Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Event-based Neural Spike Detection Using Spiking Neural Networks for Neuromorphic iBMI Systems

Chanwook Hwang, Biyan Zhou, Ye Ke, Vivek Mohan, Jong Hwan Ko, Arindam Basu

TL;DR

This work tackles the data-rate and power constraints of wireless, high-channel-count iBMIs by introducing a Spiking Neural Network Spike Detector (SNN-SPD) that operates directly on sparse event streams generated via delta modulation and pulse count modulation. The model uses a two-layer LIF-based SNN with no spike reset, trained with a surrogate gradient, and evaluated in Non-Stream and Stream modes to exploit temporal features. It achieves a peak spike-detection accuracy of $A=95.72\%$ at noise level $\sigma=0.2$, while requiring only $0$ multiplications, about $25.23$ accumulations, and $418$ weight parameters (roughly $26.6\%$ of the ANN-SPD), translating to substantial on-implant efficiency and data compression ($\sim361.96\times$ vs raw data). These results demonstrate a favorable balance between performance and hardware efficiency, enabling real-time, neuromorphic iBMI processing with reduced bandwidth and power, and paving the way for future hardware implementations and validations with real neural recordings.

Abstract

Implantable brain-machine interfaces (iBMIs) are evolving to record from thousands of neurons wirelessly but face challenges in data bandwidth, power consumption, and implant size. We propose a novel Spiking Neural Network Spike Detector (SNN-SPD) that processes event-based neural data generated via delta modulation and pulse count modulation, converting signals into sparse events. By leveraging the temporal dynamics and inherent sparsity of spiking neural networks, our method improves spike detection performance while maintaining low computational overhead suitable for implantable devices. Our experimental results demonstrate that the proposed SNN-SPD achieves an accuracy of 95.72% at high noise levels (standard deviation 0.2), which is about 2% higher than the existing Artificial Neural Network Spike Detector (ANN-SPD). Moreover, SNN-SPD requires only 0.41% of the computation and about 26.62% of the weight parameters compared to ANN-SPD, with zero multiplications. This approach balances efficiency and performance, enabling effective data compression and power savings for next-generation iBMIs.

Event-based Neural Spike Detection Using Spiking Neural Networks for Neuromorphic iBMI Systems

TL;DR

This work tackles the data-rate and power constraints of wireless, high-channel-count iBMIs by introducing a Spiking Neural Network Spike Detector (SNN-SPD) that operates directly on sparse event streams generated via delta modulation and pulse count modulation. The model uses a two-layer LIF-based SNN with no spike reset, trained with a surrogate gradient, and evaluated in Non-Stream and Stream modes to exploit temporal features. It achieves a peak spike-detection accuracy of at noise level , while requiring only multiplications, about accumulations, and weight parameters (roughly of the ANN-SPD), translating to substantial on-implant efficiency and data compression ( vs raw data). These results demonstrate a favorable balance between performance and hardware efficiency, enabling real-time, neuromorphic iBMI processing with reduced bandwidth and power, and paving the way for future hardware implementations and validations with real neural recordings.

Abstract

Implantable brain-machine interfaces (iBMIs) are evolving to record from thousands of neurons wirelessly but face challenges in data bandwidth, power consumption, and implant size. We propose a novel Spiking Neural Network Spike Detector (SNN-SPD) that processes event-based neural data generated via delta modulation and pulse count modulation, converting signals into sparse events. By leveraging the temporal dynamics and inherent sparsity of spiking neural networks, our method improves spike detection performance while maintaining low computational overhead suitable for implantable devices. Our experimental results demonstrate that the proposed SNN-SPD achieves an accuracy of 95.72% at high noise levels (standard deviation 0.2), which is about 2% higher than the existing Artificial Neural Network Spike Detector (ANN-SPD). Moreover, SNN-SPD requires only 0.41% of the computation and about 26.62% of the weight parameters compared to ANN-SPD, with zero multiplications. This approach balances efficiency and performance, enabling effective data compression and power savings for next-generation iBMIs.
Paper Structure (17 sections, 5 equations, 2 figures, 2 tables)

This paper contains 17 sections, 5 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Event Data Generation
  4. Delta Modulation
  5. Pulse Count Modulation (PCM)
  6. Spiking Neural Network Spike Detector (SNN-SPD)
  7. SNN modes
  8. Non-Stream
  9. Stream
  10. Results
  11. Dataset
  12. Evaluation Metrics
  13. Detection Performance
  14. Operation Efficiency
  15. Detection Performance Comparison
  16. ...and 2 more sections

Figures (2)

  • Figure 1: (a) Spiking neural network spike detector (SNN-SPD) detects spikes from a pulse count signal generated by the iBMI sensors. (b) Delta Modulation employed for the event data generation, data is modulated to $\pm$1 when the amplitude change is greater than a threshold. (c) Pulse count modulation (PCM) is used to accumulate delta modulation values for each period $T_s$. (d) SNN-SPD detects the spikes from the pulse count signal array using a sliding window.
  • Figure 2: Spike detection performance comparison between PCM, APM, Ev-SPD, ANN-SPD, and SNN-SPDs (Stream and Non-Stream)