Pith. sign in

REVIEW 4 major objections 7 minor 53 references

SENMAP: Multi-objective data-flow mapping and synthesis for hybrid scalable neuromorphic systems

T0 review · 4 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read SENMap, a new open-source mapper, claims to cut simulated neuromorphic chip energy use by 40 percent by optimizing where spiking and artificial neural networks are placed on the SENECA architecture.

desk verdict A useful open-source mapper for SENECA, but the headline 40% energy claim is not pinned to SENMap's mapping choices, so the paper overstates its central result. read the letter →

arxiv 2506.03450 v2 pith:ZM65BJ6K submitted 2025-06-03 cs.NE

Prithvish V Nembhani , Oliver Rhodes , Guangzhi Tang , Alexandra F Dobrita , Yingfu Xu , Kanishkan Vadivel , Kevin Shidqi , Paul Detterer
show 5 more authors
Mario Konijnenburg Gert-Jan van Schaik Manolis Sifalakis Zaid Al-Ars Amirreza Yousefzadeh
This is my paper
classification cs.NE MSC 68T0768M2090C29
keywords SENMapneuromorphicmappingspikingneuralnetworksSENSIMSENECAmulti-objectiveoptimizationeventratedata-flowsynthesis
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 introduces SENMap, an open-source mapping and synthesis tool that decides where the neurons, weights, and biases of a pretrained spiking or artificial neural network are placed on the SENECA digital neuromorphic processor. The central claim is that searching over the number of neural processing elements per core, the core mesh shape, and the event rate with genetic algorithms produces mappings that cut simulated energy use by about 40 percent relative to SENSIM's baseline mapping in asynchronous timestep mode. The payoff would be that chip designers can explore architecture configurations and application placements in simulation before tape-out, without fabricating test chips for every option. SENMap also elevates event rate to a first-class mapping parameter, finding that fully event-driven operation (fps=0) matches a 120 fps frame-based output with correlation 1.0 and zero time shift.

What carries the argument

The argument is carried by SENMap's optimization loop over a discrete design space. Four mechanisms feed that loop: partitioning strategies (layer-wise, channel-wise, height-wise, width-wise) that distribute neurons across cores; mesh compression schemes (strict area optimal, loose area optimal, strict square) that choose the core grid shape; a per-core memory bound (Eq. 1) that constrains neuron states, weights, biases, and thresholds to fit in local memory; and a distortion check that compares the end signal of a candidate mapping to a reference via normalized cross-correlation, estimating any time shift. The Pymoo framework supplies the genetic algorithm and NSGA-II that search this space, with event rate treated as a variable alongside architectural parameters, and parallel execution on a 30-core node cutting time-to-solution from 3 days to 3 hours.

What would settle it

Measure per-inference energy on fabricated SENECA silicon (or on a validated RTL/gate-level model) for the SENSIM baseline mapping versus the SENMap-optimized mapping of the same PilotNet network; if the measured difference is not approximately 40 percent, or the output correlation degrades, the claimed improvement is an artifact of the simulator.

Watch

Extended reading notes

Core claim

SENMap's central discovery is that mapping a pretrained network onto the SENECA architecture is a multi-objective optimization problem whose solution yields roughly a 40 percent energy reduction over the baseline SENSIM mapping in asynchronous event-driven mode, without compromising output-signal fidelity. The control variables that carry the gain are the number of NPEs per core, the total core count and mesh arrangement, and the frame/event rate of the input stream. On the PilotNet driving-network example, the best configuration uses just one NPE per core and spreads the network across more cores, and the fully event-driven mapping is measured as equivalent to the 120 fps reference with normalized cross-correlation 1.0 and zero time shift. The energy improvement is read from the relative-energy-versus-mapping figure as the difference between the rightmost and leftmost configurations as event rate increases.

Load-bearing premise

The 40 percent energy gain is measured inside the SENSIM simulator, whose energy and latency parameters come from the authors' prior modeling work, so the claim holds only if SENSIM's numbers match the real SENECA chip's behavior.

