TECHNICAL PROPOSAL FOR A MIP TIMING DETECTOR IN THE CMS EXPERIMENT PHASE 2 UPGRADE

Transcription

1 LHCC-P November 2017 CERN-LHCC / LHCC-P /05/2018 TECHNICAL PROPOSAL FOR A MIP TIMING DETECTOR IN THE CMS EXPERIMENT PHASE 2 UPGRADE

2 ii

3 Editors L. Gray, T. Tabarelli de Fatis Chapter Editors A. Apresyan, J. Bendavid, A. Bornheim, N. Cartiglia, L. Gray, M. Lucchini, C. Neu, T. Tabarelli de Fatis, C. Tully, J. Varela, M. Verzocchi iii

4 iv

5 Executive summary The scientific program of the High Luminosity Large Hadron Collider (HL-LHC), which includes precision Higgs coupling measurements, studies of vector boson scattering, and searches for new heavy or exotic particle, will benefit greatly from the increased collision dataset. Achieving the required high luminosity at the HL-LHC involves having as many as 200 concurrent proton-proton interactions (pileup) per 25 ns bunch crossing. At such high-pileup levels, particle reconstruction and correct assignment to primary interaction vertices present a formidable challenge to the LHC detectors that must be overcome in order to reap that benefit. In this document, we show that the ability to reconstruct the timing of most of the final state particles provides further discrimination of the interaction vertices in the same 25 ns bunch crossing beyond spatial tracking algorithms. A timing resolution of about 30 ps offsets the performance degradation due to event pileup experienced in several observables, recovering the track purity of vertices of current LHC conditions. Global event timing can be achieved by upgrading CMS with a timing detector sensitive to minimum ionizing particles (MIPs) between the tracker and the electromagnetic calorimeters. This additional MIP Timing Detector will be specialized to provide timing for the individual tracks crossing it, while photon and hadron timing will be partly provided by the upgraded CMS calorimeters. Performance benefits at 200 collisions per beam crossing are demonstrated in object and event reconstruction, as well as in the physics reach. The added value of a timing detector is expressly quantified in terms of improved track and vertex reconstruction abilities, lepton efficiencies, diphoton vertex location, and missing transverse momentum resolution. A substantial reduction of the pileup jet rate and improved performance in b-jet identification are also demonstrated. Studies with benchmark signatures demonstrate that these benefits make a significant impact on the physics programme of CMS at the HL-LHC. For Higgs boson measurements, the gain in the efficiency for final state particles and in the track purity of the primary vertex translates into sizable acceptance gains in many decay modes. The study of the Higgs boson production via vector boson fusion, with subsequent decay into τ pairs, is also significantly boosted by the improved resolution in the missing transverse momentum and the reduction of the pileup jet rate. Searches for new massive particles will also benefit from the improved CMS performance, as demonstrated for a reference SUSY signature with missing transverse momentum in the final state and for processes with displaced vertices. The CMS Collaboration is pursuing two technologies to provide MIP time-tagging for the HL- LHC detector upgrade: scintillating crystals read out by silicon photomultipliers for lower radiation areas and silicon low gain avalanche detectors for higher radiation areas. A hermetic detector design, suitable for integration in the CMS barrel on the outer tracker support tube, and in the CMS endcaps in front of the high granularity calorimeter is proposed. The R&D plan and the timeline necessary to ascertain fully the viability of these technologies, to consolidate the measurements of radiation tolerance, and to provide the proof of principle of the major detector components in a forthcoming Technical Design Report, are discussed. v

6 vi

7 Contents 1 Introduction The case for a precision timing upgrade of the CMS detector Impact of precision timing on the HL-LHC physics program Elements of the CMS timing upgrade Precision timing in the upgraded calorimeters Additional MIP timing detector The clock distribution system Overview of the MIP timing detector MIP timing detector design Barrel Mechanical structure Sensor description and radiation hardness qualification Sensor development plans Readout electronics System aspects Endcaps Mechanical structure Sensor description and radiation hardness qualification Readout Electronics System Aspects Development plans Performance studies Introduction Simulation implementation Time-aware event reconstruction Case studies: impact on event observables Reduction of pileup tracks associated to hard interaction primary vertex Pileup jet suppression Missing transverse momentum Impact on b-tagging and displaced vertices Muon and tau charged isolation Case studies: impact on selected physics analyses Higgs physics Searches for new phenomena Summary of performance benefits and next steps Project planning and cost estimates Project planning Barrel timing layer Endcap timing layer vii

8 4.2 Cost Estimates Core cost estimate Additional information Technical options for cost reduction MTD Institutions References 77 CMS Collaboration 83 viii

9 Chapter 1 Introduction 1.1 The case for a precision timing upgrade of the CMS detector The primary goal of the Phase-2 upgrade for the High-Luminosity LHC (HL-LHC) is to maintain the excellent performance of the CMS detector in efficiency, resolution, and background rejection for all final state particles and physical quantities used in data analyses. The CMS Upgrade Technical Proposal [1] presents, and the Scope Document [2] further specifies, a detailed upgrade plan to deploy an improved CMS detector by 2026, at the start of the HL-LHC operation [3]. It identifies changes necessary to withstand radiation damage effects and describes upgrades of the CMS components needed to overcome the challenge posed by the high rate of concurrent collisions per beam crossing (pileup) at the HL-LHC. For the HL-LHC, the brightness of the beams and the new focusing scheme at the interaction point, will enable the accelerator to deliver a potential luminosity of cm 2 s 1 at the beginning of each fill [3]. However, the nominal scenario is to operate at a stable luminosity of cm 2 s 1, yielding 140 pileup collisions by continuously tuning the beam focus and the crossing profile during a fill. An ultimate scenario, with cm 2 s 1 luminosity and 200 pileup collisions per beam crossing, would provide 40% more accumulated data. At 140 or 200 pileup collisions, a hard interaction, one that probes energy scales of order TeV, occurs in less than 1% of the total number of interactions simultaneously recorded by the detector. The spatial overlap of tracks and energy deposits from these collisions can degrade the identification and the reconstruction of the hard interaction, and can increase the rate of false triggers. The choice of the luminosity leveling in CMS will be a trade-off between the integrated luminosity and the experiment performance. Pileup mitigation in CMS relies upon particle-flow event reconstruction [4], which removes from relevant quantities charged tracks inconsistent with the vertex of interest, and neutral deposits in the calorimeters with ansatz-based statistical inference techniques like PUPPI [5]. The high spatial granularity of the subdetectors will enable the upgraded CMS detector to separate vertices, to identify the hard collision, and to measure signal particles with good efficiency in the offline analyses [1, 2]. In the transition from 140 to 200 pileup events, however, the peak line density of the collision vertices grows from 1.2 to 1.9 mm 1, with a line spread along the beam axis of about 4.5 cm RMS (Fig. 1.1), the probabilities of spatial overlap grow in all subdetectors, and particle-flow algorithms begin to fail at a substantial rate. The association of a track with the primary vertex, for instance, relies on a selection on the distance of closest approach to the vertex along the beam axis. Beacuse of tracks from displaced sources that cannot always be fully identified such as secondary interactions, particles decays in flight, and resolution tails the optimal selection window is wider than what would be expected by the tracking core resolution alone. According to simulation, the optimal window is of the order of 1 mm, causing a non-negligible contamination of pileup tracks into the primary vertex for vertex densi- 1

10 ties above 1 mm 1. The resulting degradation in resolutions, efficiencies, and misidentification rates at 200 pileup events impacts on several measurements [2, 6]. While measurements relying on isolated objects will suffer mainly from an acceptance reduction, measurements relying on the missing transverse momentum (p miss T ) resolution or jet counting are significantly affected. ) -1 Line density (mm PU - start 200 PU - end 140 PU - start 140 PU - end LHC Vertex position (cm) Line density p.d.f HL-LHC 200 PU HL-LHC 140 PU LHC Line density (mm ) Figure 1.1: Spread of the vertices along the beam direction at LHC and HL-LHC with 140 and 200 pileup events. The solid (dashed) line refers to the start (end) of the fill (left). Probability density function of the vertex density along the beam axis: the modes and the means of the three distributions are 0.3, 1.2, and 1.9 mm 1 and 0.2, 0.9, and 1.4 mm 1 (right). The timing upgrade of the CMS detector will improve the particle-flow performance at high pileup to a level comparable to the Phase-1 CMS detector, exploiting the additional information provided by the precision timing of both tracks and energy deposits in the calorimeters. In the time domain, pileup collisions at the HL-LHC will occur with an RMS spread of approximately ps within the 25 ns bunch crossing structure of the colliding beams, constant during the fill and uncorrelated with the line spread. If one imagines slicing the beam spot in consecutive time exposures of 30 ps, the number of vertices per exposure drops down to current LHC pileup levels. A time resolution of this size, therefore, would reduce the effective multiplicity of concurrent collisions to a level comparable to the LHC, thereby recovering the Phase-1 quality of event reconstruction. The essential basis for the proposed concept is that the time information from charged tracks is exploited in a space-time reconstruction of tracks and vertices. Moreover, the time information from photons extracted from calorimeters is matched with time information from the vertices. At the hardware level, this approach requires a dedicated detector for precision timing of minimum ionizing particles (MIPs), in addition to the enhanced timing capabilities of the calorimeters [1]. At the software level, it requires the development of algorithms to integrate the time information in particle-flow reconstruction, and to exploit that information in the offline analyses and in the high level trigger. The event display in Fig. 1.2 (left panel) visually demonstrates the power of space-time reconstruction in 200 pileup collisions, using a time-aware extension (4D) of the deterministic annealing technique adopted in vertex reconstruction by the CMS experiment [7]. According to simulation, instances of vertex merging are reduced from 15% in space to 1% in space-time. Another quantitative measure of the performance improvement is shown in the right panel of Fig. 1.2, showing the rate of tracks from pileup vertices incorrectly associated with the hard interaction vertex as a function of the line density of vertices. The rate of incorrect associations increases with the line density, as vertices start to overlap within the optimal selection window. The addition of track-time information with 30 ps precision reduces the wrong associations to a level comparable to those observed without timing at the LHC vertex density of to about 2

11 t (ns) Simulated Vertices 3D Reconstructed Vertices 4D Reconstruction Vertices 4D Tracks z (cm) Figure 1.2: Left: Simulated and reconstructed vertices in a 200 pileup event assuming a MIP timing detector covering the barrel and endcaps. The vertical lines indicate 3D-reconstructed vertices, with instances of vertex merging visible throughout the event display. Right: Rate of tracks from pileup vertices incorrectly associated with the primary vertex of the hard interaction normalized to the total number of tracks in the vertex. 0.3 mm 1. The performance of b-jet identification, which relies on vertex reconstruction, is enhanced. The removal of pileup tracks from the isolation cones improves the identification efficiency for isolated leptons and photons, which are key signatures of many processes of interest for the HL-LHC program. Similarly, the reconstruction of spatially extended objects and global event quantities that are vulnerable to the pileup, such as jets and p miss T, is also significantly improved. At 200 pileup, the p miss T resolution improves by about 10% and the rate of reconstructed jets that are spuriously clustered particles from pileup interactions ( pileup jets ) is reduced by up to 40%, using track-time information in jet reconstruction. Chapter 3 presents thorough simulation studies of track and vertex reconstruction, of particle isolation, of jet and p miss T reconstruction, and of benchmark physics measurements and searches. These studies consistently motivate that precision timing in the barrel and in the endcaps, with about 30 ps resolution, not only offsets the performance losses in the transition from 140 to 200 pileup events, but also recovers the Phase-1 (40 pileup) performance of the CMS detector, thereby enhancing the HL-LHC physics reach. 1.2 Impact of precision timing on the HL-LHC physics program The CMS physics program at the HL-LHC will target a very wide range of measurements, including in-depth studies of the Higgs boson properties and direct searches for physics beyond the standard model (BSM). The added value of a timing detector, quantified in terms of improved vertex identification, acceptance extension for isolated objects, improved p miss T resolution, and pileup jet rate reduction, makes a significant impact on the CMS physics program across several channels. These performance gains are gauged in Chapter 3 with benchmark analyses representative of Higgs boson measurements, supersymmetry (SUSY) and other BSM searches. A synopsis is presented in Table 1.1, where detector requirements are mapped into analysis and physics impacts. The benefits are broad, as further expanded below. The characterization of the Higgs boson properties, with precision measurements of the Higgs boson couplings to standard model (SM) particles, and the search for rare SM and BSM decays, will benefit from the improved acceptance for isolated objects, and in the case of H γγ decays from improved vertex identification. The quality of the isolation discriminant relies on the 3

12 Table 1.1: Representative signals for Higgs boson measurements and SUSY searches used to map each specific detector requirement into the relative performance gain at the analysis level (analysis impact) and in the measured physical quantity (physics impact). Signal Detector requirement Analysis impact Physics impact H γγ VBF+ H ττ HH χ ± χ 0 W ± H+p miss T Long-lived particles 30 ps photon and track timing barrel: central signal endcap: improved time-zero and acceptance 30 ps track timing barrel: central signature endcap: forward jet tagging hermetic coverage: optimal p miss T reconstruction 30 ps track timing hermetic coverage 30 ps track timing hermetic coverage: p miss T 30 ps track timing barrel: central signature S/ B : +20% - isolation efficiency +30% - diphoton vertex S/ B : +30% - isolation efficiency +30% - VBF tagging +10% - mass (p miss T ) resolution signal acceptance : +20% b-jets and isolation efficiency S/ B : +40% - reduction of p miss T tails mass reconstruction of the decay particle +25% (statistical) precision on cross section +20% (statistical) precision on cross section (upper limit or significance) Consolidate HH searches +150 GeV mass reach unique sensitivity to split-susy and SUSY with compressed spectra removal of pileup contributions close in angle to the candidate signal particle. Therefore the efficiency gain is maximal with hermetic coverage, while barrel coverage is especially relevant for processes with central signatures. The gain is particularly effective in the search for di-higgs production, and consequently the direct measurement of the Higgs self coupling, which will be one of the highest priorities of the HL-LHC physics program. For example, precision timing increases the signal yields for constant background in HH bbγγ by 17% from the barrel alone, and 22% with hermetic coverage (Fig. 1.3). Similar enhancements are predicted for other important Higgs boson signatures, ranging from 15 20% for HH 4b to 20 26% for H 4µ, for constant background. These acceptance extensions will provide improved precision in the measurement of rare decay modes and of statistically limited differential distributions, with sensitivity to Higgs boson pseudo-observables [8]. In the case of H ττ in the vector boson fusion (VBF) production mode, additional substantial gains arise from the improved quality of the p miss T reconstruction and of the VBF jet identification. The improvement of 10 12% in p miss T resolution at 200 pileup events from using time information for charged tracks (Fig. 3.5) yields a proportional gain in the resolution of ττ mass reconstruction and in the signal-to-background ratio, and alone counteracts almost entirely the performance degradation observed in the transition from 140 to 200 pileup events [2]. In addition, the use of time information provides a rate reduction of pileup jets by about a factor two (Fig. 1.4), which boosts the performance of VBF tagging algorithms, both in H ττ and in other Higgs boson decay modes. The sensitivity to several searches for new phenomena is largely driven by the p miss T resolution, which determines the background level for several BSM signatures, including SUSY models. The gain in the p miss T resolution with track timing leads to a reduction by a factor of 40% in the tail of the p miss T distribution above 130 GeV (Fig. 3.5), which offsets almost entirely the performance degradation of SUSY searches observed in the transition from 140 to 200 pileup [2]. Additional benefits of the precision timing are anticipated in multi-lepton signatures of new physics, owing to the increased efficiency of the lepton isolation selection, and in signatures where a direct measurement of the time of flight (TOF) of heavy particles is exploited. For example, a TOF measurement in a detector in front of the calorimeters will reduce the model dependence in searches for Heavy Charged Stable Particles, now limited to particles that have 4

13 little interactions with the calorimeters [9]. Moreover, the track-time reconstruction opens a new avenue in searches for neutral long lived particles (LLPs), postulated in many extensions of the standard model like Split-SUSY, GMSB, RPV SUSY, Stealth SUSY, SUSY models with compressed mass spectra and many others discussed in [10] and references therein. The space-time information associated to the displaced decay vertex, constructed from the decay daughters that do not escape detection, will enable the kinematic closure of the decay and, for example, the direct measurement of the LLP mass (Fig. 1.3) even for cases in which the decays are partially invisible, dramatically boosting the sensitivity of such searches and providing a novel method to characterize any future discovery. Fraction of Events HH bbγγ ( 200 Pileup Distribution ) No Timing Barrel Timing Only Barrel+Endcap Timing Increase in HH bbγγ Yield Barrel Timing Only : 17% Barrel+Endcap Timing : 22% neutralinos / 1.5 GeV cτ = 10.0 cm cτ = 3.0 cm cτ = 1.0 cm cτ = 0.3 cm y HH m χ 0 (GeV) Figure 1.3: Impact on signal efficiency for HH bbγγ for no-timing and two timing implementation scenarios (left). Mass peak of a 700 GeV neutralino reconstructed from the kinematic closure of the secondary vertex using time information with 30 ps resolution (right) Elements of the CMS timing upgrade Precision timing can be provided by the front section of the upgraded endcap calorimeter (CE) [1] and by the barrel electromagnetic calorimeter (ECAL) with a specific upgrade of the readout electronics [11]. However, neither the CE nor the ECAL detectors will provide efficient precision timing information for minimum ionizing particles (MIPs). Therefore, global event timing with the ability to reconstruct the vertex time and exploit time information in charged particle reconstruction requires a dedicated MIP Timing Detector (MTD) covering the barrel and endcap regions. The scale of the time resolution needed in each of these detectors is primarily determined by the spread in time of the luminous region at the HL-LHC. This spread amounts to about 180 ps RMS and sets the time resolution required to achieve an effective line density of vertices comparable to LHC (0.3 mm 1 ) to around 30 ps. This required time resolution does not depend on the HL-LHC luminosity leveling options. Rather, the timing upgrade of CMS will provide an extra measure of robustness against any possible future beam-crossing scenarios that may maximize or otherwise alter the luminosity production capabilities of the HL-LHC Precision timing in the upgraded calorimeters In the barrel region ( η < 1.48), the current ECAL provides time information with a resolution of order 150 ps for high energy showers (E > 30 GeV), limited mostly by time synchronization 5

14 and calibration [12]. Direct tests with electron beams ascertained that intrinsic limitations from the time spread related to shower development and light collection inside the ECAL crystals do not exceed 30 ps [13]. The proposed upgrade to the analog and digital readout electronics will enable the ECAL to achieve a time resolution of order 30 ps for photons of E T > 30 GeV, such as those from H γγ decays [11]. At lower energies, the time resolution is limited by the electronic noise, which is expected to grow with the accumulation of radiation damage in the silicon avalanche photodiodes. The ECAL will not provide precision timing for low energy photons and MIPs, which would require detector-level upgrades that are impractical. The endcap calorimeter (1.5 < η < 3.0) front section will consist of Pb absorbers interleaved with highly segmented Si sensors [1]. The readout electronics includes a time-of-arrival (ToA) and an amplitude measurement via a preamplifier and analog-to-digital converter (ADC) or time-over-threshold (ToT) measurement, depending on the energy deposited. The ToA system will provide a time measurement from each silicon pad (SiPAD) with a precision of ps, depending on the energy deposited. The combination of multiple SiPAD hits in electromagnetic showers will provide time information for photons with high precision, limited essentially only by the relative time synchronization of different CE regions. For hadron showers, fluctuations in number and magnitude of the SiPAD hits spoil the time response precision, especially at low p T. A preliminary estimate based on the simulation of K L interactions [14] indicates that the CE calorimeter could provide timing with ps precision and more than 90% efficiency for hadrons of p T > 5 GeV. The efficiency degrades to about 70% at 2 GeV for a precision of ps, with less than 50% of the hadrons having a precision better than 30 ps. To incorporate these results in our studies, we have adopted a simpler parameterization, where a resolution of 30 ps with 100% efficiency is assumed for all the charged hadrons of p T > 2 GeV. This parameterization was proven to match, in event reconstruction, the performance of the full response model based on K L results. Moreover, while the K L study is useful to suggest the potential of the CE calorimeter in the time reconstruction of low p T hadrons, and to set a comparison point for the MTD, there are uncertainties to transport these results to the full reconstruction of charged hadrons at high pileup. Therefore, in our comparative studies, we have considered two additional response scenarios, in which precision timing is assumed to be available for p T > 5 and 10 GeV, respectively Additional MIP timing detector Calorimeter-based methods for track timing of hadrons lack universal applicability or yield insufficient precision or efficiency at low energies. A dedicated hermetic timing detector with high signal-to-noise for MIP depositions will provide efficient time vertex reconstruction, and the use of timing in track-vertex association and in the object reconstruction. Mechanical, performance and upgrade schedule constraints narrowed down the options to a thin layer between the Tracker and the calorimeters, divided in a barrel ( η < 1.5) and two endcap sections covering up to η = 3.0. The radial distance of these layers from the beam axis sets a threshold on the acceptance for charged particles at about p T = 0.7 GeV in the barrel and p = 0.7 GeV in the endcaps. A study of the rate of pileup jets provides a quantitative assessment of the complementarity between MTD and the timing information in the CE calorimeter. This observable directly affects the efficiency of the tagging algorithms for VBF processes. Figure 1.4 shows, as a function of the pseudorapidity, the rate of misidentified jets in the reference Phase-2 detector with 200 pileup events relative to the rate for a CMS detector including an MTD with 30 ps precision. For η > 2.0, three different scenarios are displayed that reflect an aggressive, a moderate, and a conservative projection of the CE time response to hadrons according to the simulation dis- 6

