Skip to content

TFHE: Fast Fully Homomorphic Encryption over the Torus

Authors: Chillotti, Gama, Georgieva, Izabachène
Year: 2016
Venue: ASIACRYPT 2016 / J. Cryptology 2020

type: scheme collection: catalogue, schemes status: auto

Links: ePrint 2018/421

Visual lesson: TFHE (CGGI) · Scheme table row: TFHE (CGGI)

Summary

Auto-extracted summary (unreviewed)

This is the full journal version of TFHE, a fully homomorphic encryption scheme over the torus that generalizes GSW and its ring variants; it shows FHEW-style gate bootstrapping can be expressed purely through the external product between GSW and LWE ciphertexts, cutting bootstrapping from 690 ms to 13 ms on a single core with a 16 MB instead of 1 GB key. It further adds a leveled mode with packed ciphertexts, look-up-table and weighted-automata evaluation, a TBSR homomorphic counter, a 137 ms circuit bootstrapping from LWE to RingGSW that makes the leveled mode composable, and a practical security analysis linking parameters to LWE error rate and key entropy, with an open-source C/C++ implementation.

Abstract

This work describes a fast fully homomorphic encryption scheme over the torus (TFHE), that revisits, generalizes and improves the fully homomorphic encryption (FHE) based on GSW and its ring variants. The simplest FHE schemes consist in bootstrapped binary gates. In this gate bootstrapping mode, we show that the scheme FHEW of [29] can be expressed only in terms of external product between a GSW and a LWE ciphertext. As a consequence of this result and of other optimizations, we decrease the running time of their bootstrapping from 690ms to 13ms single core, using 16MB bootstrapping key instead of 1GB, and preserving the security parameter. In leveled homomorphic mode, we propose two methods to manipulate packed data, in order to decrease the ciphertext expansion and to optimize the evaluation of look-up tables and arbitrary functions in RingGSW based homomorphic schemes. We also extend the automata logic, introduced in [31], to the efficient leveled evaluation of weighted automata, and present a new homomorphic counter called TBSR, that supports all the elementary operations that occur in a multiplication. These improvements speed-up the evaluation of most arithmetic functions in a packed leveled mode, with a noise overhead that remains additive. We finally present a new circuit bootstrapping that converts LWE ciphertexts into low-noise RingGSW ciphertexts in just 137ms, which makes the leveled mode of TFHE composable, and which is fast enough to speed-up arithmetic functions, compared to the gate bootstrapping approach. Finally, we provide an alternative practical analysis of LWE based schemes, which directly relates the security parameter to the error rate ⋆This work was done while I. Chillotti was a PhD student in the Laboratoire de Math´ematiques de Versailles, UVSQ (Versailles, France) and while M. Georgieva worked in Gemalto (Meudon, France). of LWE and the entropy of the LWE secret key, and we propose concrete parameter sets and timing comparison for all our constructions.
Source: pdf

Extracted details

Field Value Source Confidence
contribution gate bootstrapping via external product (GSW x LWE) running in 13 ms single core with a 16 MB key llm 0.90
contribution leveled mode with two packing methods for look-up tables and arbitrary functions over RingGSW llm 0.90
contribution leveled evaluation of weighted automata and the TBSR homomorphic counter for arithmetic llm 0.90
contribution circuit bootstrapping converting LWE to low-noise RingGSW ciphertexts in 137 ms llm 0.90
contribution practical security analysis relating security parameter to LWE error rate and secret entropy, with concrete parameter sets llm 0.90
contribution open-source implementation llm 0.90
schemes TFHE llm 0.90
schemes GSW llm 0.90
schemes FHEW llm 0.90
libraries TFHE (tfhe/tfhe) llm 0.90
tasks Boolean circuit evaluation llm 0.90
tasks bootstrapping llm 0.90
tasks look-up table evaluation llm 0.90
tasks weighted automata and arithmetic in leveled mode llm 0.90
hardware single-core CPU (unspecified) llm 0.90
interaction non-interactive FHE llm 0.90
code url https://github.com/shaih/HElib/ (URL found in the full text) regex 0.50
code url https://github.com/tfhe/tfhe llm 0.90
limitations unclear how to apply RNS / double-CRT representations to TFHE, which lacks an explicit q modulus llm 0.90
limitations extension of the framework to more complex arithmetic circuits with multiplications left open llm 0.90
llm relevance low llm 0.90
key result gate bootstrapping latency: 13 ms (down from 690 ms in FHEW) (single core, same security parameter) llm 0.90
key result bootstrapping key size: 16 MB (down from 1 GB) (gate bootstrapping) llm 0.90
key result circuit bootstrapping latency: 137 ms (LWE to RingGSW conversion) llm 0.90

