Home

Welcome to the SoK wiki 👋 This wiki accompanies our paper SoK: Fully Homomorphic Encryption Tools & Compilers (coming soon^TM) and provides additional notes and remarks on the investigated tools. The FHE compilers considered in the paper are summarized below. Items in bold are experimentally evaluated in this repository.

Tools

Libraries

Library	Code	Docs	License
concrete	GitHub	https://concrete.zama.ai/	AGPL-3.0
FHEW	GitHub	--	GPL-2.0+
FV-NFLlib	GitHub	--	GPL-3.0
HEAAN	GitHub	--	CC-BY-NC 3.0
HElib	GitHub	https://homenc.github.io/HElib/	Apache-2.0
lattigo	GitHub	--	Apache-2.0
PALISADE	Gitlab	https://palisade-crypto.org/documentation/	BSD 2-clause
SEAL	GitHub	Manual for older version	MIT
TFHE	GitHub	https://tfhe.github.io/tfhe/	Apache-2.0
cuFHE	GitHub	--	MIT
nuFHE	GitHub	https://nufhe.readthedocs.io/en/latest/	GPL-3.0

Compilers

Tool/Technique	Code	License
ALCHEMY 📄	GitHub	GPL-3.0
Cingulata1 📄 (incl. CinguParam 📄)	GitHub, GitHub	CeCILL-C
Cingulata (incl. CinguParam) + Multi Start 📄	closed source	--
Cingulata (incl. CinguParam) + Cone Rewriting 📄	closed source	--
Cingulata (incl. CinguParam) + Lobster 📄	ACM DL, GitHub	free
EVA 📄	GitHub	MIT
nGraph-HE 📄 📄	GitHub	Apache-2.0
Ramparts 📄	closed source	--
SEALion 📄	closed source	--

¹ Previously known as Armadillo.

Questions and suggestions about this work should be publicly raised in the repository's issue tracker. PRs containing extensions or corrections are welcome!

Evaluation

Evaluation Programs

The following table summarizes the programs that we use to evaluate the performance and compilation/optimization time of the considered tools.

Algorithm	Reference	Source (Paper)			Eval. Category				Remarks
		Application	Tool/Optimization	Snippet/BB	Binary	Integer	FP/CKKS	NN/MVP
End-2-End CNN inference for non-trivial but not too deep model. CIFAR-10/MNIST	EVA	●	○	○	●	○	○	○	E.g., LeNet-5 or SqueezeNet.
Logistic Regression inference: MNIST/CREDIT	e.g. used in Phishing Classification	●	○	○	○	●	○	●
CHI-Squared Test	Secure GWAS	●	○	○	○	●	○	●
Edit Distance	Cheon et al. (2016), Kim et al. (2015	●	○	○	○	●	○	●
Kernel (e.g. Laplacian Sharpening Filter)	Ramparts, EVA	○	●	○	○	●	○	●	Uses the Kreyvium stream cipher.
Cardiac Risk Factor	Cingulata	○	●	○	○	○	●	○
Chebyshev Approx. of Sigmoid	Kim et al. (2018), Blatt et al. (2020)	○	○	●	○	○	●	○
Compute p-th bit of the product of a and b (or similar)	TFHE	○	○	●	○	●	○	○

Evaluation Metrics

In the paper, we define a set of metrics to compare the optimization techniques implemented in the compilers. The metrics that we consider are:

• Expressiveness^*

  The variety of programs that can be expressed using the tool.

• Accessibility/Usability^*

  The extent of available documentation, code comments, examples, and other user-oriented documentation.

• Capabilities^*

  The features of the tool (that are not necessarily related to the program's expressiveness).

• Performance/Speedup (run time) + memory consumption

  Time and resource consumption while executing our benchmarking program.

• Compilation/Optimization time

  Time required to build the program that can be executed afterwards.

Metrics marked by a star (^*) do not require executing a program. For full explanations of the criteria, please refer to the paper.

The following symbols are used to denote fulfillment of a criteria:

●  yes, supported
◐  partially supported (see paper)
○  no, not supported
-  not applicable

Basic Information

Tool/Technique	Scheme				Language (Bindings for)	FHE Library
	BGV	BFV	CKKS	CGGI
ALCHEMY	●	○	○	○	Haskell	Λ ⚬ λ
Cingulata (incl. CinguParam)	○	●	○	●	C++	TFHE
Cingulata (incl. CinguParam) + Multi Start	○	●	○	●	C++	TFHE
Cingulata (incl. CinguParam) + Cone Rewriting	○	●	○	●	C++	TFHE
Cingulata (incl. CinguParam) + Lobster	○	●	○	●	C++	TFHE
EVA	○	○	●	○	C++ (Python)	SEAL
nGraph-HE	○	○	●	○	C++ (Python)	SEAL
Ramparts	○	●	○	○	Julia	Palisade
SEALion	○	●	○	○	C++ (Python)	SEAL