15 Relative pileup jet rate CMS Phase-2 Simulation MTD timing No track-timing PU = 200; p >30 GeV track-timing p >10 GeV T track-timing p >5 GeV T track-timing p >2 GeV T Pseudorapidity η Figure 1.4: Rate of misidentified jets at 200 PU for the reference Phase-2 detector relative to the rate for a CMS detector including an MTD with 30 ps precision. The three different curves at η > 2.0 reflect conservative, moderate and aggressive projections of the hadron timing capability of the CE calorimeter. Results for the barrel are shown for comparison. cussed in Section In the forward region, the additional MTD will reduce the rate of pileup jets by about 30% on top of the most optimistic projection, in which the CE detector is assumed to provide per-track timing with 30 ps precision and 100% efficiency from p T = 2 GeV. The rate reduction can exceed 50% for less aggressive projections of the hadron timing performance in the endcaps. This observation demonstrates the benefit of a timing detector with sensitivity to MIPs, in addition to the endcap calorimeter. The results for the barrel region, where the ECAL will not provide timing information for hadrons, are shown for the sake of completeness The clock distribution system A common R&D effort is underway across the CMS sub-projects (CE, ECAL and MTD) to achieve an RMS jitter of ps in the clock distribution, including short-term, long-term and detector-wide stability. The phase noise of the current CMS central trigger and clock distribution system (TCDS) [15] should still be fully characterized to define possible improvements in terms of jitter and long-term stability. The current two-channel distribution evaluations to the back-end systems achieve sub 10 ps RMS, with clock cleaning. Sub 10 ps RMS jitter is also achieved with the beam clock source. In the upgraded system, lpgbt and Passive Optical Network components [16] are anticipated to provide the desired performance. The PLLs in the LpGBT are expected to filter the high frequency noise, while lower frequency jitter and phase instabilities will require dedicated monitoring. Two parallel approaches are being pursued for delivery of the clock. One option uses a clock encoded within the control links using the GBT protocol up to the lpgbt chip, which then recovers the clock from the link and distributes it to the front-end ASICs. In the other option, an independent high precision path is followed, and the clock is cleaned on the front-end board with a clock fan-out based on a rad-hard clock ASIC, currently in the specification process. This fan-out distributes the clock to the ASICs. A monitoring system for low frequency phase shifts is also under development. The slow variation of the phase offsets will be calibrated from minimum bias events, where the common time-zero vertex for separated vertices will constantly monitor slow drifts with high statistical precision. The MTD design and the frontend boards are devised with a sufficient number of dedicated optical links to accommodate jet T 7

16 Figure 1.5: A simplified GEANT geometry of the timing layer implemented in CMSSW for simulation studies comprises a LYSO barrel (grey cylinder), at the interface between the Tracker and the ECAL, and two silicon endcap (orange discs) timing layers in front of the CE calorimeter. either option. 1.4 Overview of the MIP timing detector Figure 1.5 shows a simplified implementation in GEANT of the proposed layout integrated in the CMS detector. The MTD will comprise a barrel and an endcap region, with different technologies based on different performance, radiation, mechanics and schedule requirements and constraints: Cost effective design over a large area: Performance studies motivate the need of a hermetic coverage, with time resolution of order ps for charged tracks throughout the detector lifetime. Integration constraints: A single layer device between the Tracker and calorimeters, covering up to η 3, is imposed by space and integration constraints. Granularity: A channel area of order 1 cm 2 in the barrel, and varying in the endcaps down to 3 mm 2 at η 3, yields a good compromise between low time response spread within a channel, low occupancy and low channel count. The channel occupancy is limited to a few percent, ensuring both a small probability of double hits, needed for unambiguous time assignment, and a manageable data volume. Radiation tolerance: The devices must be able to operate efficiently up to an integrated luminosity of 4000 fb 1, without any maintenance intervention for the barrel detector, whereas the endcap detector may be accessible during the HL-LHC era. Table 1.2 shows the expected particle fluence and radiation doses at possible timing layer locations, between the Tracker and the ECAL calorimeter, and in front of the neutron moderator of the endcap calorimeter. Marginal impact on the Tracker performance and design: The proposed design of the barrel timing layer requires the outer radius of the tracker to be reduced by up to 8

17 Table 1.2: Radiation doses and fluences at the timing layers after 4000 fb 1 from the online BRIL radiation tool [17]. The fluence is normalized to 1 MeV neutron equivalent in silicon. Region η R (cm) z (cm) Fluence (cm 2 ) Dose (kgy) barrel barrel barrel endcap endcap endcap endcap mm. This reduction in radius will make a negligible impact on the tracking performance (Section 3.1.1), but requires attention in terms of schedule and design (Section 4.1.1). Care must also be taken to prevent mechanical and thermal instabilities that may affect the Tracker alignment. Marginal impact on the calorimeter performance: In the current CMS detector, the electron and photon energy resolution is degraded by the amount of material in front of the calorimeters. The upgraded CMS Tracker is designed with significantly less material in the fiducial volume. The additional timing layer should be maintained thin enough not to deteriorate the performance of the calorimeters. Compliance with the CMS upgrade schedule: The Tracker and calorimeters upgrades constrain the MTD schedule. In particular, the available region in the barrel, at the interface between the Tracker and the calorimeter, has to be fully instrumented early in the Tracker integration phase, currently scheduled for 2022, to enable the assembly of the Tracker on schedule for the start of the Phase-2 (2025). This constraint narrows down the choice to a mature, essentially ready, technology for the barrel timing layer. Five technologies were investigated and studied in dedicated beam tests and radiation exposures, building upon and extending long-standing R&D programs [18 28]. Crystal scintillators read out with silicon photomultipliers (SiPMs) [18 20] and silicon sensors with internal gain [21 23] emerged, respectively, as a mature technology for the barrel and a viable technology for the endcap timing layers. In the barrel, we propose to adapt the present Tracker Support Tube (TST) design by instrumenting the current location of the thermal screen with a thin, actively cooled, standalone detector, based on lutetium-yttrium orthosilicate crystals activated with cerium (LYSO:Ce) read out with SiPMs. The TST outer wall temperature will be controlled by resistive heating pads, as it is currently done in the region of the support rails. Both LYSO based scintillators and SiPMs devices are mature technologies, with production and assembly procedures well established and standardized in industry. The R&D for a precision timing application is well advanced, and small prototypes consisting of LYSO:Ce crystals read out with SiPMs have been proven capable of achieving time resolution below 30 ps [29]. Both the crystals and the SiPM are proven to be radiation tolerant up to a neutron equivalent fluence of cm 2, when cooled to below 30 C. The read-out electronics can be adapted from existing positron emission tomography applications with time-of-flight (TOF-PET) measurement [30 32], which is read out upon an external (Level-1) trigger. The barrel timing layer will cover the pseudorapidity region up to η = 1.48 with a total active surface of about 40 m 2. The fundamental detecting cell will consist of a thin LYSO:Ce crystal with about mm 2 cross-section coupled to a 4 4 mm 2 SiPM. The crystal thickness will vary between about 3.7 mm ( η < 0.7) and 2.4 mm ( η > 1.1), 9

18 to level the slant depth crossed by particles from the interaction point. The proposed layout has no impact on the upgraded designs and schedules of the Tracker and the ECAL, except that the TST structure must be in place before the construction of the upgraded Tracker begins. The selection of a robust technology that requires a limited amount of residual R&D enables the severe constraints on the construction schedule to be matched. The timing layer has a negligible impact on the performance of the tracker in the barrel, and no impact in the endcaps. A preliminary simulation study, with a 4 mm thick LYSO:Ce layer, i.e. thicker than in the reference design, also indicates no significant impact on the performance of shower reconstruction and energy resolution in the ECAL. The endcap region can be instrumented with a hermetic, single layer of MIP-sensitive silicon devices with high time resolution, with a pseudorapidity acceptance from about η = 1.6 to η = 2.9. The choice for the placement of this detector within the CMS experiment is driven by the necessity to ensure its accessibility throughout the HL-LHC operation. Therefore, we propose to place the endcap timing layer (ETL) in its own independent, thermally isolated volume on the nose of the CE detector, at a distance of about 3 m from the interaction point. Sensor modules will be mounted on two sides of two double-sided disks in order to provide hermetic coverage. The active sensor area is a little over 6 m 2 per endcap. This option would exploit the services and the geometry of the CE design. We propose to use planar silicon devices with internal gain, since the technology selected for the barrel cannot be extended to the endcap, due to radiation tolerance limitations. The silicon sensors have intrinsic gain, in order to overcome capacitance and other noise sources and achieve a low-jitter rising pulse edge and hence precision timing reconstruction for minimum ionizing particles with a resolution of order ps. As a reference design we propose to build the fast-timing detector based on Low-Gain-Avalanche-Diodes (LGAD) [22, 23], which are also considered for a fast-timing layer in the very forward region (2.4 < η < 4.8) of the ATLAS experiment [33]. Thin wafers are expected to provide the desired performance. The radiation tolerance studies, though still in progress, indicate promising performance of about 30 and 50 ps at fluences corresponding to η 2.5 and 3.0, respectively, at the end of the HL-LHC operation. Studies to ascertain the radiation tolerance of alternative silicon devices with internal gain, Deep-Depleted APDs, are still ongoing. Achieving good time performance at low-gain requires cell sizes typically less than 3 mm 2, to limit the sensor capacitance. Therefore, the LGAD option would require a larger number of sensors to cover the endcap. The total channel count will be reduced by feeding a group of discriminated signals to a time to digital converter, implemented in a dedicated readout chip, under development. While the barrel timing layer should be installed before the Tracker, the time schedule to assemble the endcap discs can extend toward the end of the LHC long shutdown LS3 (2025), thus providing additional time to complete the R&D plan. While full readout of the MTD is not possible, because of bandwidth constraints, it may be possible for the MTD to play a role in the Level-1 trigger. The MTD Level-1 trigger information can be formed in the off-detector electronics from the data of regions of interest (ROI). The read out of the ROIs can be seeded by a Level-0 request from the Tracker, muon or the calorimeter triggers. The impact of timing on the CMS triggers, as well as the technical implications for the MTD read-out electronics and for the Level-1 trigger latency, are not discussed in this document and will be evaluated in the context of the MTD TDR. 10

19 Chapter 2 MIP timing detector design 2.1 Barrel In the following, we present and discuss a detector design to instrument the CMS barrel with a precision timing layer capable of detecting minimum ionizing particles (MIPs) with nearly 100% efficiency and time resolution of about 30 ps throughout the HL-LHC. The barrel timing layer (BTL) can be attached to the carbon fiber support tube of the Tracker as a thin standalone detector with minimal impact on the neighbouring sub-detectors. The fundamental detector unit consists of a LYSO:Ce crystal tile read out with a SiPM. This type of sensor has been proven capable to achieve a time resolution better than 20 ps in test beam operation [29], when SiPMs and crystal tiles are of comparable surface, and to be radiation tolerant up to a neutron equivalent fluence of cm 2. No impediments have been identified for the integration of the BTL into the overall CMS Phase-2 detector. An overview of the mechanical structure and integration aspects is given in this section, together with a discussion of the sensor optimization plan, radiation tolerance qualification and read-out scheme. A complete description of the mechanical specifications and detector performance are forthcoming in the Technical Design Report (TDR) Mechanical structure The barrel timing layer will be attached to the carbon fibre Tracker support tube (TST), which serves as the mechanical support for the Tracker. The BTL will be a thin, cylindrical structure with an inner radial boundary of 1161 mm and a length of 5200 mm, covering the pseudorapidity region up to η 1.5 with a total surface area of about 40 m 2. This design will require the outer radius of the tracker to be reduced by up to 15 mm. (see Section 1.4) BTL in the Tracker Support Tube A carbon fibre structure, consisting of inner and outer cylinders, is used to house the BTL trays (36 in φ per side) of detector modules, as shown in Fig In this structure (similar to the one currently used for the existing TST), the two horizontal rails that separate the upper and lower half-cylinders play a major role in the stability of the whole structure and will conceptually remain as in the current TST. The outer surface temperature of the TST will be adjusted to be thermally neutral to the surrounding environment with resistive pad heaters (not shown in Fig. 2.1), both when the CMS detector is in the data-taking and in the open configuration. The coldest temperature will be practically limited by the same as the Tracker (i.e. 30 C). Discussion with the manufacturer of the current TST resulted in favouring a design that would largely maintain a single walled TST with thin rails attached to the inner surface, where the BTL trays are attached. The BTL trays would be covered with a thin inner wall, likely segmented as the trays, for mechanical protection. In this configuration the main TST wall would experience 11

20 Figure 2.1: Cross section of the TST with mechanical structures to support the mounting of the barrel timing layer. The upper and lower half-cylinders, separated by two horizontal rails for mechincal stability, are divided in 18 compartments each to house the BTL trays. a thermal gradient from 30 C to 18 C through an on/off cycle of the tracker and BTL. The deformations implied by this thermal cycle are currently being evaluated, including the fabrication of a 1 m 2 TST prototype BTL Trays The cylindrical barrel structure, as shown in Fig. 2.1, is divided into 36 compartments along φ, with every second BTL tray boundary matching the segmentation of the ECAL supermodules. Each compartment contains two trays, each of them further subdivided into 54 modules. The two trays cover the full barrel length (about 5200 mm) with a width of about 185 mm driven by the module dimensions. Such a modular design represents an attractive option for integration and commissioning of the detector as the modules can be quickly loaded inside the tray by means of a supporting rail. The tray will constitute the mechanical connection between the modules and serve as a mounting point for the services on the tray. The tray structure will also house the CO 2 cooling pipes with the tray structure designed to ensure good thermal contact between the cooling lines and the SiPMs. Thermal flow analyses are currently being performed to optimize the heat extraction from the SiPMs BTL Modules The fundamental detector cell consists of a thin LYSO:Ce crystal tile with an approximate cross section between and mm 2 coupled to SiPMs with a size of around 4 4 mm 2, 12

21 with the final dimensions to be specified in the TDR. In order to maintain a low and uniform material budget in front of the ECAL the thickness of the crystal tiles will vary with pseudorapidity from 3.75 mm (for η < 0.7) to 3.0 mm (for 0.7 < η < 1.1) and 2.4 mm (for η > 1.1). The corresponding radiation length, X 0, and slant thickness are shown in Fig This optimization of the crystal thickness also reduces the overall weight of the barrel structure as reported in Table 2.1. The choice of pseudorapidity regions with different crystal thickness is such that the amount of light produced by a charged particle will not decrease with η, as the slant thickness traversed by particles scales with 1/ sin θ, where θ is the polar angle from the beam axis. Slant thickness / x 0 Pseudorapidity Constant thickness: 3 mm Equalised thickness z (cm) Figure 2.2: Slant thickness of the barrel timing layer as a function of the z (lower abscissa) and η coordinates (upper abscissa) for tiles of variable thickness (continuous line), of 3 mm (dotted line) and of 4 mm (gray line) thickness Slant thickness (mm) Table 2.1: Crystal tile thickness and weight in different η sections of the barrel timing layer, and contribution to the weight of one module. η region Crystal thickness [mm] Crystal weight [g] Module count Module weight [g] < > By chamfering the corners of the tile, the crystal+sipm devices can be staggered in a two layer configuration with 0.5 mm overlap between edges to avoid cracks and guarantee full hermeticity, as shown in Fig The read-out ASIC electronics will be connected directly to the SiPM to allow a compact structure, similar to the one used for TOF-PET applications [30, 32, 34]. The total thickness of the crystal+sipm sensor including the read-out electronics and services is thus contained within 25 mm. The crystals are grouped together in a 16 4 array with the respective SiPMs and read-out electronics to constitute a detector module of 185 mm width (along φ), 25 mm height (in radial space r), and 47 mm length (along z), containing a total of 64 channels, as shown in Fig The panel on the top-right side of the figure shows one quarter of a module prototype, used in beam test studies BTL Services The following services will be associated with each tray: two CO 2 cooling lines, eighteen lpgbt fibres (rx/tx) and, optionally, three dedicated clock distribution fibres bundled in a high-speed 13

22 7.28 [185] 1.85 [47] Figure 2.3: Top left: 3D visualization of a single detector module, consisting of a 16 4 array of crystal+sipm cells read-out with dedicated electronics. Top right: Picture of a BTL module prototype used in beam tests. Bottom: drawing of the front and side view of the detector module and basic cell with corresponding dimensions in inches [mm]. (GHz) digital fibre ribbon, low voltage (LV) power, and SiPM bias voltage. The services run the length of the tray with pigtail cables that integrate into the outer tracker service ducts (12 in φ per side). The technology choice to provide the total necessary cooling power of about 15 kw is a twophase CO 2 system, as is planned for the CMS Tracker Phase-2 upgrade and also currently in use for the LHC Run 2 pixel detector. The cooling system will be shared with the Tracker. Detailed planning for the CO 2 cooling plant is currently under study by the CMS Phase-2 Technical Coordination group. The total cooling capacity for the BTL is about 10% of the capacity needed for the Tracker. A preliminary design for the tracker services has been developed, which incorporates sufficient capacity for power, cooling and data connection to service the BTL. Existing studies show that such a cooling plant can operate at 40 C with good pressure values and a uniform temperature along 5 m long pipes [35]. The BTL cooling will have the same input operational temperature of 33 C as the Tracker (see Fig of the Tracker TDR [36]), with the capability to lower the temperature to 40 C for additional margin. The nominal temperature difference of the sensor and the coolant is within a couple of degrees, and therefore SiPMs will below 30 C. A pattern of resistive loads will be placed on the electronics boards to make the temperature uniform within the module and to allow cooling power flexibility to compensate in case some sensors are turned off. The total heating power could also be kept constant during the whole lifetime of the detector by switching off the resistive loads while the self-heating from SiPM increases. This would avoid a dynamic heat dissipation, thereby improving the thermal stability of the whole structure. Nonetheless, particular attention will be put in order to use materials with small coefficients of thermal expansion (CTE) and dedicated studies will be made to assess the thermal uniformity and stability of the detector Sensor description and radiation hardness qualification The design of the active element of the barrel timing layer relies on the understanding of the LYSO:Ce crystal and SiPMs performance in the high rate and high radiation environment of the HL-LHC. Test beam results obtained on several geometries, similar to those of the proposed design, are reported and discussed. Such results are well reproduced by a detailed GEANT simulation, which is then used to extrapolate the timing performance of the BTL at an integrated 14

23 luminosity of 4000 fb 1 based on irradiation studies performed on several SiPMs Description of sensors The basic sensor used to instrument the barrel timing layer consists of a LYSO:Ce crystal tile read out with a SiPM glued to one of its larger sides. Sensors of geometry close to the nominal geometry for BTL have been proven capable to achieve a single MIP time resolution better than 30 ps in test beam measurements [29, 37]. In such a device a MIP is detected with 100% efficiency by the scintillation light created in a few millimeters of crystal, which is then detected and transformed into an electrical signal by the SiPM. A trade off between sensor cost, timing performance and material budget in front of the ECAL lead to the choice of an optimal thickness of the crystal tile ranging from 3.75 to 2.4 mm (Fig. 2.2). For such a thickness, simulation studies suggest that a good balance between efficient light collection, low occupancy and total channel count is obtained using sensors with an area between and mm 2. The SiPM choice is a trade off between uniform light collection over a wide surface and with high efficiency, and an active area restricted in size to limit the dark current, the power consumption and heat dissipation, and the detector capacitance. For a crystal tile wrapped with a diffusive reflector and coupled to a 4 4 mm 2 SiPM with optical glue (refractive index 1.55 < n < 1.68), an average light collection efficiency of 20% is achieved. Thus, for a photon detection efficiency (PDE) of about 15% and a charged particle traversing the minimum slant thickness of the sensors of 3.75 mm (η = 0), a signal of about 4500 photoelectrons is anticipated the energy deposit of a MIP being about 1 MeV/mm in LYSO and the scintillation yield about photons/ MeV. The properties of the SiPM in terms of PDE, power consumption and radiation tolerance, are discussed in detail in Section Beam tests results on single sensors A set of test beam campaigns was carried out to qualify the performance of different crystal geometries and SiPM technologies. During beam tests the signal from the SiPM was processed with a dedicated electronic board in which the time stamp was obtained from the discriminated signal of a NINO chip [29]. As the NINO chip is a fixed threshold discriminator [38], a correction for time walk was applied based on the amplified pulse of the SiPM recorded in a separate channel of a CAEN V1742 digitizer. Beams of muons and pions with energy in the range GeV were used, as they behave as minimum ionizing particles while traversing thin sensors. Coincidence time resolution measurements were obtained from the difference of the time stamps between two detectors. The standard deviation of a Gaussian fit to the distribution, divided by 2, is used to estimate the single detector resolution. A set of LYSO:Ce crystal tiles with geometry close to the nominal BTL sensors were tested (Fig. 2.4): a mm 3 tile with 6 6 mm 2 SiPM with 50 µm cell pitch from Hamamatsu Photonics (HPK) and a mm 3 tile with 5 5 mm 2 SiPM with 20 µm cell pitch from Fondazione Bruno Kessler (FBK). A comparison of different types of crystal wrapping (using Teflon) was also made to evaluate the achievable improvements of light collection and time resolution by optimizing the crystal surface state. The time resolution obtained in a narrow beam hitting a 3 mm wide region in the centre of the tiles is reported in Fig. 2.5 for the two configurations: 21 ps for the former and 27 ps for the latter. The difference in timing performance follows simulation predictions and can be understood in terms of the different fraction of crystal surface read out by the SiPM, which affects the light collection efficiency and thus the time resolution. These results are obtained with 3 mm thick crystals while the average slant thickness in the BTL is around 4.4 mm. An increase of 40% in the light signal is expected to improve the time resolution by up to 20%. 15

