unifying-von-neumann-hpc-neuromorphic-ebbrains

name	unifying-von-neumann-hpc-neuromorphic-ebbrains
description	统一冯诺依曼HPC与神经形态计算的EBRAINS工作流框架 - 透明跨平台执行SNN，支持异构架构无缝切换。
version	1.0.0
author	arXiv:2606.08515 (Krishna Kant Singh et al.)
date	2026-06-13T00:00:00.000Z
arxiv_id	2606.08515
activation_keywords	["neuromorphic computing","von neumann hpc","ebbrains","heterogeneous architecture","snn acceleration","workflow orchestration","brain-inspired computing","transparent execution"]

Unifying von-Neumann HPC and Neuromorphic Acceleration via EBRAINS Research Infrastructure

arXiv:2606.08515 - Submitted 7 June 2026

Authors: Krishna Kant Singh, Charl Linssen, Eric Müller, Eleni Mathioulaki, Wouter Klijn, Lena Oden

Categories: cs.DC

TL;DR

提出EBRAINS研究基础设施上的统一框架，实现冯诺依曼HPC与神经形态加速器的无缝集成，支持单一科学工作流在异构架构间透明执行。

Problem Statement

现代科学工作流日益跨越多样计算架构:

• 冯诺依曼HPC: 高精度数值模拟
• 神经形态加速器: 低功耗、实时SNN推理
• 执行单一工作流需要跨架构适配
• 缺乏统一调度和资源管理框架

Core Contribution

EBRAINS统一框架:

1. 透明跨平台执行: SNN模型自动适配不同硬件
2. 工作流编排: 混合架构任务调度
3. 资源抽象层: 统一API隐藏底层差异
4. 性能优化: 架构特定优化策略

Key Methodology

1. Architecture Abstraction

Workflow Definition
├── Task Graph (DAG)
│   ├── HPC Tasks (传统计算)
│   ├── Neuromorphic Tasks (SNN推理)
│   └── Hybrid Tasks (混合执行)
└── Resource Mapping
    ├── CPU/GPU Clusters (HPC)
    ├── SpiNNaker/BrainScaleS (神经形态)
    └── Dynamic Allocation

2. Transparent Execution Layer

• 自动任务分派
• 数据格式转换
• 异构通信管理
• 性能监控

3. EBRAINS Infrastructure Integration

• SpiNNaker neuromorphic boards
• BrainScaleS analog neuromorphic
• HPC cluster nodes
• Unified scheduling API

Key Features

Seamless SNN Deployment

# Conceptual workflow
workflow = EBRAINSWorkflow()

# Define SNN model (architecture-agnostic)
snn = SpikingModel(...)

# Transparent execution
workflow.run(
    model=snn,
    backend='auto',  # Auto-select: HPC/neuromorphic
    optimization='power',  # 或 'speed', 'accuracy'
)

Dynamic Resource Allocation

• 任务特征分析 (计算密集 vs 通信密集)
• 架构匹配 (神经形态 vs 冯诺依曼)
• 运行时迁移 (负载均衡)

Performance Metrics

• 能耗效率: 神经形态优势
• 计算吞吐: HPC优势
• 实时性: 混合优化

Implications for Neuroscience Research

Large-Scale Brain Simulation

• 百万神经元网络模拟
• 实时行为验证
• 低功耗长期运行

Hybrid Simulation-Experiment

• HPC预处理 + 神经形态实时推理
• 闭环神经接口
• 边缘计算部署

Research Workflow Modernization

• 单一框架覆盖全研究生命周期
• 降低架构切换成本
• 促进跨学科协作

Technical Details

Supported Platforms

• HPC: CPU集群、GPU加速、云平台
• Neuromorphic: SpiNNaker2, BrainScaleS-2, Intel Loihi
• Hybrid: FPGA加速、边缘设备

Data Flow Management

• 异构架构间数据传输
• 格式转换 (浮点 → 脉冲编码)
• 带宽优化 (压缩、缓存)

Fault Tolerance

• 任务级容错
• 架构切换备用
• 结果验证机制

Future Directions

1. 自适应架构选择 (基于任务特征)
2. 能效优化调度算法
3. 更多神经形态硬件支持
4. 实时工作流监控界面

Related Work

• SpiNNaker neuromorphic platform
• BrainScaleS analog neuromorphic
• NEST brain simulation -神经形态边缘计算框架

---

References

• Singh et al. (2026). arXiv:2606.08515
• EBRAINS: https://ebrains.eu/
• SpiNNaker: https://spinnakermanchester.github.io/