Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/E01613B2-11B4-4CA9-88EF-AE665767BD94 | ||
| Year | 2023 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Georgios Agoritsis, "Accelerating fully homomorphic encryption (FHE) schemes with FPGAs", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023 https://doi.org/10.26233/heallink.tuc.97377 | ||
| Appears in Collections |
This thesis delves into the exploration of FPGA acceleration possibilities for Fully Homomorphic Encryption (FHE), focusing on the OpenFHE library. Following a software profiling procedure, the thesis centers its attention on the Number Theoretic Transform (NTT) as a key acceleration point and investigates state-of-the-art optimization techniques from the existing literature. The chosen platform is the Xilinx Alveo U50 FPGA card, with the OpenFHE library as the foundation. The study demonstrates how a hardware-based project designed for the Alveo U50 card can be integrated into OpenFHE. Although this research does not introduce a novel acceleration method, it primarily focuses on showcasing the integration of an application developed in Xilinx Vitis IDE with OpenFHE. An essential contribution is the provision of a functional NTT implementation that supports various NTT sizes, complementing OpenFHE. The implementation encompasses an in-place forward NTT with the Cooley-Tukey butterfly and Harvey’s modular multiplication optimization, utilizing High-Level Synthesis (HLS). This work stands as the first hardware accelerator for OpenFHE, deployed on a Xilinx FPGA, supporting up to 2^18-point Number Theoretic Transforms without FPGA reconfiguration. The accelerator is seamlessly integrated with OpenFHE as a hardware component. While the project does not yield an immediate speedup for OpenFHE, it lays a solid foundation for future acceleration endeavors. Future work directions and optimization suggestions are identified and conclude the thesis.
Copyright © DIAS 2013