24 beam direction SiPM wrapped crystal tile Figure 2.4: Top left: Set of mm3 LYSO:Ce crystals with depolished lateral faces, before and after Teflon wrapping. Bottom left: 6 6 mm2 HPK SiPMs glued on LYSO crystals. Right: Crystal+SiPM sensors plugged on the NINO board used for test beam studies. LYSO:Ce 10x10x3 + HPK 6x6 σt = 34 ps (uncorrected) 400 TEST BEAM PRELIMINARY LYSO:Ce 11x11x3 + FBK-NUV-HD 5x5 Frequency Counts TEST BEAM PRELIMINARY 500 σt = 21 ps (t-walk corr.) 0.14 σt = 38 ps (no wrapping) σt = 30 ps (rear wrapping) 0.12 σt = 27 ps (rear + front wrapping) Δt [ns] Δt [ns] Figure 2.5: Distribution of the time difference in a pair of LYSO:Ce tiles exposed to a 3 mm wide beam of MIPs hitting the centre of the tiles. Left: Results before and after time walk correction for mm3 crystals read out with 6 6 mm2 HPK SiPMs. Right: Results for mm3 crystals read out with 5 5 mm2 FBK SiPMs under different wrapping configurations. In another investigation, the response uniformity of the tiles across their surface was studied with reference to an external microchannel plate (MCP) detector (Photek) providing a precision time reference. The time resolution of the MCP was measured to be about 15 ps and uniform over a surface of 4 cm diameter. Within a single crystal, time response variations of the order of 150 ps are expected to arise, as photons from the peripheral region of the tile undergo several reflections before being detected in the SiPM. The left panel of Fig. 2.6 shows the difference between time measurements in a mm3 tile with a 5 5 mm2 FBK SiPM and in the reference MCP, as a function of the MIP impact point on the crystal surface along a coordinate transverse to the beam direction. A maximum time difference of about 200 ps between the centre and the edges of the sensors is observed, in fair agreement with expectations. If not corrected for, this effect yields an additional contribution to the time resolution of about 50 ps RMS per channel. 16

25 Figure 2.6: Left: Difference between the time measurements in a mm 3 tile with a 5 5 mm 2 FBK SiPM and in a reference MCP as a function of the MIP impact point on the crystal surface. Right: Time resolution before and after the application of a position dependent correction. The application of a position dependent correction, obtained from an independent set of test beam data, recovers optimal time resolution, as shown in the right panel of Fig An uncertainty below 1 mm on the impact point is required to keep the contribution of this effect to the time resolution below 20 ps. According to simulation studies, track extrapolation from the Phase-2 CMS Tracker to the BTL surface, combined with the 3 mm RMS resolution provided by the BTL readout pitch, yields sufficient precision for p T > 2 GeV. For p T < 2 GeV, the impact point resolution in the z-coordinate varies between 2 and 3 mm from low to high pseudorapidity. For a robust solution at the hardware level, custom SiPMs and different geometries are being developed with several vendors to achieve uniform light collection from the crystal surface. The options under investigation include a matrix of small SiPMs and a monolithic wide-area SiPM (10 10 mm 2 ). In either case, the total active area of the SiPMs i.e. the number and size of the SiPMs cells would be kept similar to the one of a 4 4 mm 2 SiPM, thereby resulting in a similar level of dark current and count rate Radiation tolerance and longevity studies The integrated radiation levels expected for the CMS barrel region for 4000 fb 1 are reported in Table 2.2, as obtained from FLUKA simulations of the CMS Phase-2 detector [39 41]. The fluence of 1 MeV equivalent neutrons (n eq ), which is responsible for the dominant damage of the SiPMs, shows a relatively flat profile along the barrel, with a maximum integrated fluence around n eq /cm 2 at z = 240 cm and a minimum of n eq /cm 2 at z = 0 cm. The total ionizing dose is also fairly uniform in η, with a maximum of 25 kgy (2.5 MRad), and represents no threat for the BTL sensors. In the following we summarize the existing studies on radiation tolerance of the sensors and discuss the simulation used to evaluate the evolution of timing performance as a function of integrated neutron fluence, up to n eq /cm 2, corresponding to 4000 fb 1, i.e. the end of the HL-LHC operation in the ultimate luminosity scenario (200 pileup events). Irradiation studies on LYSO:Ce crystals Radiation tolerance of LYSO:Ce and of the equivalent LSO:Ce crystals has been widely tested in the past years, with exposures to fluences and doses representative of the HL-LHC conditions. 17

26 Table 2.2: Radiation levels in the BTL after 4000 fb 1 of integrated luminosity for the CMS barrel region at different η. η R Z 1 MeV n eq Charged hadrons Dose [cm] [cm] [cm 2 ] [cm 2 ] [kgy] It has been demonstrated that they can withstand a hadron fluence of cm 2 and ionizing doses of 100 MRad (1000 kgy) [42 44] with minimal transparency loss. The induced absorption coefficient measured on long LYSO crystals after these exposures was found to be around 3 m 1, which is negligible in the case of very thin tiles as those proposed for the BTL. A small increase in radio-luminescence of the samples was also observed after irradiation, but its influence on the timing performance is negligible. At the end of HL-LHC, less than about 0.3% of the SiPM cells are expected to fire within a time gate of 50 ns for mm 3 crystals read out with 4 4 mm 2 SiPM, which is a negligible fraction of the dark counts originating from the SiPM itself. Commercially available optical glues, such as Meltmount, have also been widely tested up to the radiation levels expected for the barrel timing layer, i.e. neutron fluences of n eq /cm 2 and doses of 30 kgy. No significant degradation of transparency was observed and no effect on mechanical properties occurred after irradiation and several thermal cycling tests performed at the time of the CMS ECAL assembly [45, 46]. Irradiation studies on SiPMs Radiation studies have been performed on several types of SiPMs in the context of the HCAL upgrade [47 49]. An increase of dark counts, due to radiation-induced defects, is observed after the exposure to high neutron fluences, resulting in a deterioration of timing resolution due to baseline fluctuations (noise) and to a small reduction of the PDE. In recent R&D studies, in collaboration with different manufacturers (mostly HPK and FBK), it has been shown that reducing the SiPM cell size from 50 µm to 15 µm can provide a reduction of the radiationinduced dark current by a factor 7. In addition, the dark count rate (DCR) decreases with temperature by a factor α every 10 C. The temperature coefficient α depends on the specific structure of the SiPM and was measured to be 1.65 for irradiated SiPMs from FBK (12.5 µm cell pitch) and 1.88 for irradiated SiPMs from HPK (15 µm cell pitch). The SiPM from FBK, having a surface of 1 mm 2, was irradiated to a neutron fluence of n eq /cm 2, whereas the SiPMs from HPK (surface 6 mm 2 ) were irradiated to a fluence of n eq /cm 2. The performance of these SiPMs has been measured before and after irradiation at different temperatures. As reported in Fig. 2.7, the optimal signal-over-noise ratio (SNR) is obtained at a bias over-voltage around 1 2 V, which at the same time results in moderate power consumption and self-heating. The equivalent noise of these SiPM after irradiation is also reported in Fig. 2.7 and can be used to calculate the corresponding DCR level. Extrapolation of end-of-life performance The measurements of irradiated SiPMs presented above have been used to extrapolate the performance of such devices to the operating conditions foreseen for the barrel timing layer: a nominal operating temperature of 30 C and a neutron equivalent fluence of n eq /cm 2. The results of that extrapolation are reported in Table 2.3 for the operating over-voltages of 1.5 V and for devices with active area of 4 4 and 3 3 mm 2. To define the specifications of the SiPMs and to optimize the crystal/sipm layout, the impact 18

27 Figure 2.7: Signal over noise ratio (left) and equivalent noise (right) as a function of the bias referred to the breakdown bias (V-VB) for a 6 mm 2 SiPM from HPK (2.8 mm diameter) irradiated to n eq /cm 2 (top) and for a 1 mm 2 SiPM from FBK irradiated to n eq /cm 2 (bottom). Data are shown for different temperatures. Table 2.3: End-of-life specifications for the SiPMs of the barrel timing layer compared to the performance of existing devices of 4 4 and 3 3 mm 2 area from FBK and HPK irradiated to high neutron fluences. The extrapolations of DCR, PDE and power consumption are calculated for 1.5 V bias overvoltage and operating temperature of 30 C, after n eq /cm 2. Parameter Spec FBK FBK HPK HPK Area > 9 mm 2 16 mm 2 9 mm 2 16 mm 2 9 mm 2 Cell pitch µm 12.5µm 15µm 15µm Number of cells > 30k > 100k > 56k > 71k > 39k PDE > 10% 10% 10% 15% 15% Operating voltage - 34 V 34 V 40V 40 V Current/device ma 0.5 ma 1.7 ma 0.9 ma Power consumption < 60 mw 32 mw 18 mw 70 mw 39 mw Gain > DCR/device < 70 GHz 52 GHz 29 GHz 58 GHz 32 GHz of the DCR on the time resolution was evaluated with a GEANT-based simulation. The optimal threshold for discrimination depends on the level of noise and usually corresponds to 15 photoelectrons, for non-irradiated SiPMs. The optimal threshold is expected to increase, as the level of the DCR induced by irradiation increases, up to about 100 photoelectrons after a fluence of 4000 fb 1. The simulation tool adopted for this prediction reproduces the test beam results obtained with non-irradiated crystals coupled to SiPMs of various geometries. It has also been 19

28 2 Dark count rate [ Hz / mm ] end-of-life DCR extrapolation for 2e14 n/cm² HPK 15 µm SiPM FBK 12.5 µm SiPM -30 C [ps] σ t PRELIMINARY Simulation: Crystal 12x12 + SiPM 4x4 + glue - thick PDE = 10 % PDE = 15 % PDE = 20 % PDE = 25 % PDE = 30 % FBK4x4 +1.5V *FBK3x3 +1.5V HPK4x4 +1.5V HPK4x4 +2V *HPK3x3 +2V *HPK3x3 +1.5V Temperature [ C] DCR [counts per second] 9 Figure 2.8: Left: Temperature dependence of DCR/mm 2 for HPK and FBK devices irradiated to n eq /cm 2 calculated at 1.5 V bias over-voltage. Right: Evolution of time resolution with the increase of DCR for several PDE scenarios. The dots superimposed to the simulated curves represent the extrapolated performance of HPK and FBK SiPMs at the end of the detector operation, corresponding to a fluence of n eq /cm 2. used to reproduce the timing performance of similar devices for PET applications [18]. Test beam measurements of irradiated devices are planned to consolidate these predictions. The expected level of DCR in a given SiPM has been estimated based on laboratory measurements and according to the temperature extrapolation reported in the left plot of Fig The simulation of the performance evolution for different levels of PDE and DCR is represented by the continuous lines in Fig. 2.8, for a LYSO:Ce tile of mm 3 read-out with a 4 4 mm 2 SiPM. An effective thickness of 4 mm has been used to conservatively represent the average slant thickness throughout the whole barrel pseudorapidity region, which is about 4.4 mm. The extrapolated performance for the HPK and FBK irradiated SiPMs at the end of the detector operation ( n eq /cm 2 ) is represented by the dots superimposed to the simulated curves. For the HPK SiPMs with 15 µm cell pitch, operating at 1.5 V (2 V) bias overvoltage, the PDE is about 15 (20)%. Under these conditions, the time resolution is predicted to vary from 25 ps (at the beginning of operation) to 35 ps after n eq /cm 2 and to 40 ps after n eq /cm 2 at the end of detector operation (4000 fb 1 in the ultimate luminosity scenario). These extrapolations indicate that SiPMs with performance close to the target performance are already available. The possible mild degradation in time resolution at the end of operation only marginally impacts the performance in event reconstruction. Nonetheless, further optimization of the SiPMs and of the SiPM and crystal tile arrangement is being pursued as discussed in Section Monitoring and uniformity of radiation damage No dose-rate effects on the sensor performance are expected since the active elements will mostly suffer from degradation due to integrated neutron fluence, thus no short-term compensation or correction will be needed. More likely a long-term monitoring using the MIP signal in the overlapping cells region and an adjustment of the optimal operational voltage of the SiPM will be performed to always operate at the best balance between PDE and dark current. The possibility to use minimum bias events to perform a time calibration of the sensors based on 20

29 the average time of arrival of the hits is an additional approach which is being investigated through simulations. The uniformity of radiation tolerance among different samples will be addressed by dedicated R&D and by the sensor qualification process before assembly. For crystal tiles which are cut from a bigger crystal ingot, the dopant concentration is rather uniform within the scintillator volume and the same holds for radiation tolerance. The possibility that impurities or contamination of a batch of raw material occur during the production process, which could affect the radiation tolerance of a set of tiles, is small for commercially available LYSO crystals. Quality assurance tests during production, similar to those adopted to qualify the CMS ECAL crystals [50, 51], will minimize any risk. Extensive studies of uniformity and quality assurance on a large number of SiPMs have been made for the CMS HCAL Upgrade [47, 49] on over 1400 pre-production sensors. The uniformity of response to LED light was found to be better than 1.5%, proving a good uniformity of gain and PDE. The dark current of irradiated devices was also measured within the HCAL upgrade project and found to be uniform (less than 10% spread) among a large batch of samples. The spread in variation of the SiPMs breakdown voltage after a given neutron fluence has also been measured to stay within 90 mv, similar to the spread for unirradiated devices Sensor development plans R&D on sensor design Table 2.3 reports the specifications of the SiPM identified by the simulation study previously described. Optimization of the sensors is ongoing in order to enhance time resolution and radiation tolerance. In particular, improvement of light collection by means of reflective wrapping of the crystal and/or reduction of tile section from to mm 2 is being studied. This latter option would come at the price of a reduced overlap between tiles but with no impact on detector geometry and channel count. To improve the S/N ratio, the thickness of the tile is a crucial parameter as the signal scales proportionally with the slant thickness. A 10% increase of the thickness can be considered. The SiPM choice is a trade off between uniform and efficient light collection over a wide surface, and a limited active area. Irradiation tests are underway on a broader spectrum of devices to explore comprehensively the parameter space of PDE, cell pitch, DCR and different producers. Existing devices from both FBK and HPK already show improved PDE, lower power consumption and smaller DCR due to the addition of trenches between cells (limiting cross talk). For example, preliminary results on recently produced SiPMs with 15 µm cells at FBK indicates higher PDE (15 20%) for the same power consumption than those reported in Fig In a conservative scenario, where the DCR per unit area of the SiPM will not be reduced by a better device, the total amount of DCR and power consumption can be reduced by using SiPMs with smaller active area (less cells, or smaller devices). This choice would however require an improvement in the light collection efficiency, via the optimization of the optical match between the crystal and the SiPM, and of the crystal wrapping. Custom SiPMs providing a more uniform surface coverage, for a constant number of active cells, are also being investigated, including SiPMs matrices and monolithic wide-area SiPMs (10 10 mm 2 ). In order to limit the total active amount of cells in the SiPMs, and thus the dark current, the latter option implies having sparse cells on the SiPMs with an intercell distance about twice as in the current devices, albeit produced with the same size and technology. This R&D is being pursued with several manufacturers and a final proof of concept of the technology is expected for the TDR. 21

30 Optimization of the module layout In addition to the baseline design for the BTL module with two-layers of staggered sensors, two alternative geometries are being evaluated as viable options to reduce the crystal volume and thus the detector cost. A planar geometry without crystal overlap, where crystals are arranged in a flat array, would eliminate the 1 mm overlap between sensors and the need of chamfered corners. In this configuration each crystal would have a size of mm 2, providing a reduction of the total crystal volume by about 20%. This would imply a reduction in crystal costs of about 10 15% (part of the cost being due to crystal cutting and polishing). This geometry would introduce inefficiencies at the crystal boundaries in φ and η, but would simplify the assembly of the module and its mechanical structure. The impact on the global performance is under study. The second alternative geometry under consideration, tilted geometry, consists of a single layer of sensors tilted along η with respect to beam axis and partly pointing towards the interaction point. The tilt angle is limited to a maximum of 13 to remain within a maximum envelope defined by the two-layer staggered geometry. In this configuration, the crystal size can vary from to mm 2 depending on η. The overall volume reduction is equivalent to the previous case (20%), but the quasi-pointing geometry can provide hermetic coverage along η. The efficiency loss would be limited to the φ projection. The cost implication of the more complex tile mounting in this tilted geometry is still to be studied in detail. In summary, a technical optimization with partial or marginal impact on the performance might provide a reduction of the crystal cost, corresponding to about 5 10% of the BTL cost. A final decision on the exact BTL geometry will be made for the TDR Readout electronics Dedicated electronics based on a new ASIC derived from the TOFPET2 ASIC, developed for TOF-PET applications, will be used to read out the crystal+sipm array contained in each module [30, 31]. The TOFPET2 chip [32] shows good timing performance and can be adapted to match the requirements of the barrel timing layer with small modifications. The specification requirements for the BTL electronics are mainly the capability to measure MIP timing with a precision better than 20 ps, to provide a measurement of the signal amplitude with 2% precision for time-walk corrections, and to have a power consumption lower than 20 mw per channel. The current TOFPET2 chip is a 64 channel ASIC based on CMOS 110 nm technology with timing and energy branches for each channel and a dynamic range configurable between 150 and 1500 pc. This can allow a precise measurement of the MIP signal, which will be in the range between 5000 and 8000 photoelectrons (depending on the slant thickness and assuming a 15% PDE), corresponding to pc (for a SiPM gain of ). The timing branch consists of an amplifier, discriminators and a TDC with 20 ps time binning, whereas the energy branch includes an amplifier, a charge integrator and a 10-bit ADC. The TDC is based on four-fold Time to Amplitude Converters (TAC) followed by a Wilkinson ADC. The maximum conversion time is of the order of 1 µs. Four TACs per channel allow de-randomizing the input signals. The Charge Integrator has a similar four-fold structure as the TACs followed by one Wilkinson ADC. The maximum rate per channel of 0.6 Mhits/s is limited by the output links (3.2 Gb/s). The time resolution of the chip is 23 ps RMS as determined from test pulses at the input of the pre-amplifier emulating the LYSO+SiPM signal amplitude (Fig. 2.9). Before offline calibration, the TDC differential non linearity (DNL) is smaller than 0.1 LSB and the integral non linearity (INL) is smaller than 1 LSB. The ADC DNL is smaller than 0.05 LSB and the INL is smaller than 1.6 LSB. The ADC noise is 0.65 LSB (RMS) determined from test 22

31 Counts Mean : ± 0.14 Sigma: : ± Time Resolution(ps) Figure 2.9: Distribution of the time resolution measured at each phase of the test pulse when scanning the clock period in steps of 16 ps. The jitter of the test pulse is not de-convoluted. pulse integration. A power consumption of about 10 mw was measured for each channel of the chip TOFHIR chip The TOFPET2 chip will require some modifications in order to match the specifications of the barrel timing layer. A reduction of the number of input channels per ASIC from 64 to 16 will be made such that each ASIC covers less detector area and the length of SiPM traces is smaller than 2 cm. On the other hand the ASIC output rate is kept below 320 Mb/s. The interpolation TAC used in TOFPET2 provides excellent time resolution and low power consumption with non-expensive 110 nm CMOS technology. In order to cope with the high input rate, a 10-bit 4 MHz SAR ADC will replace the Wilkinson ADC. The conversion time of 250 ns would allow operating at a hit rate per channel of about 3.2 Mhit/s, corresponding to 10% occupancy and 32 MHz bunch crossings. In order to reduce the effect of the pulse tails and to filter baseline fluctuations due to a large dark count rate, the SiPM input signals will be differentiated by a RC network either off-chip or in-chip. Tests of the AC coupling performed with TOFPET2 indicate an improvement of the time resolution by 15%. The TOFPET2 is implemented in 110 nm CMOS technology from the United Microelectronics Corporation (UMC) foundry. Measurements of the radiation effects on 130 nm CMOS n- and p- channel MOSFETs from three different manufacturers have been reported in the literature [52]. Even though the effects of the total ionising dose are qualitatively similar, requiring special care in the design of the circuits, the amount of degradation is shown to vary considerably from foundry to foundry, being stronger for UMC. Recently, irradiations with X-rays of the TOFPET2 ASIC have been performed. After 0.5 MRad irradiation, the TDC range was reduced by a large factor implying a degradation of the time resolution by a factor three. The effect was interpreted as due to increased leakage in the TAC-write transistor. This effect can be cured by a redesign with larger saturation margin to accommodate the V th variation due to radiation. The strategy being pursued in the new TOFHIR chip for BTL is two-fold, namely the revision of the re-used TOFPET2 blocks for radiation tolerance and the translation to the CMOS 130 nm technology provided by TSMC, which is less sensitive to radiation. A first version of the TOFHIR chip will be implemented in UMC 110 nm, allowing the development of final detector modules and the validation of system integration within a short schedule. In parallel 23

