July 2022 Summaries

Concrete is an open-source framework designed to simplify the use of Fully Homomorphic Encryption (FHE) for developers without requiring in-depth cryptographic knowledge. It addresses the historical challenges of FHE, such as speed, usability, and computational limitations, by combining leveled and bootstrapped operations to support both approximate and exact evaluations of arbitrary functions on booleans and integers. Concrete introduces programmable bootstrapping and automates parameters and noise management, offering a high-level Rust frontend to abstract complexity while allowing fine-grained control through lower-level libraries if needed. The framework also features a compiler based on MLIR for optimizing FHE parameters, specifically supporting Python and Numpy, to enhance performance for machine learning applications. Despite improvements, Concrete aims to further reduce FHE latency and throughput with new computing platforms and anticipates the release of commercial FHE accelerators within two years, alongside continuous feature enhancements.

Concrete-core V1.0.0-gamma, the latest release in the development of Zama's low-level crypto library, introduces significant advancements towards the upcoming 1.0.0 release, notably adding Cuda acceleration for improved performance on Nvidia GPUs, which are prevalent in machine learning environments. This version includes a new Cuda backend dedicated to accelerating programmable bootstrap and keyswitch operations, multi-platform support through a restructured Core backend, and serialization support for entities, enabling better compatibility and performance across different hardware platforms. Additionally, two new APIs, a C API and a WASM API, have been introduced to extend the library's usability and facilitate a broader range of applications. The release aims to optimize user experience by integrating cost and noise models into Zama's Optimizer and Compiler, which will help users select optimal parameters and hardware configurations. The benchmarking results indicate superior performance of the Low Latency Bootstrap for small numbers of bootstraps and competitive performance of the Amortized Bootstrap for larger sets, largely benefiting from the new Cuda acceleration.

Zama has unveiled a significant update to its Homomorphic Encryption library, Concrete, marking an important step in its mission to make fully homomorphic encryption more accessible and efficient. This release includes the first official launch of the Concrete Framework, an open-source tool that allows developers to integrate homomorphic encryption into their applications without needing a deep understanding of cryptography. Additionally, the team has introduced version 1.0-gamma of Concrete Core, which is a crucial milestone towards the full release of Zama's low-level crypto library. This update features a new Cuda backend designed to enhance the performance of programmable bootstrap and keyswitch operations on Nvidia GPUs, along with other improvements.