Pith. sign in

REVIEW 3 major objections 32 references

One modular architecture generates configurable SNN hardware so the same cores can run both feedforward classifiers and multi-FPGA recurrent networks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 04:13 UTC pith:CRQLEAC4

load-bearing objection Solid dual-workload FPGA SNN generator with real multi-FPGA NEST match; evidence is honest but still LIF-only and short-horizon, with the big datapath fixes still on the roadmap. the 3 major comments →

arxiv 2607.03191 v1 pith:CRQLEAC4 submitted 2026-07-03 cs.AR cs.ET

AIGOR: A Modular, Event-Driven Neuromorphic Architecture for Configurable SNN Inference

Pierpaolo Perticaroli , Roberto Ammendola , Andrea Biagioni , Ottorino Frezza , Francesca Lo Cicero , Michele Martinelli , Pier Stanislao Paolucci , Elena Pastorelli
show 5 more authors
Luca Pontisso Cristian Rossi Francesco Simula Piero Vicini Alessandro Lonardo
This is my paper
classification cs.AR cs.ET
keywords spiking neural networksneuromorphic architectureFPGA acceleratorevent-driven computingconfigurable hardwareSNN inferencepacket-switched interconnecttimestep synchronization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Spiking neural network hardware is usually built for one neuron model, one interconnect style, and one workload class, so feedforward inference chips and large recurrent simulators rarely share a substrate. This paper presents AIGOR, an event-driven architecture assembled from parameterized compute, memory, and communication blocks whose neuron model, numeric format, folding, and partitioning are fixed only when an instance is generated from a single declarative specification. A prototype on Versal FPGAs maps a snnTorch MNIST classifier and a NEST Brunel-style recurrent network onto the same cores: the classifier matches reference accuracy, and the recurrent network matches NEST at spike-level precision across cores on two boards. Multi-node timestep synchronization is checked in simulation up to a thousand cores. Measured limits point to the synaptic-delivery path and the global barrier, which the same configurable structure is meant to absorb as refinements rather than a redesign.

Core claim

AIGOR shows that a library of parameterized IP blocks, driven by one YAML specification that emits cores, neuron kernels, and synaptic-memory images, can realize two deliberately different SNN regimes on the same timestep-synchronized, packet-switched hardware: a feedforward image classifier that reproduces its software accuracy, and a multi-core multi-FPGA recurrent balanced network that matches its NEST reference at spike-level precision for the exercised interval.

What carries the argument

Configuration-as-architecture: a declarative specification expands into timestep-synchronized processing cores that host replaceable neuron kernels, accumulate spikes in circular delay buffers, and exchange spike/sync packets over a packet-switched fabric, so model, precision, folding, and partitioning become instance parameters rather than design-time commitments.

Load-bearing premise

That matching accuracy on a small fully connected MNIST net and exact spikes for only a few milliseconds of a recurrent network is enough evidence that the same cores are a general substrate across SNN regimes.