Scheme card: TFHE (CGGI)

Group 3. Bit-wise / lookup-table FHE (GSW, FHEW, TFHE family)
Kind Scheme
Hardness assumption Torus LWE / GLWE
Plaintext type Bits / small integers via LUT
SIMD packing No
Bootstrapping Yes, programmable bootstrapping in ms
Post-quantum Yes
Status Widely used
Fit for LLM inference Main CKKS rival for non-linearities: exact activations by LUT. Slow for large linear algebra. Concrete ML targets quantized models.
Follow-ups Programmable bootstrapping for DNNs (2021/091); floating point in TFHE (2025/257).
Libraries TFHE-rs, TFHE (C++), Concrete ML, OpenFHE
Follow-up papers in the archive Programmable Bootstrapping Enables Efficient Homomorphic Inference of Deep Neural Networks
TFHE Gets Real: An Efficient and Flexible Homomorphic Floating-Point Arithmetic

System card: TFHE (CGGI)

Group 4. Alternative FHE schemes
Family Torus LWE + programmable bootstrapping
Arithmetic Bits / small integers via LUT
Interaction Non-interactive
Model scale Small CNNs, tiny Transformers
Status Widely used
Relation to the SoK Exact LUT evaluation of any non-linearity; no approximation error.
Note Main CKKS alternative for non-linear functions. Lacks CKKS-style SIMD, so large matrix products are slow.
Code https://github.com/zama-ai/tfhe-rs
Systems page TFHE (CGGI)

Keyword mentions

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

  • scheme: GSW 20 [p.1], TFHE 18 [p.1], Leveled / somewhat HE 5 [p.1], NTRU 4 [p.3], FHEW 3 [p.1], BFV 2 [p.58], BGV 2 [p.3], RSA 2 [p.62]
  • library: Microsoft SEAL 1 [p.60], HElib 1 [p.61], Concrete (Zama) 1 [p.52]
  • model: Decision tree 3 [p.25]
  • technique: Bootstrapping 137 [p.1], Lookup table 40 [p.1], Key switching 35 [p.19], Rotations 17 [p.19], Rescaling 6 [p.37], SIMD packing 5 [p.3], RNS 3 [p.58]
  • assumption: LWE 71 [p.1], Lattice-based 37 [p.2], RLWE 9 [p.3], Post-quantum 2 [p.60]
  • topic: Homomorphic encryption 38 [p.1], Open source 2 [p.54]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/catalogue/4-054_tfhe-fast-fully-homomorphic-encryption-over-the-torus.pdf catalogue 62 928.7 KB 23fa887ff217
references/schemes/s22-0_tfhe-fast-fully-homomorphic-encryption-over-the-torus.pdf schemes 62 928.7 KB 23fa887ff217 byte-identical to references/catalogue/4-054_tfhe-fast-fully-homomorphic-encryption-over-the-torus.pdf

Database key: tfhe-fast-fully-homomorphic-encryption-over-the-torus · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set tfhe-fast-fully-homomorphic-encryption-over-the-torus COLUMN VALUE