Nuclear Inst. and Methods in Physics Research, A 1032 (2022) 166628 Contents lists available at ScienceDirect Nuclear Inst. and Methods in Physics Research, A journal homepage: www.elsevier.com/locate/nima A High-Granularity Timing Detector for the ATLAS Phase-II upgrade M.P. Casado, on behalf of the ATLAS HGTD Group 1 Departament de Física, Universitat Autònoma de Barcelona, Institut de Física d’Altes Energies and Barcelona Institute of Science and Technology, Spain ARTICLE INFO ABSTRACT Keywords: The increase of the particle flux at the HL-LHC with instantaneous luminosities up to 𝐿 = 7.5 × 1034 cm−2 HL-LHC s−1 will have a severe impact on the ATLAS detector reconstruction and trigger performance. The end-cap ATLAS and forward region where the liquid Argon calorimeter has coarser granularity and the inner tracker has HGTD poorer momentum resolution will be particularly affected. A High Granularity Timing Detector will be installed pp collisions in front of the liquid Argon end-cap calorimeters to help in charged-particle reconstruction and luminosity measurement. This low angle detector is introduced to augment the new all-silicon Inner Tracker in the pseudo-rapidity range from 2.4 to 4.0. Two silicon-sensor double-sided per end-cap will provide precision timing information for minimum-ionizing particles with a resolution as good as 30 ps per track in order to assign each particle to the correct vertex. Readout cells have a size of 1.3 mm × 1.3 mm, leading to a highly granular detector with 3.7 million channels. The Low Gain Avalanche Detectors technology has been chosen as sensor as it provides excellent timing performance. The requirements and overall specifications of the High Granularity Timing Detector are presented as well as the technical design and the project status. The on-going R&D effort carried out to study the sensors, the readout ASIC, and the other components, supported by laboratory and test beam results, are also presented. 1. Introduction from |𝜂| = 2.4 to |𝜂| = 4 by exploiting the track timing information (see Fig. 1) . More generally, this detector will improve the performance of The High-Luminosity LHC (HL-LHC) [1] is expected to start opera- the ATLAS detector in the forward region to a level comparable to that tion in 2027. It will achieve an instantaneous luminosity of 7.5 × 1034 for |𝜂| < 2.4 and provide a powerful tool for luminosity measurement. cm−2 s−1 and will collect an integrated luminosity of 4 000 fb−1 . With These proceedings briefly describe the HGTD geometry and running such beam conditions the average number of interactions per collision conditions in Section 2, discuss the Low Gain Avalanche Detectors will be typically 200, about one order of magnitude larger than during (LGAD) sensors and the detector in Section 3, ASICs in Section 4, Run 2. performance and a few physics cases in Section 5, and conclude in The ATLAS detector [2] will undergo a number of upgrades to Section 6. cope with the HL-LHC running environment. The Data Acquisition (DAQ) and Trigger systems will improve their granularity and will have faster electronics achieving an output rate of 10 kHz. The current 2. Detector inner tracker will be replaced by a fully silicon based detector (ITk), and its acceptance will be extended up to pseudo-rapidity equal to The HGTD, illustrated in Fig. 2, will be built in front of the end- 4. The calorimeters will be partially upgraded with more granular cap calorimeters cryostats of ATLAS, at a distance of ±3.5 m of the and faster electronics. In the end-cap region the electronics of the interaction point and will cover the pseudo-rapidity region from 2.4 to muon chambers will also be replaced to cope with the new conditions. 4. The space to build this detector is quite limited, its active radius goes Finally, a High Granularity Timing Detector (HGTD) [3] is planned for from 120 mm to 640 mm and its thickness should not exceed 125 mm the forward region of the ATLAS detector where the performance of including 50 mm of moderator to protect the ITk and the HGTD from the calorimeters and ITk is particularly degraded. HGTD will perform back-scattered neutrons. In order to ensure a timing resolution of 30 timing measurements of charged tracks and help in the suppression of ps per track over all the HL-LHC (which might be degraded to 50 ps at pile-up. the end of HL-LHC), two layers of active sensors will be built with an The High Granularity Detector will provide a time resolution of 30 additional effective layer at low radius. This additional effective layer ps for tracks in the forward region and will be able to reject pile-up will provide 2.6 (2.0) hits