Editorial extensions

If this is right

  • Chip designers can use SENMap before fabrication to co-optimize the SENECA configuration (core count, NPEs per core, mesh) and the network mapping for a target energy and latency budget.
  • In asynchronous event-driven mode, the optimal mapping uses fewer NPEs per core and more cores, so SENECA-class designs should support a broad range of core counts to exploit this regime.
  • Event rate must be reported and controlled in neuromorphic benchmark studies, because the same network maps best at different rates, with fully event-driven operation matching the 120 fps output at correlation 1.0.
  • The open-sourced SENMap and SENSIM stack provides a reusable simulation-based design-space exploration flow for large pretrained SNNs and ANNs before silicon is available.
  • Parallel metaheuristics make the search practical: reducing mapping time from 3 days to 3 hours on a 30-core cluster, so design exploration can be iterated.

Reading between the lines

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

  • If SENSIM faithfully represents SENECA, the 40 percent figure implies mapping choices matter as much as architectural choices, so existing neuromorphic chips may be wasting a comparable share of energy simply from default placements.
  • The correlation-based distortion check suggests a portable test for any neuromorphic mapper: compare fully event-driven output against the highest-rate frame-based output; a correlation near 1.0 with zero shift is a necessary condition for a temporally faithful mapping.
  • SENMap's discrete variables (core count, mesh, NPEs, rate) map naturally onto other 2D-mesh SIMD neuromorphic designs, and the 'loose area optimal' trick for prime core counts is a general way to keep 2D layouts efficient.
  • The paper estimates accuracy via end-signal correlation rather than task-level accuracy; a direct extension would be to run full accuracy benchmarks at each candidate mapping, grounding the energy-accuracy trade-off in classification or regression outcome.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper presents SENMap, a mapping and synthesis tool that extends the SENSIM simulator to map pre-trained SNNs and ANNs onto the SENECA neuromorphic architecture. The tool supports multiple clustering/partitioning strategies, compression schemes, memory-bound constraints, and single- and multi-objective metaheuristic optimization via PyMOO. The central quantitative claim is that SENMap enables 40% energy improvements for a baseline SENSIM operating in timestep asynchronous mode. The paper also proposes an inter-spike-distortion metric based on cross-correlation of end signals and reports results on the PilotNet network across different frame rates and NPE configurations.

Significance. If the central claim is validated against a well-defined baseline, SENMap would be a useful open-source design-space exploration tool for SENECA-style neuromorphic chips, and the paper's integration of event-rate parameters and architectural co-optimization is timely. The manuscript has clear strengths: the tool is open-sourced, it supports parallel metaheuristics, it considers multiple mapping objectives, and it is evaluated on a realistic network (PilotNet) rather than only synthetic topologies. However, the current evidence for the headline claim is not yet convincing: the 40% improvement is not clearly attributed to SENMap's mapping algorithm, accuracy is not measured directly, and all numbers come from a simulator whose parameters are inherited from prior work [19].

