密码学院中文 密码学院中文

科研成果|国科大密码学院最新成果被ICPADS 2024(CCF-C)接收

  • 文/密码学院
  • 日期:2024-11-05
  • 702

国科大密码学院博士生边毅,针对数论变换非友好型格密码算法的优化实现技术开展研究。在30th International Conference on Parallel and Distributed Systems (ICPADS 2024)发表了题为“TensorPolyMul: Accelerating Polynomial Multiplication in NTT-unfriendly Lattice-based Cryptography Using Tensor Cores”的研究论文。2024年10月,边毅参加了ICPADS 2024国际会议,并在此会议上就该项研究工作进行了全英文汇报。

 


Abstract:The urgent demand for computing power in Artificial intelligence (AI) technology has driven the rapid development of dedicated accelerators. Meanwhile, the threat posed by quantum computing to traditional public-key cryptography has prompted the emergence of post-quantum algorithms, such as lattice-based cryptography. However, performance issues with these algorithms have raised concerns within the industry about the transition to quantum-safe solutions. In this paper, we propose a novel universal framework for NTT-unfriendly lattice-based post-quantum algorithms, leveraging NVIDIA’s AI accelerator Tensor Core to address this challenge. By employing techniques such as polynomial matrixization and multi-precision representation, we effectively transform the primary workload (i.e., polynomial multiplication) into a series of small coefficient matrix multiplications that can be directly accelerated by Tensor Cores. This approach effectively bridges the gap between typical Tensor Core workloads and the core workloads of lattice-based post-quantum cryptography. As a case study, we implemented a prototype called TensorPolyMul to provide an implementation of Saber, a quantum-safe Key Encapsulation Mechanism (KEM). The experiments showcase that TensorPolyMul surpasses the state-of-the-art Tensor Core-based work, achieving remarkable speed-ups of 1.53×, 1.33×, 1.62×, and 1.22× for Inner Product, MatrixVecMul, Encaps, and Decaps, respectively.


论文信息:Yi Bian, Fangyu Zheng, and Jiwu Jing: “TensorPolyMul: Accelerating Polynomial Multiplication in NTT-unfriendly Lattice-based Cryptography Using Tensor Cores,” in International Conference on Parallel and Distributed Systems (ICPADS 2024).