Optimisation of a silicon-tungsten electromagnetic calorimeter energy response to photons

Y. Shi, V. Boudry · 2026

Read Optimisation of a silicon-tungsten electromagnetic calorimeter energy response to photons by Y. Shi, V. Boudry (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-27853.txt.

Introduction

Shi and Boudry study the SiW electromagnetic calorimeter, developed by CALICE and ILD for over two decades and mature for linear machines. The device aims to meet set needs for future machines using Particle Flow with fine devices mapping showers. With round machines, better chips, and rapid machine learning growth, the design needs rework. They develop ML-based rebuild giving about 20 percent better sharpness at low levels and fixing leakage at high levels. The design is then retuned on this basis.

Catalog record

CREATORY. Shi, V. Boudry
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22270
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.27853

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Optimisation of a silicon-tungsten electromagnetic calorimeter energy response to photons Y. Shi, V. Boudry Abstract An innovative path for the detectors at future colliders to achieve higher performances is to use a Particle Flow approach, which requires highly granular calorimeters to image individual showers. The silicon–tungsten electromagnetic calorimeter (SiW-ECAL) aims at fulfilling all the expected physical and technical requirements. SiW-ECAL has been developed by the CALICE and ILD collaborations for more than two decades and is now reaching maturity, for linear machines. However, with the tendency towards circular machines, the progress of electronics and the rapid advancement of machine learning (ML) techniques, the SiW-ECAL design needs to be reoptimised to enhance its performance. This study develops ML-based reconstruction approaches for SiW-ECAL, achieving an approximate 20 % 20\text{\,}\% improvement in energy resolution in the low-energy range and effectively correcting energy leakage in the high-energy range. Subsequently, the SiW-ECAL design is re-optimized based on this method. 1 Introduction A key performance for experiments at future e + e - Higgs, Electro-Weak and Top (HET) factories, such as FCC-ee, CEPC, ILC, CLIC, is the jet energy resolution. A typical example is the separation of the ZZ and WW final states, which requires a Jet Energy Resolution(JER) of 3.8\text{\,}\% , or better [ 2 , 7 , 1 ] . The use of Particle Flow Algorithms (PFA) [ 4 , 5 ] is an effective approach to achieve this goal, by reconstructing all final-state particles individually, using the optimal sub-detector for each type. Charged particles are generally best measured with the trackers, in which case the calorimeters are used only for neutral particles. The PFA requires a high-precision tracking and high granularity calorimeters, to topologically separate contributions of particle showers, using a centimetric segmentation [ 14 ] . The PFA is particularly well suited for the HET factories, where the produced bosons predominantly decay in multi-jets of medium energy ( 45\text{\,}\mathrm{G}\mathrm{e}\mathrm{V}120\text{\,}\mathrm{G}\mathrm{e}\mathrm{V} ). High-granularity calorimeters have been studied by the CALICE collaboration for many years, with the silicon–tungsten electromagnetic calorimeter (SiW-ECAL) being one of them. The SiW-ECAL was initially designed for experiments at the ILC [ 3 , 13 ] ; however, it has become increasingly necessary to reoptimise its design, driven by the following three factors: 1. The particles in a jet are concentrated in the low-energy region; the neutral and low-energy particles exhibit particular behaviours w.r.t. their high-energy counterparts and require adequate treatments; e.g. it has been known for some time [ 11 ] that for photons in highly granular calorimeters, such as the SiW-ECAL, the energy resolution obtained by counting hits (cells above a given threshold) outperforms that obtained by summing the energies of the hits when the particle energy is below \sim$2\text{\,}\mathrm{G\text{e\kern-0.6458ptV}}$ , but this has somehow never been fully integrated in the optimisation. 2. The SiW-ECAL was previously optimized based on PFA’s [ 14 ] and some specific ( \tau ’s) final states [ 15 ] . With the rapid advancement of machine learning (ML) techniques, energy measurements can now benefit significantly from ML models such as GNN, PointNet, DGCNN, ParticleNet and so on [ 12 , 9 , 8 , 10 ] . It is therefore natural to reoptimise the SiW-ECAL to take full advantage of ML methods. 