major comments (4)
  1. [Section V, Figure 9, Abstract] The 40% energy improvement is not attributed to SENMap's mapping algorithm. The text states: 'We measure a 40% energy improvement from right to left in Figure 9 as we increase the rate and an optimal mapping requiring just 1 NPE and an increase in cores.' This conflates changes in frame rate and NPE/core configuration with the effect of SENMap's mapping optimization. No operational definition of 'baseline SENSIM' is given, and there is no matched comparison between SENMap's optimized mapping and SENSIM's default layer-wise clustering at the same frame rate and NPE count. The paper must define the baseline mapping and report energy at matched operating points to support the abstract's claim.
  2. [Section VII, Section IV-E, Table III] The conclusion that SENMap achieves 'around 40% energy efficiency without compromising accuracy' is unsupported. Accuracy is never measured on a task metric; the only evidence is the cross-correlation of normalized end signals defined in Eq. (2) and reported in Table III. No validation is provided that a correlation value such as 0.88 or 0.90 corresponds to acceptable inference accuracy for PilotNet. The authors should report a task-level accuracy or error metric (e.g., steering-angle error) for the compared mappings, or explicitly state that the correlation is only a proxy and not a substitute for accuracy.
  3. [Section V, Table II] No statistical analysis is reported for the metaheuristic optimizations. GA and NSGA-II are stochastic, yet the paper reports single energy/latency numbers without error bars, multiple seeds, or convergence information. The claimed 40% improvement could be within run-to-run variability. At minimum, the authors should report the number of independent runs and the variance of the energy estimates, or clearly state that the results are from a single illustrative run.
  4. [Section II, Section V] All energy and latency estimates are produced by SENSIM, whose parameters are taken from the same authors' prior work [19]. No comparison to measured hardware is presented. This is an external-validity limitation, but it is load-bearing for the quantitative claim. The paper should either add a validation of SENSIM's estimates against silicon measurements or explicitly frame the contribution as a simulation-only design-space exploration and soften the headline claim accordingly.
minor comments (7)
  1. [Section VI] There is a typo in the first paragraph: 'apping large pretrained SNNs' should be 'mapping large pretrained SNNs'.
  2. [Affiliation list] The affiliation for Guangzhi Tang reads 'Mastritch University-DACS'; this should be 'Maastricht University-DACS'.
  3. [Section IV-E, Eq. (2)] The notation in Eq. (2) should define x(t), y(t), the correlation operator, and the normalization exactly; currently corr is not formally defined, and the denominator appears identical to the numerator apart from the square root notation.
  4. [Table III] The table would benefit from a note explaining whether 'Time shift' is an absolute delay in milliseconds and why corr(fps=0, fps=60) and corr(fps=60, fps=120) both report a shift of 65 ms; the current presentation leaves the units and protocol ambiguous.
  5. [Section V, Figures 7-9] The x-axis of Figure 9 is labeled only as 'mapping'; the paper should specify what the axis enumerates (e.g., different NPE counts, frame rates, or SENMap configurations) and provide a legend that is legible in print.
  6. [Section V] The statement that 'dataset size and energy efficiency minimally affected results' is not supported by any figure, table, or statistical measure; either present the supporting data or remove the claim.
  7. [Section II] Since the central energy numbers depend entirely on parameters from [19], a brief summary of how those parameters were extracted (e.g., from RTL simulation, synthesis, or measurements) would help readers assess the credibility of the reported numbers.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the mapping optimization and simulator evaluation are logically independent, though the headline 40% energy gain is not tied to a defined baseline and the energy model comes from self-cited prior work.

full rationale

The derivation chain in this paper is not circular. SENMap's contribution is a mapping and synthesis search over architectural and event-rate parameters; its output is evaluated by SENSIM, an event-based simulator whose energy, communication, timing, architecture, and simulation parameters are taken from prior work [19]. That prior work is self-cited, but the parameters are described as 'extracted from lower-level hardware measurements,' which is external, falsifiable evidence rather than a restatement of the present result. The optimization loop (SENMap proposes mappings; SENSIM scores them) is a normal simulator-in-the-loop design-space exploration: the energy model is an input, not the predicted quantity. The abstract's '40 percent energy improvements for a baseline SENSIM' is, however, not demonstrated as stated. Section V says 'We measure a 40% energy improvement from right to left in Figure 9 as we increase the rate and an optimal mapping requiring just 1 NPE and an increase in cores,' which attributes the gain to changes in frame rate and NPE/core configuration, not to a head-to-head comparison of SENMap's mapping against SENSIM's default layer-wise clustering at matched operating points. That is an internal-validity flaw in the headline comparison, but it is not a circular derivation: no quantity is defined in terms of itself, no fitted parameter is renamed as a prediction, and no load-bearing argument reduces to a self-citation chain. The main limitation is external validity (simulated versus measured hardware), which falls outside circularity. Therefore the paper receives a low score for circularity, with the caveat that the central 40% claim needs a properly controlled baseline before it can be attributed to SENMap.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the accuracy of the SENSIM simulator (self-cited), a cross-correlation proxy for accuracy, and several algorithm hyperparameters chosen by hand. No new physical entities are introduced.

free parameters (4)
  • GA population size = 30
    Chosen by hand; no sensitivity analysis is provided to show the result is robust to this choice.
  • GA mutation eta = 3.0
    Algorithm hyperparameter; affects search behavior but is not tuned to the energy outcome in a documented way.
  • NSGA2 population size = 40
    Chosen by hand; no sensitivity analysis is provided.
  • NSGA2 offspring count = 10
    Chosen by hand; no sensitivity analysis is provided.
assumptions (3)
  • domain assumption SENSIM accurately models SENECA energy and latency
    All energy numbers come from SENSIM, whose parameters are from the self-cited prior work [19]. No hardware validation is presented in this paper.
  • ad hoc to paper Cross-correlation of normalized end signals is a valid proxy for inference accuracy
    Equation (2) defines z(t) as normalized cross-correlation; the paper uses this to claim accuracy is preserved without measuring task accuracy directly.
  • domain assumption Frame rate (event rate) is a critical parameter for mapping
    The paper's main argument, motivated by the PilotNet experiments but not derived from first principles.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SENMAP: Multi-objective data-flow mapping and synthesis for hybrid scalable neuromorphic systems." pith.science (2026). https://pith.science/paper/ZM65BJ6K

