The Influence of Temperature on the Electromagnetic Leakage in Side Channel Attack Based on Deep Learning

This paper investigates the impact of temperature on the quality of electromagnetic leakage signals in the context of deep learning-based attacks. The Simeck32/64 algorithm is employed as a case study to exemplify this phenomenon. The proposed methodology involves training a deep learning model to perform attacks on electromagnetic leakage signals collected at three distinct temperature levels: −20°C, 20°C, and 60°C. The experimental results demonstrate that the model achieves optimal attack performance under a low-temperature scenario, exhibiting a test accuracy of 100% and successfully compromising all 64-bit keys of the Simeck32/64 algorithm. However, as the temperature increases, the model's attack success rate diminishes significantly, yielding test accuracies of 94.5% and 74.5% in normal and high-temperature scenarios, respectively.