32 a TSMC version of the chip will be pursued. The configuration logic of the TOFPET2 chip is already protected against single event upset (SEU). Triple Module Redundancy (TMR) is used in all configuration data and state-machine flip-flops, with refreshing in case of bit-flip. The data path itself is not protected since the rate of data corruption due to SEUs is negligible. The TOFHIR chip will have additional logic to perform data filtering, following a Level-1 (L1) trigger. This will reduce the output data rate to 80 Mbit/s and allow using one e-link for output data in the front end chip. One lpgbt can be used as Data Concentrator of 24 TOFHIR chips connected to 24 e-links up to 320Mbit/s bandwidth in the lpgbt input Front End card The Front End (FE) card will house 4 TOFHIR chips for a total of 64 channels and is connected to four 16-channel SiPM modules. Six FE cards connect to one Concentrator Card (CC). The CC will house one lpgbt and the power and bias voltage distribution. Taking into account that L1-Accept filtering is applied in the front-end, the average rate of the data link from the TOFHIR ASIC to the concentrator is 80 Mbit/s (for 16 channels, 10% occupancy, 50 bit per hit, 1 MHz L1 rate). An estimate of the average rate of DAQ link in the concentrator board yields about 1.9 Gb/s, which allows operating the lpgbt at its lower output bandwidth of 5.12 Gb/s. The TOFHIR ASIC will use 1.2 V power supplies and is expected to consume 150 ma in the analog sector and 20 ma in the digital sector, for a total of 13 mw per channel. The voltages should be stable within 10%. In addition, for the operation of the front-end, TDC and ADC circuits, the TOFHIR requires reference voltages of 500 mv and 800 mv. The reference voltages will be provided in one of the two possible ways, either by radiation-tolerant external linear regulators or by an internal bandgap. In particular an option is under study based on a Point Of Load (POL) philosophy, for which a series of rad-hard linear regulators will be located close to the chips [53]. Another main source of power consumption in the electronics is the lpgbt, which operates at 1.2 Volt and uses mw power, depending on speed. Given that each lpgbt chip serves 384 channels the required power per channel is less than 2 mw. We plan to use the step-down DC-DC converter module FEASTMP CLP with a radiation tolerant ASIC (FEAST) developed at CERN. Given the tight space constraints we are studying the possibility of using a slimmer converter with a solenoid coil instead of the standard toroid. We estimate that this option would allow a gain of 3 mm in height. Measurements of FEASTMP with a solenoid coil show an efficiency of 76 80% for output currents of 1 to 3 A, which is about the same as with the toroidal coil. However the solenoid shows increased electromagnetic radiation by about 10 db (measurements without shielding). The optimization of the converter module for use in BTL is under way Clock distribution The distribution of a precise clock to the front-end system is a major requirement for the BTL. The LHC clock available in the backend system is recovered from the lpgbt down links. The high frequency clock noise is expected to be filtered by the PLL in the lpgbt. Low frequency clock jitter and possible phase instability, in particular arising from temperature variations or low-frequency response of the clock chain, may require special attention. A dedicated R&D effort across CMS sub-projects is being done to find the best solution for precise clock distribution at the level of the whole CMS detector. While waiting for the results of these investigations, we foresee the need to accommodate three dedicated clock fibres per tray in case the option of using a tree of separate clock path turns out to be necessary. Three special concentrator cards evenly located along the tray will house the clock fibre receiver and a dedicated radiation tol- 24

33 CMS Phase 2 Simulation probability < eta < < eta < < eta < < eta < E thresh (GeV) Figure 2.10: Occupancy of the barrel timing layer in 200 pileup events as a function of the energy threshold for a planar geometry of 4 mm. Results are averaged over the φ coordinate and displayed for different η ranges. The MIP most probable energy deposition corresponds to about 4 MeV for normal incidence. erant clock fan-out chip to distribute the clock over short coaxial cables to two neighbouring standard concentrator cards. The BTL would require in total 216 clock optical fibres System aspects BTL granularity and occupancy In the design geometry, four sensors (crystal+sipm units) correspond roughly to the granularity of a single ECAL Barrel PbWO 4 crystal. Figure 2.10 shows the occupancy at 200 pileup events as a function of the energy threshold for a planar geometry of 4 mm. Results are averaged over the φ coordinate and displayed for different η ranges. The MIP most probable energy deposition corresponds to about 4 MeV for normal incidence. According to simulation, the channel occupancy at 200 pileup collisions remains below 3% across the entire BTL, for signals of amplitude higher than 50% of the MIP most probable energy deposition. While the exact threshold may still be subject of optimization, this value provides a reasonable working point for the optimization of the granularity and for the estimate to the data rates (see Section ). At no threshold, there are many low energy hits, leading to an occupancy of order 10% or more, mostly due to out of time interactions originating from backscattered particles (e.g. low energy photons) from the ECAL. These hits will not affect the timing performance as they are sufficiently delayed with respect to direct tracks to not pile up on the rising edge of the signal where the time stamp is obtained. Furthermore, the average energy deposits from such particles in the LYSO crystals is relatively small peaking around 250 kev which corresponds to less than 7% of a MIP signal and thus their impact on the amplitude measurement used for time-walk corrections is minimal. Dedicated simulation studies demonstrated that the impact of both out-of-time pile up and backscattered particles on the time resolution is below 8 ps. The impact of pileup on the timing performance of the BTL was found negligible in simulations. The probability of a random overlap of two pulses in the same cell scales, to the first or- 25

34 Figure 2.11: Left: Impact of MTD on the electron ECAL energy resolution (left) and on a shower shape variable, σ ηη, relevant for electron and photon identification (right). No information from the timing layer hit is included in the shower reconstruction. der, with the square of the single cell occupancy. The double hit occupancy, which would cause an ambiguous assignment of the time information and would also distort the pulse shape, is below 0.1%. Track reconstruction will identify hits associated with more than one track. The probability of pileup of pulses from previous collisions scales with the channel occupancy, but the impact on the pulse distortion is reduced by the time delay. The residual effect of pulse shape distortion on this small number of events is effectively reduced using a dynamic pedestal subtraction, or an AC coupling at the input stage of the chip Impact of BTL material budget on the performance of ECAL A uniform barrel timing layer consisting of 4 mm thick LYSO:Ce tiles (X ), located between the tracker outer radius and ECAL at a radial distance of 1170 mm from the beam axis, has been implemented in the CMS event data processing framework, CMSSW, [54] using the default CMS Phase 2 geometry. This can be considered a first approximation of the full detector design, which would consist of thinner crystals and would include the dead material from electronics and mechanical structure, and in total would still be equivalent to < 0.4X 0. However, this simulation with a constant crystal thickness along the whole barrel length overestimates the amount of dead material in the high η region. One of the objectives of the crystal thickness leveling discussed in Section is indeed to reduce the material budget at large pseudorapidity to limit the impact on ECAL resolution. The impact of the barrel timing layer on the ECAL energy resolution and shower shapes has been studied for both photons and electrons using events from H γγ and Z e + e samples, respectively. The studies were done with no pileup, so any difference observed may also be partly washed out by the known degradation in resolution from in-time pileup [11]. Using a raw energy sum of crystals in ECAL, without high level energy correction, no effect is noticeable for photons and a modest effect (up to 1.3% in quadrature) is observed for electrons at η > 0.9, as shown in Fig No impact is observed on the shower shapes relevant for electron and photon identification, such as the σ ηη distribution, shown in Fig The small changes observed in the ECAL performance are likely to diminish or disappear with the updated leveled geometry of the barrel timing layer. Further studies are also in progress to evaluate the possibility to recover ECAL energy resolution using the hit registered in the timing layer for energy regression (through correlations with ECAL observables). 26

35 Total power consumption The total channel count is The per-sipm and the total power consumption are shown in Table 2.4, for some existing SiPM devices. At the beginning of operation the power consumption of the barrel timing layer will be driven by the electronics with 29 mw/channel including the ASIC and lpgbt power consumption, the efficiency of the DC-DC conversion on the trays, the voltage drop in the LV regulators and the power lost in cables. At the end of detector operation, the contribution to the total power consumption originating from the SiPM becomes comparable to that from the electronics. While the R&D will still focus on the minimization of the power consumption, existing devices already match or are close to a target of less than 10 kw for the SiPMs, which would yield a total power budget for BTL, after 4000 fb 1, around 18 kw. Table 2.4: Estimate of dark current levels and power budget for different SiPM size at the end of the detector operation (4000 fb 1 ) for an operating bias overvoltage of 1.5 V and nominal temperature of 30 C. SiPM Beginning of operation End of detector life Total (0 fb 1 ) (4000 fb 1 ) (4000 fb 1 ) Dark current HPK 4 4 mm 2 < ma 1.7 ma Dark current FBK 4 4 mm 2 < ma 0.9 ma Dark current HPK 3 3 mm 2 < ma 0.9 ma Dark current FBK 3 3 mm 2 < ma 0.5 ma Power cons. HPK 4 4 mm mw 70 mw 17.4 kw Power cons. FBK 4 4 mm mw 32 mw 7.9 kw Power cons. HPK 3 3 mm mw 39 mw 9.7 kw Power cons. FBK 3 3 mm mw 18 mw 4.5 kw Total weight The weight of the total structure is mostly driven by the crystal volume. Given the LYSO crystal density of about 7.4 g/cm 3 and the tile thickness varying between 3.75 and 2.4 mm (Table 2.1), the total weight of the barrel timing layer is around 1200 kg (806 kg from crystal volume and the rest from photodetectors, electronics and mechanical structure). The contribution of different components to the total weight of the timing layer barrel structure is given in Table 2.5. Table 2.5: Summary of contribution from single components to the total weight of the barrel timing layer structure. Material Density Thickness Total weight [g/cm 3 ] [mm] [kg] Crystal Silicon PCB Insulating foam Cooling bar Fiberglass Support Cooling system tbd Cables tbd Total 1206 The total load of the BTL structure is distributed uniformly on the inside of the outer wall of the TST. The TST is mounted on the CMS barrel hadron calorimeter (HCAL) via four metallic 27

36 brackets, using the same method as the present TST. The tracker systems are several times more massive than the BTL and derive their alignment from the horizontal rail system of the TST, which will remain unchanged relative to the present system. The incremental weight of the timing layer compared to the previous thermal screen system does not impose a significant additional load and will not affect the stability of the Tracker detector. The weight of the individual trays would be less than 20 kg, which allows us to handle them, possibly by means of a dedicated frame similar to the ECAL enfourneur. 2.2 Endcaps In this section we present and discuss a detector design to instrument the CMS endcap region with a precision timing layer (endcap timing layer ETL) capable of detecting MIPs with near- 100% efficiency and providing a time resolution of 30 ps in the rapidity interval 1.6 < η < 3.0. In the ETL reference design the detector is attached on the CE nose, in a separate cold volume, on the interaction side of the neutron moderator, as presented in Fig The ETL detector is thus protected, like the Tracker, from the back splash from the calorimeter. Figure 2.12: Endcap timing layer placement on the CE detector nose. The ETL detector (shown in yellow) is placed on the side of the interaction point, to the left of the neutron absorber (blue). The services for the ETL are routed through the CE cold volume, in two regions at 3 o clock and at 9 o clock, as indicated in the top right corner of this figure. 28

37 The ETL detector is based on planar silicon detectors with internal gain. The intrinsic gain enhances the MIP signal and provides an adequate signal-to-noise ratio for good timing precision. Two planar silicon technologies have been proven to provide sufficient timing resolution to meet the ETL requirements: Ultra-fast Silicon Detectors (UFSD) [55, 56] which are timingoptimized Low-Gain Avalanche Detectors (LGAD) [22], and Hyper-Fast Silicon (HFS), which is based on deep-depleted APDs [21]. Of the two, only the UFSD technology has been demonstrated so far to be able to withstand the radiation dose expected during the HL-LHC lifetime and to continue to provide acceptable time resolution. Therefore we base our design, presented in the following, on this sensor technology implemented in 5 10 cm 2 arrays of 1 3 mm 2 pixels. At the innermost radius of the ETL the particle fluence expected after an integrated luminosity of 4000 fb 1 corresponds to a dose equivalent to n eq /cm 2. The results of the radiation hardness qualification of the UFSD detectors are discussed in detail in Section Results from test beam studies of this past year have informed the direction of the currently ongoing R&D program, resulting in an intense prototyping schedule for 2018 to study large-area pixellated sensors and a corresponding ASIC. In the next sections we will outline the ETL design. This is organized with an overview of the sensor technology and readout scheme, radiation qualification studies, system-scaling aspects including power consumption, and finally a road map for additional R&D and production. This reference design tries to exploit as much as possible the R&D already done for other CMS detectors (outer and inner tracker, CE) in the areas of mechanical supports, cooling, and readout and control electronics to minimize the risks associated with new developments and exploit as much as possible the expertise already gained within CMS Mechanical structure The endcap timing layer (ETL) detector is placed on the CE nose, on the interaction side of the neutron moderator, as presented in Fig. 2.12, in a cold volume separate from the CE. In this configuration the ETL will cover the pseudorapidity region 1.6 < η < 2.9, with an active surface area of 12.1 m 2 for the two endcaps. The separation of the ETL and CE cold volumes allows servicing the ETL detector during regular shutdowns. A late installation of the ETL, as long as all services are already in place, would also be possible, in the case that the construction of the sensors and modules is delayed. The ETL modules described below are mounted in rings on flat aluminum support structures, which are split into 45 wedges, as shown in Fig The aluminum wedges are made from 0.25 MIC6 R cast aluminum plates, which provide a very stable, flat, and reproducible support structure. In order to ensure that the modules overlap in phi near the edges of the wedges, wedges will be machined to have the recessed and extruding regions on the edges, as shown on the left in Fig There is no offset in radius, the wedges are mated in the region of the recess, ensuring a continuous coverage in phi. Three wedge types for each disk are needed to get the desired edge overlaps (6 wedge types total). Stacked wedges are then assembled together into a wedge-unit, which are stiffer than a single wedge alone. These wedge-units are then assembled along φ to form one double-sided double-disk. To ensure hermetic coverage modules are mounted on both sides of two wedges: modules for odd (even) numbered rings are mounted on the wedge closest to (farthest from) the IP. In the azimuthal direction full coverage is obtained alternating the modules on the two sides of each wedge. A mechanical implication of the hermetic phi coverage of the modules is that the wedges will be machined to have the recessed and extruding regions on the edges, demonstrated in Fig There is no offset in radius, the wedges are mated in the region of the recess, ensuring a continuous coverage in phi. This arrangement of modules is shown in Fig The ETL double-disk is then mounted 29

38 on the CE nose, as shown in Fig Two double-disk units will be produced, one for each endcap. Single wedge Stacked wedges Figure 2.13: The wedge structure (Left) and arrangement of modules (Right) on one disc of an ETL double-disk structure, showing modules belonging to odd numbered rings mounted on the first wheel of the ETL. The even-numbered rings (not pictured here) are on a separate disk located 2 mm in z closer to the interaction region and fill the the radial space between each ring, including and inner and outer ring beyond the first and fifth rings shown here. The ETL detector is populated with modules that constitute the basic unit, similar to the tracker model. The conceptual design of these modules is presented in Fig The dimensions chosen for the module size are identical to those of the PS modules [36] of the CMS Phase 2 Outer Tracker (OT) detector, which would allow for reusing the design of some of the tooling developed for the OT modules assembly. Like in the Macro Pixel Array (MPA) [36] part of the PS modules of the OT detector, the sensor is bump-bonded to the read out chips (ROC), and the ROCs are wire-bonded to a flexible circuit. Unlike the PS modules, the ETL modules are single layer detectors. This allows for placing the hybrid circuit on top of the sensor, avoiding the need for an external support frame. This structure is similar to the one used for the Phase 1 and Phase 2 CMS pixel detectors, and the ETL modules are mounted directly on the aluminum wedges. The dimensions of the sensors are optimized for production on 6 (150 mm) wafers, similar to the OT. Two ETL sensors can be produced from a single 6 wafer, as shown in Fig The total number of modules required to cover one double-sided double-disk is 1312, with the modules arranged in 11 rings. The ETL modules are attached to the support plates via module end-holders that are screwed into the aluminum. To improve the thermal contact between the modules, in particular the ROCs, and the aluminum plates a thermal paste is used. This is a mounting scheme identical to the one used for the Phase 1 CMS forward pixel detector. The heat generated on the module, where the end-of-life module power from Table 2.8 is used, is evacuated via the aluminum support to the CO 2 cooling pipes, keeping the silicon sensors on ETL modules at an operating temperature below 25 C. This is the scheme used for cooling the CE and the inner tracker sensors. To ensure efficient heat removal, the components that generate a large amount of heat, such as the DC-DC converters and optical transceivers, are mounted directly on the support structure next to the modules. Unless serial powering is used there will be no hot-spots on the modules, as the DC-DC converter is placed on the disk away from the sensors and ASICs. In the case of serial powering the current shunt for an inactive ASIC could generate a hot-spot, and this will be studies for the TDR. Wires and optical fibers are routed radially outwards on 30

39 Figure 2.14: Left: A 3D visualization of the ETL module, consisting of the UFSD sensor sandwiched between 16 readout chips (bottom part) and the flexible hybrid circuit (top part) wirebonded to the readout chips. The sensor is bump-bonded to the readout chips. Right: A schematic of a 6 wafer with 2 UFSD sensors for the ETL detector. each wedge, above and between the modules. Thin protection frames are used to constrain the services and protect the modules. Figure 2.15: Left: Side view of the ETL wedge, showing a CO 2 cooling tube running inside the aluminum plate. Right: Routing of the CO 2 cooling pipe over the half of a wedge. The other half of the wedge is the mirror image of the one shown in the figure, with the left edge of the image being the symmetry axis. The power dissipated by the modules is removed by a network of low mass cooling pipes fed by the CO 2 cooling system, powered by a CO 2 cooling plant that can be operated separately from those for the other CMS detectors. The CO 2 flows inside cooling pipes that run within the grooves carved inside the aluminum plates, as shown in Fig The thermal performance of ETL modules was studied using finite element analysis (FEA) simulations based on CAD models using ANSYS. The wedge plate is made of MIC6 aluminum, into which round-bottom grooves are machined in order to place the stainless steel cooling tubes. Inside the tube, it is assumed that the two-phase CO 2 coolant is at 33 C at the location of the first module, which is the warmest location along a cooling loop. A sheet of aluminum is glued to the plate to seal over the cooling grooves. A heat transfer coefficient of 5000 W/m 2 /K is assumed for the heat 31

40 Figure 2.16: The FEA solution for a full ETL wedge. Only the temperatures on the sensors are shown. The routing of the cooling pipes is shown in Fig transfer from the cooling pipe inner wall to the CO 2. Heat loads of individual components used in the simulation are listed in Table 2.8. These power estimates are obtained from simulations of the sensor response, from assumptions on the readout chip discussed below in Section , and from Ref. [36]. The layout of the cooling pipes and the results of the FEA simulation are shown for the full wedge in Figs and As these results indicate, temperatures below 27 C, below our target of 25 C, are achieved on all sensors in the ETL detector, even in the most challenging locations, such as at the edges of a wedge, where some modules are supported and cooled down only in part by the aluminum. The total amount of space required along the beam axis for the ETL detector is 42 mm, plus 20 mm required for the additional thermal screen between the ETL and CE cold volumes. The development of a thinner DC-DC converter, already discussed in the case of the BTL in Section 2.1.4, would allow a reduction of the space required for the ETL to well below 40 mm. A preliminary study indicates that this space, and the space needed for the additional thermal screen separating the ETL and CE cold volumes, can be accommodated on the nose of CE without changing the current envelope of the CMS endcap detector. This requires reducing the thickness of the neutron moderator from 18 cm, the value used in the tracker TDR [36], to 12 cm. This change of the neutron moderator thickness generates an increase in the particle flux in the tracker and in the silicon layers of the CE, as shown in Fig This increase in dose is small enough that it does not induce a major performance impact after 4000 fb 1. Also shown is the radial dependence of the particle flux in the ETL location, which decreases from n eq /cm 2 at the largest rapidity to n eq /cm 2 at the largest radius. The services for the ETL have to be routed inside the cold volume of CE. Initial studies indicated that enough space is available between the cable trays of the HCAL Barrel (HB) and Patch panel 1 (PP1) in two specific locations (3 and 9 o clock positions), allowing the routing of the ETL cooling pipes up to the external area of the first endcap return yoke (YE1) disk. The routing of the ETL cooling lines requires a reduction of the rapidity coverage of the first two layers of the EM section of CE in the same area. Further studies are required for the power cables and the optical fibers. Estimates of the total surface required for the ETL services in each endcap are presented in Table 2.6 for the two scenarios (scenario A: one lpgbt per module, no serial powering; scenario B: one lpgbt every two to four modules, serial powering of ROCs within 32

41 Figure 2.17: Left: Ratio of the neutron equivalent doses observed in the CMS detector for the proposed neutron moderator thickness of 12 cm and the value corresponding to tracker TDR setup, 18 cm. Right: Radial profile of the dose in the ETL detector location. Table 2.6: ETL service requirements for each endcap in the two detector configurations discussed in the text. A packing efficiency of 70% is assumed when calculating the total required area. Service Required area (cm 2 ) Scenario A Scenario B CO 2 Cooling 46 Optical Fibers 16 5 Power Cables Total one module, two or four modules powered in parallel in rings 1 2 and 3 11, respectively) Module Specifications The size of the individual pixels on the ETL sensors is determined by several requirements: keeping the occupancy below 5%, keeping the pad capacitance low enough to achieve the target time resolution with gain 20, keeping the pad volume small enough to have low leakage current at the target integrated luminosity. The pixel size of 1 3 mm 2, chosen for the ETL sensors, satisfies these requirements, maintaining low capacitance of around 7 pf. At smaller pseudorapidity (1.6 < η < 2.1) three adjacent pixels will be logically combined together to reduce the number of read-out channels: the pixels will be ganged together inside the ROC, after the pre-amplifier comparator stage, to create larger read-out units. The combination of pixels after the discriminator stage retains the full benefit of small pixels, as the pertinent noise value is that of the small pixels and not that of the larger read-out unit, see Section The pixel size has an important implication on the detector dead area: the p + gain layer, i.e. the pad active area, is surrounded by an n ++ deep implant, roughly 20 µm wide, and by a p-stop implant, 15 µm wide. The sum of these two factors implies a 35 µm dead area around each pad, causing a dead area for a 1 3 mm 2 pad of 9%. The ganging of the discriminator pulses will be a configurable parameter of the ROC, 33

