Optimization of the local reconstruction in a high granular calorimeter using a heterogeneous computing model

The High-Luminosity phase of the Large Hadron Collider (HL-LHC) at the European Organisation for Nuclear Research (CERN) will deliver an unprecedented dataset of proton-proton collisions at the highest centre-of-mass energies to date. With over 140 simultaneous interactions per bunch crossing and luminosity of around $10^{34} ~\text{cm}^{-2} ~\text{s}^{-1}$, both the CMS detector and its computing infrastructure face significant challenges. These include severe radiation damage, high particle fluxes, and the need for fast and precise reconstruction to isolate physics signals of interest, particularly for precision measurements. To meet these demands, the CMS experiment is undergoing major upgrades, including the replacement of the endcap calorimeters with a high-granularity sampling calorimeter (HGCAL). HGCAL will provide combined energy and time measurements from approximately six million channels, enabling accurate particle flow reconstruction under extreme pileup. The upgrade features a silicon-based sampling structure with fine spatial segmentation, enhancing both spatial and temporal resolution.

This thesis explores the development and validation of a local reconstruction and calibration framework using data from HGCAL beam tests. A key contribution is an automated method to extract the most probable value (MPV) of energy deposits from minimum ionising particles (MIP) across individual silicon sensors, a critical step in calorimeter energy calibration. Essential processing steps include pedestal subtraction, noise correction, and timing alignment. The calibration algorithms were tested and validated under real test beam conditions. This work contributes to a per-channel calibration method that will be essential for the accurate calibration of HGCAL in preparation for its deployment in the HL-LHC era.