@misc{pith2026250603450,
  author       = {Pith},
  title        = {Pith review of: SENMAP: Multi-objective data-flow mapping and synthesis for hybrid scalable neuromorphic systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZM65BJ6K}},
  note         = {Machine review of arXiv:2506.03450}
}
read the original abstract

This paper introduces SENMap, a mapping and synthesis tool for scalable, energy-efficient neuromorphic computing architecture frameworks. SENECA is a flexible architectural design optimized for executing edge AI SNN/ANN inference applications efficiently. To speed up the silicon tape-out and chip design for SENECA, an accurate emulator, SENSIM, was designed. While SENSIM supports direct mapping of SNNs on neuromorphic architectures, as the SNN and ANNs grow in size, achieving optimal mapping for objectives like energy, throughput, area, and accuracy becomes challenging. This paper introduces SENMap, flexible mapping software for efficiently mapping large SNN and ANN applications onto adaptable architectures. SENMap considers architectural, pretrained SNN and ANN realistic examples, and event rate-based parameters and is open-sourced along with SENSIM to aid flexible neuromorphic chip design before fabrication. Experimental results show SENMap enables 40 percent energy improvements for a baseline SENSIM operating in timestep asynchronous mode of operation. SENMap is designed in such a way that it facilitates mapping large spiking neural networks for future modifications as well.

Figures

Figures reproduced from arXiv: 2506.03450 by the authors.

Figure 2
Figure 2. SENECA SDK [20] [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Mapping of PilotNet SNN Models to SENSIM [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Output comparison of Inter-spike distorted end signal [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Analytics and data collection framework for Pre and Post optimization [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: SENSIM output comparison (#260-#290) frames for different mapping schemes, event/frame rate & neural processing elements [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Comparing the output for fps=30,60,120,0 (event-based) for(#0-#500) frames for different mapping & number of NPEs [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Relative energy vs mapping for (#260-#290) frames (PilotNet) at [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 49 canonical work pages

  1. [19]

    SENSIM: An Event- driven Parallel Simulator for Multi-core Neuromorphic Systems,

    P. Nembhani, K. Vadivel, G. Tang, M. Tahghighi, G.-J. van Schaik, M. Sifalakis, Z. Al-Ars, and A. Yousefzadeh, “SENSIM: An Event- driven Parallel Simulator for Multi-core Neuromorphic Systems,” in 2024 International Joint Conference on Neural Networks (IJCNN), pp. 1– 6, June 2024. ISSN: 2161-4407

    work page 2024

  2. [1]

    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,” Nov

    work page

  3. [2]

    A Scalable Multicore Architecture With Heterogeneous Memory Structures for Dynamic Neuromorphic Asynchronous Processors (DYNAPs),

    S. Moradi, N. Qiao, F. Stefanini, and G. Indiveri, “A Scalable Multicore Architecture With Heterogeneous Memory Structures for Dynamic Neuromorphic Asynchronous Processors (DYNAPs),”IEEE transactions on biomedical circuits and systems, vol. 12, pp. 106–122, Feb. 2018

    work page 2018

  4. [3]

    Neurogrid: A Mixed-Analog-Digital Multichip System for Large-Scale Neural Simulations,

    B. V . Benjamin, P. Gao, E. McQuinn, S. Choudhary, A. R. Chan- drasekaran, J.-M. Bussat, R. Alvarez-Icaza, J. V . Arthur, P. A. Merolla, and K. Boahen, “Neurogrid: A Mixed-Analog-Digital Multichip System for Large-Scale Neural Simulations,”Proceedings of the IEEE, vol. 102, pp. 699–716, May 2014. Conference Name: Proceedings of the IEEE

    work page 2014

  5. [4]

    Gyro: A Digital Spiking Neural Network Architecture for Multi-Sensory Data Analytics,

    F. Corradi, G. Adriaans, and S. Stuijk, “Gyro: A Digital Spiking Neural Network Architecture for Multi-Sensory Data Analytics,” inProceedings of the 2021 Drone Systems Engineering and Rapid Simulation and Performance Evaluation: Methods and Tools Proceedings, (Budapest Hungary), pp. 9–15, ACM, Jan. 2021

    work page 2021

  6. [5]

