Potential of Unsupervised Deep Learning for Detection of EM Side-Channel Attacks

The growing concern over side-channel emissions and their impact on the confidentiality of computer applications and systems has led to increased attention towards the vulnerability of the backside of the Si substrate. This research paper proposes an innovative potential approach to bolster hardware security and mitigate vulnerabilities against electromagnetic side-channel attacks. The approach involves the fusion of Multi-physics simulation and deep learning techniques to accurately identify the most susceptible locations for side-channel leakage on the backside of silicon substrates within integrated circuits. By employing contextual feature extraction and an attention mechanism, the study aims to enhance chip-level security analysis and automate the detection of points of interest that are prone to side-channel attacks. The presents potential method overcomes limitations present in previous works, leading to improved quality and accuracy of the results.