What would settle it

Map a third, non-LIF workload (for example adaptive-exponential neurons or a denser multi-layer recurrent net) through the same single-specification flow onto the prototype cores and check whether spike-level or accuracy agreement with the software reference still holds over a longer simulated window than the reported ~2 ms exact-match regime.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Feedforward classifiers and recurrent neuroscience models can share one FPGA core library and generation flow instead of separate accelerators.
  • Changing neuron model, fixed-point width, or spatial-versus-temporal folding becomes a specification edit rather than a redesign of the datapath.
  • Multi-FPGA partitions of a network can reuse the same core design, with the torus and sync protocol carrying spikes across boards.
  • Bottlenecks identified in synaptic delivery and the global barrier can be attacked as configuration-compatible refinements to the same cores.
  • Once banking, fused spike/sync, and neighbor-local sync are in place, accuracy–area–latency–energy trade-offs can be swept as a measurable design space on one device.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the generation flow stays consistent, design-space exploration of SNN hardware could become a systematic Pareto sweep rather than a collection of one-off accelerators.
  • Streaming inference at detector front-ends is a natural next regime for the same cores, because the architecture already decouples compute from a low-latency packet fabric.
  • Long-horizon recurrent fidelity will hinge on fixed-point policy and barrier cost more than on the modular packaging itself; those axes are the real scalability test.
  • A banked synaptic accumulator that removes the per-worker merge could matter more for dense fully connected layers than for sparse random connectivity, so workload class will still shape the best configuration.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The paper presents AIGOR, a modular event-driven neuromorphic architecture for SNN inference assembled from parameterized compute, memory, and communication IP blocks. Neurons are organized into timestep-synchronized cores that exchange spikes as packets over a packet-switched fabric (Apeiron); neuron model, numeric format, spatial/time-multiplexed folding, and partitioning across cores/workers are configuration axes resolved from a single YAML specification that generates core RTL, HLS neuron kernels, and synaptic-memory images. A Versal VPK180 prototype is validated on two workloads mapped onto the same cores: a feedforward snnTorch MNIST classifier (256 o128 o10 LIF) that reproduces ~95% reference accuracy, and a Brunel-style recurrent NEST network (~2048 LIF neurons) that matches NEST at spike-level precision across up to 8 cores on two FPGAs (exact spike times for ~2 ms simulated time under 32-bit/12-integer fixed-point). Post-implementation utilization (Table 2), measured classifier throughput (568 samples/s), and SystemC multi-node sync validation to 1000 cores on a 10 imes10 imes10 torus are reported; measured bottlenecks motivate unimplemented refinements (banked synaptic accumulator, fused spike/sync, GALS).

Significance. If the dual-workload result holds, AIGOR is a concrete contribution to configurable FPGA SNN systems: one generation flow and the same cores support both a machine-learning feedforward classifier and a multi-FPGA neuroscience-style recurrent network, with external-reference correctness (snnTorch accuracy; NEST spike-level match) rather than self-defined metrics. The configuration axes (Table 1), hierarchical loader, and explicit localization of the synaptic-delivery bottleneck give a usable platform for design-space exploration. Strengths include measured multi-FPGA spike-level agreement, a utilization sweep, and a clear roadmap that treats refinements as changes to the same cores. The work is incremental relative to Spiker+ and related generators, and the general-substrate claim is only partially evidenced (LIF-only hardware, short-horizon exact NEST match, simulation-only 1000-core sync, unimplemented Section 7 datapaths), but the dual-regime prototype is still a solid systems result for the FPGA neuromorphic community.

major comments (3)
  1. §6.1 and Fig. 4: spike-level NEST agreement is exact only for ~2 ms of simulated time (0.1 ms timestep) under the representative fixed-point setting, after which trajectories diverge; population statistics then agree. For a multi-core multi-FPGA recurrent claim this horizon is short. The manuscript should either (i) quantify divergence (e.g., spike-time error / ISI / rate error vs. simulated time and bit-width) or (ii) state the claim as short-horizon exact match plus longer-horizon statistical agreement, and note that a full fixed-point sweep is deferred to §7.5.
  2. §5.1 and §3.3: both published hardware workloads use LIF only (snnTorch Leaky and NEST iaf_psc_delta). Adaptive-exponential and IAF are listed as supported kernels, but no FPGA result exercises them. The abstract and contribution bullets frame AIGOR as spanning SNN regimes via configurable models; either add at least one non-LIF on-FPGA check or narrow the claim to LIF with model configurability as a generation-flow property not yet hardware-validated beyond LIF.
  3. §6.2 and §7: the central throughput claim localizes the bottleneck in the synaptic-delivery datapath and global barrier and motivates banked accumulator, fused spike/sync, and GALS as the next stage. None of these are implemented or measured. The paper is acceptable as a prototype-plus-roadmap, but the abstract and §1 should not imply that the refinements are part of the validated system; keep them clearly labeled as proposed, and avoid performance claims that depend on them.

