Pith. sign in

REVIEW 5 major objections 5 minor 53 references

SpikeX: Exploring Accelerator Architecture and Network-Hardware Co-Optimization for Sparse Spiking Neural Networks

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read SpikeX co-designs a sparse-SNN accelerator with its network, cutting energy-delay product by 15x to 150x at equal accuracy.

desk verdict Solid incremental extension of PTB with a serious unaddressed Vmem carry gap that undermines the headline EDP numbers until fixed. read the letter →

arxiv 2505.12292 v1 pith:C5YTEMLS submitted 2025-05-18 cs.NE cs.AIcs.AR

Boxun Xu , Richard Boone , Peng Li This is my paper · ORCID
classification cs.NEcs.AIcs.AR
keywords spikingneuralnetworkssystolicarrayacceleratorhardware-awaretrainingnetwork-hardwareco-designenergy-delayproductsparsityneuromorphichardwarearchitecturesearch
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that spiking neural network inference can be made much cheaper by co-designing the network and the accelerator rather than optimizing either in isolation. It proposes SpikeX, a systolic-array accelerator with two sparsity-exploiting mechanisms: agile dispatch of spatio-temporal workload units and activation-induced weight tailoring that skips weights for silent inputs. On top of the hardware, it adds a hardware-aware training loss $\mathcal{L}_{\mathrm{HW}}$ that penalizes an energy-delay-product proxy, and a hardware architecture search that picks the time-window size per layer during training. The central quantitative claim is a 15.1x to 150.87x reduction in energy-delay product over the parallel-time-batching baseline without accuracy loss, with up to 99% latency and 96% energy reduction. If correct, this makes low-power real-time SNN deployment substantially more practical, and it suggests that SNN hardware gains come as much from training and search as from the chip itself.

What carries the argument

The central object is the Neuro-Temporal Work Unit (NTWU), the unit of work assigned to one processing element: synaptic integration for one post-synaptic neuron over one time window of TWS timesteps. Activity tags at time-window, time-block, and time-stride granularity mark which units contain any spikes, so zero-tag units and their associated weights are skipped. Three levels of weight reuse—inside a time window, across time windows for the same neuron, and across post-synaptic neurons sharing a filter—are realized by the agile spatio-temporal dispatch scheduler, which switches between high-temporal and high-spatial density modes. The co-optimization side is carried by a differentiable hardware loss $\mathcal{L}_{\mathrm{HW}}(W)=\mathrm{EDP}(\mathrm{Sp}(W))$, a piecewise-linear fit of simulated EDP to the time-window sparsity $\mathrm{Sp}$, plus a hypernet that relaxes discrete time-window-size choices into continuous selection probabilities so gradient-based architecture search can pick per-layer TWS.

What would settle it

Implement SpikeX and the PTB baseline in RTL or on an FPGA with identical memory hierarchy, array size, and energy model, run DVS-Gesture Medium/Large and N-MNIST, and measure EDP; if the ratio is not within 15.1x–150.87x, the claim is refuted. A cheaper falsification is to recompute the final EDP comparison after re-fitting the piecewise-linear hardware-loss model directly to full cycle-level simulation traces without the sparsity proxy; if the layerwise EDP estimates shift enough to shrink the margin below 15.1x, the proxy is the weak link.

Watch

Extended reading notes

Core claim

The paper's central claim is that unstructured spatiotemporal spike sparsity can be turned into a scheduling resource instead of being treated as an obstacle. Because well-trained SNNs fire rarely and in clusters, the paper packs time into time windows and neurons into neuro-temporal work units (NTWUs), then dispatches these units across a systolic array so the same multi-bit weights are reused within a window, across windows, and across post-synaptic neurons. Zero-tag work units are skipped, and weights whose input channels contain no spikes are never loaded. The paper further claims that training the network with a hardware-loss term—a piecewise-linear model of simulated EDP as a function of time-window sparsity—and searching over per-layer time-window sizes yields the same model accuracy with 15.1x to 150.87x better EDP than the PTB baseline on DVS-Gesture (medium and large) and N-MNIST.

Load-bearing premise

The load-bearing premise is that the cycle-level simulator, with its 8x8 systolic array, 54KB global buffer, 2KB local buffer, and CACTI 32nm energy models, measures both SpikeX and the PTB baseline faithfully; because the hardware-loss function is fitted to that same simulator, any error in modeled memory energy, latency, or tag overhead flows directly into the claimed 15.1x–150.87x EDP reduction.