per track at small (large) radii. This increase E-mail address: casado@ifae.es. 1 Copyright 2021 CERN for the benefit of the ATLAS Collaboration. Reproduction of this article or parts of it is allowed as specified in the CC-BY-4.0 license. https://doi.org/10.1016/j.nima.2022.166628 Received 31 December 2021; Received in revised form 22 February 2022; Accepted 8 March 2022 Available online 24 March 2022 0168-9002/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). M.P. Casado Nuclear Inst. and Methods in Physics Research, A 1032 (2022) 166628 Fig. 1. Visualization of the truth interactions in a single bunch crossing in the 𝑧–𝑡 plane, showing the simulated Hard Scatter (HS) 𝑡𝑡̄ event interaction (red) with pile-up interactions superimposed (black) for ⟨𝜇⟩ = 200. Source: From Ref. [3]. Fig. 3. Maximum fluence the modules will have to face in their Lifetime as a function of the radial position in the detector. The three regions reflect the different replacement scenarios (every 1/fb for the inner ring, every 2/fb for the middle ring, while the outer ring will not be replaced). Source: Figure from Ref. [3]. Fig. 2. Illustration of the HGTD. The blue (green) region represents the active (off detector electronics) part of the detector. Source: From Ref. [4]. in the number of hits will be achieved by increasing the overlap of the modules for R < 230 mm from 20% to 70%. Due to the high particle rate in the HGTD region it is crucial for the detector to be radiation hard. At the end of the HL-LHC the maximum neutron-equivalent fluence at a radius of 120 mm will reach 2.1 × 1015 neq /cm2 (see Fig. 3) including a safety factor of 1.5. That fluence decreases exponentially as function of the radius. The radiation dose (see Fig. 4) will damage the electronics and the sensors, leading to a Fig. 4. Ionizing dose for HL-LHC with scale factor applied and considering decrease of the timing resolution as a function of integrated luminosity. replacements. To maintain a sufficiently precise timing resolution the sensors and the Source: From Ref. [3]. ASICs will be replaced every 1 000/fb for the inner ring (120 mm < 𝑅 < 230 mm), every 2 000/fb for the middle ring (230 mm < 𝑅 < 470 mm), while the outer ring will not need to be replaced. In this scenario, the is the key ingredient to get an excellent time resolution for energy maximum fluence that the sensors have to sustain is 2.5 × 1015 neq /cm2 deposited by Minimum Ionizing Particles (MIP). The expected currents (including a safe factor of 1.5). for different irradiation levels (therefore different gains) are presented in Fig. 5(b). For large gain, the rise time is about 500 ps and the signal 3. Sensors duration is approximately 1 ns. When the irradiation neutron fluence increases, the charge is smaller and the rise time and the signal duration The HGTD sensors will be made of LGADs [5]. LGADs are n-on-p are shorter. silicon detectors containing an extra highly-doped p-layer below the n- After amplification in the gain layer, the height of the LGAD signal is p junction to create a high field which causes internal amplification as displayed in Fig. 5(a). When a charged particle crosses the detector, an proportional to the gain. On the other hand, the slope dV/dt depends on initial current is created by the drift of the electrons and holes in the the thickness of the sensor, favouring thin sensors since the electronics silicon. When the electrons reach the amplification region, new elec- jitter scales as the inverse of the slope [3]. However, the jitter also tron/hole pairs are created and the holes drift towards the p+ region depends linearly on the detector pad capacitance, therefore limiting the and generate a large current. This charge amplification is referred to as potential use of very thin sensors. Consequently, the optimal thickness the gain of the LGAD. This current, larger than in a standard diode, relies strongly on the performance of the read-out ASIC. The baseline 2 M.P. Casado Nuclear Inst. and Methods in Physics Research, A 1032 (2022) 166628 Fig. 5. Cross section of an LGAD (a) and simulated signal current in LGADs at start and after full integrated neutron fluence (b). Source: From Ref. [3]. 