Cryptology ePrint Archive: Report 2016/725

Tile-Based Modular Architecture for Accelerating Homomorphic Function Evaluation on FPGA

Mustafa Khairallah and Maged Ghoneima

Abstract: In this paper, a new architecture for accelerating homomorphic function evaluation on FPGA is proposed. A parallel cached NTT algorithm with an overall time complexity O(sqrt(N)log(sqrt(N)) is presented. The architecture has been implemented on Xilinx Virtex 7 XC7V1140T FPGA. achieving a 60% utilization ratio. The implementation performs 32-bit 2^(16)-point NTT algorithm in 23.8 us, achieving speed-up of 2x over the state of the art architectures. The architecture has been evaluated by computing a block of each of the AES and SIMON-64/128 on the LTV and YASHE schemes. The proposed architecture can evaluate the AES circuit using the LTV scheme in 4 minutes, processing 2048 blocks in parallel, which leads to an amortized performance of 117 ms/block, which is the fastest performance reported to the best of our knowledge.

Category / Keywords: FHE, Homomorphic, FPGA, Virtex, NTT, CRT

Original Publication (with major differences): 2016 IEEE 59th International Midwest Symposium on Circuits and Systems

Date: received 22 Jul 2016, last revised 30 Jul 2016

Contact author: khairallah at ieee org

Available format(s): PDF | BibTeX Citation

Version: 20160731:055921 (All versions of this report)

Short URL: ia.cr/2016/725

Discussion forum: Show discussion | Start new discussion


[ Cryptology ePrint archive ]