Circularity Check

0 steps flagged

No circularity: dual-workload claims are empirical measurements against external snnTorch/NEST references, not quantities forced by the architecture's own definitions or self-citation chain.

full rationale

AIGOR is an architecture/systems paper whose load-bearing claims are hardware measurements, not first-principles derivations. The classifier accuracy (~95% MNIST) is checked against an independent snnTorch reference; the recurrent network is checked spike-for-spike against an independent NEST reference under identical stimuli. Resource utilization and throughput are post-implementation measurements on VPK180 silicon. Multi-node sync to 1000 cores is a simulation exercise of the barrier protocol, not a fitted prediction. Self-citations (Apeiron transport [2], companion ePIC study [32]) supply infrastructure context and do not define or force the accuracy or spike-match results. Section 7 refinements are proposed future work, not claimed outcomes. There is no self-definitional loop, no fitted parameter renamed as prediction, no uniqueness theorem imported from the authors, and no ansatz smuggled in via citation. The derivation chain is simply: configure instance from YAML → synthesize → run → compare to external software. Score 0.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 3 invented entities

This is a systems/architecture paper, not a theory derivation. Load-bearing content is engineering assumptions (discrete-time barrier sync, packet-switched spike delivery, HLS/SystemC generation fidelity) plus configuration choices (precision, network sizes) rather than fitted physical constants. Invented entities are the architecture and proposed datapath modules; independent evidence for the prototype is partial (FPGA runs), for the refinements is none yet.

free parameters (4)
  • Fixed-point format (32 total bits, 12 integer)
    Representative numeric setting used for all reported hardware agreement; not derived from first principles. Exact spike match window depends on this choice.
  • MNIST encoding window (25 timesteps) and 16×16 on-chip downsample
    Hand-chosen encoding/preprocessing that defines the classifier workload and measured 568 samples/s operating point.
  • Recurrent network scale (~1638 E / 410 I, p=0.1, 800 Hz Poisson)
    Brunel-style parameters chosen for the NEST benchmark; validation horizon and multi-core partition depend on this instance.
  • Worker/core folding (e.g., 8 workers × 32 neurons spatial for classifier; W×n sweeps in Table 2)
    Mapping knobs selected for the prototype; utilization and throughput claims are for these points, not a full Pareto sweep.
axioms (5)
  • domain assumption Discrete-time, globally timestep-synchronized SNN execution with barrier (sync) events preserves correct spike ordering across cores.
    Core execution model in Section 3.1; multi-node correctness rests on this barrier semantics.
  • domain assumption Event-driven synaptic fanout from on-chip synaptic memory plus circular delay buffers correctly implements delayed synaptic delivery for the supported neuron models.
    Section 3.2 delivery path; functional match to NEST/snnTorch assumes this microarchitecture is faithful.
  • domain assumption Apeiron packet-switched fabric provides reliable low-latency intra- and inter-FPGA transport transparent to cores.
    Section 2–3 transport substrate; multi-FPGA results inherit Apeiron properties.
  • ad hoc to paper Parameterized SystemC/Catapult + Vitis HLS + VHDL generation from one YAML yields mutually consistent RTL, kernels, and synaptic images.
    Section 4 generation flow; the 'single specification' claim depends on toolchain fidelity not independently audited here.
  • domain assumption snnTorch Leaky and NEST iaf_psc_delta LIF conventions can be reproduced by configurable kernels for the exercised cases.
    Section 5.1; dual-tool validation assumes kernel parity with those reference models.
invented entities (3)
  • AIGOR architecture (timestep-synchronized cores + configurable workers + generation flow) no independent evidence
    purpose: Provide a modular, retargetable FPGA substrate for multiple SNN workload classes from one declarative description.
    Primary contribution; evidence is prototype runs and simulation, not external independent reimplementation.
  • Banked synaptic accumulator with P parallel RMW lanes and optional scheduled crossbar no independent evidence
    purpose: Relieve measured synaptic-delivery bottleneck of per-worker routing and round-robin merge.
    Section 7.1 design proposal; not implemented or measured on FPGA in this paper.
  • Fused single-word spike/sync protocol and neighbor-local (GALS) torus synchronization no independent evidence
    purpose: Reduce per-timestep barrier cost and improve multi-core scaling.
    Sections 7.3–7.4; roadmap items without hardware results here.

