% 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”.

@PHDTHESIS{Kleijnen:1032287,
      author       = {Kleijnen, Robert},
      title        = {{N}eu{C}o{NS} and {S}tacked-{N}et: {F}acilitating the
                      {C}ommunication for {A}ccelerated {N}euroscientific
                      {S}imulations},
      volume       = {106},
      school       = {Duisburg-Essen},
      type         = {Dissertation},
      address      = {Jülich},
      publisher    = {Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag},
      reportid     = {FZJ-2024-06126},
      isbn         = {978-3-95806-788-2},
      series       = {Schriften des Forschungszentrums Jülich Reihe Information
                      / Information},
      pages        = {xx, 110, xxi-xxxiv},
      year         = {2024},
      note         = {Dissertation, Duisburg-Essen , 2024},
      abstract     = {Investigating the inner workings of the brain runs into
                      multiple challenges. A couple of those challenges are the
                      level of detail which can be obtained by in vivo experiments
                      and the time required to investigate long-term processes. A
                      potential solution to thesechallenges is the use of
                      simulators which run at an accelerated speed compared to
                      biological real time and allow to probe physiological
                      features that are inaccessible otherwise. Computer systems
                      can simulate parts of the brain, but traditional computer
                      architectures achieve neither the speed-up factor nor the
                      scale desired. The main challenges here are the massively
                      parallel operation of the brain, the decentralisation of
                      memory and the high level of connectivity between neurons.
                      The eld of Neuromorphic Computing attempts to solve these
                      challenges by developing computer systems with a
                      fundamentally dierent architecture. By using the knowledge
                      obtained by neuroscience, the architecture of the system can
                      be based on the structures found in the biological brain to
                      better mimic these characteristics. One of the challenges of
                      such a system is the high throughput, low latency
                      communication of spike events through the system. This work
                      focuses on these challenges and investigates the
                      communication trac generated on a neuromorphic system when
                      running biologically representative large-scale spiking
                      neural networks in order to come up with a suitable
                      solution. The investigation of the communication trac is
                      done using a Python based network simulator, which is
                      presented in this work as well. This simulator analyses the
                      communication trac with respect to both, the amount of
                      communication data as well as the latency. To prove the
                      correct functionality of the simulator, simulation data are
                      compared against results obtained with already existing
                      models as well as experimental data. This comparison not
                      only proves the correct functionality of the simulator, but
                      also shows o some of the advantages of this tool. The
                      simulator oers a higher level of detail than existing models
                      while also being able to handle more complex heterogeneous
                      connectivity models. This last feature is unique for this
                      tool and is especially important in this work as the
                      heterogeneity is a key characteristic in biological neural
                      networks. Simultaneously, thisalso allows the evaluation of
                      neuron mapping algorithms by the simulator. To better
                      understand the impact of dierent network designs, the tool
                      is used to evaluate the performances resulting from a
                      variation of dierent design aspects such as the topology,
                      the routing algorithm, casting protocol and node size. The
                      goal of this study is to develop a novel communication
                      network concept that can facilitate the communication in a
                      large-scale neurmorphic system next to providing the tooling
                      for its examination. To achieve this goal, the knowledge
                      obtained during the simulation study is used to
                      conceptualize a new stacked network topology. This network
                      topology shows a reduction of the network load over a factor
                      of 10 and a reduction of the latency up to a factor of 3,
                      while hardly increasing the hardware cost of the network.},
      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)3 / PUB:(DE-HGF)11},
      urn          = {urn:nbn:de:0001-20241209140418012-7496251-2},
      doi          = {10.34734/FZJ-2024-06126},
      url          = {https://juser.fz-juelich.de/record/1032287},
}