Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Architectural Exploration of Hybrid Neural Decoders for Neuromorphic Implantable BMI

Vivek Mohan, Biyan Zhou, Zhou Wang, Anil Bharath, Emmanuel Drakakis, Arindam Basu

TL;DR

This work addresses the challenge of decoding neural signals for dense, wireless Neu-iBMI systems by employing an event-based neural sensing pipeline with a tunable EvFilter that dramatically reduces processing load. It systematically explores hybrid decoders—NN, Segmented Time-bins NN (ST-NN), and SNN-based decoders—operating directly on NCNS-generated events, with a high-performing LSTM baseline using EvFilter-SPD inputs achieving up to $R^2 \\approx 0.73$. The results show that the SNN-Decoder can reach $R^2 \\\approx 0.70$ while consuming 5–23× fewer compute and memory resources than ANN- or LSTM-based decoders, and the ST-NN-Decoder offers comparable performance with finer temporal resolution. Overall, the proposed pipeline eliminates spike recovery, detection, and sorting stages, enabling low-power, wireless, edge-based iBMI implementations with high channel counts and practical deployment prospects on neuromorphic hardware like Loihi or DYNAP-SE.

Abstract

This work presents an efficient decoding pipeline for neuromorphic implantable brain-machine interfaces (Neu-iBMI), leveraging sparse neural event data from an event-based neural sensing scheme. We introduce a tunable event filter (EvFilter), which also functions as a spike detector (EvFilter-SPD), significantly reducing the number of events processed for decoding by 192X and 554X, respectively. The proposed pipeline achieves high decoding performance, up to R^2=0.73, with ANN- and SNN-based decoders, eliminating the need for signal recovery, spike detection, or sorting, commonly performed in conventional iBMI systems. The SNN-Decoder reduces computations and memory required by 5-23X compared to NN-, and LSTM-Decoders, while the ST-NN-Decoder delivers similar performance to an LSTM-Decoder requiring 2.5X fewer resources. This streamlined approach significantly reduces computational and memory demands, making it ideal for low-power, on-implant, or wearable iBMIs.

Architectural Exploration of Hybrid Neural Decoders for Neuromorphic Implantable BMI

TL;DR

This work addresses the challenge of decoding neural signals for dense, wireless Neu-iBMI systems by employing an event-based neural sensing pipeline with a tunable EvFilter that dramatically reduces processing load. It systematically explores hybrid decoders—NN, Segmented Time-bins NN (ST-NN), and SNN-based decoders—operating directly on NCNS-generated events, with a high-performing LSTM baseline using EvFilter-SPD inputs achieving up to . The results show that the SNN-Decoder can reach while consuming 5–23× fewer compute and memory resources than ANN- or LSTM-based decoders, and the ST-NN-Decoder offers comparable performance with finer temporal resolution. Overall, the proposed pipeline eliminates spike recovery, detection, and sorting stages, enabling low-power, wireless, edge-based iBMI implementations with high channel counts and practical deployment prospects on neuromorphic hardware like Loihi or DYNAP-SE.

Abstract

This work presents an efficient decoding pipeline for neuromorphic implantable brain-machine interfaces (Neu-iBMI), leveraging sparse neural event data from an event-based neural sensing scheme. We introduce a tunable event filter (EvFilter), which also functions as a spike detector (EvFilter-SPD), significantly reducing the number of events processed for decoding by 192X and 554X, respectively. The proposed pipeline achieves high decoding performance, up to R^2=0.73, with ANN- and SNN-based decoders, eliminating the need for signal recovery, spike detection, or sorting, commonly performed in conventional iBMI systems. The SNN-Decoder reduces computations and memory required by 5-23X compared to NN-, and LSTM-Decoders, while the ST-NN-Decoder delivers similar performance to an LSTM-Decoder requiring 2.5X fewer resources. This streamlined approach significantly reduces computational and memory demands, making it ideal for low-power, on-implant, or wearable iBMIs.
Paper Structure (15 sections, 6 equations, 3 figures, 2 tables)

This paper contains 15 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Event Filter
  4. Neural Decoders for Neuromorphic iBMI
  5. NN-Decoder
  6. Segmented Time-bins NN-Decoder (ST-NN-Decoder)
  7. SNN-Decoder
  8. Results
  9. Dataset
  10. Evaluation Metrics
  11. Decoding Performance
  12. Decoder Model Training and Testing
  13. Decoder Performance Analysis
  14. Hardware Implications for Decoder
  15. Conclusion

Figures (3)

  • Figure 1: (a) Neural decoding pipeline for NCNS-based (ISCAS2023NCE2025) Neu-iBMI. Events from NCNS are passed through a spike detector or event filter, which allows events corresponding to the spike. Event-based feature extraction involves binning events for some ANN-based decoder models, while filtered event data or spikes may be directly streamed through SNN-Decoder. $\mathrm{R^2}$ score is the performance metric used. (b) Logical block diagram of an event filter for NCNS background activity suppression. A spike detector can be realized as a special case of the event filter by choosing an appropriate period of refraction.
  • Figure 2: Input feature extraction for (a) NN-Decoder and (b) ST-NN-Decoder (c) SNN-Decoder (right: network architecture).
  • Figure 3: Compression ratio improvement with EvFilter/EvFilter-SPD.