ArborDorp: An Area Efficient CKKS Accelerator
Recommended citation: ArborDorp: An Area Efficient CKKS Accelerator. Xianglong Deng, Shengyu Fan, Zhicheng Hu, Zhuoyu Tian, Aisen Xie, Liang Chang, Meng Li, Rui Hou, Dan Meng, Mingzhe Zhang. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD). 2026.
Abstract
Fully Homomorphic Encryption (FHE) is recognized as a promising technology for privacy preservation, as it enables computation on encrypted data. However, the significant performance overhead has hindered its widespread adoption in real systems. To address this issue, several ASIC-based accelerators have been proposed, which significantly enhance FHE performance. Unfortunately, these accelerators require a substantial circuit area, making FHE acceleration prohibitively expensive for broad adoption in real-world systems. In response to these challenges, this paper introduces ArborDorp, a novel FHE accelerator based on ASIC. ArborDorp accelerates FHE applications while mitigating the increase in circuit area. We incorporate a series of innovative optimization techniques in ArborDorp: 1) We optimize the number-theoretic transform (NTT) algorithm by merging the modular operation with its procedural multiplication and accumulation (MAC) arithmetic. This approach reduces the area of modulo hardware, enabling performance improvement by increasing the parallellism of computational components. 2) By performing algorithm transformations on various kernels and extracting common arithmetic operators, ArborDorp enables IP, and BConv to reuse the NTT hardware component. This results in reduced circuit area and improved circuit utilization. ArborDorp reduces the circuit area by 10.7% compared to SHARP while delivering competitive performance. ArborDorp-PLUS, a larger implementation that doubles the number of lanes, further achieves an average performance improvement of 1.34× over SHARP. ArborDorp significantly improves FHE acceleration while mitigating the associated overheads, making it a promising solution for practical applications in real-world systems.
