Empirical evaluation of multi-device profiling side-channel attacks

Abstract

Side-channel analysis of cryptographic systems can allow for the recovery of secret information by an adversary even where the underlying algorithms have been shown to be provably secure. This is achieved by exploiting the unintentional leakages inherent in the underlying implementation of the algorithm in software or hardware. Within this field of research, a class of attacks known as profiling attacks, or more specifically as used here template attacks, have been shown to be extremely efficient at extracting secret keys. Template attacks assume a strong adversarial model, in that an attacker has an identical device with which to profile the power consumption of various operations. This can then be used to efficiently attack the target device. Inherent in this assumption is that the power consumption across the devices under test is somewhat similar. This central tenet of the attack is largely unexplored in the literature with the research community generally performing the profiling stage on the same device as being attacked. This is beneficial for evaluation or penetration testing as it is essentially the best case scenario for an attacker where the model built during the profiling stage matches exactly that of the target device, however it is not necessarily a reflection on how the attack will work in reality. In this work, a large scale evaluation of this assumption is performed, comparing the key recovery performance across 20 identical smart-cards when performing a profiling attack.

Understanding Screaming Channels: From a Detailed Analysis to Improved Attacks

“…In this paper we mainly use profiled correlation attacks, therefore we focus on trace normalization rather than on machine learning based approaches. Literature shows that, while interdevice variations and other differences have a huge impact on the performance of vanilla template attacks, using simple normalization techniques (e.g., offset removal and variance normalization) can significantly reduce the problem [MBTL13,EG12,CK14,HOTM14,CK14]. Templates built from multiple devices are more robust, but they are less effective for a single device [RSV + 11, HOTM14,CK14].…”

“…The main disadvantage of z-score normalization is that the average and standard deviation are hard to estimate on a small number of attack traces. For this reason, it might be better to normalize each trace individually by subtracting it with its average value [HOTM14,CK14], or to normalize the attack set using the mean and standard deviation estimated on the profile set [HOTM14].…”

Practical Approaches Toward Deep-Learning-Based Cross-Device Power Side-Channel Attack

“…To the best of our knowledge, both data sets consist of traces captured from one single device, which are not suitable for current work. Hence, we collected new traces from 30 different 8-bit AVR microcontrollers running [10], [30], [19], [18], [20], [31] Cross-device Attack Template Attack [14], [13], [32], [15] Neural Networks [21], [22], This Work the AES-128 algorithm using the ChipWhisperer platform [24] (Figure 2). Although 8-bit microcontrollers are becoming less preferred for encryption engines nowadays, recent body of work ( [13], [18], [37], [38], [39]) investigated performance of Profiled SCA attack using datasets gathered from 8-bit microcontrollers.…”

“…Table 1 summarizes the related works. As can be seen, most of the template attacks were evaluated on the same device, whereas only in a few cases, ( [14], [13], [32], [15]) attacks were performed on a different device. This article significantly improves on [21] to present a generalized Deep Learning based cross-device SCA attack on 30 different devices, utilizing DTW and PCA along with the multi-device training.…”

“…Traditionally, profiling attacks are performed by generating templates (hence called template attack [8], [13], [14], [15]) for different keys by utilizing a multi-variate Gaussian distribution approximation of the pre-identified points-of-interests (POIs). Recently, the hardware security research community has focused its attention to the machine learning (ML) based profiled attacks using Support Vector Machine (SVM) [16], [17], Random Forest (RF) [17] as well as deep learning based attacks [10], [18], [19], [20]).…”