42 avoiding the need for two types of ASICs during the detector construction. The ROCs in the outer 8 rings of the ETL have 32 channels per ROC, due to the 3-to-1 ganging of the pixels. The total number of pixels per mm 2 ETL sensor is 1536 for the inner rings and 512 for the outer rings. In the simplest read-out and powering scheme, each module will be serviced by 16 ROCs and one aggregator ASIC that receives the clock, trigger and control data from the lpgbt, distributes these signals to the 16 ROCs, and merges the output data from the 16 ROCs into a single stream. The aggregator ASIC is mounted on the flexible circuit and connected to all the ROCs. In this design, each module will be connected to the DAQ backend via two lpgbt fibers that are used to read out the data from the detector, and to send clock, triggers, and configurations to the detector. As discussed in Section one additional fiber per module may be necessary to distribute the clock signal. One two-stage DC-DC converter per module is used to provide power to each individual module. Further details of the electronics and readout system are given in Section A reduction of the services can be achieved by sharing the read-out and control fibers between modules, by powering modules in parallel, and by using serial powering of the ROCs within one module. A possible realization of this scheme is discussed below in Section Sensor description and radiation hardness qualification Description of the sensors and test beam results UFSD is a concept in silicon detector design that merges the best characteristics of standard silicon sensors with the main feature of APDs, illustrated in Fig The overarching idea is to design silicon detectors with signals that are large enough to assure excellent timing performance by achieving a large slew rate dv/dt, but to keep the gain as low as possible [22, 56 61]. Figure 2.18: Schematic of a traditional silicon diode (left) and of a Low-Gain Avalanche Diode (right). The additional p + layer underneath the n ++ electrode creates, when depleted, a large electric field that generates charge multiplications. Timing optimized, thinned, UFSD sensors have been successfully produced by three vendors: Centro Nacional de Microelectrónica (CNM), Barcelona [22, 55, 62], Fondazione Bruno Kessler (FBK) [58, 63], and Hamamatsu Photonics (HPK) [64, 65]. Multiple production techniques have been used in the fabrication of the sensors (Fig. 2.19), including variation of substrate and gain layer dopants. This has resulted in sensors that are capable of surpassing the radiation tolerance needs of the endcap timing layer while providing suitable timing resolution through the end-of-life of the detector. In the following, we review the most relevant results from the R&D program on UFSD, and then discuss future developments aimed at building sensors with larger surface area and smaller 34

43 dead regions between pixels. Further improvements of the radiation hardness and of the operational parameters of the sensors will also be investigated. Most of the studies done so far have been performed with discrete amplifiers and waveform digitizers connected to the sensors via wire-bonds. In the future we plan to use dedicated ASICs developed for the ETL, as soon as prototypes become available, and to move towards larger devices connected to the readout via bump-bonds. Figure 2.19: Left: Production wafer from FBK. Right: The layout of the CNM wafer, with the indication of the test structures for specific detectors. Both CNM and FKB are capable of producing sensors with Gallium as an alternative dopant, which is being studied for possible further radiation hardness improvements. Sensor wafers from HPK are not pictured, and HPK cannot produce Gallium-doped sensors Results from 50 µm thick UFSD Several beam tests using 50 µm thick UFSD have been performed in the last couple of years. The tests have been carried out both at CERN and at FERMILAB, employing new and irradiated sensors. The sensors employed in these tests were manufactured by CNM, HPK and FBK, and were read out by custom made boards designed at Santa Cruz (Ca), Kansas University (KU) and FERMILAB [55, 66]. The sensors from FBK were manufactured with gain layers doped with Boron, Gallium, and Carbon-enriched Gallium. The combinations of these beam tests aimed at establishing the performance of UFSD sensors, the uniformity in their response, and the possible difference among manufacturers. Several CNM mm2 UFSD have been tested [55] at the CERN North Area using a 180 GeV π-meson beam. A fully custom broadband amplifier and a trigger board comprising a SiPM coupled to a quartz bar were used; this beam test not only established the feasibility of UFSD for timing, but also validated the simulation package WF2 [67], used to simulate the response of the sensor. In the beam test at Fermilab [66] a 120 GeV proton beam was used, in combination with a high precision pixel tracking detector. In this test, UFSD sensors manufactured by CNM and HPK were studied, and measurements of the uniformity of time resolution, signal amplitude, and charged particle detection efficiency across the sensor surface were performed. LGAD uses high-resistivity silicon bulk, with a level of p-dopant around 1012 atoms/cm3, and highly doped electrodes, n 1019 atoms/cm3. Some sensors had part of the front surface metallized, giving us the opportunity to study the difference in response between area with and without 35

44 the metal cover. The uniformity of the sensor response in pulse height before irradiation was found to have a 2% spread. The efficiency and timing resolution before irradiation were found to be 100% and ps, respectively. The measurements of the time difference between the reference timestamp and the timestamps of the HPK sensors are shown in Fig. 2.20, exhibiting an offset of about 20 ps between the metallized area and the non-metallized area of the sensor. The metallized area extends about 1.5 mm into the sensor active area, as shown in Fig The feature is present in all 4 types of the HPK PIX sensors and appears to be statistically consistent in shape and magnitude. Time resolution across the whole sensor area is shown on the right panels of Fig. 2.20, where we observe values around 40 ps across the entire sensor area. t [ns] t [ns] t [ns] t [ns] HPK 50A-PIX HPK 50B-PIX HPK 50C-PIX HPK 50D-PIX x-coordinate [mm] Time resolution [ps] Time resolution [ps] Time resolution [ps] Time resolution [ps] HPK 50A-PIX HPK 50B-PIX HPK 50C-PIX x-coordinate [mm] HPK 50D-PIX x-coordinate [mm] Figure 2.20: HPK 50A/B/C/D indicate different boron concentrations in the gain layer of the device, with the concentration increasing from A to D. Two classes of HPK devices were tested with these dopant variations, 1x1mm2 pads and 2x2 arrays of 3x3mm2, the latter are given the suffix?pix?. t (left) and time resolution (right) measurements as a function of the X position of the beam particle for the HPK 50A-, 50B-, 50C-, and 50D-PIX sensors. Regions between the vertical dashed lines on the left figure indicate the metalized areas of the sensors. The same effect has also been measured in CNM sensors. After a neutron fluence of n eq /cm 2, the CNM sensor exhibits a gain variation of a factor of 2.5 when comparing metallized and non-metallized sensor areas. An irradiated CNM sensor achieved a time resolution of 30 ps for the metallized area and 40 ps for the non-metallized area, while the HPK sensor achieved a 30 ps time resolution. This difference in response can be problematic, and no decisive differences have been seen among UFSDs from different foundries that would indicate a better radiation resistance related to this effect. Therefore the HPK design with a fully metallized sensor surface has been chosen as a reference. Given that the ETL design comprises a single UFSD layer, the dead area between pads needs to be kept as small as possible, since it directly impacts ETL active area. Using the beam test high-precision tracking, the no-response area between pads was measured to be about 70 µm for CNM and 110 µm for HPK sensors, as shown in Fig In the CERN North Area, 50 µm UFSD sensors manufactured by FBK with different gain layer compositions have been tested using π-mesons with a momentum of 180 GeV. The detectors were 1 1 mm 2 pads, mounted on a small PC board, that provided the filtering for the bias 36

45 LGAD Sensor: HPK 50C-PIX LGAD Sensor: CNM W9HG11 Efficiency Efficiency µm Channel µm Channel Channel Channel x-coordinate [mm] x-coordinate [mm] Figure 2.21: Efficiency measurement across the x-axes of the HPK 50C- PIX (left and CNM W9HG11 (right) sensors. The scan along y-axis shows a similar size of the dead area between pixels. The HPK sensor is operated at 450 V, and CNM sensor is operated at 180 V. The data points in blue are those from a pixel, and while those in red are from the neighboring pixel. The arrows indicate the distance the turn-on points of the fitted efficiency curves. voltage. Each sensor is wire-bonded to a strip leading to a SMA connector where a Cividec 40db current amplifier was connected. The data were stored in a 4 GHz, 40 Gs LeCroy digital scope. The study of these sensors prior to irradiation demonstrates that with equal bias and gain the same resolution is achieved regardless of the composition of the gain layers. Initial tests of these sensors are promising, with the sensor variations demonstrating uniform and very stable pulse shapes. Results after irradiation on these sensors, with different gain layer compositions, will be available at the beginning of Results of an irradiation campaign on 50 µm UFSD sensors In this section we focus on the results obtained in the irradiation campaign of 50 µm thick UFSD sensors manufactured by HPK [68]. The sensors have a Boron-doped gain layer, therefore this study shows how to operate UFSD sensors in case further R&D on radiation resistance will not demonstrate a better alternative. The left panel of Fig shows the summary of how the gain changes with irradiation, and how the external bias voltage should be increased to compensate for the loss of active doping in the gain layer. All measurements are taken at 20 C. As a reference value, the initial bias has to be raised from 150 V to 750 V to maintain a gain of 10. The fastest changes are for fluences between and n eq /cm 2. Above n eq /cm 2 the gain is obtained mostly in the field generated by the external bias. The right panel of Fig shows the time resolution for the irradiation points presented in the left panel. The main feature shown by the data is that the only relevant parameter for the time resolution (assuming saturated drift velocity) is the gain value, regardless of the fluence, and that the resolution is indeed getting better for gain up to 20. However, in the plot it is possible to identify an additional interesting feature: at low gain there are two different groups, with the first one formed by the lower-fluence data up to n eq /cm 2 and the second by higherfluence data. For a fixed value of gain, 5 for example, the lower-fluence group has a worse time 37

46 Gain Gain0vs.0Bias0,T0=00<200 o C0&0T0=0<300 o C00 HPK050D0pre<rad00T0=0<20C HPK050D01e140T0=00<20C HPK050D03e1400T0=0<20C HPK050D06e140T0=00<20C HPK050D01e150T0=00<20C HPK050D03e150T0=0<20C HPK050D06e150T0=0<20C HPK050D01e1500T0=0<30C HPK050D06e1500T0=0<30C Gain Resolution0[ps] HPK050<micron0sensors Gain0vs.0Bias0,T0=00<200 o C0&0T0=0<300 o C00 Bias0Voltage0[V] HPK050D0pre<rad00T0=0<20C HPK050D01e140T0=00<20C HPK050D03e1400T0=0<20C HPK050D06e140T0=00<20C HPK050D01e150T0=00<20C HPK050D03e150T0=0<20C HPK050D06e150T0=0<20C HPK050D01e1500T0=0<30C HPK050D06e1500T0=0<30C HPK050D01e14 HPK050D03e14 HPK050D06e Gain HPK050D01e15 40 resolution than the higher-fluence group. HPK050D03e15 This fact is due to the different type of multiplication 20 HPK050D06e15 mechanism in the two groups: below a fluence of n eq /cm 2, the multiplication occurs 0 mainly in1 the gain layer, 10 while for higher 100 fluences the multiplication happens everywhere in Gain Resolution0[ps] Bias0Voltage0[V] HPK050:micron0sensors050D0 Time0Resolution0vs0Gain00T0=0:20C HPK050D0pre:rad HPK050D01e14 HPK050D03e14 HPK050D06e14 HPK050D01e15 HPK050D03e15 HPK050D06e15 Figure 2.22: Gain as a function of the bias voltage of the UFSD sensor irradiated to the indicated HPK050:micron0sensors050D0 neutron100fluences Time0Resolution0vs0Gain00T0=0:20C at 20 C and 30 C, showing the need for increasing the bias of irradiated HPK050D0pre:rad sensors to reach adequate gain. the bulk, allowing for the multiplication process to start earlier. This fact leads to two changes in the pulse shape that improve the time resolution: the multiplication is smoother and the rise time decreases. The results shown up to now have been reported at the bias voltages that minimize the time resolution. These values tend to be lower than the maximum voltage reachable, where the current becomes excessive and the resolution deteriorates, largely due to increased noise. Given the large number of ETL sensors, that might be grouped together in a given bias line, it is useful to measure the sensitivity of UFSD performance to the exact bias voltage. In order to study this effect, we investigated the sensor performance at 10% reduced bias, indicated as headroom voltage V HR. The results indicate the capability of UFSD of working at a reduced bias voltage: for the majority of voltages the resolution worsened by 10% while for some voltages did not change significantly. This result is important since, as explained in Section , the sensor will be exposed to non-uniform irradiation and therefore part of it could be working at a voltage lower than optimal to avoid voltage break-down at the other end of the sensor. Table 2.7: Time resolution at the operating voltage (OV) and headroom voltage (HR). T [ C] Fluence V OP [V] Gain σ t [ps] V HR [V] Gain σ t [ps] [n eq /cm 2 ] V OP V OP V HR V HR Power consumption, biasing scheme and sensor dimension A necessary parametrization to compute the ETL power requirement is the leakage current as a function of fluence. It has been proven that the leakage current increases in UFSD devices at the same rate as standard silicon sensors [69], as this process is due to basic properties of the silicon bulk. Figure 2.23 on the left side shows the leakage current increase as a function 38

47 of fluence per cm 2 at a temperature of 20 C. This picture shows the interesting feature that the rapid change of bias voltage with radius happens early on in the HL-LHC lifetime at small radii, and then moves to larger radii with increasing fluences, where similar effects are seen in other experiments [70]. At about 1/2 of the HL-LHC lifetime, the innermost sensors reach a stable biasing situation, that will extend up to radius 70 cm at the end of HL-LHC. The combination of the left and right sides of Fig allows us to compute the power per cm 2 at each location of ETL by multiplying the expected bias voltage by the expected leakage current at a given fluence. This is shown on Fig. 2.24, right side, assuming a constant gain of 20 for the whole ETL. Bias"[V] Evolution"of"the"bias"working"point"for"HPK"sensors y"="$3e$43x 3 +"4E$28x 2 +"3E$13x"+" E E E E+16 Fluence"[n/cm2] I"["uA] Current"increase"vs"fluence"for"a"UFSD" 1"cm2,"50&micron"thick,"gain"20,"T"="&20C y"="7e&14x"& E E E+15 Fluence"[n/cm2] Figure 2.23: Left: Operating voltage vs fluence for 50 µm thick HPK sensors. Right: Leakage current per cm 2 as a function of fluence at 20 C. Figure 2.24 shows, on the left, a parametrization of the fluence as a function of radius for 3 different eras of HL-LHC running: 1/4, 1/2 and full lifetime. Using these curves to determine the fluence at each position of ETL, the power needs for the UFSD can be computed as a function of position assuming a constant gain of 20 throughout the entire system. This is shown on the right side of Fig Fluence [10^14 neq/cm2] I"["uA] Current"increase"vs"fluence"for"a"UFSD" Fluence vs radius 1"cm2,"50&micron"thick,"gain"20,"T"="&20C End of lifetime 1/2 Lifetime 1/4 Lifetime y"="7e&14x"& E E E+15 Radius [cm] Fluence"[n/cm2] Figure 2.24: Left: Dose as a function of radius for 3 moments of the HL-LHC lifetime. Right: Power requirements as a function of radius for UFSD sensors at 20 C assuming a constant gain of 20 for the whole ETL. An important consequence of the steep change in bias condition as a function of radius is shown 39

48 in Fig. 2.25, left: assuming a 10 cm long sensor, the difference in bias voltage at the two edges of the sensor exceeds 190 V. Even though UFSD sensors can be operated at a reduced bias voltage without a dramatic loss of performance, see Table 2.7, the requirement of 190 V of headroom requires dedicated R&D. Clearly allowing for smaller sensors will reduce this problem, as shown on the right side of Fig. 2.25: a 2.4 cm wide sensor (the width of 2 read-out chips) will be required to have a headroom of 70 V, which is already possible in current sensors. The results of the R&D program aimed at increasing the headroom of UFSD will inform the final design of the modules. We are aiming to have single sensor modules with a 10 cm sensor, as this simplifies the detector construction process, but we also consider as a fall-back solution the possibility of building modules with the same size and multiple (two or four) sensors. Having multiple sensors per module will result in small additional dead areas at the boundary between two sensors: the space needed for guard rings (500 µm per sensor), and an additional 100 µm to allow for misalignment of sensors and imperfections in the sensor dicing. This will result in additional dead areas of 1 3% for modules built out of 2 or 4 sensors, respectively. 200 Bias2working2point2difference2between2each2 10Ecm2wide2ring 200 Bias2working2point2difference2between2each22 2.4Fcm2wide2ring Voltage22[V] End2of2lifetime2 1/22Lifetime 1/42Lifetime2 Voltage22[V] End2of2lifetime2 1/22Lifetime 1/42Lifetime Radius2[cm] Radius2[cm] Figure 2.25: Left: Difference in bias voltage at the edge of each 10-cm wide ring. Right: A zoom of the left picture for 2.4-cm wide rings Readout Electronics Initial studies are underway for the design of a dedicated ROC for the ETL detector, that should provide a time measurement with no capacitative load from the sensor with a precision around 25 ps, and an 8-bit ADC to provide a capability for time-walk correction. Additionally, to keep the heat load of the detector at a manageable level, we aim at a power consumption of 100 mw/cm 2 for the ROC ( 3 mw per channel). Two complementary design activities are ongoing as part of the project of designing this new ROC. Both activities are using the 65 nm technology, with the goal of reusing, where possible, functional blocks that have been already designed and tested in the development of other ASICs for the LHC experiments by the RD53 Collaboration [71] and within the lpgbt project. In the first of the two design activities, the overall architecture of the ROC is investigated from the point of view of achieving the required 25 ps time measurement precision while minimizing the power consumption. In parallel the analogue front-end of the ROC (preamplifier and discriminator) from the TOFFEE Collaboration is being ported to the 65 nm technology in view of its usage in the ROC. This second design activity builds on the significant R&D in the area of time measurement that has been carried out in TOF-PET community. A recent report from the TT-PET Collaboration at TREDI 2017 demonstrated the second round of the ASIC prototype that achieved the goals that we pursue for ETL [65]. The goal of the TT-PET Collaboration is to equip about 600,000 channels with 40

49 about 2112 chips which are synchronized at the 10 ps level. They have achieved the goal of 140 mw/cm 2 with discrete component amplifiers, and have lowered this to 14 mw/cm 2 with their ASICs. Additionally, they developed a 10 ps TDC with 1 mw/channel. The translation of the front-end part of the ASIC design from 130 nm to 65 nm technology is being pursued in parallel. We have been investigating various architectures to achieve the required 25 ps time measurement precision while minimizing the power consumption. The design has so far focused on ensuring that the clock can be distributed over a large area without introducing a jitter that would spoil the precision of the measurement. The front-end design being pursued can be included at a later stage in the overall ASIC architecture being developed. The current (and still evolving) overall design foresees using a cascade of three different measurement to achieve this precision. First a 320 MHz clock, obtained either from an internal PLL or directly from the lpgbt, is used for a coarse measurement, dividing the LHC clock in ns bins. The 320 MHz clock is then fed into a delay-locked loop (DLL) with 16 elements. Each delay element differs from the previous by a phase of 195 ps (3.125s ns/16). Finally each phase from the DLL is fed into a a 3-bit linear phase interpolator (LPI) in each pixel, to obtain the final 10 bit precision that corresponds to a time measurement binning better than 25 ps. Simulations of the routing of the 320 MHz clock and of the signals from the 16 DLL phases into the pixels indicate that the phase differences between the signals to the LPIs can be kept under 10 ps, which gives a negligible contribution to the time measurement precision. In addition to the 10 bit time measurement, a charge measurement will be performed with a precision of 8 bits to allow for time-walk corrections. The possibility of using a second DLL together with the LPI in each pixel is also being considered. The data from each bunch crossing is stored locally in each pixel in a FIFO, provided there is an hit in the pixel. The cell in which the data is stored depends on an internal pointer that is incremented at every 40 MHz clock cycle. Data output from the FIFO is available at all times at the input of the readout control and it depends on the address presented by the external trigger. Once there is a trigger, when the token arrives at the pixel, the row and column address, the 10 bits of TDC data, and the 8 bits of ADC data are placed on a readout bus. The readout architecture also has the ability to bypass the analog front end and force a readout on this pixel by downloading a Force Readout bit through the slow control interface. It is also possible to download Inject and Kill information to either test or block the outputs of the analog front-end. The data from all the ROCs in a module is transferred in parallel using E-links at 320 Mbit/s to the aggregator ASIC, where they are serialized into a single data stream that is sent to the lpgbt. The data rate expected for the modules in the innermost ring of the detector is 300 Mbit/s. The aggregator chip is also responsible for the fan-out of the CMS clock to the individual ROCs in a module. The two functions of the aggregator ASIC are a subset of the functionality of the CE concentrator ASIC. Given the timeline for the development of large area sensors requiring more than 1 ROC and the timeline for the development of the CE concentrator, we do not plan to start the work on the aggregator for another 18 months. At that time, it should be possible to quickly develop this ASIC starting from the libraries developed for the lpgbt and the CE ASICs. The largest data rate from the ETL occurs in the innermost ring, and reaches the level of 1.4 Gbits/s per module. In the simplest readout scheme where a single lpgbt is used to read out a single module, 110 fiber bundles are required to read out one detector end, with each fiber bundle connected to 12 modules. The small data rate suggests reducing the number of lpgbt and fiber connections, and a solution with one lpgbt every two to four modules is also being considered. This reduces the number of fiber bundles to 48 or less, with individual fibers hav- 41

