New Approaches for Efficient Fully Homomorphic Encryption
PublicDownloadable Content
open in viewer<P>In the last decade, cloud computing became popular among companies for outsourcing some of their services. Companies use cloud services to store crucial information such as financial and client data. Cloud services are not only cost effective but also easier to manage since the companies avoid maintenance of servers. Although cloud has its advantages, maintaining the security is a big concern. Cloud services might not have any malicious intent, but attacks targeting cloud systems could easily steal vital data belong to the companies. The only protection that assures the security of the information is a strong encryption. However, these schemes only protects the information but prevent you to do any computation on the data. This was an open problem for more than 30 years and it has been solved recently by the introduction of the first fully homomorphic encryption (FHE) scheme by Gentry. The FHE schemes allow you to do arbitrary computation on an encrypted data by still preserving the encryption. Namely, the message is not revealed (decrypted) at any given time while computing the arbitrary circuit. However, the first FHE scheme is not practical for any practical application. Later, numerous research work has been published aiming at making fully homomorphic encryption practical for daily use, but still they were too inefficient to be used in everyday practical applications.</P> <P>In this dissertation we tackle the efficiency problems of fully homomorphic encryption (FHE) schemes. We propose two new FHE schemes that improve the storage requirement and runtime performance. The first scheme (Doröz, Hu and Sunar) reduces the size of the evaluation keys in existing NTRU based FHE schemes. In the second scheme (F-NTRU) we designed an NTRU based FHE scheme which is not only free of costly evaluation keys but also competitive in runtime performance.</P> <P>We further proposed two hardware accelerators to increase the performance of arithmetic operations underlying the schemes. The first accelerator is a custom hardware architecture for realizing the Gentry-Halevi fully homomorphic encryption scheme. This contribution presents the first full realization of FHE in hardware. The architecture features an optimized multi-million bit multiplier based on the Schönhage-Strassen multiplication algorithm. Moreover, a number of optimizations including spectral techniques as well as a precomputation strategy is used to significantly improve the performance of the overall design. The other accelerator is optimized for a class of reconfigurable logic for somewhat homomorphic encryption (SWHE) based schemes. Our design works as a co-processor: the most compute-heavy operations are offloaded to this specialized hardware. The core of our design is an efficient polynomial multiplier as it is the most compute-heavy operation of our target scheme. The presented architecture can compute the product of very-large polynomials more efficiently than software implementations on CPUs.</P> <P>Finally, to assess the performance of proposed schemes and hardware accelerators we homomorphically evaluate the AES and the Prince block ciphers. We introduce various optimizations including a storage-runtime trade-off. Our benchmarking results show significant speedups over other existing instantiations. Also, we present a private information retrieval (PIR) scheme based on a modified version of Doröz, Hu and Sunar’s homomorphic scheme. The scheme is capable of privately retrieving data from a database containing 4 billion entries. We achieve asymptotically lower bandwidth cost compared to other PIR schemes which makes it more practical.</P>
- Creator
- Contributors
- Degree
- Unit
- Publisher
- Language
- English
- Identifier
- etd-061417-201844
- Keyword
- Advisor
- Committee
- Defense date
- Year
- 2017
- Date created
- 2017-06-14
- Resource type
- Rights statement
Relations
- In Collection:
Items
Permanent link to this page: https://digital.wpi.edu/show/0p096706r