CoChat

HiAER-spike software-hardware reconfigurable platform for event-driven neuromorphic computing at scale

Open access

The DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index; 5 of the 12 sources its claims cite were checked for support.

Abstract

In this work, we present HiAER-Spike, a modular, reconfigurable, event-driven neuromorphic computing platform designed to execute large spiking neural networks with up to 160 million neurons and 40 billion synapses - roughly twice the neurons of a mouse brain at faster than real time. This system, assembled at the UC San Diego Supercomputer Center, comprises a co-designed hard- and software stack that is optimized for run-time massively parallel processing and hierarchical address-event routing (HiAER) of spikes while promoting memory-efficient network storage and execution. The architecture efficiently handles both sparse connectivity and sparse activity for robust and low-latency event-driven inference for both edge and cloud computing. A Python programming interface to HiAER-Spike, agnostic to hardware-level detail, shields the user from complexity in the configuration and execution of general spiking neural networks with minimal constraints in topology. The system is made easily available over a web portal for use by the wider community. In the following, we provide an overview of the hard- and software stack, explain the underlying design principles, demonstrate some of the system’s capabilities, and solicit feedback from the broader neuromorphic community. Examples are shown demonstrating HiAER-Spike’s capabilities for event-driven vision on benchmark CIFAR-10, DVS event-based gesture, MNIST, and Pong tasks.

Overview

CoChat Summary

TL;DR

HiAER-Spike is a modular, reconfigurable FPGA-based event-driven neuromorphic platform hosted at the San Diego Supercomputer Center and accessible via the Neuroscience Gateway, designed to scale to 160 million neurons and 40 billion synapses. Initial single-core benchmarks demonstrate conversion and execution of MLPs, LeNet-5, and spiking CNNs on MNIST, DVS Gesture, CIFAR-10, and Pong with low reported energy and latency.

Key findings

  • Planned full system: 40 FPGAs, each intended to support up to 4 million neurons and 1 billion synapses, yielding up to 160 million neurons and 40 billion synapses total — more than twice the neurons of a typical mouse brain and at faster-than-real-time simulation speed. The paper’s experiments use only a single core on one FPGA.
  • MNIST: all LeNet-5 variants and the larger MLP variant achieved over 97% test accuracy; all MLP and LeNet-5 variants with input size (1, 28, 28) achieved identical PyTorch and HiAER-Spike test accuracies. Table 3 reports HiAER-Spike at 138 neurons: 96.59% accuracy, 1.1 μJ, 4.2 μs; and at 5814 neurons: 98.14% accuracy, 17.1 μJ, 48.6 μs.
  • DVS Gesture: converted HiAER-Spike versus post-quantization SpikingJelly test accuracies ranged from perfect match to a 1.74% discrepancy. The largest deployed spiking CNN had 109,615 neurons, occupying approximately 86% of a single-core system and approximately 0.07% of anticipated maximum system capacity. Table 4 reports HiAER-Spike at 1115 neurons: 54.51% accuracy, 79.8 μJ, 184.9 μs; and at 17,709 neurons: 68.75% accuracy, 510.7 μJ, 1156.2 μs.
  • CIFAR-10 and Pong: CIFAR-10 HiAER-Spike accuracy was 80.49% versus 82.56% software, with 4770.7 μJ and 10,508.5 μs; Pong score was 20.36 versus 20.74 software, with 149.3 μJ and 425.7 μs. For DVS Gesture spiking CNNs, scaling regressions were Energy (μJ) = 0.0294x − 30.293 (R² = 0.994) and Latency (μs) = 0.0658x − 53.031 (R² = 0.995), where x is neuron count, with n = 5.

Methods

  • Design: modular, reconfigurable, event-driven neuromorphic platform using FPGA-based computing architecture, hierarchical address-event routing (HiAER), and a co-designed hardware/software stack. Synaptic connectivity is stored in HBM using adjacency lists; spike events and membrane potentials use on-chip BRAM/URAM. The software interface is implemented in C++ and Python via the `hs_api` library and is accessible through the Neuroscience Gateway portal.
  • Data/sample: trained MLPs, LeNet-5 variants, and spiking CNNs using PyTorch and SpikingJelly on binarized MNIST, DVS Gesture, CIFAR-10, and a DVS-based Pong task. Networks were quantized to 16-bit integers and converted to run on a single core of one HiAER-Spike FPGA card.
  • Analysis: compared software and HiAER-Spike inference accuracy; approximated energy from the number of HBM accesses reported by the FPGA and latency from FPGA-reported clock cycles; performed linear regressions for energy and latency scaling within a single core.

