Integration of a hardware accelerator for CKKS (fully homomorphic encryption) into the NEORV32 CFS.

[Sam-Vervaeck]

Hey @stnolting! It has been a while since I posted a new discussion topic but here I am again. This may well be my final one since I am finished with my master's thesis project.

I recently uploaded my project code to GitHub to make it open-source available: NEORV32_AlohaHE_thesis. You can find a short description of how I used the NEORV32 in that repo. The CFS came in very handy!

I want to mention again how grateful I am for your support of the NEORV32, for being patient with new users and for answering all my questions. You also got a shout-out in the introduction of my thesis text :).

For the lazy among us I include the same short description here:

During my Master's thesis of 2023-2024 at the COSIC group at the KU Leuven, I had a topic on the CKKS algorithm for fully homomorphic encryption (FHE). I made a demo to show the integration of a hardware accelerator for CKKS and a RISC-V core. For the accelerator I chose the state-of-the-art Aloha-HE accelerator from Florian Krieger et al. at that time. For the RISC-V core I chose the NEORV32 by Stephan Nolting.

This repository contains a Vivado project showcasing a CKKS encryption demo. The Aloha-HE accelerator is integrated into the Custom Fuction Subsystem (CFS) of the NEORV32 and has an input and output channel that propagates all the way to the toplevel op the core. A block diagram contains the NEORV32 and a BRAM which stores the test data for the CKKS computations through a .coe file initialisation. The program execution is controlled by NEORV32 and the CKKS computations are performed by Aloha-HE by offloading the data to the CFS block containing the accelerator.

[stnolting]

Hey @Sam-Vervaeck.

Coincidentally, I'm also in the process of learning more about homomorphic encryption - so thank you very much for sharing this! 👍

Will your actual thesis be published as well? I would be really interested in reading it.

[Sam-Vervaeck]

I'll ask about the confidentiality.

[stnolting]

Cool, thanks!

[Sam-Vervaeck]

Hey @stnolting ,

I am happy to say I can share it with you, apparently I have full control over my text :). I will attach the PDF to this message.

A bit of context as well:

• I took out some unrelated KUL pages from the text.
• Sadly, I had to submit my text already at the end of August while I was still working on my final design. I got that better design working just in time for my thesis defense presentation in September but it is not included in this text. I added a snippet of my defense presentation that explains the final design and the final performance and area results.
• The best (most useful) parts of my text are the introduction into FHE and the CKKS algorithm if you are learning about those.

thesis_github.pdf
final_design.pdf