pith-pipeline@v1.1.0-grok45 · 17491 in / 3853 out tokens · 35158 ms · 2026-07-12T04:13:56.208405+00:00 · methodology

0 comments
read the original abstract

Spiking neural networks (SNNs) run today on a fragmented landscape of hardware: dedicated neuromorphic processors, application-specific FPGA accelerators, and large-scale neuroscience simulators, each typically built around a fixed neuron model, execution strategy, or workload class. We present AIGOR, a modular, event-driven neuromorphic architecture for spiking neural network inference. AIGOR organizes neurons into timestep-synchronized processing cores that exchange spikes as packets over a packet-switched communication layer, and it is assembled from a library of parameterized compute, memory, and communication IP blocks rather than as a one-off design for a single network. The neuron model, numeric precision, the folding of neurons onto hardware, and the partitioning across cores are configured per instance rather than committed at design time; a single declarative specification then generates the cores, neuron kernels, and synaptic-memory images that realize a given network. We validate a first prototype on the AMD Versal VPK180 across two deliberately different workloads mapped onto the same cores: a feedforward image classifier trained in snnTorch and a recurrent bal anced random network modeled in NEST. The classifier reproduces its snnTorch reference accuracy, and the recurrent network matches its NEST reference at spike-level precision across multiple cores spanning two FPGAs. We report post-implementation resource utilization and validate the multi-node synchronization scheme in simulation up to one thousand cores on a three-dimensional torus. The prototype's measured limits localize the throughput bottleneck in the synaptic-delivery datapath and the global timestep barrier, and motivate a set of datapath refinements, now in development, that the configurable structure of the architecture admits as changes to the same cores.

Figures

Figures reproduced from arXiv: 2607.03191 by Alessandro Lonardo, Andrea Biagioni, Cristian Rossi, Elena Pastorelli, Francesca Lo Cicero, Francesco Simula, Luca Pontisso, Michele Martinelli, Ottorino Frezza, Piero Vicini, Pierpaolo Perticaroli, Pier Stanislao Paolucci, Roberto Ammendola.

Figure 1
Figure 1. Figure 1: AIGOR prototype deployed across two AMD Versal Premium [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Prototype SNN core with local per-worker synaptic routing. Events [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AIGOR configuration-driven generation flow. A single YAML [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Functional validation on the recurrent benchmark. [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Design space for the banked synaptic-accumulator redesign. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The two configurable routing modes over the [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

32 extracted references · 3 canonical work pages

  1. [1]

    Boahen,Point-to-point connectivity between neuromorphic chips using address events, IEEE Trans

    K. Boahen,Point-to-point connectivity between neuromorphic chips using address events, IEEE Trans. Circuits Syst. II 47 (2000) 416, doi:10.1109/82.842110

    work page doi:10.1109/82.842110 2000

  2. [2]

    Ammendola et al.,APEIRON: a Framework for High Level Program- ming of Dataflow Applications on Multi-FPGA Systems, EPJWebConf

    R. Ammendola et al.,APEIRON: a Framework for High Level Program- ming of Dataflow Applications on Multi-FPGA Systems, EPJWebConf. 295 (2024) 11002 (CHEP 2023), doi:10.1051/epjconf/202429511002

    doi:10.1051/epjconf/202429511002 2024

  3. [3]

    Duarte et al.,Fast inference of deep neural networks in FPGAs for particle physics, JINST 13 (2018) P07027

    J. Duarte et al.,Fast inference of deep neural networks in FPGAs for particle physics, JINST 13 (2018) P07027. 18

    2018

  4. [4]

    Aarrestad et al.,Fast convolutional neural networks on FPGAs with hls4ml, Mach

    T. Aarrestad et al.,Fast convolutional neural networks on FPGAs with hls4ml, Mach. Learn.: Sci. Technol. 2 (2021) 045015

    2021

  5. [5]

