Encrypted Computing Software Development Kit

Overview

Fully Homomorphic Encryption (FHE) allows computation to be performed on encrypted data without having to decrypt the data which brings in profound and beneficial implications for data privacy and data confidentiality. However, these benefits come with a significant performance cost which has so far confined FHE based applications to specific use case archetypes primarily in use by the regulated industries and government.

Most of the FHE schemes of today perform the computation using very large polynomial rings, thus requiring considerable compute power and data movement between main memory and the CPU's registers, which introduces large computational overheds. New FHE hardware accelerator architectures aim at accelerate the computation over large polynomial rings while minimizing the data movement between the memory and the compute elements.

The Encrypted Computing SDK introduces a multistage transformation (compiler) pipeline that breaks down the large polynomial computation into the various abstraction layers to make it easier for developers to target different hardware platform architectures as well as to develop new implementations of FHE schemes, and integrate with existing 3rd Party libraries, compilers and transpilers. The SDK is inspired by the LLVM Compiler Infrastructure, and adopts a modular approach based on language independent intermediate representations (IRs) that promotes the separation of concerns at each stage of the pipeline, allowing for dedicated transformations and optimizations.

Our first target hardware accelerator is Intel’s HERACLES accelerator technology which introduces a new Polynomial Data type which does not exist in today's traditional CPUs. For this new polynomial data type, it supports a new set of novel and fundamental instructions, the Polynomial Instructions Set Architecture (P-ISA), that operates directly on large polynomials in a SIMD fashion.