Editorial extensions

If this is right

  • SpikeX with hardware-aware training and architecture search reports 24.38x, 4.91x, and 2.07x energy reduction and 10.29x, 5.19x, and 7.25x latency reduction over the PTB baseline on DVS-Gesture Medium, DVS-Gesture Large, and N-MNIST, respectively.
  • Hardware-aware training alone increases time-window sparsity by 11.3% to 29.9% and reduces high-density neurons by 11.2x to 33.2x while preserving accuracy, meaning sparsity learned during training is a large part of the gain.
  • Per-layer time-window size is not a fixed optimum: the search picks different TWS values by layer and by input sparsity, and the paper shows the chosen TWS moves the energy/latency balance, sometimes trading a slight energy rise for a 14.2% latency drop.
  • The co-optimization techniques are architecture-general: applying hardware-aware training alone to the PTB baseline also improves it, so the methodology is not tied to SpikeX's specific datapath.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension not in the paper: use the same sparsity-proxy hardware loss to search over other reconfigurable accelerator parameters, such as bit-width, buffer capacity, or array shape, since the $\mathcal{L}_{\mathrm{HW}}$ interface only requires a differentiable EDP-vs-sparsity relation.
  • The dependence on clustered temporal sparsity suggests the method could transfer to other event-stream workloads—neuromorphic sensors, event cameras, or sparse graph activity—that can be packed into time windows, even outside SNN layers.
  • The paper's simulator is the only evidence for the headline EDP range; without RTL or silicon, a sympathetic reader should treat the 15.1x–150.87x figures as an estimate that physical implementation overheads in tag logic and scheduling are likely to erode.
  • A concrete next experiment the paper does not report: fit $\mathcal{L}_{\mathrm{HW}}$ to EDP measurements from an FPGA implementation rather than the simulator and check whether the learned per-layer TWS values change; if they shift, hardware-in-the-loop training would be needed.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. This manuscript proposes SpikeX, a systolic-array SNN accelerator, together with a network/hardware co-optimization methodology. The architecture is built around neuro-temporal work units (NTWUs), hierarchical activity tags, three levels of weight reuse, and activation-induced weight tailoring. The optimization side introduces a hardware-aware training loss LHW(W)=EDP(Sp(W)) based on a piecewise-linear fit to simulated EDP, and a hardware architecture search (SpikeX-HAS) over layer-wise time-window sizes. The paper claims a 15.1x to 150.87x reduction in energy-delay-product (EDP) over the PTB baseline, up to 99% latency reduction and 96% energy reduction, without compromising model accuracy, based on a cycle-level simulator with CACTI 32nm models and evaluated on DVS-Gesture and NMNIST.

Significance. If the claims are substantiated, the paper would be a useful contribution to SNN accelerator design: it targets the under-explored problem of unstructured spatiotemporal sparsity, proposes a concrete weight-reuse dataflow for binary activations and multi-bit weights, and introduces a differentiable sparsity-based proxy that enables hardware-aware training and architecture search. The evaluation covers multiple network sizes and datasets, and the energy breakdowns by memory level and by access type are informative. The work also makes a falsifiable prediction: sparser networks with layer-specific time windows are faster on the described array. However, the central EDP gains are currently demonstrated only in the authors' own simulator, and one load-bearing dataflow detail concerning the LIF recurrence is not specified, so the evidence does not yet fully support the headline quantitative claims.