50 ing at most a rate of 3.0 Gbit/s. The total amount of data per detector end is 1.4 Tbit/s. This data can be read out connecting 72 pairs of fibers (6 fiber bundles) for one octant to a single Advanced Telecommunications Computing Architecture (ATCA) board that will receive 170 Gbit/s from the detector. The data would then be sent to one of two DAQ and TTC Hub (DTH) boards in the ATCA crate and from there it would proceed via 100 GBit/s Ethernet links to the HLT farm. The entire ETL system would then require 2 ATCA crates, each equipped with 8 boards that have a performance similar to the outer tracker DTC or to the CE backend, and one or two DTH boards. The performance requirements for the ATCA boards for the ETL backend are smaller in terms of data input to those of the outer tracker DTC, that has smaller bandwidth to the DTH. The corresponding boards for the inner tracker require a larger bandwidth to the High-Level Trigger (HLT). The requirements for the CE and the ETL readout boards is very similar. In the simplest powering scheme high voltage (HV) and low voltage (LV) power will be provided individually to each module using the same powering scheme as that for the OT, using a CAEN modular system (LV+HV on the same board, no sensing) to be placed in UXC. As is evident, ETL benefits from many common aspects with the OT, for issues such as mechanical structure, cooling, lpgbt and VTrx+, DC-DC converters. To minimize the longitudinal space required for the ETL, special versions of the DC-DC converters will be required, with a thinner coil, as already foreseen for other detectors (BTL, CE). In order to minimize the requirements on the space for services (that have to be routed on the periphery of CE), we are also considering a configuration of the detector where we use a single lpgbt every four modules and where we use serial powering of a chain of 8 ROCs and avoid using one DC-DC converter on each module. DC-DC converters are still needed to provide power to the lpgbt chips and to the aggregator chip on each module, but their number can be significantly reduced. The clock distribution system for the ETL will be similar to the one already discussed in Section The impact of the additional aggregator ASIC in the clock distribution tree will have to be studied to ensure that it does not introduce additional jitter in the time measurement. Service space estimates allow for an increase of the number of fiber bundles by 50% in the case that the clock distribution needs to be done separately from trigger and control signals System Aspects Power Consumption The total module count assuming the geometry described above is 1312 per endcap. Each module in the outer 8 rings contains 512 channels, while those in the inner 3 rings contain 1536 channels per module. The power consumption of the sensor is estimated as 15.9 mw/cm 2 after irradiation, which is rounded up to roughly 0.73 W for the full sensor at the end-oflife. The read-out ASIC power is estimated as the sum of the pre-amplifier, discriminator, and the TDC, which drive the majority of the power consumption on the ROC. Extrapolating the power consumption of the pre-amplifier of TT-PET chip [65] of 140 µw per channel, we assume a factor of 10 worse power consumption for the ETL chip s preamplifier. We estimate 0.23 mw/channel for the discriminator based on other similar ASICs [65, 72, 73], and the basis of 2 mw/channel power consumption of the TDC is taken from other similar developments, e.g. the PICOTDC [73]. The power consumption of the remainder of the active components of the ETL module are taken from the estimates of the PS modules of the outer tracker, where the same components are used, except for the aggregator ASIC where we scale down the power requirements of the tracker concentrator ASIC by a factor 6. As an example, the DC-DC converters take up about as much as half of the power that they provide. Hence, the power con- 42

51 sumed by the DC-DC converters is equal to half of the total power consumed by the readout ASIC, concentrator ASIC, lpgbt, and VTRx+. Table 2.8: Power consumption per ETL module after 4,000 fb 1. Component Small pixels Large pixels Sensor high-rad 0.7 W 0.7 W Read-out ASIC 5.6 W 3.5 W Aggregator ASIC 0.1 W 0.1 W lpgbt (high BW) 0.75 W 0.75 W VTRx+ 0.3 W 0.3 W DC-DC converter 3.6 W 2.6 W Total W 8.0 W Taking into account the power consumed per module, and the ohmic losses in the wires used to distributed the power, we estimate the power consumption of the full ETL detector after 3,000 fb 1 to be 10.8 kw per detector end. This number can be reduced by 4 kw per detector end when using serial powering and reducing the number of lpgbts. This reduction comes mostly from almost completely removing the DC-DC converters from the power distribution. A small number of DC-DC converters is still needed to provide power to the lpgbt, VTRx+, and aggregator ASICs Impact on Other Sub-detectors As discussed above, the amount of material added for the ETL in front of the CE is partially compensated, in terms of radiation lengths, by the reduction in the thickness of the neutron moderator ( 14 mm of aluminum plus 2.4 mm of silicon versus 32 mm of polyethylene). We do not expect any significant change in the resolution of the calorimeter due to the small increase in material (0.2 X 0 ). The ETL is mounted to the front of the CE, and therefore its services must be routed past the CE. Preliminary work has begun to define the required services and to determine how to route them from behind the CE to the ETL. The current understanding as of the writing of the CE TDR [14] of the requirements placed on the CE by the ETL are summarized here, but since this is an area of active work, the requirements and the designs to meet them are likely to evolve substantially. Depending on the powering and readout scenario, the ETL will dissipate between 6.8 kw and 10.8 kw per endcap. The power deposited in the CE by resistive losses in the ETL electrical cables is estimated be between 150 W and 450 W per endcap. The CO 2 cooling for the ETL subdivided into 4 cooling loops per endcap requiring supply and return lines of 8 mm and 14 mm diameter respectively. It is currently proposed to combine supply and return lines into a vacuum jacketed coaxial line with a diameter of 32 mm (4 per endcap) to ensure no thermal coupling with the CE. The need for the vacuum jacket for the piping run within the CE thermal screen is being evaluated. The cross-sectional area required for the electrical and optical services as they pass through the CE is between 85 cm 2 and 235 cm 2 including an allowance for a 70% filling factor. Including the four vacuum jacketed cooling lines results in a range of area required between 130 cm 2 and 280 cm 2. It is currently proposed to route the ETL services inside the CE thermal screen, concentrating them in two 5 28 cm 2 channels at the 3 o clock and 9 o clock positions. It is furthermore proposed that these channels be placed outside the radius reserved for the CE services to avoid 43

52 interference between the two systems, as shown in Fig This takes advantage of gaps near the horizontal axis between Tracker patch panels (PP1) in the region adjacent to the magnet cryostat and between HB cable trays in the 53 region. These gaps are reserved for rails used to install the Tracker, which are removed before the detector is closed and therefore the gaps can, in principle, be used for to route the ETL services. This space, however, lies outside the current envelope designated for the endcap and a formal configuration change would need to be approved by the CMS Integration Office before this proposed solution can be implemented. Figure 2.26: Proposed location of channels to route the ETL services outside the CE but inside the CE thermal screen. Left: Transverse view of the endcap showing the proposed routing of the ETL cooling tubes. Right: detail of the channel at 3 o clock that would contain the ETL cooling pipes and electrical and optical services. Figure 2.27(a) shows the current concept of how the ETL services would be routed from the channels shown in Fig into the ETL thermal volume. At the front of the CE, the thermal screen is split to provide a separate thermal volume for the ETL from that which contains the CE and the neutron moderator. The front surface of the thermal screens are at the envelope surface which defines the maximum longitudinal and radial extent of the endcap. The outer radius of the moderator is reduced by about 125 mm relative to its pre-etl size to provide a volume in which to route the ETL services within the ETL volume. The ETL services pass through the CE thermal screen just above the moderator. Design work in ongoing to ensure that no condensation occurs at the warm end of the feedthroughs when the ETL is open and the CE is cold. To reach the feedthrough, they are routed radially inward through a channel that is formed by a local reduction in the outer radius of the first two CE-E cassettes (first four detector layers) in 28 cm wide regions at the 3 o clock and 9 o clock positions. Taking account of the finite number of silicon module shapes that are used, the region in which CE-E active detector must be eliminated to permit the routing of the ETL services is 4.4 modules, as shown in Fig. 2.27(b). The physics impact of the modifications to this region must be studied. In the current concept the region in which CE-E active detector must be eliminated to permit the routing of the ETL services is 3.5 modules, as shown in Fig. 2.27(b). The affected region extends from η = 1.48 to η = 1.63, corresponding to trigger tower 18 and most of trigger tower 19, over about 10% of the azimuth. Design work is being done to understand how to minimize 44

53 the volume of CE-E that must be modified. (a) (b) Figure 2.27: Current concept of the CE-ETL interface. (a) The region of the feedthrough of the ETL services from the CE to the ETL cold volume. (b) One quadrant of the first CE-E layer showing the region of silicon sensors that would be eliminated from the first four detector layers (two cassettes) to accommodate the routing of the ETL services as shown in (a) Development plans This section details some major points towards completing the ETL system design, a complete list of items to achieve can be found in Section Sensor optimization and ASIC prototyping Several studies are planned to finalize the parameters of the detector. The primary goal is to complete the detailed study of radiation damage in UFSD sensors. Up to now, results with irradiated sensors have been attained with boron doping for the gain layer and several methods of mitigating the drop in timing resolution as a function of radiation dose have been proposed. In particular, the gallium gain layer and carbon-enriched boron layer productions have been produced in 2017 and are being studied, in addition to new runs investigating large-area UFSD sensor arrays and reducing the inter-pad dead region that will be available in early A series of test beams are already planned at Fermilab and other locations to investigate these new device types and to consolidate the information on the precision-timing properties of irradiated UFSD sensors. Furthermore, this campaign will yield knowledge of the appropriate operating conditions for production-like devices as they age in the detector and the bias voltage of the sensor is raised. As an outcome of this program the final design of the radiation-hard sensors will be selected by the end The active sensor area size is also subject to optimization. The sensor geometry plays a nonnegligible role in the determination of the fill factor for a complete sensor. The final choice of geometry needs to optimize both the active area of the sensor and the timing resolution of the devices. There are also technical improvements foreseen by CNM such that the fieldtermination gap between the sensors can be significantly reduced by using trenches between sensor pads, resulting in at least a 30% decrease in the inter-sensor gap and more in the corresponding dead area, depending on geometry. Particularly relevant for the final detector configuration are the maximum sensor size and the width of the HV plateau where each detector can be operated. If sensors with a size of mm 2 cannot be produced with a sufficient yield, 45

54 or if the width of the HV plateau does not allow, after irradiation, to operate a sensor of this size (because the voltage required at the inner radius is above the breakdown voltage for the part of the sensor at the outer radius), smaller sensor devices will have to be used. Both problems will be addressed via the R&D program we plan to pursue in the next two years. If only the second problem is an issue, we could foresee rotating the modules by 90, which involves building a larger number of modules to cover the same area, with an estimated cost increase of 20%. If there are fabrication issues that result in a low yield for large sensor devices, we could foresee fabricating smaller devices, which would be read out by a smaller number of ROCs. To avoid increasing the number of modules, we would then fabricate multi-sensor modules, possibly by adding one rigid ceramic layer to support the module during its assembly. Multiple HV connections would be necessary on each module, complicating the layout of the flexible circuit, but this would also help addressing the problem with the width of the HV plateau in the innermost rings of the detector where the radiation damage, and therefore the change in the operating point for the HV, is larger. The cost increase in this case should be limited as we expect that the extra costs in the module assembly will be partially compensated by an increase in the yields during both the sensor fabrication and the bump bonding. Multi-channel prototypes for UFSD sensors already exist in the context of R&D for the Highgranularity Timing Device (HGTD) being proposed within ATLAS, indicating that multi-cell UFSDs are feasible and maintain good timing resolution. Considering the much larger active area of our sensors and the number of ASICs being produced, careful attention will need to be given to the validation and design of the ROCs discussed earlier. This will proceed in multiple stages, with low-channel multiplicity ASIC prototypes being constructed first allowing for design optimization and characterization, finally yielding a design-spec ASIC prototype. Upon completion of the prototype ASIC, tests of a bump-bonded sensor package will proceed in order to ensure the scaling of the system in terms of ASIC features, and in particular the preservation of the sensor timing resolution and the distribution of internal clocks with low jitter. This finalized prototype will inform the final round of design decisions for the ETL ROC spec, leading to the production ASIC. The development of the ASIC foresees various small scale prototypes before the fabrication of a prototype with the full functionality and final dimensions. During the initial development of the ASIC two issues have been identified as being critical for the project: the clock distribution in the relatively large surface ASIC and ensuring low power consumption. While ASICs that achieve resolutions of a few ps have been available for a few years, there are no large area ASICs with those properties that are designed for bump bonding to a sensor element. The R&D program therefore focuses on these two specific topics. The first two prototypes will be built to test the clock distribution scheme described in Section 2.2.3, test the analogue front-end, and validate the estimates of the power consumption. They will be available in June and September 2018, respectively. The development costs for large area ASICs prevent us from studying the entire clock distribution chain over an ASIC with the dimensions that will be used for the detector construction. Therefore only the most critical elements of the clock distribution will be tested at first, and simulations will be used initially to demonstrate that the clock signal can be distributed over the entire area of the ASIC without introducing additional jitter. The first prototype will contain a 320 MHz PLL and and global 200 ps DLL. The aim of this prototype is to test the time resolution that is achieved in the digital part of the final ASIC, considering that the final LPI interpolation is not going to introduce additional jitter in the time measurement [74]. Even if the 65 nm technology has been demonstrated to be radiation hard to doses larger than the ones expected in the ETL, we will measure the timing response of the ASIC before and after irradiation. For example for the first prototype we will check the jitter on the outputs of the 46

55 DLL. The second prototype will contain the front-end amplifier and a discriminator. It will also reuse blocks from the first prototype to demonstrate that the time over threshold (ToT) method using the 200 ps DLL can be used for time walk compensation. The third prototype, available in March 2019 will contain the LPI block allowing for timing measurements with a binning of 20 ps. It will have a complete circuitry for reading out 1x1 mm 2 pixels, including the complete chain consisting of front-end amplifier, discriminator, ToT ADC, TDC. It will also have a SRAM memory distributed in the pixels, allowing for simultaneous read and write capability, as well as all the logic required to control the behaviour of the ASIC. Minimal readout of individual pixels, without zero suppression, will be available. This ASIC will be bump-bonded to a 2x2 matrix of 1x1 mm 2 pixels. From the following prototype onward the full functionality of the final ASIC will be implemented. The fourth prototype, available in September 2019 will implement a 4x6 matrix of 1x3 mm 2 pixels. It will be followed six months later by a full-size ASIC. These two last ASICs will be bump bonded to UFSD sensors, allowing for integrated tests on large area devices. The development of the aggregator ASIC can be delayed until late 2019, and it can exploit the same technology as the lpgbt in order to benefit from the extensive engineering that has gone into the clock distribution, controls, and e-links, which are the primary components of the aggregator schematic. This ASIC could be available for performing tests on modules with multiple ROCs at the same time as the final readout chip. 47

56 48

57 Chapter 3 Performance studies 3.1 Introduction Studies have been performed on the impact that the MTD can have on the physics deliverables of CMS, in terms of object reconstruction and of the cumulative effect of timing across all aspects of event reconstruction. Where possible, the studies have been performed as a function of interaction density in the LHC. Current LHC conditions correspond to a vertex line density of 0.4 events/mm. The typical density for the HL-LHC scenario corresponding to 140 pileup (PU) interactions is 1.3 events/mm, and at 200 pileup the density is 1.9 events/mm. In some instances, the reconstruction performance has been studied as a function of the assumed timing resolution in the range ps to evaluate the dependence of performance on the ultimate timing resolution achievable in a MTD. Full-analysis studies are benchmarked assuming the 200 PU scenario. Only a selected number of full analyses are presented that touch on some of the most important signatures for the CMS physics program at the HL-LHC. The Phase-2 CMS detector including the upgraded Tracker, calorimeters, and muon systems is used to derive the reference performance to which compare the differential gains from the MTD. Unless explicitly stated, up-to-date algorithms optimized for the reconstruction of 200 pileup events with the upgraded CMS detector have been used as well Simulation implementation A full simulation with the detailed geometry of the dedicated timing layers and full reconstruction of the signals is being developed. Preliminary studies of the performance impact of precision timing rely on a simpler implementation in the CMS simulation workflow based on smearing of the simulated time from other subsystems. For charged particles, the time is assigned directly from the time of the simulated track at the production vertex, assuming a 30 ps resolution and applying a minimum p T threshold of 0.7 GeV in the barrel and a p threshold of 0.7 GeV in the endcap, and covering the expected MTD fiducial region of η < 3. For photons in the central region of the detector, the time of each reconstructed hit is assigned from the energy-weighted average of each simulated hit time-stamp in the ECAL barrel in a thin transverse slice of the crystal near the shower maximum. For photons in the forward region, the time is assigned based on the energy-weighted average of the simulated hit time-stamp in a CE cell. In order to approximate the eventual behaviour of the readout electronics in the CE, which includes a Time-of-Arrival circuit, the time stamps of cells are assigned an energy-dependent timing resolution with a constant term of 50 ps, which is achieved at about 20 MIPs, depending on the CE silicon thickness. Separated simulation studies with a preliminary GEANT implementation of the MTD geometry within CMS have been performed to ascertain the validity of the adopted approximation. In particular, dedicated studies were conducted to verify if non-gaussian tails could affect the 49

58 time measurement in the detector or the extrapolation of the measured time to the vertex (timeof-flight correction). All of these effects were found to be negligible when compared to the 30 ps Gaussian smearing adopted in the simplified model. From a parametric model of the signal pulse shape and the full simulation of 200 pileup events in CMS, we estimate that pulse shape distortions due to pileup from previous and concurrent collisions and from backscattered particles from the calorimeters affect the time measurement with a contribution below 8 ps (Section ). Similarly, overlaps in the same MTD cell of spatially correlated tracks, belonging to the same jet, have a negligible effect (Section ). The time-of-flight correction was found to have an RMS uncertainty of 2 3 ps in a dedicated study where the hit in the MTD associated to the track is back propagated to the vertex. Tails of order 1 mm observed in track reconstruction, originated by interactions in the tracker or by secondary particles from decays, translate into time inaccuracies of order 3 ps, which are insignificant compared to the 30 ps resolution of the proposed detector. It has also been verified that the additional MTD layers do not impact significantly on the performance of the calorimeters (Section ) and of the Tracker. In particular, the design of the BTL will require the radius of the outer Tracker layer to be reduced by up to 15 mm (Section 1.4). The impact of this modification on the tracker performance will be marginal. The momentum resolution would change from δp T /p T = 0.540% and 0.918% at 10 and 100 GeV with the nominal geometry [36] to and 0.936% with the modified geometry. In summary, the approximation adopted provides a correct framework to study the impact of precision timing on the event reconstruction and on the physics processes with the CMS detector at the HL-LHC Time-aware event reconstruction Vertex reconstruction in the tz-plane, i.e. in time and position along the beam line, was studied using a time-aware extension of the deterministic annealing technique adopted in vertex reconstruction by the CMS experiment [7]. Since this technique can be extended to more than three dimensions, it is the natural choice for finding vertices in the dense environment of 200 pileup events. As outlined in Chapter 1 (Fig. 1.2), the introduction of time information significantly improves the performance of the vertex reconstruction algorithm in associating tracks with vertices. For example, instances of vertex merging are reduced from 15% in space to 1% in space-time, at 200 PU. Moreover, the space-time reconstruction capability has dramatic impact on several observables (e.g., pileup jet ID, p miss T, b tagging, lepton isolation), as further discussed in the next sections. This reconstruction method is further extended to associate the individual photons reconstructed in the calorimeters to a collision vertex. The neutral particle is assigned a straight trajectory with time-of-flight timing corrections corresponding to the z-vertex positions. This generates a set of compatible vertices, which are a function of the time-of-flight and vertex t 0 information. This capability would open several possibilities, including an improved identification of the diphoton vertex in H γγ decays, improved sensitivity to long-lived particles decaying into photons, improved measurement of the particle isolation from electromagnetic deposits, and improved resolution in the calorimetric missing transverse energy. A guiding illustration of how this approach works is provided in Fig. 3.1, showing how a pair of photons in a diphoton event is matched to a space-time vertex reconstructed from tracks. The green lines in the figure show, for each photon, the vertex time that would be needed for a vertex at the position in z. The figures visually show that a pair of photons arriving from a common vertex provides enough timing information alone to reconstruct the vertex z-position. The vertex 50

59 Figure 3.1: Space-time diagrams illustrating the concept of hermetic timing for H γγ events with two photons separated by a large rapidity gap (top) and a small rapidity gap (bottom). The reconstructed time for the photons at each vertex (green open dots), with error bars from the uncertainty on the time measurement of photons, can be cross referenced with the time information of the 4D vertices. The green straight lines are drawn to guide the eye. The pileup is reduced to an average of 20 in this case, to improve clarity. For photons with a small rapidity gap, the coincidence with a 4D-vertex is necessary to enable vertex location. 51