    Accelerated Analog Neuromorphic Computing

    J. Schemmel, S. Billaudelle, P. Dauer, and J. Weis, “Accelerated Analog Neuromorphic Computing,” Mar. 2020. arXiv:2003.11996 [cond-mat, q-bio]

    work page Pith review arXiv 2020

  7. [6]

    Lava Software Framework — Lava documentation

    “Lava Software Framework — Lava documentation.”

    work page

  8. [7]

    PyCARL: A PyNN Interface for Hardware-Software Co-Simulation of Spiking Neural Network

    A. Balaji, P. Adiraju, H. J. Kashyap, A. Das, J. L. Krichmar, N. D. Dutt, and F. Catthoor, “PyCARL: A PyNN Interface for Hardware-Software Co- Simulation of Spiking Neural Network,” May 2020. arXiv:2003.09696 [cs]

    work page Pith review arXiv 2020

Show all 53 references
  1. [8]

    Nengo: a Python tool for building large-scale functional brain models,

    T. Bekolay, J. Bergstra, E. Hunsberger, T. DeWolf, T. Stewart, D. Ras- mussen, X. Choo, A. V oelker, and C. Eliasmith, “Nengo: a Python tool for building large-scale functional brain models,”Frontiers in Neuroinformatics, vol. 7, 2014

    2014

  2. [9]

    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, pp. 1659–1671, Dec. 1997

    1997

  3. [10]

    Towards artificial general intelligence with hybrid Tianjic chip architecture,

    J. Pei, L. Deng, S. Song, M. Zhao, Y . Zhang, S. Wu, G. Wang, Z. Zou, Z. Wu, W. He, F. Chen, N. Deng, S. Wu, Y . Wang, Y . Wu, Z. Yang, C. Ma, G. Li, W. Han, H. Li, H. Wu, R. Zhao, Y . Xie, and L. Shi, “Towards artificial general intelligence with hybrid Tianjic chip architect...

    2019

  4. [11]

    Darwin: A neuromorphic hardware co-processor based on spiking neural networks,

    D. Ma, J. Shen, Z. Gu, M. Zhang, X. Zhu, X. Xu, Q. Xu, Y . Shen, and G. Pan, “Darwin: A neuromorphic hardware co-processor based on spiking neural networks,”Journal of Systems Architecture, vol. 77, pp. 43–51, June 2017

    2017

  5. [12]

    sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker,

    O. Rhodes, P. A. Bogdan, C. Brenninkmeijer, S. Davidson, D. Fellows, A. Gait, D. R. Lester, M. Mikaitis, L. A. Plana, A. G. D. Rowley, A. B. Stokes, and S. B. Furber, “sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker,”Frontiers in Neuroscience, vol. 12, 2018

    2018

  6. [13]

    Mapping Spiking Neural Networks to Neuromorphic Hardware,

    A. Balaji, F. Catthoor, A. Das, Y . Wu, K. Huynh, F. G. Dell’Anna, G. In- diveri, J. L. Krichmar, N. D. Dutt, and S. Schaafsma, “Mapping Spiking Neural Networks to Neuromorphic Hardware,”IEEE Transactions on V ery Large Scale Integration (VLSI) Systems, vol. 28, pp. 76–86, Jan. 2020

    2020

  7. [14]

    DFSynthesizer: Dataflow-based Synthesis of Spiking Neural Networks to Neuromorphic Hardware,

    S. Song, H. Chong, A. Balaji, A. Das, J. Shackleford, and N. Kandasamy, “DFSynthesizer: Dataflow-based Synthesis of Spiking Neural Networks to Neuromorphic Hardware,” Aug. 2021. arXiv:2108.02023 [cs]

    2021 arXiv

  8. [15]

    Deep Spiking Networks,

    P. O’Connor and M. Welling, “Deep Spiking Networks,” Nov. 2016. arXiv:1602.08323 [cs]

    2016 arXiv

  9. [16]

    Going Deeper With Directly-Trained Larger Spiking Neural Networks,

    H. Zheng, Y . Wu, L. Deng, Y . Hu, and G. Li, “Going Deeper With Directly-Trained Larger Spiking Neural Networks,” Dec. 2020. arXiv:2011.05280 [cs]