    P. Perticaroli et al.,Achieving Low-Latency, High-Throughput On- line Partial Particle Identification for the NA62 Experiment Us- ing FPGAs and Machine Learning, Electronics 14 (2025) 1892, doi:10.3390/electronics14091892

    work page doi:10.3390/electronics14091892 2025

  6. [6]

    Anderson et al.,FELIX: a high-throughput network approach for interfacing to front-end electronics for ATLAS experiments, J

    J. Anderson et al.,FELIX: a high-throughput network approach for interfacing to front-end electronics for ATLAS experiments, J. Phys.: Conf. Ser. 664 (2015) 082050

    2015

  7. [7]

    Vallarino et al.,The dual-radiator RICH detector of the ePIC exper- iment at the EIC, Nucl

    S. Vallarino et al.,The dual-radiator RICH detector of the ePIC exper- iment at the EIC, Nucl. Instr. Meth. A (2024) 168834

    2024

  8. [8]

    Maass,Networks of spiking neurons: The third generation of neural network models, Neural Networks 10 (1997) 1659

    W. Maass,Networks of spiking neurons: The third generation of neural network models, Neural Networks 10 (1997) 1659

    1997

  9. [9]

    Gerstner, W.M

    W. Gerstner, W.M. Kistler, R. Naud, L. Paninski,Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition, Cambridge University Press, 2014

    2014

  10. [10]

    Eshraghian, M

    J.K. Eshraghian, M. Ward, E.O. Neftci, X. Wang, G. Lenz, G. Dwivedi, M. Bennamoun, D.S. Jeong, W.D. Lu,Training Spiking Neural Net- works Using Lessons From Deep Learning, Proc. IEEE 111 (2023) 1016, doi:10.1109/JPROC.2023.3308088

    doi:10.1109/jproc.2023.3308088 2023

  11. [11]

    Gewaltig, M

    M.-O. Gewaltig, M. Diesmann,NEST (NEural Simulation Tool), Schol- arpedia 2 (2007) 1430

    2007

  12. [12]

    Golosio, G

    B. Golosio, G. Tiddia, C. De Luca, E. Pastorelli, F. Simula, P.S. Paolucci,Fast Simulations of Highly-Connected Spiking Corti- cal Models Using GPUs, Front. Comput. Neurosci. 15 (2021) 627620, doi:10.3389/fncom.2021.627620

    work page doi:10.3389/fncom.2021.627620 2021

  13. [13]

    Brunel,Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons, J

    N. Brunel,Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons, J. Comput. Neurosci. 8 (2000) 183

    2000

  14. [14]

    K. Roy, A. Jaiswal, P. Panda,Towards spike-based machine intelligence with neuromorphic computing, Nature 575 (2019) 607

    2019

  15. [15]

    Schuman, S.R

    C.D. Schuman, S.R. Kulkarni, M. Parsa, J.P. Mitchell, P. Date, B. Kay, Opportunities for neuromorphic computing algorithms and applications, Nat. Comput. Sci. 2 (2022) 10

    2022

  16. [16]

    Davies et al.,Loihi: A neuromorphic manycore processor with on- chip learning, IEEE Micro 38 (2018) 82

    M. Davies et al.,Loihi: A neuromorphic manycore processor with on- chip learning, IEEE Micro 38 (2018) 82. 19

    2018

  17. [17]

    Orchard et al.,Efficient Neuromorphic Signal Processing with Loihi 2, in 2021 IEEE Workshop on Signal Processing Systems (SiPS), 2021, p

    G. Orchard et al.,Efficient Neuromorphic Signal Processing with Loihi 2, in 2021 IEEE Workshop on Signal Processing Systems (SiPS), 2021, p. 254

    2021

  18. [18]

    Merolla et al.,A million spiking-neuron integrated circuit with a scalable communication network and interface, Science 345 (2014) 668

    P.A. Merolla et al.,A million spiking-neuron integrated circuit with a scalable communication network and interface, Science 345 (2014) 668

    2014

  19. [19]

    Pei et al.,Towards artificial general intelligence with hybrid Tianjic chip architecture, Nature 572 (2019) 106

