BELFORT

Enabling Fully Homomorphic Encryption through Hardware Acceleration

a KU Leuven spin-off

Belfort FHE Accelerator

The most advanced TFHE accelerator, available! — To showcase our technology, we’ve developed a demo for Trivium-based transciphering application, ideal for blockchain applications. It seamlessly converts symmetric key ciphertexts to the FHE domain or provides encrypted randomness, enhancing efficiency and functionality on-chain.

HW Accelerated   1 FPGA
>300x

FASTER

SW Optimised   1 Thread

It is a Fully Fledged Accelerator

Integrated into TFHE-rs. Enable the FPGA — you are good to go. No special compiler is required! No fuss.


Supports any TFHE parameter set; both booleans and integers. Acceleration for TFHE-rs’ default parameter sets are already available. With one-click, we can provide versions for any other parameter set too.

Versatile. Acceleration through off-the-shelf AMD Alveo FPGAs. No need for custom hardware.

Can operate on both local servers and AWS, Azure cloud services, ensuring flexibility and accessibility.

Acceleration in Numbers

Here is a performance comparison for 64-bit integer multiplication.

Besides, FPGA are the most cost effective. You can check AWS Pricing Calculator for the most up to date cost comparison of renting these instances on cloud.

PlatformCPU / GPU / FPGA CountExecution Time
AMD EPYC9R14192366 ms
Intel Xeon Siver 4208 x2323754 ms
NVIDIA H1001456 ms
Belfort1559 ms
Belfort 4254 ms
(These measurements are taken over TFHE-rs 0.6. TheAMD CPU performance is obtained from TFHE-rs benchmarks. NVIDIA GPU case is calculated from the interpolation of acedemic works. The rest are the measurements taken on our development systems.)

Superior Technology

In-depth architectural optimization: We’ve meticulously optimized our architecture for low latency and high throughput, ensuring optimal performance.

Entirely compute-bound bootstrapping throughput: 100% utilization of arithmetic units and only small bootstrapping key cache

Noise-trimmed fixed-point representations: Our innovative patent-pending approach reduces representation size by up to 50% compared to prior floating-point or integer FFTs, maximizing efficiency without compromising accuracy.

Belfort FHE Accelerator is ready to speed up your FHE applications: It is now much more capable than its earlier published version.

Contact Us

We can schedule a demonstration of our technology.

Contact Us