Cross-Device Profiled Side-Channel Attack with Unsupervised Domain Adaptation

Deep learning (DL)-based techniques have recently proven to be very successful when applied profiled side-channel attacks (SCA). In a real-world SCA scenario, attackers gain knowledge about the target device by getting access similar prior attack. However, most state-of-the-art literature performs only proof-of-concept attacks, where traces intended for profiling and attacking are acquired consecutively on same fully-controlled device. This paper reminds that even small discrepancy between attack (regarded as domain discrepancy) can cause single-device completely fail. To address issue of discrepancy, we propose Cross-Device Profiled Attack (CDPA), which introduces an additional fine-tuning phase after establishing pretrained model. The is designed adjust pre-trained network, such it learn hidden representation not discriminative but also domain-invariant. order obtain domain-invariance, adopt maximum mean (MMD) loss constraint term classic cross-entropy function. We show MMD easily calculated embedded in standard convolutional neural network. evaluate our strategy both publicly available datasets multiple devices (eight Atmel XMEGA 8-bit microcontrollers three SAKURA-G evaluation boards). results demonstrate CDPA improve performance DL-based orders magnitude, significantly eliminates impact caused different devices.