Back to search

University of British Columbia

Prototype machine learning for pion reconstruction in ATLAS Level-0 Global FPGA trigger for HL-LHC

Abstract

dc:description

The Large Hadron Collider (LHC) will be upgraded to the High-Luminosity LHC (HL-LHC) during Long Shutdown 3 starting in mid-2026. In the HL-LHC era, a high-pileup environment resulting in up to 200 proton-proton interactions per bunch-crossing is expected in the ATLAS detector. Identifying and calibrating physics objects in the ATLAS calorimeters under these conditions requires a more capable hardware trigger system, which ATLAS will deploy as part of its Phase-II upgrade. Recent studies, based on detector conditions from previous data-taking periods, have shown that machine learning outperforms the existing algorithms in classifying particle shower types and calibrating energy deposits by pions, the most abundantly produced final-state hadrons, in the ATLAS calorimeters. This thesis extends these studies and presents the performance of the DeepSets neural network for this classification and calibration under HL-LHC conditions. A custom asymmetric loss function is developed, enabling the model to learn the out-of-cluster and dead material energies, achieving a correct calibration above 1 GeV of deposited energy and a resolution improvement above 5 GeV. This performance is retained by a model of only 3 thousand parameters and with cell energies restricted to the GEP 6-40-4 reduced-precision format. This thesis also presents a first hardware prototype of the DeepSets model on a Zynq-7020 FPGA. A novel systolic architecture, MEISSA-TMAC, is developed to eliminate the inter-layer stalls incurred by existing designs in multi-layer networks. The prototype design is successfully placed and routed using less than half of the available resources and within the 2 μs latency budget. The next design iteration will need to parallelize this implementation for integration into the VP1802 FPGA of the ATLAS Level-0 Global trigger.

Degree

thesis:*
Name thesis:degree_name
Master of Science - MSc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Physics
Grantor dc:publisher
University of British Columbia
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Leong, Chin Chong (Kelvin)
Contributors dc:contributor
  • Gay, Colin
  • Swiatlowski, Maximilian

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/95357

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Leong, Chin Chong (Kelvin). Prototype machine learning for pion reconstruction in ATLAS Level-0 Global FPGA trigger for HL-LHC. master's thesis, University of British Columbia, 2026. http://hdl.handle.net/2429/95357