% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@INPROCEEDINGS{Kleijnen:906212,
      author       = {Kleijnen, Robert and Robens, Markus and Schiek, Michael and
                      van Waasen, Stefan},
      title        = {{A} {N}etwork {S}imulator for the {E}stimation of
                      {B}andwidth {L}oad and {L}atency {C}reated by
                      {H}eterogeneous {S}piking {N}eural {N}etworks on
                      {N}euromorphic {C}omputing {C}ommunication {N}etworks},
      publisher    = {IEEE},
      reportid     = {FZJ-2022-01296},
      pages        = {320-327},
      year         = {2021},
      abstract     = {Observing long-term learning effects caused by neuron
                      activity in the human brain in vivo, over a period of weeks,
                      months, or years, is impractical. Over the last decade, the
                      field of neuromorphic computing hardware has grown
                      significantly, i.e. SpiNNaker, BrainScaleS and Neurogrid.
                      These novel many-core simulation platforms offer a practical
                      alternative to study neuron behaviour in the brain at an
                      accelerated rate, with a high level of detail. However, they
                      do by far not reach human brain scales yet as in particular
                      the massive amount of spike communication turns out to be a
                      bottleneck. In this paper, we introduce a network simulator
                      specifically developed for the analysis of bandwidth load
                      and latency of different network topologies and
                      communication protocols in neuromorphic computing
                      communication networks in high detail. Unique to this
                      simulator, compared to state of the art network models and
                      simulators, is its ability to simulate the impact of
                      heterogeneous neural connectivity by different models as
                      well as the evaluation of neuron mapping algorithms. We
                      cross-check the simulator by comparing the results of a run
                      using a homogeneous neural network to the bandwidth load
                      resulting from comparable works, but simultaneously show the
                      increased level of detail reached with our simulator.
                      Finally, we show the impact heterogeneous connectivity can
                      have on the bandwidth and how different neuron mapping
                      algorithms can enhance this effect.},
      month         = {Dec},
      date          = {2021-12-20},
      organization  = {2021 IEEE 14th International Symposium
                       on Embedded Multicore/Many-core
                       Systems-on-Chip (MCSoC), Singapore
                       (Singapore), 20 Dec 2021 - 23 Dec 2021},
      cin          = {ZEA-2},
      cid          = {I:(DE-Juel1)ZEA-2-20090406},
      pnm          = {5234 - Emerging NC Architectures (POF4-523)},
      pid          = {G:(DE-HGF)POF4-5234},
      typ          = {PUB:(DE-HGF)8},
      UT           = {WOS:000788295800046},
      doi          = {10.1109/MCSoC51149.2021.00054},
      url          = {https://juser.fz-juelich.de/record/906212},
}