Paper 2026/2921
CKKS Acceleration for Next-Wave Homomorphic Encryption
Abstract
Fully Homomorphic Encryption (FHE) is a powerful cryptographic tool that enables computation directly on encrypted data without first decrypting it. It allows privacy-preserving computing where sensitive data is processed in the cloud or in other untrusted environments. Emerging functional bootstrapping and coefficient-based matrix-multiplication techniques offer opportunities to accelerate nonlinear and linear computations, respectively. However, there is still a gap between their theoretical advantages and achievable hardware performance gains. We present GEAR, an algorithm–architecture co-design for efficient FHE acceleration supporting recent CKKS advancements while preserving full conventional CKKS functionality for existing workloads. On the algorithmic side, we propose FusedFBS, a fused functional bootstrapping algorithm that combines coefficient-based matrix multiplication with grafting-enabled functional bootstrapping to enable high-precision, hardware-friendly FHE computation. On the hardware side, we propose the GEAR architecture, with a Modular Systolic Array and a Mixed-Radix Constant-Geometry NTT Unit, to accelerate key kernels. These mechanisms balance efficient execution with flexibility for evolving FHE algorithms, making it a versatile FHE accelerator. We demonstrate FusedFBS through encrypted Deep Learning inference. Across representative workloads, GEAR achieves more than 3000$\times$ speedup over the CPU baseline. Compared with state-of-the-art CKKS accelerators, it delivers significant speedup under similar area and power budgets.
Metadata
- Available format(s)
-
PDF
- Category
- Implementation
- Publication info
- Preprint.
- Keywords
- Accelerator
- Contact author(s)
-
xuw009 @ ucsd edu
sanjay deshpande1 @ northwestern edu
jaehk @ cs stanford edu
jakub szefer @ northwestern edu
tajana @ ucsd edu
plonga @ microsoft com - History
- 2026-11-21: approved
- 2026-11-17: received
- See all versions
- Short URL
- https://ia.cr/2026/2921
- License
-
CC BY
BibTeX
@misc{cryptoeprint:2026/2921,
author = {Xuan Wang and Sanjay Deshpande and Jaehyung Kim and Jakub Szefer and Tajana Rosing and Patrick Longa},
title = {{CKKS} Acceleration for Next-Wave Homomorphic Encryption},
howpublished = {Cryptology {ePrint} Archive, Paper 2026/2921},
year = {2026},
url = {https://eprint.iacr.org/2026/2921}
}