major comments (5)
  1. [Section III-C with Eqs. (2)-(3)] The default high-temporal-density dispatch schedules NTWUs for the same postsynaptic neuron onto different PEs concurrently. Section III-C states that 'input activities of different time windows are fed into different columns (PEs) of a row from the top,' and Figure 3 shows NTWU(n,tw) and NTWU(n,tw+1) on different columns. However, the LIF model in Eqs. (2)-(3) makes the membrane voltage at the start of time window tw+1 depend on the final membrane voltage of time window tw. Section III-A describes only an intra-PE sequential Vmem update over the time points within a single time window. No register transfer, systolic pass, or scheduling constraint is described that carries the final Vmem of NTWU(n,tw) into the PE executing NTWU(n,tw+1). If the simulator overlaps these NTWUs, it may be evaluating dynamics different from those that produced the reported accuracy; if it serializes them, the claimed PE utilization and the resulting latency/EDP improvements are unsupported. The manuscript must specify the state-carry mechanism or provide a correctness analysis of the overlapping schedule.
  2. [Section IV-A with Section VI-C] The hardware-aware training loss is LHW(W)=EDP(Sp(W)), where EDP(.) is a piecewise-linear fit to 'well-evaluated simulation data' from the same cycle-level simulator that is later used to compute the final energy and latency improvements in Section VI-C. Minimizing this fitted proxy and then reporting EDP gains from the same simulator is circular to the extent that the simulator is treated as ground truth. Please provide a validation of the piecewise-linear proxy against the full simulator on held-out layers or settings, and ideally an independent check such as RTL/FPGA measurements or a second, independently written simulator. Without such validation, the claimed 15.1x-150.87x EDP range is not an independent test of the co-optimization methodology.
  3. [Section VI-C, Figure 13] The paper states that the proposed optimization techniques can also be applied to the PTB baseline, but it does not report PTB+HT or PTB+HAS results. The reported comparison 'SpikeX+HT+HAS versus PTB' therefore conflates the effect of the training/search methodology with the effect of the SpikeX hardware dataflow. Please include PTB+HT, PTB+HAS, and SpikeX without HT/HAS so that the hardware architecture contribution and the training/search contribution can be separated. This ablation is necessary to support the claim that the architecture itself, rather than only the sparsity-inducing training, drives the reported gains.
  4. [Abstract, Section VI-C, Figure 13] The reported EDP range is not consistent with the per-network energy and latency improvements shown in Figure 13. Multipling the presented energy and latency improvements gives approximately 250x for DVS-Gesture Medium (24.38 x 10.29), 25x for DVS-Gesture Large (4.91 x 5.19), and 15x for NMNIST (2.07 x 7.25). The abstract and conclusion claim a range of 15.1x to 150.87x, which does not match these numbers. Please clarify how the headline EDP range is computed, or correct the reported values.
  5. [Section V-C.2] The PTB baseline is described in a single paragraph without a quantitative specification of the memory hierarchy, systolic array size, scheduling policy, or the exhaustive-search range used to select the baseline time-window sizes. Because the PTB baseline is the only comparator and the claimed EDP gains are relative to it, the comparison could be sensitive to these unspecified choices. Please provide a full baseline configuration table and, if possible, additional baselines from other groups to increase confidence that the reported improvements are not artifacts of baseline setup.
minor comments (5)
  1. [Abstract] The phrase 'without comprising model accuracy' should read 'without compromising model accuracy.'
  2. [Sections II-B and III-B] There are typos in 'sparisity' and 'firing sparisty'; these should be corrected to 'sparsity' throughout.
  3. [References] Reference [3] and reference [4] appear to be duplicates of the same CVPR paper, and reference [8] appears to be a garbled citation of the MNIST paper. The bibliography should be cleaned up.
  4. [Figure 5 caption] The caption contains the phrase 'Outside-Network Activations (Independent on Networks),' which is not defined or explained in the text; please clarify its meaning.
  5. [Section VI-A] The synthetic firing-rate results in Figure 9 and Figure 10 should state the number of random trials and report variance or confidence intervals, since the input activations are generated randomly.

Circularity Check

1 steps flagged · score 6.0 of 10

EDP gains are partly the optimized objective itself: LHW is a fit to the same simulator that produces the headline EDP numbers.

  1. fitted input called prediction [Section IV-A (LHW definition, Eq. 4) and Section VI-C (EDP Evaluation); simulator defined in Section V-B]
    "we fit a piecewise linear model to the approximate EDP per filled time window based on well-evaluated simulation data. ... Thus, we evaluate LHW as LHW (W ) = EDP(Sp(W )). ... EDP Evaluation- We use energy-delay-product(EDP) to simultaneously consider the latency and energy efficiency of the overall system. ... this present work can have a further 15.1x to 150.87x improvement on [26]."

    The training objective Ltot = Lacc + beta*LHW is minimized with LHW(W) = EDP(Sp(W)), where EDP(.) is a piecewise-linear fit to simulation data produced by the paper's own cycle-level/CACTI simulator. The final EDP numbers in Section VI-C are then generated by that same simulator. Therefore the EDP reduction credited to SpikeX-HT and SpikeX-HAS is not an independent prediction: it is the value of the objective being optimized. Sparse networks appear faster partly because the loss was explicitly constructed to make them so. The architecture-level weight-reuse and dispatch speedups are simulated separately and are less circular, but the headline 15.1x-150.87x EDP range includes the co-optimization components that reduce to the fitted objective.

full rationale

The circularity is partial, not total. The SpikeX architecture itself (agile NTWU dispatch, three-level weight reuse, activation-induced weight tailoring) is evaluated by the simulator independently of the training objective; Figure 10's latency-versus-firing-rate sweep is a genuine architectural comparison, and the accuracy results in Figure 7 are empirical. What is circular is the EDP contribution credited to hardware-aware training and architecture search: the networks and TWS choices are selected by minimizing LHW(W) = EDP(Sp(W)), where EDP(.) is a piecewise-linear fit to 'well-evaluated simulation data' from the same simulator that later reports the 15.1x-150.87x EDP improvement. Thus a large part of the headline reduction is the optimized objective re-measured, not an independent forecast. The self-cited PTB baseline [26] is implemented in the simulator rather than merely assumed, so that self-citation is not load-bearing. Separately, the paper provides no RTL/FPGA/silicon validation of the simulator, and the high-temporal-density dispatch of consecutive time windows of one neuron onto different PEs is not shown to carry Vmem across windows (Eqs. 2-3); these are correctness and validation risks, not circularity, and they do not raise the score further. Score 6 reflects partial circularity in the central EDP range, not a fully tautological derivation.

