Efficient Homomorphic Encryption using FPGA-Acceleration

Abstract: The data being handled by many applications is in most cases confidential. Furthermore, the applications handling such data are nowadays often offloaded to remote data centers in the cloud. Computation on data in this environment is not secure since data is only encrypted when stored, and not while it is operated on. A more secure way is to operate on the data without decrypting it using Homomorphic Encryption. However high computational complexity associated with Homomorphic Encryption limits its practical application. We developed a hardware/software codesign for the purpose of accelerating operations on encrypted data using reconfigurable hardware. We implemented a top-level application for encrypted search using the TFHE open source library. Profiling our application found that the FFT-based polynomial multiplication consumes the majority of computational time (90%). We performed a hardware/software task partition of the TFHE library and implemented the polynomial multiplication using FPGA hardware on the heterogeneous Xilinx PYNQ-Z2 SoC Evaluation Board. Hardware Synthesis showed that our hardware accelerated approach is significantly faster than its software equivalent (23x).