February 13, 2026 Master Thesis Open Access

Baptist, Frank Michael

MARC21 XML Export

<?xml version='1.0' encoding='UTF-8'?>
<record xmlns="http://www.loc.gov/MARC21/slim">
  <leader>00000nam##2200000uu#4500</leader>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">GNN-ETM</subfield>
  </datafield>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">GNN</subfield>
  </datafield>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">Belle II</subfield>
  </datafield>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">Trigger</subfield>
  </datafield>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">GravNet</subfield>
  </datafield>
  <datafield tag="653" ind1=" " ind2=" ">
    <subfield code="a">Object Condensation</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">thesis</subfield>
    <subfield code="b">master-thesis</subfield>
  </datafield>
  <datafield tag="520" ind1=" " ind2=" ">
    <subfield code="a">&lt;p&gt;At the Belle II experiment, collisions occur at a rate higher than 200 MHz. Due to bandwidth and storage limitations, storing this amount of data for offline analysis is not feasible. That is why the experiment relies on a two-level trigger system as a preselection before passing the events to subsequent software-based processing and analysis. The first level, known as the Level 1 Trigger (L1 Trigger), is a hardware-based system that makes decisions within a few microseconds. The buffered data from the detector can only be stored for a limited time, which imposes strict latency requirements on the L1 Trigger.&lt;/p&gt;

&lt;p&gt;One approach to enhance the L1 Trigger is the development of the GNN-ETM. It utilizes dynamic graph building based on GravNet and can predict a varying number of clusters per event using Object Condensation. The GNN-ETM is implemented on field-programmable gate array (FPGA) hardware, which allows for low-latency inference and is suitable for deployment on the L1 Trigger. The current parasitic implementation of the GNN-ETM in the L1 Trigger has shown promising results in terms of precision and latency. Nevertheless, further improvements, especially in terms of latency, are necessary to develop a potential replacement for the clustering algorithm currently deployed on the ECL L1 Trigger.&lt;/p&gt;

&lt;p&gt;In this thesis, I present the optimization of the GNN-ETM for reduced latency, including modifications to the model architecture and training procedure.&lt;/p&gt;</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="a">Ferber, Torben</subfield>
    <subfield code="u">KIT/ETP</subfield>
    <subfield code="4">ths</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="a">Klute, Markus</subfield>
    <subfield code="u">KIT/ETP</subfield>
    <subfield code="4">ths</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="a">Haide, Isabel</subfield>
    <subfield code="u">KIT/ETP</subfield>
    <subfield code="4">ths</subfield>
  </datafield>
  <datafield tag="245" ind1=" " ind2=" ">
    <subfield code="a">Training and Optimization of a Graph Neural Network for Deployment on FPGA Hardware in the Belle II Level 1 Trigger</subfield>
  </datafield>
  <controlfield tag="001">22417</controlfield>
  <controlfield tag="005">20260324150544.0</controlfield>
  <datafield tag="856" ind1="4" ind2=" ">
    <subfield code="s">11387675</subfield>
    <subfield code="u">https://publish.etp.kit.edu/record/22417/files/Frank_Masterthesis-3.pdf</subfield>
    <subfield code="z">md5:e79ef60462966d2deef2030cca66da26</subfield>
  </datafield>
  <datafield tag="909" ind1="C" ind2="O">
    <subfield code="o">oai:publish.etp.kit.edu:22417</subfield>
    <subfield code="p">user-belle2</subfield>
    <subfield code="p">user-hardware</subfield>
    <subfield code="p">user-etp</subfield>
  </datafield>
  <datafield tag="260" ind1=" " ind2=" ">
    <subfield code="c">2026-02-13</subfield>
  </datafield>
  <datafield tag="041" ind1=" " ind2=" ">
    <subfield code="a">eng</subfield>
  </datafield>
  <datafield tag="100" ind1=" " ind2=" ">
    <subfield code="a">Baptist, Frank Michael</subfield>
    <subfield code="u">KIT/ETP</subfield>
  </datafield>
  <datafield tag="542" ind1=" " ind2=" ">
    <subfield code="l">open</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">user-belle2</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">user-etp</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">user-hardware</subfield>
  </datafield>
</record>