Assumptions & free parameters 2 free parameters · 4 assumptions · 2 invented entities

The quantified claims rest on a small set of fitted or assumed components: the simulator itself, the piecewise-linear EDP proxy used as the training loss, the tag-based zero-skipping assumption, and the representative-sparsity assumption. No code is released. The architectural ideas are clearly specified, but the numbers cannot be reproduced from the paper.

free parameters (2)
  • piecewise linear EDP proxy coefficients = not reported
    Section IV-A states a piecewise linear model is fit to approximate EDP per filled time window from simulation data; the coefficients are not given, but they define the hardware loss used in training and thus shape all EDP numbers.
  • hardware/accuracy tradeoff beta = swept 5e-05 to 1e-01
    Equation (4) weights hardware loss by beta; the paper sweeps it but does not state the chosen value for the final EDP comparison, making the reported accuracy/EDP tradeoff hard to reproduce.
assumptions (4)
  • domain assumption The LIF model (equations 1-3) is the exact computation performed by each processing element, and an activity tag of zero implies no presynaptic spikes in that window, so skipping zero-tag work units preserves network semantics.
    Section III-B defines hierarchical tags as bitwise OR of lower-level tags and uses zero tags to skip work; the correctness of this skipping is assumed rather than proven under all neuron states.
  • domain assumption The cycle-level simulator with CACTI 32nm memory energy models reliably estimates energy and latency for both SpikeX and the PTB baseline.
    Sections V-A and V-B describe the simulator, but no validation against RTL, FPGA, or measured silicon is provided; all quantitative claims rely on this.
  • domain assumption Sparsity statistics from the training/validation distribution remain representative at deployment, so hardware-aware training transfers to actual inputs.
    The hardware loss is computed from spiking activity on known data at training time (Section IV-A); the paper does not analyze distribution shift.
  • ad hoc to paper The fitted piecewise-linear EDP-per-active-window proxy is an adequate differentiable surrogate for full cycle-level EDP during optimization.
    Section IV-A introduces this fit specifically to make training tractable; its accuracy is asserted from 'well-evaluated simulation data' but no fit quality metrics are reported.
invented entities (2)
  • Neuro-temporal work unit (NTWU)
    purpose: Groups a post-synaptic neuron over a time window into a schedulable work unit to enable weight reuse and zero-skipping.
    The NTWU is defined and evaluated only within the paper's simulator (Section III-B/C); no external benchmark or implementation demonstrates its benefit independently.
  • Hierarchical activity tags (TW, TB, TS, SP-MB)
    purpose: Enables coarse-to-fine zero-skipping and activation-induced weight tailoring across time blocks and spatial memory blocks.
    The tags are paper-introduced abstractions; their overhead and benefit are measured only in the custom simulator, not in silicon or an independent implementation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SpikeX: Exploring Accelerator Architecture and Network-Hardware Co-Optimization for Sparse Spiking Neural Networks." pith.science (2026). https://pith.science/paper/C5YTEMLS

@misc{pith2026250512292,
  author       = {Pith},
  title        = {Pith review of: SpikeX: Exploring Accelerator Architecture and Network-Hardware Co-Optimization for Sparse Spiking Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C5YTEMLS}},
  note         = {Machine review of arXiv:2505.12292}
}
read the original abstract

Spiking Neural Networks (SNNs) are promising biologically plausible models of computation which utilize a spiking binary activation function similar to that of biological neurons. SNNs are well positioned to process spatiotemporal data, and are advantageous in ultra-low power and real-time processing. Despite a large body of work on conventional artificial neural network accelerators, much less attention has been given to efficient SNN hardware accelerator design. In particular, SNNs exhibit inherent unstructured spatial and temporal firing sparsity, an opportunity yet to be fully explored for great hardware processing efficiency. In this work, we propose a novel systolic-array SNN accelerator architecture, called SpikeX, to take on the challenges and opportunities stemming from unstructured sparsity while taking into account the unique characteristics of spike-based computation. By developing an efficient dataflow targeting expensive multi-bit weight data movements, SpikeX reduces memory access and increases data sharing and hardware utilization for computations spanning across both time and space, thereby significantly improving energy efficiency and inference latency. Furthermore, recognizing the importance of SNN network and hardware co-design, we develop a co-optimization methodology facilitating not only hardware-aware SNN training but also hardware accelerator architecture search, allowing joint network weight parameter optimization and accelerator architectural reconfiguration. This end-to-end network/accelerator co-design approach offers a significant reduction of 15.1x-150.87x in energy-delay-product(EDP) without comprising model accuracy.

Figures

Figures reproduced from arXiv: 2505.12292 by the authors.

