Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OFHE: An Electro-Optical Accelerator for Discretized TFHE

Mengxin Zheng, Cheng Chu, Qian Lou, Nathan Youngblood, Mo Li, Sajjad Moazeni, Lei Jiang

TL;DR

DTFHE enables multi-bit homomorphic encryption but imposes heavy 32/64/128-bit polynomial multiplications and FFT/IFFT workloads that strain prior accelerators. OFHE introduces a photonic-electro-optical accelerator with a 64-point photonic FFT engine and CMOS control, supporting reconfigurable $32$-, $64$-, and $128$-bit datapaths and decomposing large FFTs into 64-point kernels to deliver substantial gains. It achieves $8.7\%$ lower latency, $57\%$ higher throughput, and $94\%$ higher throughput per Watt against prior accelerators, while maintaining scalable performance for 64- and 128-bit operations and improving general-purpose circuit latency by about $58\%$ on average. The architecture highlights the practical potential of photonic accelerators for complex DTFHE workloads, enabling faster, more energy-efficient, and versatile homomorphic computing in real-world applications.

Abstract

This paper presents \textit{OFHE}, an electro-optical accelerator designed to process Discretized TFHE (DTFHE) operations, which encrypt multi-bit messages and support homomorphic multiplications, lookup table operations and full-domain functional bootstrappings. While DTFHE is more efficient and versatile than other fully homomorphic encryption schemes, it requires 32-, 64-, and 128-bit polynomial multiplications, which can be time-consuming. Existing TFHE accelerators are not easily upgradable to support DTFHE operations due to limited datapaths, a lack of datapath bit-width reconfigurability, and power inefficiencies when processing FFT and inverse FFT (IFFT) kernels. Compared to prior TFHE accelerators, OFHE addresses these challenges by improving the DTFHE operation latency by 8.7\%, the DTFHE operation throughput by $57\%$, and the DTFHE operation throughput per Watt by $94\%$.

OFHE: An Electro-Optical Accelerator for Discretized TFHE

TL;DR

DTFHE enables multi-bit homomorphic encryption but imposes heavy 32/64/128-bit polynomial multiplications and FFT/IFFT workloads that strain prior accelerators. OFHE introduces a photonic-electro-optical accelerator with a 64-point photonic FFT engine and CMOS control, supporting reconfigurable -, -, and -bit datapaths and decomposing large FFTs into 64-point kernels to deliver substantial gains. It achieves lower latency, higher throughput, and higher throughput per Watt against prior accelerators, while maintaining scalable performance for 64- and 128-bit operations and improving general-purpose circuit latency by about on average. The architecture highlights the practical potential of photonic accelerators for complex DTFHE workloads, enabling faster, more energy-efficient, and versatile homomorphic computing in real-world applications.

Abstract

This paper presents \textit{OFHE}, an electro-optical accelerator designed to process Discretized TFHE (DTFHE) operations, which encrypt multi-bit messages and support homomorphic multiplications, lookup table operations and full-domain functional bootstrappings. While DTFHE is more efficient and versatile than other fully homomorphic encryption schemes, it requires 32-, 64-, and 128-bit polynomial multiplications, which can be time-consuming. Existing TFHE accelerators are not easily upgradable to support DTFHE operations due to limited datapaths, a lack of datapath bit-width reconfigurability, and power inefficiencies when processing FFT and inverse FFT (IFFT) kernels. Compared to prior TFHE accelerators, OFHE addresses these challenges by improving the DTFHE operation latency by 8.7\%, the DTFHE operation throughput by , and the DTFHE operation throughput per Watt by .
Paper Structure (12 sections, 10 figures, 4 tables, 1 algorithm)

This paper contains 12 sections, 10 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Discretized TFHE (DTFHE) Basics
  4. DTFHE Operations and Implementations
  5. Related Work and Motivation
  6. OFHE
  7. Experimental Methodology
  8. Results and Analysis
  9. Comparison Against Prior Accelerators
  10. Performance of 64- & 128-bit Datapaths
  11. General Purpose Application Performance
  12. Conclusion

Figures (10)

  • Figure 1: The latency breakdown of DTFHE operations (i.e., gate, functional, full-domain bootstrappings, and TRLWE tensor product).
  • Figure 2: The power breakdown of prior TFHE hardware accelerators including MATCHA Jiang:DAC2022, FPT Beirendonck:ARIX2022 and Strix Putra:MICRO2023.
  • Figure 3: The components and pipeline of OFHE, when processing $m$-bit inputs $i_{m-1},i_{m-2},\ldots,i_{0}$.
  • Figure 4: PFFTE extinction ratio with varying input #.
  • Figure 5: Fraction bit # vs output approx. noise.
  • ...and 5 more figures