Towards FPGA Acceleration of CKKS in Microsoft SEAL: A Microkernel-Based Architectural Exploration

HomomorphicEncryptionenablescomputationonencrypted

data, making it a key technology for privacy-preserving cloud comput-

ing and artificial intelligence workloads. Among modern schemes, CKKS

is well suited for approximate numerical computation, but its adoption

remains limited by the cost of polynomial arithmetic over large residue-

number-system moduli. Microsoft SEAL, a widely adopted open-source

library,implementsCKKSthroughreusablearithmeticprimitives.Based

on this decomposition, this paper presents a preliminary architectural

study of FPGA-based microkernels to accelerate these arithmetic op-

erations. The work targets modular addition, modular multiplication,

and forward and inverse Number Theoretic Transforms as the core mi-

crokernels for CKKS ciphertext operations and investigates their imple-

mentation on pipelined and parallel FPGA architectures. The proposed

microkernels have been implemented and evaluated on a Xilinx Ultra-

Scale+ device to assess the performance, scalability, and feasibility of an

FPGA accelerator tightly integrated with Microsoft SEAL.