Figure 1
Figure 1. (a) The proposed SpikeX hardware Architecture (b) The schematic design of Processing Element (PE) in the systolic array (c) A comprehensive weight [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Four temporal granularities: time point, time window (TW), time block [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Agile SpatioTemporal Dispatch. (Left) High temporal density mode, where NTWUs across TW grids are dispatched onto distinct PE columns. (Right) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Activation-induced weight tailoring during determining necessary [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: An analysis of realistic layer-wise firing activities of LIF neurons and structured sparsity of packed spiking time windows(TWs) with and without [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The backpropagation of hardware-aware training via sparsity. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Accuracy, active spike density and active time window(TW) density across different hyperparameter [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: (a) The conventional SNN training methodology (b) Our proposed training methodology Spike-HAS which involves hardware-aware training and [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: The normalized energy consumption and latency under different spike [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 12
Figure 12. Figure 12: (a) and (b) illustrate the energy consumption breakdown of SpikeX when executing fully-connected layers. By component, memory accounts for 62% of the total energy consumption, with a detailed distribution as follows: RF (39.5%), LBUF (13.4%), GLB (7.3%), and DRAM (1.7…
Figure 10
Figure 10. Figure 10: The normalized latency under different spike firing rate of CONV2 [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: The impact of time window size on normalized energy in CONV2 [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 13
Figure 13. Figure 13: The normalized latency and energy with- and without SpikeX, SpikeX-HT and SpikeX-HAS, in each dataset. [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 37 canonical work pages

  1. [26]

    Parallel time batching: Systolic-array acceleration of sparse spiking neural computation,

    J.-J. Lee, W. Zhang, and P. Li, “Parallel time batching: Systolic-array acceleration of sparse spiking neural computation,” in 2022 IEEE International Symposium on High-Performance Computer Architecture (HPCA), 2022, pp. 317–330

    work page 2022

  2. [1]

    Truenorth: Design and tool flow of a 65 mw 1 million neuron programmable neurosynaptic chip,

    F. Akopyan, J. Sawada, A. Cassidy, R. Alvarez-Icaza, J. Arthur, P. Merolla, N. Imam, Y . Nakamura, P. Datta, G.-J. Nam, B. Taba, M. Beakes, B. Brezzo, J. B. Kuang, R. Manohar, W. P. Risk, B. Jackson, and D. S. Modha, “Truenorth: Design and tool flow of a 65 mw 1 million neuron programmable neurosynaptic chip,” IEEE Transactions on Computer- Aided Design o...

    work page 2015

  3. [3]

    A low power, fully event-based gesture recognition system,

    A. Amir, B. Taba, D. Berg, T. Melano, J. McKinstry, C. Di Nolfo, T. Nayak, A. Andreopoulos, G. Garreau, M. Mendoza, J. Kusnitz, M. Debole, S. Esser, T. Delbruck, M. Flickner, and D. Modha, “A low power, fully event-based gesture recognition system,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2017, pp. 7388–7397

    work page 2017

  4. [4]

    A low power, fully event-based gesture recognition system,

    A. AMIR, B. TABA, D. BERG, T. MELANO, J. MCKINSTRY , C. DI NOLFO, T. NAYAK, A. ANDREOPOULOS, G. GARREAU, M. MENDOZA, and OTHERS, “A low power, fully event-based gesture recognition system,” in PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION , 2017, pp. 7243–7252

    work page 2017

  5. [5]

    Hard- ware/software co-design with adc-less in-memory computing hardware for spiking neural networks,

    M. P. E. Apolinario, A. K. Kosta, U. Saxena, and K. Roy, “Hard- ware/software co-design with adc-less in-memory computing hardware for spiking neural networks,” IEEE Transactions on Emerging Topics in Computing, vol. 12, no. 1, pp. 35–47, 2023

    work page 2023

  6. [6]

    Are snns truly energy-efficient?—a hardware perspective,

    A. Bhattacharjee, R. Yin, A. Moitra, and P. Panda, “Are snns truly energy-efficient?—a hardware perspective,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 13 311–13 315

    work page 2024

  7. [7]

    Efficient biologically-plausible training of spiking neural networks with precise timing,

    R. Boone, W. Zhang, and P. Li, “Efficient biologically-plausible training of spiking neural networks with precise timing,” in International Conference on Neuromorphic Systems 2021 , 2021, pp. 1–8

    work page 2021

  8. [8]

    Comparison of classifier methods: a case study in handwritten digit recognition,

    L. Bottou, C. Cortes, J. Denker, H. Drucker, I. Guyon, L. Jackel, Y . LeCun, U. Muller, E. Sackinger, P. Simard, and V . Vapnik, “Comparison of classifier methods: a case study in handwritten digit recognition,” in Proceedings of the 12th IAPR International Conference on Pattern Recog- nition, Vol. 3 - Conference C: Signal Processing (Cat. No.94CH3440-5) ...

    work page 1994

Show all 53 references
  1. [9]

    Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,

    Y .-H. Chen, T. Krishna, J. S. Emer, and V . Sze, “Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,” IEEE Journal of Solid-State Circuits , vol. 52, no. 1, pp. 127–138, 2017

    2017

  2. [10]

    Taking neuromorphic computing to the next level with loihi 2,

    M. Davies, “Taking neuromorphic computing to the next level with loihi 2,” https://download.intel.com/newsroom/2021/new-technologies/ neuromorphic-computing-loihi-2-brief.pdf

