Year: 2025
Type: conference-paper
Abstract: As a high-performance computing platform, the GPU is much suitable for cryptographic applications in cloud computing environment. Unfortunately, recent studies found that GPU-based AES implementations are also vulnerable to cache-timing attacks. Although those architectural-level and algorithmic-level countermeasures specifically tailored for power/electro-magnetic (EM) analysis are effective to thwart the cache-timing attacks, they are either of high cost or require customized hardware. In this work, we propose a masking scheme, which is intrinsically device-independent, to prevent GPU-based AES implementations from cache-timing attacks with the full considerations of low cost and provable security. The scheme reduces the requirement for random from linear to constant number by taking the advantage of cache-timing leakage model and, at the same time, maintains its security. The security of the masking scheme is formally proved under realistic assumptions. In addition, we also present a typical GPU-based prototype implementation of the scheme and evaluate the performance of the implementation on CUDA-enabled GPUs of different architectures. The results show that the performance of the implementation varies from architecture to architecture. Nonetheless, all of them outperform previous schemes.
Authors: Guojian Bi, Yiwen Gao, Yongbin Zhou
Institution: Nanjing University of Science and Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Awards: G2333442876, G4061967014, G4236585819
OpenAlex ID: W4411672659