Skip to main content
QUICK REVIEW

[论文解读] Energy-Efficient Neuromorphic Computing for Edge AI: A Framework with Adaptive Spiking Neural Networks and Hardware-Aware Optimization

Olaf Yunus Laitinen Imanov, Derya Umut Kulali|arXiv (Cornell University)|Feb 2, 2026
Advanced Memory and Neural Computing被引用 0
一句话总结

NeuEdge 通过结合混合时序脉冲编码、硬件感知的协同优化和自适应阈值,提供在多个人神经形态平台上实现高精度与实时延迟的能效边缘AI。

ABSTRACT

Edge AI applications increasingly require ultra-low-power, low-latency inference. Neuromorphic computing based on event-driven spiking neural networks (SNNs) offers an attractive path, but practical deployment on resource-constrained devices is limited by training difficulty, hardware-mapping overheads, and sensitivity to temporal dynamics. We present NeuEdge, a framework that combines adaptive SNN models with hardware-aware optimization for edge deployment. NeuEdge uses a temporal coding scheme that blends rate and spike-timing patterns to reduce spike activity while preserving accuracy, and a hardware-aware training procedure that co-optimizes network structure and on-chip placement to improve utilization on neuromorphic processors. An adaptive threshold mechanism adjusts neuron excitability from input statistics, reducing energy consumption without degrading performance. Across standard vision and audio benchmarks, NeuEdge achieves 91-96% accuracy with up to 2.3 ms inference latency on edge hardware and an estimated 847 GOp/s/W energy efficiency. A case study on an autonomous-drone workload shows up to 312x energy savings relative to conventional deep neural networks while maintaining real-time operation.

研究动机与目标

  • 通过利用神经形态计算解决边缘AI的能耗与延迟瓶颈。
  • 开发一个联合优化编码、训练、映射和运行时自适应的SNN整体框架。
  • 在多个人神经形态平台上进行真实世界部署并进行全面的能量测量。
  • 展示硬件感知的协同优化,最大化利用率同时最小化能量。

提出的方法

  • 引入混合时序编码,将速率与精确定时脉冲相结合,以将脉冲数量降低4.7倍。
  • 提出硬件感知网络设计器,联动优化网络拓扑和片上映射以实现高核心利用率。
  • 开发自适应阈值机制,根据输入统计动态调整神经元发放阈值。
  • 提供端到端的 NeuEdge 训练算法,结合代理梯度和硬件感知的损失组件。
  • 在Intel Loihi 2、IBM TrueNorth、树莓派 4 和 NVIDIA Jetson Nano 上进行能量与延迟分析的验证。

实验结果

研究问题

  • RQ1混合时序脉冲编码是否能在边缘神经形态硬件上提升特征表征同时降低脉冲数量?
  • RQ2如何协同优化网络设计与硬件映射以最大化芯片利用率、最小化跨核心通信?
  • RQ3自适应阈值在低活动边缘场景下是否显著降低能耗且不牺牲精度?
  • RQ4在多个人神经形态平台与边缘设备上的真实世界能耗与延迟收益有哪些?
  • RQ5NeuEdge 在边缘设备上能在视觉和音频任务上达到多大程度的先进效率?

主要发现

MethodPlatformAccuracy (%)Latency (ms)Power (mW)Energy/Inf (mJ)Efficiency (GOp/s/W)
NeuEdgeLoihi 292.44.22871.21412
Standard SNNLoihi 288.78.93803.38127
ANN-SNNLoihi 291.212.34105.0498.3
Hybrid encoding (CIFAR-10 baseline)Jetson Nano92.118.4342062.912.4
Quantized DNNRaspberry Pi91.347.2184086.85.8
DVS Gesture — Standard SNNLoihi 294.83.73241.20284
DVS Gesture — ANN-SNNLoihi 295.95.13561.82218
DVS Gesture — NeuEdgeLoihi 296.72.32410.55847
MobileNetV2Jetson Nano94.714.2298042.318.9
Standard SNN — TrueNorthTrueNorth91.46.8780.53312
NeuEdge — TrueNorthTrueNorth93.24.1670.27524
  • NeuEdge 在视觉与音频任务上实现了91-96%的准确率。
  • 能效达到847 GOp/s/W,边缘端延迟为2.3 ms。
  • 在 Loihi 2 上,硬件利用率提升至89%(相比 naive 映射的47%),突触记忆提升至78%。
  • 混合编码将脉冲数量减少4.7x(CIFAR-10 基线从4.8M降至1.9M)。
  • 自适应阈值在低活动场景下将功耗降低约67%。
  • NeuEdge 相对于 GPU 基线在边缘CPU上实现高达312x 的能耗提升,且相对于传统神经网络达到89x 的提升。

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。