    2021

  3. [11]

    Loihi: A neuromorphic manycore processor with on-chip learning,

    M. Davies, N. Srinivasa, T.-H. Lin, G. Chinya, Y . Cao, S. H. Choday, G. Dimou, P. Joshi, N. Imam, S. Jain et al., “Loihi: A neuromorphic manycore processor with on-chip learning,” IEEE Micro, vol. 38, no. 1, pp. 82–99, 2018

    2018

  4. [12]

    A large-scale model of the functioning brain,

    C. Eliasmith, T. C. Stewart, X. Choo, T. Bekolay, T. DeWolf, Y . Tang, and D. Rasmussen, “A large-scale model of the functioning brain,” Science, vol. 338, no. 6111, pp. 1202–1205, 2012. [Online]. Available: https://www.science.org/doi/abs/10.1126/science.1225266

    2012 doi

  5. [13]

    Neural architecture search: A survey,

    T. Elsken, J. H. Metzen, and F. Hutter, “Neural architecture search: A survey,” Journal of Machine Learning Research , vol. 20, no. 55, pp. 1–21, 2019. [Online]. Available: http://jmlr.org/papers/v20/18-598.html

    2019

  6. [14]

    Audio and image cross-modal intelligence via a 10tops/w 22nm soc with back-propagation and dynamic power gating,

    Z. Fan, H. An, Q. Zhang, B. Xu, L. Xu, C.-W. Tseng, Y . Peng, A. Cao, B. Liu, C. Lee, Z. Wang, F. Liu, G. Wang, S. Jiang, H.-S. Kim, D. Blaauw, and D. Sylvester, “Audio and image cross-modal intelligence via a 10tops/w 22nm soc with back-propagation and dynamic power gating,” ...

    2022

  7. [15]

    Aimmi: Audio and image multi-modal intelligence via a low-power soc with 2-mbyte on- chip mram for iot devices,

    Z. Fan, Q. Zhang, H. An, B. Xu, L. Xu, C.-W. Tseng, Y . Peng, A. Bejarano-Carbo, P. Abillama, A. Cao et al. , “Aimmi: Audio and image multi-modal intelligence via a low-power soc with 2-mbyte on- chip mram for iot devices,” IEEE Journal of Solid-State Circuits , 2024

    2024

  8. [16]

    Sparse coding,

    P. Foldiak and D. Endres, “Sparse coding,” Scholarpedia, vol. 3, no. 1, p. 2984, 2008, revision #145589

    2008

  9. [17]

    The spinnaker project,

    S. B. Furber et al., “The spinnaker project,” Proceedings of the IEEE , vol. 102, no. 5, pp. 652–665, May 2014

    2014

  10. [18]

    The spinnaker project,

    S. B. Furber, F. Galluppi, S. Temple, and L. A. Plana, “The spinnaker project,” Proceedings of the IEEE , vol. 102, no. 5, pp. 652–665, 2014

    2014

  11. [19]

    Highly efficient neuromorphic learning system of spiking neural network with multi-compartment leaky integrate-and-fire neurons,

    T. Gao, B. Deng, J. Wang, and G. Yi, “Highly efficient neuromorphic learning system of spiking neural network with multi-compartment leaky integrate-and-fire neurons,” Frontiers in Neuroscience , vol. 16,

  12. [20]

    A survey of fpga-based neural network accelerator,

    K. Guo, S. Zeng, J. Yu, Y . Wang, and H. Yang, “A survey of fpga-based neural network accelerator,” 2018

    2018

  13. [21]

    A quantitative description of membrane current and its application to conduction and excitation in nerve,

    A. Hodgkin and A. Huxley, “A quantitative description of membrane current and its application to conduction and excitation in nerve,” Journal of Physiology, vol. 117, pp. 500–544, 1952

    1952

  14. [22]

    Hybrid macro/micro level backpropagation for training deep spiking neural networks,

    Y . Jin, P. Li, and W. Zhang, “Hybrid macro/micro level backpropagation for training deep spiking neural networks,” 05 2018

    2018

  15. [23]

    In-datacenter performance analysis of a tensor processing unit,

    N. P. Jouppi, C. Young, N. Patil, D. Patterson, G. Agrawal, R. Bajwa, S. Bates, S. Bhatia, N. Boden, A. Borchers et al. , “In-datacenter performance analysis of a tensor processing unit,” in Proceedings of the 44th annual international symposium on computer architecture , 2017...

    2017

  16. [24]

    Understanding reuse, performance, and hardware cost of dnn dataflow: A data-centric approach,

    H. Kwon, P. Chatarasi, M. Pellauer, A. Parashar, V . Sarkar, and T. Krishna, “Understanding reuse, performance, and hardware cost of dnn dataflow: A data-centric approach,” ser. MICRO ’52. New York, NY , USA: Association for Computing Machinery, 2019, p. 754–768. [Online]. Ava...

