Dual-Leak: Deep Unsupervised Active Learning for Cross-Device Profiled Side-Channel Leakage Analysis

Deep Learning (DL)-based side-channel analysis (SCA), as a new branch of SCA attacks, poses a significant privacy and security threat to implementations of cryptographic algorithms. Despite their impacts on hardware security, existing DL-based SCA attacks have not fully leveraged the potential of DL algorithms. Therefore, previously proposed DL-based SCA attacks may not show the real capability to extract sensitive information from target designs. In this paper, we propose a novel cross-device SCA method, named Dual-Leak, that applies Deep Unsupervised Active Learning to create a DL model for breaking cryptographic implementations, even with countermeasures deployed. The experimental results on both the local dataset and publicly available dataset show that our Dual-Leak attack significantly outperforms state-of-the-art works while no labeled traces are required from victim devices (i.e., unsupervised learning). Countermeasures are also discussed to assure hardware security against new attacks.