Skip to content

TFHE Gets Real: An Efficient and Flexible Homomorphic Floating-Point Arithmetic

Authors: L. Bergerat; I. Chillotti; D. Ligier; J.-B. Orfila; S. Tap
Year: 2025
Venue: IACR ePrint 2025/257

type: scheme collection: catalogue, schemes, sok-references status: auto

Links: ePrint 2025/257

Summary

Auto-extracted summary (unreviewed)

The paper designs homomorphic floating-point arithmetic over TFHE, introducing new operators (WoP-PBS-based MiniFloats, a homomorphic floating-point encoding, encrypted ciphertext selection and carry propagation) that compose into addition, subtraction, multiplication and division for IEEE754-like and custom precisions, plus special values and function approximations such as ReLU and sigmoid. With parameters for 32- and 64-bit precision, 32-bit additions run in 2.5 s and multiplications in about 1 s in a multi-threaded setting, at least 10x faster than prior approaches; the implementation is released as a tfhe-rs artifact.

Abstract

Floating-point arithmetic plays a central role in computer science and is used in various domains where precision and computational scale are essential. One notable application is in machine learning, where Fully Homomorphic Encryption (FHE) can play a crucial role in safeguarding user privacy. In this paper, we focus on TFHE and develop novel homomorphic operators designed to enable the construction of precise and adaptable homomorphic floating-point operations. Integrating floating-point arithmetic within the context of FHE is particularly challenging due to constraints such as small message space and the lack of information during computation. Despite these challenges, we were able to determine parameters for common precisions (e.g., 32-bit, 64-bit) and achieve remarkable computational speeds, with 32-bit floating-point additions completing in 2.5 seconds and multiplications in approximately 1 second in a multi-threaded environment. These metrics provide empirical evidence of the efficiency and practicality of our proposed methods, which significantly outperform previous efforts. Our results demonstrate a significant advancement in the practical application of FHE, making it more viable for realworld scenarios and bridging the gap between theoretical encryption techniques and practical usability. ∗Ilaria Chillotti and Damien Ligier contributed to this work during a previous employment at Zama. TFHE Gets Real: an Efficient and Flexible Homomorphic Floating-Point Arithmetic Contents
Source: pdf

Extracted details

Field Value Source Confidence
contribution homomorphic floating-point encoding and building-block operators for TFHE (MiniFloats, ciphertext selection, carry propagation) llm 0.80
contribution homomorphic addition, subtraction, multiplication and division with custom precision llm 0.80
contribution handling of special values and function approximations including ReLU and approximate sigmoid llm 0.80
contribution parameter sets for common precisions (32-bit, 64-bit) llm 0.80
contribution open-source implementation in TFHE-rs llm 0.80
schemes TFHE llm 0.80
libraries TFHE-rs llm 0.80
tasks homomorphic floating-point arithmetic llm 0.80
tasks activation functions (ReLU, sigmoid) llm 0.80
interaction non-interactive FHE llm 0.80
code url https://github.com/zama-ai/tfhe-rs/tree/artifact_tches_2025 (URL found in the full text) regex 0.50
code url https://github.com/zama-ai/tfhe-rs.59 (URL found in the full text) regex 0.50
code url https://github.com/zama-ai/tfhe-rs/tree/artifact_tches_2025 llm 0.80
limitations not yet practical for general-purpose applications; a considerable gap to cleartext floating-point performance remains llm 0.80
llm relevance low llm 0.80
key result 32-bit floating-point addition latency: 2.5 s (multi-threaded) llm 0.80
key result 32-bit floating-point multiplication latency: about 1 s (multi-threaded) llm 0.80
key result speed-up over previous homomorphic floating-point methods: at least 10x (stated in introduction) llm 0.80

System card: TFHE floating point

Group 4. Alternative FHE schemes
Family TFHE
Arithmetic Floating-point emulation
Interaction Non-interactive
Model scale -
Status Active SOTA
Relation to the SoK True floating point instead of CKKS fixed point.
Note SoK cites it as evidence that FHE floating point is now feasible but costly.
Systems page TFHE floating point

SoK reference [58]

L. Bergerat, I. Chillotti, D. Ligier, J.-B. Orfila, and S. Tap, “TFHE gets real: an efficient and flexible homomorphic floating-point arithmetic,” Cryptology ePrint Archive, Report 2025/257, 2025. [Online]. Available: https://eprint.iacr.org/2025/257

Listed on the reference page of the site.

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: TFHE 122 [p.1], BFV 4 [p.3], FHEW 4 [p.3], BGV 3 [p.3], CKKS 2 [p.3]
  • library: TFHE-rs 5 [p.47], Concrete (Zama) 2 [p.56], Zama 2 [p.1]
  • hardware: Intel Xeon 2 [p.7]
  • technique: Programmable bootstrapping 70 [p.2], Bootstrapping 46 [p.3], Lookup table 19 [p.11], ReLU 13 [p.2], Sigmoid 12 [p.2], Rotations 7 [p.11]
  • assumption: LWE 62 [p.3], RLWE 8 [p.3], Lattice-based 4 [p.58]
  • topic: Homomorphic encryption 26 [p.1], Benchmark 4 [p.9], Open source 2 [p.47]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/catalogue/4-056_tfhe-gets-real-an-efficient-and-flexible-homomorphic-floatin.pdf catalogue 59 2.4 MB 8544b4f2a8c0
references/schemes/s22-2_eprint-2025-257.pdf schemes 59 2.4 MB 8544b4f2a8c0 byte-identical to references/catalogue/4-056_tfhe-gets-real-an-efficient-and-flexible-homomorphic-floatin.pdf
references/sok-references/058_tfhe-gets-real-an-efficient-and-flexible-homomorphic-floatin.pdf sok-references 59 2.4 MB 8544b4f2a8c0 byte-identical to references/catalogue/4-056_tfhe-gets-real-an-efficient-and-flexible-homomorphic-floatin.pdf

Database key: tfhe-gets-real-an-efficient-and-flexible-homomorphic-floating-point · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set tfhe-gets-real-an-efficient-and-flexible-homomorphic-floating-point COLUMN VALUE