Federated WGAN-GP for Privacy-Preserving Synthetic Side-Channel Trace Generation and Cryptographic Security Evaluation
Year: 2026
Type: preprint
Abstract: Side-channel attacks threaten embedded cryptographic systems by exploiting physical leakages such as power consumption, timing behavior, and electromagnetic emissions. To address the privacy risks of centralized trace collection, we propose a federated WGAN-GP framework for decentralized synthetic side-channel trace generation. Raw traces remain on client devices while local WGAN-GP training and federated averaging are performed across 20 heterogeneous clients drawn from ASCAD, ChipWhisperer, and DPAV4 in a 7-7-6 configuration. Experimental result over 10 random seeds, 12 communication rounds, 2 local epochs, and 40,000 traces show stable convergence, with a final quality score of 0.8167 ± 0.0004 and a synthetic-to-real attack success rate of 0.5% ± 0.84%. ChipWhisperer clients achieved the strongest local performance, outperforming ASCAD by 10.58%. The results confirm that balanced heterogeneous federation can support privacy-preserving trace synthesis with a significant privacy–utility trade-off.
Source: Research Square
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236796833