60 resolution is, however, heavily dependent on how separated the two photons are in pseudorapidity. As further detailed in Section , a triple coincidence with a vertex t 0 is required for the correct association with a vertex of a substantial fraction of the photon pairs. 3.2 Case studies: impact on event observables Reduction of pileup tracks associated to hard interaction primary vertex An essential step in rejecting pileup is to exclude from relevant quantities charged particles which are not associated with the hard interaction. This step is critical to maintain the performance of charged isolation sums for leptons, b-tagging, and the charged component of jets and missing transverse momentum. One method commonly used in CMS to associate tracks with the hard primary vertex, PV, is a simple selection on the distance between the track and the vertex, z(track, PV) < 1 mm, which maintains high efficiency for charged particles actually originating from the hard interaction. To optimize this seleciton, the isolation sum of charged particles was studied in 200 pileup collisions for z ranging between 0.2 and 2 mm. For particles at any pseudorapidity within the barrel acceptance, a selection window of 1 mm maximizes the identification performance (measured by the ROC curve). In the endcaps, a slightly looser selection may provide a better performance, with 1 mm being close to optimal. Hence, for an interaction density in excess of 1.0 event/mm, corresponding to operations at approximately 100 pileup and above, some or many of the resulting tracks will be associated with the hard primary vertex, contaminating the set of tracks used to calculate the relevant physics quantities and degrading the performance. This effect is most directly quantified as a function of the line density of events along the beam line, which drives the probability that additional vertices will be close enough in space to the hard interaction to contaminate the set of associated tracks with those from pileup. This association method can be extended with precision timing, by adding a requirement on the time distance t(track, PV) < N σt track, where N will eventually be properly optimized. In this study, a track-time resolution σt track = 30 ps and N = 3 were assumed, corresponding to the requirement t(track, PV) < 90 ps. Using this association criterion, the efficiency for reconstructing and associating tracks from pileup interactions in tt events is shown in Fig. 3.2 (left), both with and without precision timing, as a function of the event line density. In order to establish the process-independence of this metric, the same study has been performed on Z µµ, with results shown in Fig. 3.2 (right). These results suggest a generic reduction in the effective amount of pileup by a factor of approximately 4 5 for physics quantities constructed from charged particles. The effect of timing on signal tracks for these processes has been studied as well, and found to have no resulting penalty in the association efficiency. In order to assess the dependence of this gain on the acceptance of the precision timing detector, the track-vertex association is compared for pileup tracks in Z µµ events under several different scenarios in Fig. 3.3 (left). For tracks which fall outside of the precision timing acceptance, the association with the primary vertex relies only on z. Therefore the overall reduction in associated pileup tracks decreases, as the benefit is removed outside of the acceptance. Figure 3.3 (right) shows that pileup tracks in a restricted acceptance region ( η < 1.5 in this example) are effectively removed from the primary vertex by a timing detector covering the same acceptance region. Neither study indicates any appreciable loss in signal-track efficiency. This observation indi- 52

61 # of pileup tracks/signal PV CMS Simulation preliminary 40 tt event tracks 35 p >0.9 GeV T HL-LHC BS, 3D vtx, PU=140 HL-LHC BS, 3D vtx, PU=200 HL-LHC BS, 4D vtx, PU=140 HL-LHC BS, 4D vtx, PU= TeV # of pileup tracks/signal PV CMS Simulation preliminary 40 zµµ event tracks 35 p >0.9 GeV T HL-LHC BS, 3D vtx, PU=140 HL-LHC BS, 3D vtx, PU=200 HL-LHC BS, 4D vtx, PU=140 HL-LHC BS, 4D vtx, PU= TeV Density (events/mm) Density (events/mm) Figure 3.2: Number of tracks associated with pileup incorrectly associated with the hard primary vertex in tt (left) and Z µµ (right) events as a function of the pileup density, shown with (4D vtx) and without (3D vtx) precision timing. # of pileup tracks/signal PV CMS Simulation preliminary 40 zµµ event tracks 35 p >0.9 GeV, η <4.0 T No timing, PU=200 w/ Timing η <4.0, PU=200 w/ Timing η <3.0, PU=200 w/ Timing η <1.5, PU= TeV # of pileup tracks/signal PV CMS Simulation preliminary 40 zµµ event tracks 35 p >0.9 GeV, η <1.5 T No timing, PU=200 w/ Timing η <4.0, PU=200 w/ Timing η <3.0, PU=200 w/ Timing η <1.5, PU= TeV Density (events/mm) Density (events/mm) Figure 3.3: Number of pileup tracks in Z µµ events incorrectly associated with the hard primary vertex as a function of pileup density, shown without and with precision timing for several different acceptance scenarios, considering tracks within the full Tracker acceptance (left) and just in the central part (right) of the detector. 53

62 cates that the reduced acceptance, aside from effects on global event quantities such as missing transverse momentum, does not affect localized quantities such as isolation sums or jets within the remaining acceptance. This conclusion could change slightly with more sophisticated track-vertex association algorithms, which make more explicit use of the reconstructed pileup vertices, since the efficiency for reconstructing and resolving minimum bias vertices is expected to be more sensitive to the acceptance because of the lower multiplicity and softer momentum spectrum of the charged particles Pileup jet suppression In the presence of pileup, soft jets and underlying event activity from multiple pileup interactions may overlap and be clustered into a higher energy jet, serving as an additional background for e.g. Vector Boson Fusion (VBF) tagging and other final states with jets. For present LHC Run 2 conditions, this background can be largely suppressed by cleaning the charged particles based on spatial association with the primary vertex, and pileup jets are mainly an issue beyond the present tracking acceptance. Although the tracking acceptance of the CMS Phase-2 detector will be extended, the higher pileup density and corresponding increase in charged particles from pileup associated to the primary vertex leads to a non-negligible rate of pileup jets. The rate of pileup jets is studied in Z µµ events with 200 pileup conditions, with and without precision timing for the charged particles, using jets reconstructed with the PUPPI algorithm [5] and anti-k T clustering with 0.4 distance parameter [75], with a reconstructed jet p T > 30 GeV. The PUPPI algorithm currently uses the first primary vertex in the reconstructed collection. In order to avoid events where the incorrect primary vertex is used, two reconstructed muons with p T > 20 GeV and associated with the first primary vertex are required for events entering the study. Tracks are associated with the primary vertex with z(track, vertex) < 1 mm; an additional requirement of t(track, vertex) < 3 σt track is applied when considering the precision timing case. Signal jets are defined as reconstructed jets that are matched to a generator-level jet with p T > 4 GeV within a cone of R < 0.2, while pileup jets are defined as reconstructed jets that are not matched to a generator-level jet with p T > 4 GeV within a cone of R < 0.6. The relative rate of both signal jets and pileup jets with and without precision timing for the charged particles is shown in Fig. 3.4, where the addition of precision timing reduces the rate of pileup jets by 25 50% depending on pseudorapidity with minimal effect on the signal jet rate. These projections use a more recent release and a different working point that those used for the study of Section The working point and the rate reduction from timing may be further optimized. This reduction in pileup jet rate is expected to have a significant impact on signal extraction cases that rely on jet multiplicity event categorization, central or forward jet vetoes, or forward jet tagging, as in the case of VBF topologies Missing transverse momentum The performance of p miss T reconstruction is studied with and without the addition of precision timing information for the charged particles entering the p miss T calculation. This is done for PUPPI missing p T, constructed from both the charged and neutral particles, with an optimized weighting scheme [5]. A standard technique for the characterization of the performance is to study the scale and resolution of the hadronic recoil against the Z boson in Z µµ events. For the purposes of this study, tracks are associated with the hard interaction vertex by requiring z(track, vertex) < 1 mm, with an additional requirement of t(track, vertex) < 90 ps in the case that precision timing is available. It has been verified that timing selection for the charged 54

63 CMS Phase-2 Simulation s=14tev CMS Phase-2 Simulation s=14tev Relative Signal Jet Rate Z ll, <PU> = 200 Jet p > 30GeV t No MTD MTD Relative Pileup Jet Rate Z ll, <PU> = 200 Jet p > 30GeV t No MTD MTD Jet η Figure 3.4: Rate of signal jets (left), and pileup jets (right) reconstructed with the PUPPI algorithm with anti-k T 0.4 clustering and reconstructed p T > 30 GeV with precision timing for the charged particles relative to the no timing case. particles does not change the scale for PUPPI missing p T, indicating that the charged particles of the jet are not being removed by the additional requirement of compatibility in time with the hard interaction vertex. In the absence of precision timing, the PUPPI missing p T exhibits a degradation in resolution as a function of pileup density (Fig. 3.5-left), due to additional charged particles from nearby (in z) pileup interactions contaminating the hadronic recoil sum. The addition of precision timing for the charged particles reduces the slope and improves the resolution for high event density. The resolution at the average vertex density corresponding to 200 PU is improved by 10 15% with the MTD, and is equivalent to the resolution without a timing detector at 140 PU. Jet η ) (GeV) /p )/(u σ (u T CMS Phase-2 Simulation preliminary no MTD MTD Quadratic difference (GeV) Relative difference (%) PU = Line density (events/mm) ) -1 (GeV Normalized dn/d u Ratio CMS Phase-2 Simulation preliminary PU = 200 no MTD MTD u Figure 3.5: Left: Resolution of the hadronic recoil component perpendicular to the Z boson p T as a function of pileup density, which characterizes the contribution to the resolution of noise, including pileup. The dotted lines show linear fits to each set of data points; the dashed and the dot-dashed lines show, respectively, the difference in resolution in quadrature and in percent between the timing and no-timing case. Right: Distribution from simulation at 200 PU of the PUPPI hadronic recoil component transverse to the Z boson p T, with and without precision timing in Z µµ events with no real missing momentum. In order to examine the effect of the resolution improvement on the tails of the distribution, 55

64 often relevant for new physics searches involving particles invisible to the detector, the distributions of the PUPPI transverse recoil component in Z µµ events with no real missing momentum are shown in the right panel of Fig This distribution show a reduction of at least 40% in the rate of events with > 130 GeV, with the addition of precision timing. This is relevant towards reducing the background for searches and helping reduce high level trigger rates. Further studies and optimization of the core versus tail resolution are envisioned; larger samples are needed to better characterize the behaviour of the tails. Beyond the improvements to the performance with the addition of precision timing for charged particles, significant improvements are also expected from the addition of precision timing for the neutral electromagnetic component of the missing momentum. Preliminary studies suggest that approximately 45% of photons in the barrel have at least one hit in the barrel timing layer, due to conversions in the Tracker or in the timing layer volume itself. The MTD therefore will provide precise timing information even at lower energies, where the time measurement from upgraded ECAL barrel is less precise. Furthermore, the efficient and precise measurement of the hard interaction time enabled by the precision timing layers will allow timing information for photons from the CE to be maximally exploited Impact on b-tagging and displaced vertices In very high pileup conditions, secondary vertex b-tagging is degraded by the formation of spurious secondary vertices caused by pileup tracks, reducing the ability to distinguish signal from background. This degradation is seen clearly in Fig. 3.6, and depends on the average pileup, and pileup density. In order to mitigate this problem, the secondary vertexing algorithms were updated to be aware of timing information from the MTD. By requiring tracks to be within 3.5σ t of the selected primary vertex, the number of spurious reconstructed secondary vertices was reduced by 30%. This causes the ROC curves in Fig. 3.6 to improve significantly, especially for tighter working points where near-zero-pileup performance is achieved and the dependence of b-tagging efficiency on the pileup density is removed. These performance benefits have a particularly important impact on the signal yield in acceptance-sensitive analyses, such as di-higgs boson production in the HH bbbb and HH bbγγ final states. Further gain is expected from retraining the b-tagging discriminants for 200 PU conditions, exploiting the additional information from timing in a consistent manner. Figure 3.6: Secondary vertex tagging ROC curves for light and charm jets for η < 1.5 (left) and for 1.5 < η < 3.0 (centre), and b-tagging efficiency vs. average pileup density, with a constant light-jet efficiency of 0.01 (right). Results with and without timing are compared to the zero pileup case. 56

65 3.2.5 Muon and tau charged isolation Since charged particles comprise the largest fraction of hadronic activity in the event, the charged isolation is the most important contribution to isolation sums in the context of identifying isolated leptons and photons. Since these charged isolation sums rely on the association of tracks with the primary vertex, they are affected by the same issues described above, and similarly mitigated by the addition of precision timing. Studies have been performed examining the charged isolation efficiency for both muons and hadronically decaying taus. In the study of muons, prompt authentic muon candidates are considered from a Z µµ sample, each of which is matched to a true generator-level muon. Non-prompt muon candidates come from legitimate muons buried within a generator-level matched jet coming from events in a tt sample, most of which originate from semileptonic decays of heavy-flavour (HF) hadrons. Muon candidates are required to have p T > 20 GeV, η < 2.8 and a point-of-closestapproach to the primary vertex that is within 1 mm in the z direction. Tracks with p T > 5 GeV and within 1 mm in z of the primary vertex are counted for the isolation sum if they are within a cone R < 0.3 of the muon candidate. In the with-timing case, the additional requirement of t(track, PV) < 3 σt track is applied, for the assumed time resolution σt track = 30 ps. The isolation sum is computed excluding the muon candidate track, and then the ratio of the sum to the candidate p T is calculated. A selection requirement applied in this relative isolation corresponds to a point in the plane of the efficiency for prompt muons, ɛ prompt, versus non-prompt muons, ɛ non prompt. The ROC curve for a spectrum of possible isolation thresholds is shown in Fig. 3.7 (left). A clear benefit can be seen in terms of a reduced non-prompt efficiency in the with-timing case throughout the range ɛ prompt > The efficiency as a function of pileup density for a representative selection is shown in Fig. 3.8 (left). The impact of precision timing is evident at high event densities, with an acceptance gain of about 6% at the average event density of 1.4 mm 1 with 200 pileup collisions. The isolation efficiency at 200 PU with timing is equivalent to the isolation efficiency of current-era LHC pileup densities without timing, at constant background. A similar study was performed on hadronic taus, considering the so-called 3-prong, 1-prong, and 1-prong+π 0 tau categories, related to the decay multiplicity of the tau candidate. Here authentic tau candidates as well as jets-as-fakes come from a Z+jets sample. All tau candidates are required to have p T > 20 GeV, η < 2.4, and are reconstructed using the fixed-strip approach [76]. Here the raw charged track isolation is computed in a cone of R = 0.4. The results can be seen in the right-hand panels of Figs. 3.7 and 3.8. In terms of the charged isolation efficiency for real taus, there is an improvement of performance at 200 PU with timing that exceeds that of the current-era LHC pileup densities without timing. A comparison of the dependence of the charged isolation efficiency on the event density for various possible time resolution values is shown for muons and taus in Fig One can see the overall benefits in terms of recovered prompt candidate efficiency tracks with time resolution. The efficiency gain is still sizeable at 50 ps resolution. 3.3 Case studies: impact on selected physics analyses Higgs physics One of the highest priorities of the HL-LHC physics program will be the precision characterization of the Higgs boson. This campaign will require sensitivity to rare Higgs boson production mechanisms and/or decays (for example, H µµ, di-higgs boson production, etc.), as well 57

66 Figure 3.7: ROC curves calculated for a cut-off scan in relative isolation for muon candidates (left) and in raw isolation for tau candidates (right), comparing the no- and with-timing cases. Figure 3.8: The efficiency for prompt and fake muons (left) and authentic tau candidates (right) as a function of the event density for a representative operating point selection. Charged-isolation Efficiency CMS Phase-2 Simulation 0.2 Z ττ, PU = 200 No Timing MTD, σ t = 30ps 0.1 σ t = 50ps σ t = 70ps σ t = 90ps Line density (mm ) Figure 3.9: Muon efficiency for relative isolation cut-off of 0.05 (left) and hadronic tau efficiency for absolute isolation cut-off of 2.5 GeV (right) for different timing resolution assumptions, as a function of line density. as high-statistics samples of the signatures upon which the current-era Higgs boson campaign is built (for example, H ZZ 4l, H γγ, VBF Higgs boson production and subsequent 58

67 decay to taus, etc.). The cumulative impact of the MTD has been studied in the context of enhanced signal yields at constant background rejection for prominent Higgs boson processes for the HL-LHC era. Fraction of Events HH bbbb ( 200 Pileup Distribution ) No Timing Barrel Timing Only Barrel+Endcap Timing Increase in HH bbbb Yield Barrel Timing Only : 14% Barrel+Endcap Timing : 18% Fraction of Events Higgs ZZ 4l ( 200 Pileup Distribution ) 0.4 No Timing 0.35 Barrel Timing Only Barrel+Endcap Timing 0.3 Increase in Higgs ZZ 4l Yield Barrel Timing Only : 19% 0.25 Barrel+Endcap Timing : 26% y HH y Higgs Figure 3.10: Projections for yield enhancement in HH bbbb (left) and H ZZ 4l (right) as a function of the rapidity. The distributions are normalized to the no-timing case. For instance, in the HH bbγγ analysis with the CMS Phase-2 detector without MTD [11], the b-tagging selection corresponds to a 1% misidentification probability for the light quarks, and the photon selection to about 90% charged isolation efficiency. The MTD based reconstruction would provide, for the same background of misidentified b jets or photons, an increase in the b-tagging (photon) efficiency in the range 4 6% (5 7%), depending on the pseudorapidity. This gain in signal acceptance provides an increase in the HH bbγγ signal yield by 17% from the BTL alone, and 22% from the combined power of BTL and ETL, at constant background rate (see Fig. 1.3-left). Similar performance enhancements are predicted for other important Higgs boson signatures such as HH bbbb (14% BTL, 18% BTL+ETL) and H ZZ 4µ (19%, 26%). Figure 3.10 shows these increases as a function of the Higgs or di-higgs boson rapidity for these two processes. Both the BTL and ETL contribute significantly to the signal gain. These projections correspond to a 1% rejection power for false-positive b tags from light quarks and to a charged isolation efficiency of 90% per muon without the MTD, consistent with other CMS Phase-2 studies [36, 77]. The full optimization of the working points is left for future studies. A summary of these gains can be found in Table 3.1. Generally, across the Higgs boson campaign channels envisioned for the HL-LHC era, the improved acceptance is expected to yield a gain in the performance, expressed as signal over square-root of the background, between 10% and 20% depending on the decay mode. Without the MTD, a similar gain in performance would require an increase in the integrated luminosity at least proportional to the acceptance gains provided by the MTD. Below, more detailed studies are presented on two Higgs boson modes, for which specific additional benefits are expected from the improved vertex reconstruction capability (H γγ) and the improved pileup jet rejection and p miss T resolution (VBF Higgs boson production and subsequent decay to taus) with the MTD. 59

68 Table 3.1: Projections for Higgs boson yield increases when exploiting the timing information provided by the MTD and the resultant increased efficiency for tagged b jets or isolated leptons for a given fake probability. Projections are based on existing Phase-2 studies and scaled according to the ROC curves in Figs. 3.7(left) and 3.6 (left, center). Signal increase (%) Channel BTL BTL+ETL Relevance HH bbγγ Higgs self-coupling HH bbbb Higgs self-coupling H ZZ 4l Mass, width, spin+parity, differential cross sections, EFTs Higgs boson decay into photon pairs The clean diphoton signature of the H γγ decay channel makes it one of the most important channels to characterize the Higgs boson. The sensitivity of the measurement depends on the invariant mass resolution of the diphoton pair and on the quality of the photon identification. Specific upgrades of the CMS calorimeters have been defined for the Phase-2 and reconstruction algorithms are under study, targeting a performance in photon identification and energy resolution similar to Run 2 despite the increased pileup [11]. The invariant mass resolution depends on the energy and opening angle resolution of the two photons. If the longitudinal position of the diphoton vertex is known to better than about 10 mm, the opening angle resolution contributes negligibly to the diphoton mass resolution [78]. This measurement will benefit from the improved acceptance for isolated objects and improved vertex identification capability provided by track and photon timing information. At low vertex multiplicities, the decay vertex can be identified using the kinematic properties of the tracks associated with the reconstructed vertices and their correlation with the diphoton kinematics [78]. During LHC operations in , with an average multiplicity of 20 vertices, the probability of correct vertex identification was about 80% [79]. At high multiplicities, vertex identification is a major challenge. According to simulation, at 140 pileup events, the efficiency drops below 40% for H γγ events produced via gluon-gluon fusion, and it degrades to about 30% at 200 pileup events. This efficiency loss can be offset by means of a precise measurement of the photon time, which enables the vertex position along the beam direction to be determined via triangulation. The performance of this method, which depends on the opening angle between the two photons and scales with the time resolution, has been quantified in a dedicated study with simulated H γγ events for photons of p T > 30 GeV and η < 2.5, with selections that mimic those applied in Ref. [78]. Results presented in the Phase-2 upgrade of the CMS barrel calorimeters TDR [11] show that, for a resolution of 30 ps on each photon, the longitudinal vertex position can be located within 1 cm of the true vertex via triangulation only for a subset of the H γγ decays, in which the pseudorapidity gap between the two photons satisfies the condition η > 0.8. In the complementary, and equally populated, sample of photons ( η < 0.8), the resolution of the triangulation method becomes comparable to the size of the luminous region. For this sample, photon timing alone does not provide sufficient information to locate the H γγ decay vertex. The ability to correctly identify the vertex in events with a small pseudorapidity gap between the two photons is recovered by additionally requiring a triple coincidence between the photon time calculated at the location of each track-reconstructed vertex and the vertex time-zero, as visually illustrated in Fig A quantitative measure of the compatibility of the photon pair with the space-time position of 60

