End-to-end NTT on FPGA accelerates FHE operations
Original title: End-to-End NTT on FPGA for FHE Acceleration
The design transforms 2^16 coefficients every 5.12 μs at 400 MHz and accelerates FHE by 1.8-2.2x.
FHE incurs high overhead in polynomial operations dominated by NTT/INTT. The streaming kernel is floorplanned across three dies of an AMD Alveo V80. It integrates into Lattigo for end-to-end FHE acceleration in key switching, bootstrapping, and neural network inference.
- Fully unrolled delay-commutator streaming NTT/INTT kernel
- 400 MHz throughput of 2^16 coefficients every 5.12 μs
- Up to 22x faster than CPU per transform
- 1.8-2.2x speedup in core FHE operations
- 1.8-2.1x for neural network inference
- Products
- AMD Alveo V80
From the original article
1 min read at arxiv.orgFully homomorphic encryption (FHE) enables privacy-preserving computation on encrypted data without decryption. Despite its strong security guarantees, FHE incurs substantial computational and memory overhead.
By Mohamed Allam · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.