3. Circular colliders are proposed for future Higgs factories, the maximum centre of mass energy of them is lower than for linear colliders, which requires less material budget of ECAL. This study develops an ML-based energy reconstruction method, with an emphasis on the low-energy region, and reoptimised the SiW-ECAL with this method. This article is organized as follows: section 2 introduces the basic chain to derive calorimeter performance. Section 3 presents the ML-based energy reconstruction approach. Section 4 discusses the reoptimisation of the SiW-ECAL, and section 5 provides the conclusions and outlook. 2 Energy linearity and resolution Energy linearity and resolution are the key performance indicators of calorimeters. In this study, the SiW-ECAL performance was obtained through the basic chain consisting of three procedures: simulation, reconstruction, and evaluation. The simulation was performed using DD4hep [ 6 ] , a software framework built with ROOT and Geant4. A flexible SiW-ECAL prototype geometry was implemented, as illustrated in fig. 1 . It consists of 120 layers, each comprising a tungsten absorber of 0.7\text{\,}\mathrm{mm} , approximately 0.2\text{\,}\mathrm{X}_{0} in radiation length, and the silicon of 0.75\text{\,}\mathrm{m}\mathrm{m} . Including carbon fibre, cathodes, glue, and PCB, the total material of the prototype reaches about 26\text{\,}\mathrm{X}_{\mathrm{0}} . In the transverse direction, the prototype measures 200\text{\times}200\text{\,}\mathrm{mm} — approximately 2\text{\,}\mathrm{R}_{M} —which is sufficient to contain electromagnetic (EM) showers. The silicon is finely segmented into 1\text{\times}1\text{\,}\mathrm{mm} cells. Such an ultra-segmented geometry is designed to provide a convenient way to simulate less-segmented ECAL geometries by selecting readout layers and combining Si cells without rerunning the simulation. For instance, a geometry with 30 sampling layers and 5\text{\times}5\text{\,}\mathrm{mm} Si cells can be obtained by reading 1 in each 4 layers and combining 25 cells into 1. For reference, this 30-layer, 5\text{\times}5\text{\,}\mathrm{mm} configuration is used as the baseline geometry in this study, while alternative geometries are discussed in section 4 . Figure 1: Schematic diagram of the SiW-ECAL prototype geometry. The generation includes two kinds of photon datasets: one with continuous energies ranging from 0\text{\,}\mathrm{G}\mathrm{e}\mathrm{V}70\text{\,}\mathrm{G}\mathrm{e}\mathrm{V} , and the other with discrete energies of 0.10.250.51.02.05.010.020.0\dots 0.1\text{\,}\mathrm{G\text{e\kern-0.6458ptV}} . The continuous-energy data set contains 1.25 million events, of which 1 million are used for training and the rest for validation. The discrete-energy data set, with 50 thousand events per each energy point, is employed to evaluate the ECAL performances from different reconstruction methods or geometry configurations. It is referred to as the test sample, in contrast to the training sample and the validation sample. All photons are generated along the vertical direction without any angular spread. A threshold of 0.1\text{\,}\mathrm{M}\mathrm{I}\mathrm{P} was applied to calorimeter hits, where \mathrm{M}\mathrm{I}\mathrm{P} denotes the energy deposited by a minimum-ionizing particle traversing a Si cell. The reconstruction employed different energy reconstruction methods. Traditional approaches, “Sum E” and “N Hits,” are introduced as baselines in this section, while ML-based methods are presented later in section 3 . The Sum E method reconstructs the particle energy from the total deposited energy, whereas the N Hits method relies on the number of hits. The distributions of deposited energy and number of hits both deviate from a Gaussian form at low energies. Several fitting functions were tested to describe this behavior, and the gamma function provides the best modeling—fitting well at low energies and converging to Gaussian at high energies. Therefore, the gamma fit is adopted in this study. Calibration was carried out using the validation sample. For the Sum E method, a linear fit was applied to the deposited energy as a function of the particle energy, while for the N Hits method both a linear fit and a spline fit were used. The resulting calibration curves were then applied to the test sample to reconstruct the particle energy. The reconstructed energy distributions were fitted with a gamma function to extract the energy linearity and resolution. The obtained energy linearity and resolution are shown in fig. 2 . For the N Hits method, the uncorrected case corresponds to the linear fit calibration, whereas the corrected case corresponds

Read the complete text