Abstract
Lattice-based cryptography has experienced significant advancements in recent years due to its versatility and simplicity. The ring learning with errors (RLWE) problem is widely adopted in lattice-based cryptography. However, the polynomial multiplication is the performance bottleneck of RLWE-based cryptography, which requires further examination. In this brief, a lightweight and efficient encryption/decryption coprocessor for RLWE-based cryptography is proposed. The time complexity of the Schoolbook polynomial multiplication (SPM) is reduced from n2 to n2/8 by enhancing multiplication parallelism. Moreover, an optimized structure for the Compressed cumulative distribution table (CDT) Gaussian sampler is proposed, resulting in 22.2% reduction in storage resource. The proposed SPM structure demonstrates a 2.3× performance speedup and 2.7× hardware efficiency for the encryption core, compared with state-of-the-art SPM accelerators. Additionally, it achieves a 2.4× performance speedup and 3.2× improvements on hardware efficiency for the decryption core.
Funding Information
  • National Natural Science Foundation of China (62176206)
  • National Key Research and Development Program of China (2023YFB4403500)