    2020 arXiv

  10. [17]

    Spiking Deep Residual Network,

    Y . Hu, H. Tang, and G. Pan, “Spiking Deep Residual Network,” June

  11. [18]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” Dec. 2015. arXiv:1512.03385 [cs]

    2015 arXiv

  12. [20]

    Yousefzadeh, G.-J

    A. Yousefzadeh, G.-J. van Schaik, M. Tahghighi, P. Detterer, S. Traferro, M. Sifalakis, M. Hijdra, J. Stuijt, M. Hartmann, F. Catthoor, I. Ocket, and M. Konijnenburg,SENeCA: Next generation Neuromorphic Processor for EdgeAI. Institute of Electrical and Electronics Engineers (I...

    2022

  13. [21]

    SENECA: building a fully digital neuromorphic processor, design trade- offs and challenges,

    G. Tang, K. Vadivel, Y . Xu, R. Bilgic, K. Shidqi, P. Detterer, S. Traferro, M. Konijnenburg, M. Sifalakis, G.-J. van Schaik, and A. Yousefzadeh, “SENECA: building a fully digital neuromorphic processor, design trade- offs and challenges,”Frontiers in Neuroscience, vol. 17, 2023

    2023

  14. [22]

    Open the box of digital neuromorphic processor: Towards effective algorithm-hardware co-design,

    G. Tang, A. Safa, K. Shidqi, P. Detterer, S. Traferro, M. Konijnenburg, M. Sifalakis, G.-J. van Schaik, and A. Yousefzadeh, “Open the box of digital neuromorphic processor: Towards effective algorithm-hardware co-design,” Mar. 2023. arXiv:2303.15224 [cs, eess]

    2023 arXiv

  15. [23]

    Optimizing event-based neural networks on digital neuromorphic architecture: a comprehensive design space exploration,

    Y . Xu, K. Shidqi, G.-J. van Schaik, R. Bilgic, A. Dobrita, S. Wang, R. Meijer, P. Nembhani, C. Arjmand, P. Martinello, A. Gebregiorgis, S. Hamdioui, P. Detterer, S. Traferro, M. Konijnenburg, K. Vadivel, M. Sifalakis, G. Tang, and A. Yousefzadeh, “Optimizing event-based neura...

    2024

  16. [24]

    CARLsim 3: A user-friendly and highly optimized library for the creation of neurobiologically detailed spiking neural networks,

    M. Beyeler, K. D. Carlson, T.-S. Chou, N. Dutt, and J. L. Krichmar, “CARLsim 3: A user-friendly and highly optimized library for the creation of neurobiologically detailed spiking neural networks,” in2015 International Joint Conference on Neural Networks (IJCNN), pp. 1–8, July...

    2015

  17. [25]

    Mapping spiking neural networks onto a manycore neuromorphic architecture,

    C.-K. Lin, A. Wild, G. N. Chinya, T.-H. Lin, M. Davies, and H. Wang, “Mapping spiking neural networks onto a manycore neuromorphic architecture,”SIGPLAN Not., vol. 53, pp. 78–89, June 2018

    2018

  18. [26]

    SNEAP: A Fast and Efficient Toolchain for Mapping Large- Scale Spiking Neural Network onto NoC-based Neuromorphic Platform,

    S. Li, S. Guo, L. Zhang, Z. Kang, S. Wang, W. Shi, L. Wang, and W. Xu, “SNEAP: A Fast and Efficient Toolchain for Mapping Large- Scale Spiking Neural Network onto NoC-based Neuromorphic Platform,” Mar. 2020. Number: arXiv:2004.01639 arXiv:2004.01639 [cs]

    2020 arXiv

  19. [27]

    Mapping Deep Neural Networks on SpiNNaker2,

    F. Kelber, B. Wu, B. V ogginger, J. Partzsch, C. Liu, M. Stolba, and C. Mayr, “Mapping Deep Neural Networks on SpiNNaker2,” in Proceedings of the Neuro-inspired Computational Elements Workshop, (Heidelberg Germany), pp. 1–3, ACM, Mar. 2020

    2020

  20. [28]

    Run-time Mapping of Spiking Neural Networks to Neuromorphic Hardware,

    A. Balaji, T. Marty, A. Das, and F. Catthoor, “Run-time Mapping of Spiking Neural Networks to Neuromorphic Hardware,” June 2020. Number: arXiv:2006.06777 arXiv:2006.06777 [cs]

