In the mid-1990s the design of the CMS detector was being finalized, culminating in the relevant Technical Design Reports at the end of the decade. This was followed by the start of construction. The main constraints at the time were radiation tolerance and immunity to the effects of pileup. The detectors were designed to withstand the radiation (neutron fluence and ionizing dose) from accumulating ~500 fb-1 of 14 TeV proton-proton collisions (spread over ~10 years of operation) with a pileup of 20-25 events. The LHC is well on its way to accumulating hundreds of fb-1 and may reach the 500 fb-1 by LS3. But the pileup is already averaging about double the original design. It is a testament to the original detector designs and the ingenuity of the physicists/engineers that operate and maintain them that this huge pileup can be mitigated and the underlying physics uncovered, even at the trigger level. The challenges of the High-Luminosity phase of the LHC are similar, but far more extreme: CMS must perform well after an integrated luminosity of ~3000 fb-1 with a pileup of between 140 and 200. Whilst some of the existing detectors – mainly those in the barrel region – can withstand the increased radiation, the challenge of the pileup – particularly at trigger level – requires new electronics for all detectors. And in the endcaps (pseudorapidity > 1.5) only the forward calorimeter (known as HF)  and the muon chambers will survive the radiation and continue to perform well. The silicon tracker (pixels and strips) will be replaced in LS3, along with the main endcap calorimeters: the ECAL (homogeneous electromagnetic calorimeter, based largely on lead tungstate scintillating crystals with vacuum phototriode light detection) and HCAL (sampling hadronic calorimeter based on plastic scintillating tiles, wavelength-shifting fibres and SiPM light detection). In 2015 the CMS Collaboration decided to replace these detectors with the High Granularity Calorimeter (HGCAL), to be installed in LS3.
Overview of HGCAL
The HGCAL is one of the most ambitious detector projects ever undertaken, due to the combination of extremely high readout and trigger granularity coupled with the harsh radiation environment of the CMS endcaps during the high-luminosity phase of LHC. The starting point was to identify radiation-tolerant materials. As the radiation field changes by 4-5 orders of magnitude over the Z-η region covered by HGCAL (reaching ~1016 cm-2 1 MeV-neutron equivalent and 2 MGy in the hottest region at the end of HL-LHC), two materials have been selected: silicon in the high-fluence (and dose) region; plastic scintillator tiles in the less harsh regions. To mitigate the effects of radiation damage to the silicon, it must be cooled to about -30oC: the coolant chosen is bi-phase CO2, profiting from the same developments for phase-2 Trackers etc. Cooling the entire HGCAL to this temperature is also beneficial to the scintillator part, as it allows on-tile silicon photomultipliers (SiPMs) to be used for light detection.
The HGCAL will be realized as a 52-layer sampling calorimeter, as shown schematically in figure 1 along with relevant parameters.
Figure1: Schematic transverse slice of the HGCAL.
The first 28 layers form the electromagnetic section of the calorimeter and are based on hexagonal modules, comprising hexagonal silicon sensors (maximizing the useable surface of 8” circular silicon wafers) divided into hexagonal cells, glued to high-density copper-tungsten alloy (25%:75%) baseplates on one side, and PCBs containing the front-end ASICs on the other side. Three different thicknesses of silicon are used: 120μm in the highest-fluence regions; 200 μm in the medium-fluence regions, and 300 μm in the remainder. The sensors are divided into hexagonal cells of ~0.5 cm2 (for the thinnest sensors) and ~1.1 cm2 for the others. These cell sizes are chosen as a compromise between cell capacitance (which influences the noise) and channel count. The modules are attached to either side of a copper plate that has an embedded pipe carrying the CO2, in order to cool the silicon sensors and evacuate the heat generated by the front-end electronics. Motherboards, containing data-concentrator ASICs and links to the off-detector electronics, are plugged into the hexagonal modules. Lead plates, sandwiched between thin steel sheets, are placed on either side of the double-sided module-copper-module assembly, to form self-supporting “cassettes”. Each cassette covers 60o in phi and has an overall thickness (including absorbers, cooling, modules and services) of about 25 mm – one of the major challenges of HGCAL.