69 ) 2 pdf(χ CMS Phase-2 Simulation PU = 140 η γ γ < 0.8 H γ γ vertex Pileup vertex Vertex χ 2 R γ H γ Event fraction with R CMS Phase-2 Simulation PU = 140 η γ γ > 0.8 η γ γ < R (vertex rank by χ ) Figure 3.11: Left: Distribution of the χ 2 of H γγ (red histogram) and pileup vertices (blue histogram) for 30 ps resolution in the calorimeters, 20 ps resolution in vertex timing, and a η < 0.8 cut-off on the photon pair. Right: Fraction of events in which the diphoton vertex has a rank equal or lower than the rank in the horizontal axis, for 140 pileup events. each reconstructed vertex is obtained from a χ 2 statistics. A vertex time-zero resolution of 20 ps, from track timing, and a resolution of 30 ps on each photon are assumed. The χ 2 distributions for the true diphoton vertex, known from simulation, and for all the other vertices are shown in the left panel of Fig Results are displayed only for events with a small pseudorapidity gap between the two photons ( η < 0.8). The overlap of the distributions at low χ 2 indicates that there is a finite probability for a random pileup vertex to have a χ 2 lower than the true diphoton vertex. However, the distributions are sufficiently separated to enable vertex ranking according to the χ 2 value, and reject vertices of high rank. The rejection power is illustrated in the right panel of Fig. 3.11, both for the sample with large and small pseudorapidity gaps between the photons. The graphs show, for 140 pileup events, the fraction of events in which the diphoton vertex has a rank equal or lower than the rank in the horizontal axis. In 95% of the cases the diphoton vertex ranks amongst the first 10 ( for η > 0.8) or 20 (for η < 0.8) reconstructed vertices. Even in the least favourable case of diphotons with η < 0.8, this method provides a fivefold reduction of the effective multiplicity of collisions, for a marginal loss in the efficiency. The independent analysis of the vertex kinematic properties can thus be applied to a restricted list of vertices, comparable in size to (or lower than) at the present LHC, and is therefore expected to provide similar (or better) performance. The effect on the H γγ fiducial cross section measurement at HL-LHC is estimated in the context of projections of the most recent Run 2 results [79]. The impact on the invariant mass resolution and on the signal-to-background ratio is estimated for different CMS upgrade scenarios. At 140 pileup without timing information, the primary vertex selection efficiency is reduced to 40% ( S2+ no timing ). With precision timing available only for the photons from the calorimeters, the vertex is correctly identified for a fraction of the events via triangulation, corresponding to the sample with η > 0.8, and the total vertex selection efficiency is estimated to be about 55% ( S2+Calorimeter timing ). With precision timing available both for the photons in the calorimeters as well as for the charged particles in the event, a fivefold reduction in the effective pileup is assumed for all of the events, and the total primary vertex selection efficiency is increased to 75% ( S2+Calorimeter and MTD timing ), nearly fully recovering the present Run 2 performance ( S2 ). The signal lineshapes for the four scenarios are shown in Fig The improvement from precision timing with the full upgrade scope corresponds to 61

70 around 15% reduction in the statistical uncertainty of the fiducial cross section measurement. Although there are still significant systematic uncertainties on this measurement at 3000 fb 1, the statistical-only uncertainty can serve as a proxy for finely binned differential cross sections and/or differential cross sections measured as a ratio to the fiducial cross section, for which most systematic uncertainties will cancel. For such statistically limited measurements, the improvements from precision timing correspond to an approximately 30% increase in equivalent integrated luminosity. These projections do not include the acceptance gain from improved isolation with track timing, which are estimated to be of order 10% per photon, and would therefore increase by another 20% the equivalent integrated luminosity. Figure 3.12: Lineshape for the H γγ signal in each of the four considered scenarios. Since this is a combination over several analysis categories, the individual category contributions are weighted according to the signal to background ratio in order to be representative of their contribution to the final result VBF H ττ The VBF production channel of the Higgs boson, with subsequent decay via H ττ, is a key signature for the characterization of the Higgs boson. Already in Run 2 this channel is the most sensitive among the suite of H ττ analyses [80] and provides an important benchmark for performance capabilities of the future experiment. Further, VBF and vector boson scattering (VBS) topologies in general will play an important role in HL-LHC searches, given that many models of beyond Standard Model (BSM) physics predict an enhancement of the rate of such events, or adjustment of the kinematics of the decay products in such events, or both. As discussed in the CMS Upgrade Scope Document [2], for measurements of the VBF Higgs to ττ final state, where the ττ mass is reconstructed using the missing transverse momentum, the performance of the analysis expressed as signal over the square-root of background is degraded by about 15%, following a degradation of 15% in the resolution in the transition from 140 to 200 pileup. This effect alone translates into a 40% increase in the luminosity needed to achieve the equivalent result at 200 pileup. In addition the rate of jets reconstructed from pileup energy depositions which is observed to increase up to 30% in the endcaps with the Phase-2 CMS detector reduces the signal yield and increases the background from Drell Yan production, further degrading the analysis performance by 25%. Track-timing with MTD provides a reduction of the pileup jet rate by 40-50% in the endcap region (Fig. 3.4) and an improvement of 10 15% in p miss T resolution (Fig. 3.5), which offset entirely the performance 62

71 degradations observed in Ref. [2], for the Phase-2 detector without time information. The signal over the square-root of background at 200 pileup is estimated to increase by 10% and 30% for these two effects, respectively. In addition, precision timing will affect the overall signal efficiency through improvements to τ h isolation. The performance estimate of the impact of this effect exploits the extrapolation of the yields for signal and major backgrounds from the Run 2 version of the analysis, assuming that the di-tau invariant mass shape is consistent from Run 2 to the HL-LHC era. The predicted sensitivity of the no-timing analysis is obtained from scaling the Run 2 yields to 300 fb 1. This integrated luminosity was chosen such that figures-of-merit such as the expected upper limit on VBF Htt and significance of observation are relevant. The VBF H ττ signal will be established well before accumulating 3000 fb 1 in the HL-LHC era, hence a smaller integrated luminosity is used here for this study. For the with-timing analysis, a working point is selected from the ROC curve in Fig. 3.7 that maintains the same efficiency for authentic τ h candidates as in the no-timing case, and achieves a reduced fake probability. This new fake probability is used to scale the jet-borne fake τ h background from QCD in the extrapolation to 300 fb 1. The sensitivities of the two cases are compared in Table 3.2, where the expected upper limit on VBF H ττ is a factor of 25% better in the with-timing case than in the timing-absent scenario. The significance of observation is also presented, showing that the with-timing case is 20% better than in the timing-absent analysis. Similar performance enhancements are projected for the expected uncertainty on the signal strength modifier µ and the fermion coupling strength modifier κ f. These performance gains are factorized to those provided by the pileup jet mitigation and the improved p miss T resolution. Table 3.2: Comparison of expected 95% CL upper limits, significance of observations and corresponding p values for VBF H ττ in a data sample corresponding to 300 fb 1 of integrated luminosity in the HL-LHC era, for both the no-timing and with-timing scenarios. Yields for signal and background are scaled from Run 2 estimates. No MTD With MTD µ = σ/σ SM upper limit Significance of observation Uncertainty on µ, for SM signal-injected case (%) +26.2/ / 20.6 Uncertainty on κ f, for SM signal-injected case (%) +29.7/ / 34.3 A more complete assessment of the impact of the MTD on the analysis as a whole is in progress, in which the effects of pileup jet mitigation and improved p miss T resolution and the improvements on τ h isolation are incorporated in the same analysis Searches for new phenomena p miss T -based Search for Chargino-Neutralino Production One of the most sensitive channels for electroweak SUSY chargino neutralino production is via χ 0 2 χ± 1 WH χ0 1 χ0 1. For this channel, the most sensitive signature includes a charged lepton from W boson decay, two b quarks from the decay of the Higgs boson and significant missing transverse momentum, mostly attributable to the escaping neutralinos. This search is critically dependent on the p miss T resolution due to the use of a minimum transverse mass cut-off that is very effective in discriminating between SM backgrounds, which have a kinematic endpoint at the W boson mass, and the SUSY signal, which favours large p miss T and therefore large values of M T, as shown in Fig

72 (GeV) 0 1 χ m σ Discovery Reach fb Phase II, 140PU fb Phase II, 200PU 14 TeV, PU = 140/200 CMS Phase II Delphes Simulation fb Phase II, 140PU, No Tracker Extension χ ± χ W χ Hχ m ± = m 0 (GeV) Figure 3.13: Left: M T distribution for SM backgrounds and several χ 0 2 χ± 1 WH χ0 1 χ0 1 signal hypotheses. Right: Effect of the degradation due to pileup on the sensitivity for the SUSY search for the W ± H+ (from Ref. [2]). The kinematic endpoint becomes smeared out as the p miss T resolution degrades, hence the improvement from the additional charged track cleaning provided by the MTD, as discussed above, will have a significant benefit on the reach of this important search. In the CMS Upgrade Scope Document [2] it is shown that the effect of the degraded resolution on SUSY searches in the W ± H+ final state implies a significant reduction of the discovery potential. In a dataset of 3000 fb 1, the mass reach for discovery is reduced from 940 GeV at 140 pileup to about 800 GeV at 200 pileup, as shown in Fig (right). With precision timing, the resolution of the lower pileup operation is restored (see Fig. 3.5), the discovery potential recovered, and extended to beyond 1 TeV when the full integrated luminosity of the ultimate HL-LHC scenario (4000 fb 1 ) is exploited. The improvement provided by the MTD will also impact other p miss T based searches, including searches for WIMP-like dark matter production or stable massive dark sector particles Long-Lived Particle Searches Searches for long lived particles (LLPs) are theoretically well motivated. Many BSM extensions allow or require long lifetimes of particles due to higher-dimension operators, very small couplings, heavy mass scales, or suppressed phase space. More detailed discussion on LLPs can be found in Ref. [10] and references therein. The MTD provides new, powerful information in searches for long-lived particles. A precision MIP timing detector allows one to assign timing for each reconstructed vertex and to measure the time of flight of LLPs between primary and secondary vertices. Using the measured displacement between primary and secondary vertices in space and time, the velocity of an LLP in the laboratory frame, β LAB P (and γ P ), can be measured. In such scenarios, the LLP can decay to fully-visible or partially-invisible systems. Using the measured energy and momentum of the visible portion of the decay, one can calculate its energy in the LLP rest frame and reconstruct the mass of the LLP, assuming that the mass of the invisible system is known. The benefits of precision timing with the MTD on such LLP searches can be illustrated in three representative SUSY examples. The first example is a gauge-mediated SUSY breaking (GMSB) scenario where the χ 0 1 couples to the gravitino G via higher-dimension operators sensitive to the SUSY breaking scale; in such χ 1 χ 2 64

73 scenarios, the χ 0 1 may have a long lifetime [81]. It is produced in top-squark pair production with t t + χ 0 1, χ0 1 Z + G, and Z e + e. The decay diagram is shown in Fig (left). p t χ 0 1 t Z(e + e ) G electrons / neutralinos / 1.5 GeV cτ = 10.0 cm cτ = 3.0 cm cτ = 1.0 cm cτ = 0.3 cm p t χ 0 1 G Z(e + e ) t η m χ 0 (GeV) Figure 3.14: Diagram for top-squark pair production and decay (left), η distribution for electrons from the secondary vertex (center), and distribution of the mass of χ 0 1 (right) reconstructed from the final state kinematics for decays with M( t) = 1000 GeV and M( χ 0 1 ) = 700 GeV. The mass distributions are shown for various values of the cτ of the χ 0 1. The events were generated with Pythia8 [82]. The masses of the top-squark and neutralino were set to 1000 GeV and 700 GeV, respectively. Generator level quantities were smeared according to the expected experimental resolutions. A position resolution of 12 µm in each of the three spatial directions was assumed for the primary vertex [7]. The secondary vertex position for the e + e pair was reconstructed assuming 30 µm track resolution in the transverse direction. The momentum resolution for electrons was assumed to be 2%. And finally, the time resolution of charged tracks at the displaced vertex were assumed to be 30 ps. The mass of the LLP was reconstructed from the final state kinematics and the time meaurements as explained above, assuming that the gravitino is massless. The right panel of Fig shows the distribution of the reconstructed mass of the neutralino for various cτ values of the LLP. The fraction of events with separation between primary and secondary vertices exceeding 3σ in both space and time as a function of the MTD resolution is shown in Fig (left). The mass resolution, defined as half of the shortest mass interval that contains 68% of events with 3σ displacement is shown in Fig (right), as a function of the MTD resolution. The second example is a SUSY scenario where the two lightest neutralinos and light chargino are higgsino like. The light charginos and neutralinos are nearly mass degenerate [83] and may become long-lived as a consequence of the heavy higgsinos [84]. Neutralino-chargino χ 0 2 χ± 1 pairs in proton-proton collisions at s = 14 TeV were generated with Pythia8. The χ 0 2 and χ ± 1 were forced to decay into the χ0 1 (LSP) and a virtual Z boson or a W, respectively. The masses of the χ 0 2 and χ± 1 were set to 400 GeV. The mass of the χ0 1 was set to 390 GeV. The virtual Z was forced to decay into an e + e pair. The generator level quantities were smeared according to the expected experimental resolutions as described above. In the limit where the light charginos and neutralinos are degenerate in mass ( M = M( χ 0 2 ) M( χ 0 1 ) 0), the energy of the e+ e (visible) system in the LLP rest frame provides a direct measurement of the mass splitting. The left panel of Fig shows the distribution of reconstructed M for various cτ values of the LLP. The fraction of events with separation between the primary and secondary vertices exceeding 3σ in both space and time, as a function of the MTD resolution for this decay, is very similar to the one from the example discussed above. The mass resolution, defined as a half of the shortest 1 65

74 CMS Preliminary ~ ~ 0 t t t (bw) χ (Z [e e ] G ~ ) CMS Preliminary ~ ~ 0 t t t (bw) χ (Z [e e ] G ~ ) secondary vertex selection cτ = 10.0 cm cτ = 3.0 cm cτ = 1.0 cm cτ = 0.3 cm resolution (%) 0 1 χ 1 σ M cτ = 10.0 cm cτ = 3.0 cm cτ = 1.0 cm cτ = 0.3 cm MTD resolution σ t [ps] MTD resolution σ t [ps] 2 Figure 3.15: Efficiency (left) and mass resolution (right) as a function of the timing resolution of the MTD for reconstruction of the χ 0 1 mass in the SUSY GMSB example of χ0 1 G + e e +, with M( χ 0 1 ) = 700 GeV, considering events with a separation of primary and secondary vertices by more than 3σ in both space and time. electrons / M resolution (%) 1 σ CMS Preliminary 0 ± ± 0 χ χ Z* (e e ) χ W χ cτ = 10.0 cm cτ = 3.0 cm cτ = 1.0 cm cτ = 0.3 cm η MTD resolution σ t [ps] Figure 3.16: Mass difference between the χ 0 2 and χ0 1 states as reconstructed from the final state kinematics (left), η distribution of electrons from the secondary vertex (center), and mass resolution as a function of the timing resolution of the MTD for the reconstruction of the chargino neutralino mass splitting M (right), for various values of the cτ of the LLP. 2 66

75 mass interval that contains 68% of events with 3σ displacement, is shown in Fig (right), as a function of the MTD resolution. The two examples described above have an electron positron pair as the visible part of the LLP decay. The corresponding e ± pseudorapidity distribution is shown in the central panels of Figs and Most of the events have electrons in the central region, with η < 1.5, which emphasizes the need for the barrel portion of the MTD in the reconstruction of these signatures. A third such signature has been considered as well. In the GMSB benchmark scenario [85] used as the reference in this search, the lightest neutralino ( χ 0 1 ) is the next to lightest supersymmetric particle, can be long lived and decays to a photon and a gravitino ( G), which is the LSP. Figure 3.17 (left) shows a diagram of a possible gluino pair-production process that results in a diphoton final state. Figure 3.17: Left: Diagrams for a SUSY process that results in a diphoton final state through gluino production at the LHC. Right: Sensitivity to GMSB χ 0 1 G + γ signals expressed in terms of neutralino lifetimes for 300, 180 and 30 ps resolution, corresponding to the current detector, the Phase-2 detector with photon timing without MTD and with MTD, respectively. For a long lived neutralino, the photon from the χ 0 1 G + γ decay is produced at the χ 0 1 decay vertex, at some distance from the beam line, and reaches the detector at a later time than the prompt, relativistic particles produced at the interaction point. The time of arrival of the photon at the detector can be used to discriminate the signal from the background. The time of flight of the photon inside the detector is the sum of the time of flight of the neutralino before its decay and the time of flight of the photon itself, until it reaches the detector. Since the neutralino is a massive particle the latter is clearly negligible with respect to the former. In order to be sensitive to short neutralino lifetimes of order 1 cm, the performance of the measurement of the photon time of flight is a crucial ingredient of the analysis. Therefore, the excellent resolution of the MTD apparatus can be exploited to determine with high accuracy the time of flight of the neutralino, and similarly the photon, also in case of a short lifetime. An analysis has been performed at generator level in order to evaluate the sensitivity power of a search for displaced photons at CMS in the scenario where a 30 ps timing resolution is available from the MTD. The events were generated with Pythia8, exploring neutralino lifetimes (cτ) explored in the range cm. The values of the Λ scale parameter were considered in the range TeV, which is relevant for this model to be consistent with the observation of 67

76 a 125 GeV Higgs boson [86]. After requiring the neutralino decaying within the CMS ECAL acceptance and the photon energy being above a trigger like threshold, the generator-level photon time of flight was smeared according to the expected experimental resolutions. A cutoff at a photon time greater than 3σ of the time resolution is applied and the signal region is assumed to be background free. The signal efficiency of such a requirement is computed and translated, assuming the theoretical cross sections provided in Ref. [85], to an upper limit at 95% CL on the production cross section of the χ 0 1 G + γ process. Figure 3.17 (right) shows the analysis sensitivity in terms of the Λ scale (and therefore of the neutralino mass) and lifetime for three different assumptions on the timing resolution. The 300 ps resolution is representative of the time-of-flight resolution (TOF) for these events with current CMS detector performance [87]. The 180 ps resolution is representative of the TOF resolution of the upgraded CMS detector without the MTD, in which the TOF measurement will be dominated by the time spread of the luminous region. The vertex timing provided by the MTD detector will bring the TOF resolution to about 30 ps. As visible in the figure, a full scope upgrade of the CMS detector with photon and track timing will provide a dramatic increase in sensitivity at short lifetimes and high masses, already after the first 300 fb 1 of integrated luminosity. 3.4 Summary of performance benefits and next steps The addition of a precise timing detector, able to efficiently tag charged particles, significantly suppresses the effect of pileup on all object-level observables. This suppression yields significant and democratic improvements to many physics analyses by increasing signal efficiencies or reducing the width of residual distributions for discriminating variables. Performance benefits at 200 collisions per beam crossing are demonstrated over a wide spectrum of cases and expressly quantified in terms of improved track and vertex reconstruction performance, lepton efficiencies, diphoton vertex location, missing transverse momentum resolution, b jet identification performance, and in substantial reductions of the pileup jet rate. Studies with benchmark signatures demonstrate that these benefits make a significant impact on the physics program of CMS at the HL-LHC, and motivate that the MTD covers both the barrel and the endcap regions. For Higgs boson measurements, the gain in the efficiency for final state particles and in the track purity of the primary vertex translates into acceptance gains ranging between 20% and 60% depending on the process. The study of the Higgs boson production via vector boson fusion, with subsequent decay into τ pairs, is also significantly boosted by the improved resolution in the missing transverse momentum and the reduction of the pileup jet rate. Moreover, the MTD provides an entirely new handle in the search for beyond Standard Model physics, with unique sensitivity for long lived particles. All of these studies are performed examining the impact on the offline reconstruction. Further benefits are anticipated by including timing information in the high-level trigger, and improving the reconstruction online as well as offline. The online reconstruction can differ from the offline in terms of required reconstruction quality, but the ability to maintain low thresholds is critically important for the quality of all physics analysis. Finally, it may be possible for the MTD to play a role in the Level-1 trigger, where improvements via rate reduction are even more important, as discussed in Section 1.4. The impact of timing on the CMS triggers, as well as the technical implications for the MTD read-out electronics and for the Level-1 trigger latency, will be evaluated in the context of the MTD TDR. 68

77 Chapter 4 Project planning and cost estimates 4.1 Project planning The upgrade of the CMS experiment for the LHC Phase 2 with a MIP Timing Detector will imply the construction of a barrel and two endcap timing layers exploiting different technologies to achieve a timing resolution on charged tracks of about ps. The schedule for the work on the timing layers is given in Fig. 4.1, with several different phases separated by major decision points or deliverables. The R&D phase to demonstrate technologies and the detector concept is in full progress (Chapter 2) and is planned to continue throughout 2018, with the Technical Design Report (TDR) anticipated for the end of An engineering and prototyping phase, culminating in the delivery of the Engineering Design Reports (EDR), will follow. Given the different maturity of the respective technologies, the BTL EDR is anticipated for the end of 2019, while the ETL will have an engineering phase extended up to the end of The preproduction, production and installation phase will follow. The BTL will be mounted into a double walled Tracker Support Tube. The construction and installation of the BTL is planned to be completed early in the Tracker integration phase, currently scheduled for The ETL installation on the CE nose will follow the completion of the CE calorimeter in A global contingency of nine months exists on both the BTL and ETL plans. Figure 4.1: Preliminary construction timeline overview of the MIP Timing Detector Barrel timing layer The barrel timing layer will involve the development and construction of a modular structure, based on LYSO:Ce crystals read out by SiPMs, to be operated at 30 C. The steps planned to reach these goals are outlined in Table 4.1, identified with MTD.B.xx, with technically driven milestones dates. The effort includes the completion of the R&D phase with the definition of the detector concept in 2018, followed by the engineering of the modules and the definition of the assembly procedure, and by the production and installation phase. While we believe this schedule is maintainable, provided an early decision on the project, ways to gain contingency are under consideration. R&D phase: The R&D plan to complete the milestones for technology demonstration proceeds in parallel for sensors, read-out electronics and modules (MTD.B.01 08). 69