Research on Cross-Device Bypass Attacks Based on Deep Learning

SPECK algorithm is a typical lightweight algorithm with ARX structure, which is widely used in resource-constrained portable devices. In this paper, we address cross-device scenarios and explore the feasibility of accuracy as an index for evaluating the efficiency of attacks based on the bypass analysis of SPECK encryption algorithm. Three different scenarios are considered, i.e., different measurement locations for the same device, different IC sockets for the same chip, and different keys for different chips, to explore the impact of changes in the environment on the attack efficiency. The experiments show that using different devices and keys or changing chip IC sockets during the analysis and attack phases will seriously affect the accuracy metrics compared to changing the measurement locations of the probes. It is concluded that while accuracy is the most commonly used metric for monitoring and evaluating neural networks, it leads to a significant underestimation of the efficiency of attacks in cross-device conditions, and in fact models with low accuracy can still successfully attack the attacking device.