Efficient general-purpose computation with fully homomorphic encryption

Abstract

Fully homomorphic encryption (FHE) is a powerful type of cryptography that allows for meaningful computation directly on encrypted data. This unique property has important ramifications for secure outsourced computing and allows FHE to solve key confidentiality issues in a cloud computing context. However, the technology has seen little adoption outside of research circles due to its slow speeds relative to plaintext computation and complex, non-intuitive programming model. The novel methodologies encompassed in this dissertation overcome the shortcomings of FHE and pave the way for widespread adoption of the technology. ☐ First, initial strides are presented that solve both key issues in FHE in the form of two end-to-end frameworks for general-purpose FHE computation leveraging the CGGI cryptosystem. Additionally, a standardized benchmark suite and universal compiler targeting state-of-the-art FHE libraries are introduced to allow users to rapidly compare different FHE implementations in order to find the best suited library and scheme for their applications. ☐ Next, the HELM framework is presented, which utilizes two distinct oblivious lookup table mechanisms and a bridging technique to dynamically change the underlying plaintext space of FHE ciphertexts is proposed to achieve efficient general-purpose computation. HELM can consume off-the-shelf HDL designs to generate and execute FHE applications, achieving a speedup of 1-2 orders of magnitude relative to prior works. Lastly, a hardware-accelerated, bespoke system for arbitrary encrypted computation called ArctyrEX is introduced that allows programmers to express their intents in a high-level language and seamlessly scales to multi-GPU architectures. ArctyrEX achieves speedups of over 40x relative to the current state-of-the-art for a variety of realistic workloads.