On the Feasibility of Exploiting the USB Power Line for Side-Channel Attacks against Samsung Smartphone Models

When plugged to a malicious USB charging station, smartphones may be exposed to a side-channel attack (SCA) that analyzes current fluctuations in the USB power line, and infers the location of screen animations displayed when pressing buttons associated to sensitive information (e.g., unlocking PIN). However, the literature showed this SCA against old phone models (e.g., Samsung Galaxy Nexus, Motorola G4), and thus there is a lack of investigations on recently released ones. Our work fills this gap by conducting, on recently released models (Galaxy S22 and S23), a thorough investigation of SCA based on artificial intelligence classifiers that attempt to infer PIN buttons pressed. We go several steps further compared to related work. First, we not only investigate the accuracy of classifiers on devices with PIN button press animations, but also on devices without such animations, which are the key elements exploited. In addition, we investigate classifiers created by capturing power signal of button presses using two different approaches, named sequential, in which a single button is pressed several times consecutively to collect samples, and random, where PIN buttons are pressed randomly. The first approach, although being unrealistic in practice, is intuitive and thus may be adopted in the literature to evaluate SCA. Our findings show that in such scenario, classifiers are biased, and even presenting accuracy higher than 90%, they are not useful in practice because users commonly adopt random PIN sequences. Our results also reveal that in a real attack scenario (with random PIN sequences), classifiers have accuracy of 12% or even lower, showing that SCA fails against the recently released phone models. We confirm our results by capturing power signals with two different voltage readers. Finally, we analyze captured datasets, providing valuable indicatives of why SCA fails in practice.