    2019

  17. [25]

    Reconfigurable dataflow optimization for spatiotem- poral spiking neural computation on systolic array accelerators,

    J.-J. Lee and P. Li, “Reconfigurable dataflow optimization for spatiotem- poral spiking neural computation on systolic array accelerators,” in 2020 IEEE 38th International Conference on Computer Design (ICCD) , 2020, pp. 57–64

    2020

  18. [27]

    Cifar10-dvs: An event-stream dataset for object classification,

    H. Li, H. Liu, X. Ji, G. Li, and L. Shi, “Cifar10-dvs: An event-stream dataset for object classification,” Frontiers in Neuroscience , vol. 11,

  19. [28]

    H2learn: High-efficiency learning accelerator for high-accuracy spiking neural networks,

    L. Liang, Z. Qu, Z. Chen, F. Tu, Y . Wu, L. Deng, G. Li, P. Li, and Y . Xie, “H2learn: High-efficiency learning accelerator for high-accuracy spiking neural networks,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems , vol. 41, no. 11, pp. 4782–4796, 2021

    2021

  20. [29]

    Sparse compressed spiking neural network accelerator for object detection,

    H.-H. Lien and T.-S. Chang, “Sparse compressed spiking neural network accelerator for object detection,” IEEE Transactions on Circuits and Systems I: Regular Papers , vol. 69, no. 5, pp. 2060–2069, may 2022. [Online]. Available: https://doi.org/10.1109%2Ftcsi.2022.3149006

    2022

  21. [30]

    Networks of spiking neurons: The third generation of neural network models,

    W. Maass, “Networks of spiking neurons: The third generation of neural network models,” Neural Networks , vol. 10, no. 9, pp. 1659–1671,

  22. [31]

    Spinnaker 2: A 10 million core processor system for brain simulation and machine learning,

    C. Mayr, S. Hoeppner, and S. Furber, “Spinnaker 2: A 10 million core processor system for brain simulation and machine learning,” arXiv preprint arXiv:1911.02385, 2019

    1911 arXiv

  23. [32]

    Cacti 6.0: A tool to model large caches,

    N. Muralimanohar, R. Balasubramonian, and N. Jouppi, “Cacti 6.0: A tool to model large caches,” HP Laboratories, 01 2009

    2009

  24. [33]

    Spinalflow: An architecture and dataflow tailored for spiking neural networks,

    S. Narayanan, K. Taht, R. Balasubramonian, E. Giacomin, and P.-E. Gaillardon, “Spinalflow: An architecture and dataflow tailored for spiking neural networks,” in Proceedings of the ACM/IEEE 47th Annual International Symposium on Computer Architecture , ser. ISCA ’20. IEEE Pres...

    2020

  25. [34]

    Converting static image datasets to spiking neuromorphic datasets using saccades,

    G. Orchard, A. Jayawant, G. K. Cohen, and N. Thakor, “Converting static image datasets to spiking neuromorphic datasets using saccades,” Frontiers in Neuroscience , vol. 9, pp. 77–82 vol.2, 1994. [Online]. Available: https://www.frontiersin.org/articles/10.3389/fnins.2015.00437

    1994

  26. [35]

    Spinnaker: A 1-w 18-core system-on-chip for massively-parallel neural network simulation,

    E. Painkras, L. A. Plana, J. Garside, S. Temple, F. Galluppi, C. Patterson, D. R. Lester, A. D. Brown, and S. B. Furber, “Spinnaker: A 1-w 18-core system-on-chip for massively-parallel neural network simulation,” IEEE Journal of Solid-State Circuits , vol. 48, no. 8, pp. 1943–...

    1943

  27. [36]

    Convolutional spiking neural networks for spatio-temporal feature extraction,

    A. Samadzadeh, F. S. T. Far, A. Javadi, A. Nickabadi, and M. H. Chehreghani, “Convolutional spiking neural networks for spatio-temporal feature extraction,” CoRR, vol. abs/2003.12346, 2020. [Online]. Available: https://arxiv.org/abs/2003.12346

    2003 arXiv

  28. [37]

    Scale- sim: Systolic cnn accelerator,

    A. Samajdar, Y . Zhu, P. Whatmough, M. Mattina, and T. Krishna, “Scale- sim: Systolic cnn accelerator,” 10 2018

    2018

  29. [38]

    Slayer: Spike layer error reassignment in time,

    S. B. Shrestha and G. Orchard, “Slayer: Spike layer error reassignment in time,” arXiv preprint arXiv:1810.08646 , 2018

    2018 arXiv

  30. [39]

    Deep learning in spiking neural networks,

    A. Tavanaei, M. Ghodrati, S. R. Kheradpisheh, T. Masquelier, and A. Maida, “Deep learning in spiking neural networks,” Neural Networks , vol. 111, pp. 47–63, 2019. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0893608018303332

    2019

  31. [40]

    Sies: A novel implementation of spiking convolutional neural network inference engine on field-programmable gate array,

    S.-Q. Wang, L. Wang, Y . Deng, Z.-J. Yang, S.-S. Guo, Z.-Y . Kang, Y .-F. Guo, and W.-X. Xu, “Sies: A novel implementation of spiking convolutional neural network inference engine on field-programmable gate array,” Journal of Computer Science and Technology , vol. 35, pp. 475–...