Limitations

  • Stated: initial benchmarks use only a single operational core on one FPGA; accuracy generally lower than other platforms; the primary goal was to test and demonstrate the hardware design, software interface, and conversion methods rather than optimize accuracy; DVS Gesture used binarized and downsampled inputs of (2, 63, 63) and (2, 90, 90) rather than full (2, 128, 128) inputs; the frame accumulation method may be improvable; scaling regressions are limited to a small sample size (n = 5); hardware microarchitecture details are deferred to a future manuscript.
  • Evident from the text: no multi-core, multi-FPGA, or full-system scaling results are reported; cross-platform comparisons differ in model size, number of time steps, and input preprocessing, so accuracy, energy, and latency comparisons are not controlled.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s44335-026-00062-8

Cites

20 of 20 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 High-level system architecture of HiAER-Spike. a Neurons and synapses form the ' gray matter ' of dense local interconnects in the system while long-range connections (between cores and FPGAs) form the ' white matter ' of sparse global interconnects. b The hardware equivalent of a as implemented in our multi-core architecture on the FPGA. The gray matter inside each core is implemented as sequentially updated integrate-and- fi re neurons, whose internal state is stored in neural membrane registers in URAM, whereas spike events are routed through synaptic look-up tables in HBM. The white matter is implemented as a hierarchical multicast bus (HiAER) interconnecting axon spike register modules that are stored in BRAM across cores.
Fig. 1 High-level system architecture of HiAER-Spike. a Neurons and synapses form the ' gray matter ' of dense local interconnects in the system while long-range connections (between cores and FPGAs) form the ' white matter ' of sparse global interconnects. b The hardware equivalent of a as implemented in our multi-core architecture on the FPGA. The gray matter inside each core is implemented as sequentially updated integrate-and- fi re neurons, whose internal state is stored in neural membrane registers in URAM, whereas spike events are routed through synaptic look-up tables in HBM. The white matter is implemented as a hierarchical multicast bus (HiAER) interconnecting axon spike register modules that are stored in BRAM across cores.From the PDF
Fig. 2 Core Architecture. a Host programming interface with SNN compiler and low-level hardware interface. b Heterogeneous memory organization within a single core, as well as the off-chip HBM storing the synaptic connectivity table. Onchip URAM and BRAM store state variables of axons and neurons. The top-left panel shows the layout of the data structure in HBM, supporting parallelism of 16 neurons per single core.
Fig. 2 Core Architecture. a Host programming interface with SNN compiler and low-level hardware interface. b Heterogeneous memory organization within a single core, as well as the off-chip HBM storing the synaptic connectivity table. Onchip URAM and BRAM store state variables of axons and neurons. The top-left panel shows the layout of the data structure in HBM, supporting parallelism of 16 neurons per single core.From the PDF
Image 3
From the PDF
Fig. 4 DVS gesture SNN size and quantization. Test accuracy of the DVS gesture network compared across different model sizes and full-precision software models, quantized software models, and hardware.
Fig. 4 DVS gesture SNN size and quantization. Test accuracy of the DVS gesture network compared across different model sizes and full-precision software models, quantized software models, and hardware.From the PDF
Image 5
From the PDF
Image 6
From the PDF

Verification

Claims the paper makes

3 Valid · 2 Not valid · 5 Not checked · 48 Own results and interpretations

14 more claims

Create a free account to see all 58 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 24
  • It may state two separate claims as one. 22

Create a free account to see these claims.

Citations

Cites

Show all 20 references

Research works printed but not linked: 2

  1. Davies, M. et al. Taking neuromorphic computing to the next level with Loihi2. Intel. Labs’ Loihi 2, 1–7 (2021).
  2. Esser, S. K., Appuswamy, R., Merolla, P., Arthur, J. V. & Modha, D. S. Backpropagation for energy-efficient neuromorphic computing. Adv. Neural Inform. Process. Syst. 28, (2015).

Metadata

Authors
Frank, G., Hota, G., Wang, K., Deng, C., Arora, K., Vins, D., Uppal, A., Olajide, O., Yoshimoto, K., Wang, Q., Yamaoka, M., Leugering, J., Deiss, S., Gibb, L., & Cauwenberghs, G.
Affiliations
  • University of California San Diego (US)
  • Fujitsu (Japan) (JP)
  • Forschungszentrum Jülich (DE)
Corresponding author
  • Frank, Gwenevere
  • Gibb, Leif
  • Cauwenberghs, Gert
Venue
npj Unconventional Computing
Published
2026-05-04
Volume
3
Issue
1
ISSN
3004-8672 (ISSN-L)
OpenAlex
W7160205547
Cited by
0
References
17
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Unconventional Computing (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by-nc-sa)
  • arXiv (Cornell University)
Funders
  • National Science Foundation (NS-1823366, CCF-2208771)
  • Office of Naval Research (N00014-24-1-2127)
Topics
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
  • Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
  • Physical Sciences › Mathematics › Theoretical Computer Science › DNA and Biological Computing
Keywords
neuromorphic computing, spiking neural networks, event-driven computing, reconfigurable hardware platform, hardware-software co-design, sparse connectivity, event-based vision, web portal

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: HiAER-spike software-hardware reconfigurable platform for event-driven neuromorphic computing at scale. https://app.cochat.ai/papers/10.1038/s44335-026-00062-8
The paper
Frank, G., Hota, G., Wang, K., Deng, C., Arora, K., Vins, D., Uppal, A., Olajide, O., Yoshimoto, K., Wang, Q., Yamaoka, M., Leugering, J., Deiss, S., Gibb, L., & Cauwenberghs, G. (2026). HiAER-spike software-hardware reconfigurable platform for event-driven neuromorphic computing at scale. npj Unconventional Computing. https://doi.org/10.1038/s44335-026-00062-8

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.