The following 8 layers are similar, forming the front part of the hadronic section of HGCAL, but are single-sided and using a lighter baseplate (carbon fibre or copper – to be decided) for the modules and stainless steel as main absorber. The final 16 layers each incorporate silicon modules in the low-radius (high radiation) region and scintillator tiles with on-tile SiPM light detection in the high-radius region. The use of both detector technologies optimizes the overall cost of the HGCAL whilst maintaining excellent long-term performance. Again, stainless steel is used as the main absorber.
Wherever possible “generic” electronics components will be used, such as the lpGBT chipset, Versatile Link++ and FEAST DC/DC system. These will be complemented by custom ASICs at the front-end, for signal processing, analog-to-digital conversion, data storage, trigger-primitive generation and data concentration.
Prototype tests, and expected performance
A program of prototype development began in 2016. Hexagonal silicon sensors, cut from 6” wafers, were produced and built into modules, to evaluate the feasibility of the overall design and study the performance in beams at FNAL and CERN. An existing front-end ASIC was used in 2016, the Skiroc2 chip originally developed for the CALICE collaboration. Although not really suited to LHC operation, the silicon modules performed well, with measured longitudinal/transverse shower shapes agreeing very well with simulation. The performance, in terms of position, energy and timing resolution all agreed with simulation, despite fewer layers being used in the beam tests than will be used in reality. In 2017 a new ASIC was available – the Skiroc2-CMS, including many of the features of the final desired front-end ASIC: ~20 ns shaping time, low noise, large dynamic range obtained through the use of “standard” amplifier stages plus a Time-over-Threshold technique for large signals, and a Time-of-Arrival circuit with ~50 ps accuracy in order to help mitigate in-time pileup. A prototype module, showing the through-hole wire-bonding used to connect the PCB to the silicon below, is shown in figure 2.
Figure 2: Prototype 6” hexagonal silicon module, with zoom showing wire bonding from PCB to silicon.
A 25 X0 electromagnetic section and 4λ hadronic section were formed from layers of silicon modules and appropriate absorbers. And for the first time a scintillator+SiPM rear hadronic section was included in the beam tests: a slightly modified version of the CALICE AHCAL. Again the performance was as expected, both for electromagnetic and hadronic showers, giving confidence in the overall design and in the simulation. Figure 3 shows the MIP signal (used for calibrating the silicon and scintillators) as well as the transverse spread (data and simulation) of hadron showers, as measured at CERN in 2017.
Figure 3: Mip signal (left) and transverse hadronic shower shape (right) as measured in beams at CERN in 2017.
The simulation was used to predict the performance of the HGCAL in CMS. Although the reconstruction algorithms are in an early stage of development, the expected performance in terms of energy resolution, particle identification and triggering are all comparable to the present CMS endcap calorimeters, even in the presence of 200 pileup events and after 3000 fb-1. The readout/triggering granularity and timing resolution for showers are key elements leading to this performance. Figure 4 shows the expected energy resolution for electrons, showing an insensitivity to pileup and a constant term (the most relevant in the endcaps) of around 1%. Also shown is a plot of the mass spectra of H→γγ events in the presence of 200 pileup, where both photons are in the HGCAL and have not converted in the Tracker.
Figure 4: Expected electron energy resolution (left) and H→γγ mass resolution (right) for the HGCAL with 200 pileup.
The HGCAL TDR (CERN-LHCC-2017-023) was approved in April 2018. The next years will be extremely challenging in terms of engineering of all types, to finalize the mechanics, on-detector electronics, services and off-detector readout/trigger boards. In parallel, the triggering, reconstruction and clustering algorithms (in 5 dimensions – X, Y, Z, E, t) will be optimized, adding to information from the new Tracker and new/existing Muon systems, to give an unprecedented “particle flow” picture of the complex events expected at HL-LHC.
Several groups in EP are playing key roles in the development of HGCAL, including EP-CMX, EP-CMG and EP-CMO inside the CMS team, as well as EP-DT, EP-ESE and EP-LCD.