Implementing an FPGA Accelerator w/Coherent Memory Access in Apache Arrow
summary
Prototype implements a Gentry-style homomorphic encryption pipeline on an FPGA and focuses on cutting data-transfer overhead by integrating Fletcher with the in-memory columnar format Apache Arrow and using IBM’s CAPI SNAP for coherent main-memory access that removes a device driver and format-copy steps. The FPGA hosts encryption/evaluation kernels to accelerate noisy integer arithmetic; current status is early testing of simple kernels and end-to-end data paths rather than a full, bootstrapped FHE system.
AI-written summary · about
chapters (6)
- 0:00Introduction and background
- 1:07Homomorphic encryption basics
- 6:12Gentry's blueprint explained
- 13:27FPGA acceleration for HE
- 15:39CAPI and Apache Arrow benefits
- 17:27Current project status
related talks
56mBuilding a compiler and processor for fast processing of encrypted data
20mConfidential Computing within an AI Accelerator
17mFHE Rollups and Co-Processor in Action
28mGenerative AI + Proprietary Data
30mEncrypting Memory at Scale
09mFormally Secure Hardware with Rust
41mMiguel de Vega, Ravital Solomon, Remi Gai & Pascal Paillier I Pragma Denver 2024
15mOsprey: Transparent and Efficient Virtual Memory for Secure Computation