Fully Homomorphic Encryption (FHE) enables the secure offloading of arbitrary computation to an untrusted server. While secure, software implementations of FHE are about 10 000x slower than their unencrypted equivalents. I will be contributing to the development of a dedicated FHE hardware accelerator that aims to bridge this gap.

In particular, I will be responsible for designing the instruction set for the accelerator. As computations on encrypted data lack data-dependent branches, I will make use of a VLIW-style architecture with explicit scheduling and data movement. I hope to provide insights that could also help accelerate FHE across a broader set of use-cases, e.g., on mobile devices or in data centers.

I’m participating in SuperUROP to gain experience leading a research project and presenting my results to others. I got particularly interested in computer architecture after taking the introductory 6.004 and the more advanced 6.823. I’m excited to work towards accelerating Fully Homomorphic Encryption as this could make practical privacy-preserving computations in the cloud.