    2020 arXiv

  21. [29]

    Compiling Spiking Neural Networks to Neuromorphic Hardware,

    S. Song, A. Balaji, A. Das, N. Kandasamy, and J. Shackleford, “Compiling Spiking Neural Networks to Neuromorphic Hardware,” inThe 21st ACM SIGPLAN/SIGBED Conference on Languages, Compilers, and Tools for Embedded Systems, pp. 38–50, June 2020. arXiv:2004.03717 [cs]

    2020 arXiv

  22. [30]

    Mapping of Spiking Neural Network Topologies on Neuromorphic Hardware|TU Delft Repository,

    K. Shreya, “Mapping of Spiking Neural Network Topologies on Neuromorphic Hardware|TU Delft Repository,” Aug. 2021

    2021

  23. [31]

    Endurance-Aware Mapping of Spiking Neural Networks to Neuromorphic Hardware,

    T. Titirsha, S. Song, A. Das, J. Krichmar, N. Dutt, N. Kandasamy, and F. Catthoor, “Endurance-Aware Mapping of Spiking Neural Networks to Neuromorphic Hardware,”IEEE Transactions on Parallel and Distributed Systems, vol. 33, pp. 288–301, Feb. 2022. Conference Name: IEEE Transa...

    2022

  24. [32]

    Optimal Mapping of Spiking Neural Network to Neuromorphic Hardware for Edge-AI,

    C. Xiao, J. Chen, and L. Wang, “Optimal Mapping of Spiking Neural Network to Neuromorphic Hardware for Edge-AI,”Sensors, vol. 22, p. 7248, Jan. 2022. Number: 19 Publisher: Multidisciplinary Digital Publishing Institute

    2022

  25. [33]

    Mapping Very Large Scale Spiking Neuron Network to Neuromorphic Hardware,

    O. Jin, Q. Xing, Y . Li, S. Deng, S. He, and G. Pan, “Mapping Very Large Scale Spiking Neuron Network to Neuromorphic Hardware,” in Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, V olume 3, ASPLOS ...

    2023

  26. [34]

    R-MaS3N: Robust Mapping of Spiking Neural Networks to 3D-NoC-Based Neuromorphic Systems for Enhanced Reliability,

    W. Y . Yerima, K. N. Dang, and A. B. Abdallah, “R-MaS3N: Robust Mapping of Spiking Neural Networks to 3D-NoC-Based Neuromorphic Systems for Enhanced Reliability,”IEEE Access, vol. 11, pp. 94664– 94678, 2023. Conference Name: IEEE Access

    2023

  27. [35]

    EdgeMap: An Optimized Mapping Toolchain for Spiking Neural Network in Edge Computing,

    J. Xue, L. Xie, F. Chen, L. Wu, Q. Tian, Y . Zhou, R. Ying, and P. Liu, “EdgeMap: An Optimized Mapping Toolchain for Spiking Neural Network in Edge Computing,”Sensors, vol. 23, p. 6548, Jan. 2023. Number: 14 Publisher: Multidisciplinary Digital Publishing Institute

    2023

  28. [36]

    Pymoo: Multi-Objective Optimization in Python,

    J. Blank and K. Deb, “Pymoo: Multi-Objective Optimization in Python,” IEEE Access, vol. 8, pp. 89497–89509, 2020. Conference Name: IEEE Access

    2020

  29. [37]

    Search biases in constrained evolutionary optimization,

    T.P. Runarsson and X. Yao, “Search biases in constrained evolutionary optimization,”IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 35, pp. 233–243, May 2005. Conference Name: IEEE Transactions on Systems, Man, and Cybernetics, Part ...