    J. Pei et al.,Towards artificial general intelligence with hybrid Tianjic chip architecture, Nature 572 (2019) 106

    2019

  20. [20]

    Schemmel, D

    J. Schemmel, D. Brüderle, A. Grübl, M. Hock, K. Meier, S. Millner,A wafer-scale neuromorphic hardware system for large-scale neural model- ing, in Proc. IEEE ISCAS, 2010, p. 1947

    2010

  21. [21]

    Furber, F

    S.B. Furber, F. Galluppi, S. Temple, L.A. Plana,The SpiNNaker Project, Proc. IEEE 102 (2014) 652

    2014

  22. [22]

    C. Mayr, S. Höppner, S. Furber,SpiNNaker 2: A 10 Million Core Processor System for Brain Simulation and Machine Learning, arXiv:1911.02385 (2019)

    Pith/arXiv arXiv 1911

  23. [23]

    S. Deng, D. Yu, C. Lv, X. Du, L. Jiang, X. Zhao, W. Tong, X. Zheng, W. Fang, P. Zhao, G. Pan, S. Dustdar, A.Y. Zomaya,Edge Intelligence with Spiking Neural Networks, arXiv:2507.14069 (2025)

    Pith/arXiv arXiv 2025

  24. [24]

    Neftci, H

    E.O. Neftci, H. Mostafa, F. Zenke,Surrogate Gradient Learning in Spik- ing Neural Networks, IEEE Signal Processing Magazine 36 (2019) 51

    2019

  25. [25]

    S.R. Kulkarni et al.,On-Sensor Data Filtering using Neuromorphic Computing for High Energy Physics Experiments, in Proceedings of the 2023 International Conference on Neuromorphic Systems (ICONS ’23), doi:10.1145/3589737.3605976

    doi:10.1145/3589737.3605976 2023

  26. [26]

    Coradin et al.,Unsupervised Particle Tracking with Neuromorphic Computing, arXiv:2307.11242

    E. Coradin et al.,Unsupervised Particle Tracking with Neuromorphic Computing, arXiv:2307.11242

    Pith/arXiv arXiv

  27. [27]

    Kauth, T

    K. Kauth, T. Stadtmann, V. Sobhani, T. Gemmeke,neuroAIx: A high- performance, energy-efficient and reconfigurable accelerator for neuro- scientific spiking neural networks, Front. Comput. Neurosci. 17 (2023) 1144143

    2023

  28. [28]

    Carpegna, A

    A. Carpegna, A. Savino, S. Di Carlo,Spiker+: A Framework for the Generation of Efficient Spiking Neural Networks FPGA Accelerators for Inference at the Edge, IEEE Trans. Emerg. Topics Comput. (2024), doi:10.1109/TETC.2024.3511676

    doi:10.1109/tetc.2024.3511676 2024

  29. [29]

    Farahani, M.R

    M. Farahani, M.R. Roshanshah, S. Safari,Flexi-NeurA: A Configurable Neuromorphic Accelerator with Adaptive Bit-Precision Exploration for Edge SNNs, arXiv:2602.18140 (2026). 20

    arXiv 2026

  30. [30]

    Al Hafiz, A

    M.I. Al Hafiz, A. Podobas,NeuroRing: Scaling Spiking Neural Networks via Multi-FPGA Bidirectional Ring Topologies and Stream-Dataflow Ar- chitectures, arXiv:2604.28059 (2026)

    Pith/arXiv arXiv 2026

  31. [31]

    Gautam, P

    A. Gautam, P. Date, S. Kulkarni, R. Patton, T. Potok,NeuroCoreX: An Open-Source FPGA-Based Spiking Neural Network Emulator with On-Chip Learning, arXiv:2506.14138 (2025)

    Pith/arXiv arXiv 2025

  32. [32]

    Perticaroli et al.,AIGOR for online data reduction at the ePIC dRICH, in preparation (2026)

    P. Perticaroli et al.,AIGOR for online data reduction at the ePIC dRICH, in preparation (2026). 21

    2026