基于GPU加速的全同态加密关键技术研究.pdf

(NumberTheoryTransformation,NTT)

(R)LWECKKSCKKS

CPU

NTTNTT

GPUCPU+GPU

CKKS

NTT

GPUGPU

NTT

-

I/ONTT

GPU

GPUHEXL

HElib300%CPU+GPU

CKKS30%108%

34%

GPU

-I-

Abstract

Asanewfactorofproduction,datamustbesharedandprocessedtoproducevalue.

Anddataprivacyprotectionregulationsrequireprivacyprotectionofsensitivedata.Full

homomorphicencryptionsupportsalgebraicoperationsonencrypteddata(ciphertext)di-

rectly,andisoneofthemainsolutionsforprivacycomputing.However,theefficiencyof

totalhomomorphicencryptionrestrictsitsfurtherapplication.ThebasicoperationNum-

berTheoryTransformation(NTT)underciphertextinvolvesalargenumberofaddition

andmultiplicationofpolynomialringswithhighdimensionalintegralcoefficients,andthe

operationsonpolynomialringsaremorecomplexthanthoseonnon-polynomialrings.

ThisdissertationfocusesonCKKS,ahomomorphicencryptionalgorithmbasedon

(R)LWE.CKKSisoneofthemostwidelyusedhomomorphicencryptionschemethat

supportsa