| 引用本文： |
-
骆泽文,王志伟,赵路坦,孟丹,侯锐.全同态加密密钥切换的软硬件协同优化研究综述[J].信息安全学报,已采用 [点击复制]
- Luo Zewen,Wang Zhiwei,Zhao Lutan,Meng Dan,Hou Rui.A Survey on Hardware and Software Collaborative Optimization for Fully Homomorphic Encryption Keyswitch[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 全同态加密作为隐私计算的核心支撑技术，能够在加密数据上直接进行任意函数运算，在数据安全领域具有重要应用价值。然而，密钥切换技术作为全同态加密的核心组件，其高昂的计算开销已成为制约技术实用化的关键瓶颈。本文以基于环上带错误学习问题的逐字型全同态加密方案为研究对象，系统综述密钥切换优化技术的演进历程。文章系统建立了全同态加密密钥切换优化的多层次技术图谱，从理论算法、软件优化到硬件加速三个维度，深入分析了各类优化技术的共性和差异性，并为不同应用场景提供软硬件协同设计的优化策略。在理论算法层面，通过解构密钥切换的计算流程并识别其性能瓶颈，深入剖析主流算法在计算复杂度与噪声增长间所采取的平衡机制。在软件实现层面，围绕数论变换、模乘和基转换等关键组件的计算优化，计算重用技术以及软件库工程实现优化，给出多样化的优化策略以供探讨。在硬件加速层面，涵盖基于GPU并行架构、FPGA流水线设计及ASIC定制化电路的多平台加速体系设计，揭示了算法特性与软硬件设计的内在耦合规律。最后，本文对近年来密钥切换技术的优化研究进行系统分析，并展望未来密钥切换软硬件协同优化的研究方向，包括算法与架构深度融合、异构协同计算机制以及能耗感知调度策略，旨在为构建实用化全同态加密系统提供理论指导与工程实现路径。 |
| 关键词: 全同态加密 隐私计算 密钥切换 硬件加速 |
| DOI： |
| 投稿时间：2025-07-08修订日期：2025-11-13 |
| 基金项目:国家重点研发计划，中国科学院战略性先导科技专项，国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| A Survey on Hardware and Software Collaborative Optimization for Fully Homomorphic Encryption Keyswitch |
|
Luo Zewen, Wang Zhiwei, Zhao Lutan, Meng Dan, Hou Rui
|
| (Institute of Information Engineering，Chinese Academy of Sciences) |
| Abstract: |
| Fully homomorphic encryption has emerged as a foundational technology for privacy preserving computation due to its ability to support arbitrary function evaluation directly on encrypted data, with significant implications for data security. However, as a core component of fully homomorphic encryption, the keyswitch technique incurs substantial computa-tional overhead that remains a principal obstacle to practical deployment. This paper provides a systematic review of evolutionary optimization techniques for keyswitch within word-wise fully homomorphic encryption schemes based on the ring learning with errors problem. We construct a multi-level technology map of fully homomorphic encryption keyswitch optimization that spans theoretical algorithms, software optimizations, and hardware accelerations, analyzing the commonalities and differences among representative techniques, and providing optimization strategies for hardware and software co-design across diverse application scenarios. At the theoretical algorithm level, we decompose the com-putational workflow of the keyswitch algorithm, identify performance bottlenecks, and examine the trade-offs between computational complexity and noise growth observed in mainstream algorithms. At the software implementation level, this paper focuses on the component-wise and engineering optimizations for critical modules, including the number the-oretic transform, modular multiplication, and base conversion, together with the computation reuse techniques and im-provements to software library engineering, thereby offering a broad spectrum of optimization strategies for future re-search. In terms of hardware acceleration, we survey multiple platform acceleration architectures involving GPU parallel-ism, FPGA pipeline designs, and customized ASIC circuits, and reveal the inherent coupling between algorithmic charac-teristics and hardware software co-design. Finally, this paper conducts a systematic analysis of recent optimization studies on keyswitch and delineates future research directions for hardware and software collaborative optimization, including deep integration of algorithms and architectures, heterogeneous computing mechanisms, and energy aware scheduling strategies, with the aim of providing both theoretical guidance and practical engineering pathways for building deploya-ble fully homomorphic encryption systems. |
| Key words: Fully Homomorphic Encryption Privacy Computing Keyswitch Hardware Acceleration |