    2020

  32. [41]

    Compsnn: A lightweight spiking neural network based on spatiotemporally compressive spike features,

    T. Wang, C. Shi, X. Zhou, Y . Lin, J. He, P. Gan, P. Li, Y . Wang, L. Liu, N. Wu, and G. Luo, “Compsnn: A lightweight spiking neural network based on spatiotemporally compressive spike features,” Neurocomputing, vol. 425, pp. 96–106, 2021. [Online]. Available: https://www.scie...

    2021

  33. [42]

    Spiking transformer hardware accelerators in 3d integration,

    B. Xu, J. Hwang, P. Vanna-iampikul, S. K. Lim, and P. Li, “Spiking transformer hardware accelerators in 3d integration,” arXiv preprint arXiv:2411.07397, 2024

    2024 arXiv

  34. [43]

    Towards 3d acceleration for low-power mixture-of-experts and multi- head attention spiking transformers,

    B. Xu, J. Hwang, P. Vanna-iampikul, Y . Yin, S. K. Lim, and P. Li, “Towards 3d acceleration for low-power mixture-of-experts and multi- head attention spiking transformers,” arXiv preprint arXiv:2412.05540 , 2024

    2024 arXiv

  35. [44]

    Trimming down large spiking vision transformers via heterogeneous quantization search,

    B. Xu, Y . Song, and P. Li, “Trimming down large spiking vision transformers via heterogeneous quantization search,” arXiv preprint arXiv:2412.05505, 2024

    2024 arXiv

  36. [45]

    Workload-balanced pruning for sparse spiking neural networks,

    R. Yin, Y . Kim, Y . Li, A. Moitra, N. Satpute, A. Hambitzer, and P. Panda, “Workload-balanced pruning for sparse spiking neural networks,” IEEE Transactions on Emerging Topics in Computational Intelligence , 2024

    2024

  37. [46]

    Mint: Multiplier-less integer quantization for energy efficient spiking neural networks,

    R. Yin, Y . Li, A. Moitra, and P. Panda, “Mint: Multiplier-less integer quantization for energy efficient spiking neural networks,” in 2024 29th Asia and South Pacific Design Automation Conference (ASP-DAC). IEEE, 2024, pp. 830–835

    2024

  38. [47]

    Sata: Sparsity-aware training accelerator for spiking neural networks,

    R. Yin, A. Moitra, A. Bhattacharjee, Y . Kim, and P. Panda, “Sata: Sparsity-aware training accelerator for spiking neural networks,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 42, no. 6, pp. 1926–1938, 2022

    1926

  39. [48]

    Gpu-based simulation of spiking neural networks with real-time performance & high accuracy,

    D. Yudanov, M. Shaaban, R. Melton, and L. Reznik, “Gpu-based simulation of spiking neural networks with real-time performance & high accuracy,” in The 2010 international joint conference on neural networks (IJCNN). IEEE, 2010, pp. 1–8

    2010

  40. [49]

    Spike-train level backpropagation for training deep recurrent spiking neural networks,

    W. Zhang and P. Li, “Spike-train level backpropagation for training deep recurrent spiking neural networks,” in Advances in Neural Information Processing Systems, 2019, pp. 7800–7811

    2019

  41. [50]

    Temporal spike sequence learning via backpropagation for deep spiking neural networks,

    ——, “Temporal spike sequence learning via backpropagation for deep spiking neural networks,” Advances in neural information processing systems, vol. 33, pp. 12 022–12 033, 2020

    2020

  42. [51]

    A digital liquid state machine with biologically inspired learning and its application to speech recognition,

    Y . Zhang, P. Li, Y . Jin, and Y . Choe, “A digital liquid state machine with biologically inspired learning and its application to speech recognition,” IEEE transactions on neural networks and learning systems , vol. 26, no. 11, pp. 2635–2649, 2015

    2015

  43. [1997]

    Available: https://www.sciencedirect.com/science/article/ pii/S0893608097000117 13

    [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S0893608097000117 13

  44. [2017]

    Available: https://www.frontiersin.org/articles/10.3389/ fnins.2017.00309

    [Online]. Available: https://www.frontiersin.org/articles/10.3389/ fnins.2017.00309

    2017

  45. [2022]

    Available: https://www.frontiersin.org/articles/10.3389/ fnins.2022.929644

    [Online]. Available: https://www.frontiersin.org/articles/10.3389/ fnins.2022.929644

    2022

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.