning

confidence: 99%

Improving Side-Channel Analysis Through Semi-supervised Learning

Picek

Heuser

Jović

et al. 2019

Smart Card Research and Advanced Applications

View full text Add to dashboard Cite

The profiled side-channel analysis represents the most powerful category of side-channel attacks. In this context, the security evaluator (i.e., attacker) gains access to a profiling device to build a precise model which is used to attack another device in the attacking phase. Mostly, it is assumed that the attacker has significant capabilities in the profiling phase, whereas the attacking phase is very restricted. We step away from this assumption and consider an attacker restricted in the profiling phase, while the attacking phase is less limited. We propose the concept of semi-supervised learning for side-channel analysis, where the attacker uses a small number of labeled measurements from the profiling phase as well as the unlabeled measurements from the attacking phase to build a more reliable model. Our results show that the semisupervised concept significantly helps the template attack (TA) and its pooled version (TAp). More specifically, for low noise scenario, the results for machine learning techniques and TA are often improved when only a small number of measurements is available in the profiling phase, while there is no significant difference in scenarios where the supervised set is large enough for reliable classification. For high noise scenario, TAp and multilayer perceptron results are improved for the majority of inspected dataset sizes, while for high noise scenario with added countermeasures, we show a small improvement for TAp, Naive Bayes and multilayer perceptron approaches for most inspected dataset sizes. Current results go in favor of using semi-supervised learning, especially self-training approach, in side-channel attacks.

show abstract

“…In this paper, we promote efficient approaches exploiting (i) a smaller running time and (ii) less knowledge on the target device compared to the Bayes classifier. Following recent papers in the side-channel community discussing machine learning models, we focus on (learning) models called random forests, template attacks and profiled attacks based on the kernel density estimation method [5], [19], [20], [22]. The main issue of profiled attacks lies in the exploitation of leakages containing a large number of dimensions.…”

Section: Introductionmentioning

confidence: 99%

Efficient Profiled Attacks on Masking Schemes

Lerman

Markowitch

2019

IEEE Trans.Inform.Forensic Secur.

Self Cite

View full text Add to dashboard Cite

Side-channel adversaries represent real-world threats against (certified and uncertified) cryptographic devices. Masking schemes represent prevailing countermeasures to reduce the success probabilities of side-channel attacks. However, masking schemes increase the implementation cost in term of power consumption, clock cycles, and random number generation. Investigation of tools evaluating the degree of resilience of cryptographic devices using masking (against side-channel attacks) represents an important aspect in certification procedures (e.g., Common Criteria, FIPS 140-2 and EMVco). Several side-channel evaluation techniques exist such as template attacks and machine learning based attacks. In this paper, we formalise results obtained in side-channel attacks community when targeting masked implementations. We report theoretical as well as practical results of parametric and non-parametric side-channel attacks on masking schemes. The theoretical part reports results based on a simulation of the execution of software devices while the practical part focuses on actual leakages measured during the execution of a software implementation in three different contexts (that contain different levels of noise).

show abstract

Template Attacks vs. Machine Learning Revisited (and the Curse of Dimensionality in Side-Channel Analysis)

Cited by 86 publications

References 22 publications

Physical attacks on block ciphers

Physical attacks on block ciphers

Improving Side-Channel Analysis Through Semi-supervised Learning

Efficient Profiled Attacks on Masking Schemes

Contact Info

Product

Resources

About