FPGA Cluster Accelerates Homomorphic Encryption Relinearization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current homomorphic encryption techniques are slow due to high computation overhead, and existing FPGA-based accelerators like HEAX face resource limitations, leading to inefficient processing and limited scalability for large parameter sets.

Innovation Solution

A cluster of FPGAs is used to accelerate homomorphic encryption by breaking down processes into parallel operations across multiple FPGA devices, each specializing in inverse number-theoretic transform, number-theoretic transform, multiplication, and residue number system functions, with optimized communication protocols to enhance throughput and resource utilization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single FPGA is used for homomorphic encryption acceleration, then device complexity is reduced, but processing speed and scalability deteriorate

Engineering Contradiction:
ImproveFPGA configurationVSAvoidprocessing speed
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent divides the homomorphic encryption processing system into multiple separate FPGA devices, each dedicated to specific computational tasks (NTT, INTT, multiplication, etc.). This segmentation allows parallel processing across multiple devices, significantly improving processing speed while keeping each individual FPGA relatively simple in configuration.

Inventive Principle:
Principle #1Segmentation

2Productivity

If more FPGA resources are allocated to handle large parameter sets, then processing capability improves, but resource limitations and scalability issues worsen

Engineering Contradiction:
Improveprocessing capabilityVSAvoidresource utilization
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent transitions from a single-dimension approach (one FPGA handling all tasks) to a multi-dimensional distributed architecture where multiple FPGAs work in parallel. This dimensional change allows the system to handle large parameter sets (N>=8192, q>=200) by distributing the computational burden across multiple devices, effectively overcoming the resource limitations of a single FPGA.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Productivity

If parallel processing is implemented across multiple FPGAs, then throughput improves, but communication overhead and system complexity increase

Engineering Contradiction:
ImprovethroughputVSAvoidcommunication protocol
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces a host computer as an intermediary that coordinates the parallel processing across multiple FPGAs. The host manages data distribution, coordinates computational tasks, and aggregates results, thereby simplifying the communication protocol requirements between FPGAs while maintaining high throughput through efficient parallel execution.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS11764944B2Field-programmable gate array (FPGA) clusters and methods of using an FPGA cluster for homomorphic encryption acceleration
Publication Date: 2023.09.19 HUAWEI TECH CO LTD
  • US11764944B2 patent drawing
  • US11764944B2 patent drawing
  • US11764944B2 patent drawing

AI summary

A field-programmable gate array (FPGA) cluster, comprising a plurality of FPGA devices, can be used to accelerate homomorphic encryption functionality. In particular, the FPGA cluster can accelerate the relinearization process used in homomorphic encryption by using multiple FPGA devices to perform portions of the relinearization process in parallel. Further, the use of the FPGA cluster provides sufficient memory resources to allow data used by the relinearization process, namely the keyswitch keys, to be stored on-chip.