    2005

  30. [38]

    Reference Point Based NSGA-III for Preferred Solutions,

    Y . Vesikar, K. Deb, and J. Blank, “Reference Point Based NSGA-III for Preferred Solutions,” in2018 IEEE Symposium Series on Computational Intelligence (SSCI), pp. 1587–1594, Nov. 2018

    2018

  31. [39]

    MOEA/D: A Multiobjective Evolu- tionary Algorithm Based on Decomposition

    Q. Zhang, S. Member, and H. Li, “MOEA/D: A Multiobjective Evolu- tionary Algorithm Based on Decomposition.”

  32. [40]

    jMetalPy: A Python framework for multi-objective opti- mization with metaheuristics,

    A. Ben ´ıtez-Hidalgo, A. J. Nebro, J. Garc ´ıa-Nieto, I. Oregi, and J. Del Ser, “jMetalPy: A Python framework for multi-objective opti- mization with metaheuristics,”Swarm and Evolutionary Computation, vol. 51, p. 100598, Dec. 2019

    2019

  33. [41]

    A parallel global multiobjective framework for optimization: pagmo,

    F. Biscani and D. Izzo, “A parallel global multiobjective framework for optimization: pagmo,”Journal of Open Source Software, vol. 5, p. 2338, Sept. 2020

    2020

  34. [42]

    Open BEAGLE: a new C++ Evolutionary Computation framework,

    C. Gagn ´e and M. Parizeau, “Open BEAGLE: a new C++ Evolutionary Computation framework,” inProceedings of the 4th Annual Conference on Genetic and Evolutionary Computation, GECCO’02, (San Francisco, CA, USA), p. 888, Morgan Kaufmann Publishers Inc., July 2002

    2002

  35. [43]

    Opt4J: a modular framework for meta-heuristic optimization,

    M. Lukasiewycz, M. Glaß, F. Reimann, and J. Teich, “Opt4J: a modular framework for meta-heuristic optimization,” inProceedings of the 13th annual conference on Genetic and evolutionary computation - GECCO ’11, (Dublin, Ireland), p. 1723, ACM Press, 2011

    2011

  36. [44]

    Efficient mapping of large scale SNN and rate-based DNN on SENeCA|TU Delft Repository

    V . N. Prithvish, “Efficient mapping of large scale SNN and rate-based DNN on SENeCA|TU Delft Repository.”

  37. [45]

    Towards Efficient Deployment of Hybrid SNNs on Neuromorphic and Edge AI Hardware

    “Towards Efficient Deployment of Hybrid SNNs on Neuromorphic and Edge AI Hardware.”

  38. [46]

    A Hybrid SNN-ANN Network for Event-based Object Detection with Spatial and Temporal Attention,

    S. H. Ahmed, J. Finkbeiner, and E. Neftci, “A Hybrid SNN-ANN Network for Event-based Object Detection with Spatial and Temporal Attention,” Mar. 2024. arXiv:2403.10173 [cs]

    2024 arXiv

  39. [47]

    Particle swarm optimization,

    J. Kennedy and R. Eberhart, “Particle swarm optimization,” inProceed- ings of ICNN’95 - International Conference on Neural Networks, vol. 4, pp. 1942–1948 vol.4, Nov. 1995

    1942

  40. [48]

    Dask: Parallel Computation with Blocked algorithms and Task Scheduling,

    M. Rocklin, “Dask: Parallel Computation with Blocked algorithms and Task Scheduling,” inProceedings of the 14th Python in Science Conference (SciPy 2015), (Austin, Texas), pp. 126–132, 2015

    2015

  41. [49]

    End to End Learning for Self-Driving Cars,

    M. Bojarski, D. Del Testa, D. Dworakowski, B. Firner, B. Flepp, P. Goyal, L. D. Jackel, M. Monfort, U. Muller, J. Zhang, X. Zhang, J. Zhao, and K. Zieba, “End to End Learning for Self-Driving Cars,” Apr. 2016. Number: arXiv:1604.07316 arXiv:1604.07316 [cs]

    2016 arXiv

  42. [50]

    Simple Explanation of the No-Free-Lunch Theorem and Its Implications,

    Y . Ho and D. Pepyne, “Simple Explanation of the No-Free-Lunch Theorem and Its Implications,”Journal of Optimization Theory and Applications, vol. 115, pp. 549–570, Dec. 2002

    2002

  43. [51]

    Challenges and recent prospectives of 3D heterogeneous integration,

    S. Zhang, Z. Li, H. Zhou, R. Li, S. Wang, K.-W. Paik, and P. He, “Challenges and recent prospectives of 3D heterogeneous integration,” e-Prime - Advances in Electrical Engineering, Electronics and Energy, vol. 2, p. 100052, Jan. 2022

    2022

  44. [2019]

    Number: arXiv:1911.02385 arXiv:1911.02385 [cs]

    1911 arXiv

  45. [2020]

    arXiv:1805.01352 [cs]

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.