Profiling side-channel attacks represent the most powerful category of side-channel attacks. There, we assume that the attacker has access to a clone device in order to profile the device. Additionally, we assume the attacker to be unbounded in power in an effort to give the worst-case security analysis.
In this paper, we start from a different premise and consider an attacker in a restricted setting where he is able to profile only a limited number of measurements. To that end, we propose a new framework for profiling side-channel analysis that we call the Restricted Attacker framework. With it, we enforce the attackers to really conduct the most powerful attack possible but also we provide a setting that inherently allows a more fair analysis among attacks.
Next, we discuss the ramifications of having the attacker with unbounded power when considering neural network-based attacks.
There, we are able to prove that the Universal Approximation Theorem can result in neural network-based attacks being able to break implementations with only a single measurement. Those considerations further strengthen the need for the Restricted Attacker framework.