4. ASICs The front-end electronics is constituted by ATLAS LGAD Time Read Out Chip (ALTIROC), an ASIC chip bump bonded to the LGAD sensor. It will have 15x15 channels of 1.3 × 1.3 mm2 , which will read the 225 corresponding pads of the LGAD sensor. The main challenge in the design of this ASIC is the fact that it needs to have a small contribution to the timing resolution, in order to match the excellent performance of the LGAD. A version of ALTIROC, the ALTIROC1v2 ASIC, with 5x5 channels, bump bonded to a LGAD sensor array (HPK 3.1) was exposed to electron beam tests at DESY in fall of 2019. The LGADs were operated with a bias voltage of 230 V, resulting in a MIP charge deposit of about 20 fC. Fig. 6. Time resolution as a function of the collected charge for neutron irradiated LGADs from different producers (HPK, FBK) with a 50 μm active thickness. These Fig. 7(a) shows the time-of-arrival (TOA) as a function of the time- measurements have been made at −30 ◦ C, in the laboratory with a 𝛽 source using over-threshold (TOT). The range of the TOT is truncated since it was not a custom electronics read-out board. The typical error bar is about 3 ps. The red line possible to measure large values of TOT because of a coupling between shows that in these conditions better than 70 ps time resolution is obtained at 4 fC. Source: Figure taken from Ref. [3]. the busy signal of the TOA TDC and the falling edge of the preamplifier output. This coupling only occurs when this signal is output on the PCB. This TOA busy signal must be output during test beam in order to active thickness has been chosen to be 50 μm, the total thickness is synchronize the data from ALTIROC and the oscilloscope used to record 300 μm and the pad size 1.3 × 1.3 mm2 . the waveforms of a Quartz+SIPM system used as timing reference. This Over the last five years LGAD sensors have been produced by signal will not be needed for the HGTD and is not used for test bench CNM/Spain, HPK/Japan, FBK/Italy and recently NDL and IME/China, measurements (it is only used for debug purposes). In the next iteration, with different doping levels, active thicknesses, pad sizes, and inter- ALTIROC1v3, the busy signal will be output as a differential signal to pad gaps. These detectors have been exposed to protons, neutrons and solve this problem. Therefore, with ALTIROC1v2, only a range of the X-rays up to the expected maximum radiation levels (including the TOT can be used in test beam and Fig. 7(a) also displays a fit in this safety factors) and intensively characterized in the laboratory (with restricted range used for the time walk correction. probe stations, 𝛽 sources and lasers) and in beam tests (at CERN, DESY, Fig. 7(b) shows the time difference between LGAD+ALTIROC hybrid Fermilab). and the reference time from the Quartz+SiPM system before and after Fig. 6 summarizes the time resolution obtained in the laboratory the time walk correction extracted from the fit in Fig. 7(a). The distri- measurements, with dedicated electronics, for sensors from different butions are Gaussian without any tails. The measured time resolution producers exposed to a neutron fluence up to 2.5 × 1015 neq cm−2 . A charge of 4 fC can be reached up to a fluence of 2.5 × 1015 neq cm−2 , decreases from 58.3 ± 1.6 ps to 46.3 ± 1.4 ps after time walk correction. providing a time resolution smaller than 70 ps per hit. The performance Subtracting the Landau contribution (about 25 ps), the remaining time of sensors from both manufacturers is similar, even if before irradiation resolution is about 39 ps containing contributions from the electronics the optimal operating bias voltage might be different because the dop- jitter, TDC and clocks. These are encouraging results, but it is important ing profile is different. With a minimal signal for hit detection by the to measure the performance of full sized devices (15 × 15 pads) before electronics of about 2 fC and 4fC for timing performance requirement, and after irradiation. The first full size ASIC prototype (ALTIROC2) has the expected hit efficiency is at least 95% [3]. been fabricated and is already being tested by the HGTD Group. 3 M.P. Casado Nuclear Inst. and Methods in Physics Research, A 1032 (2022) 166628 Fig. 7. (a): Distribution of the time-of-arrival (TOA) as a function of the time-over-threshold (TOT). The dots correspond to the mean value of the TOA distribution for a given TOT bin extracted from a Gaussian fit. The red line is a fit of the average TOA as a function of the TOT. (b): Distributions of the time difference between LGAD+ALTIROC and the Quartz+SiPM system before (red) and after (black) time walk correction together with Gaussian fits. The numbers are the fitted Gaussian widths where the time resolution of the Quartz+SiPM system has been subtracted quadratically. Source: Figures taken from Ref. [3]. 5. Performance The precise timing measurement of the HGTD can be used to greatly improve the reconstruction of different physics objects. This can be achieved by rejecting out of time pile-up that could contaminate objects thus reducing their reconstruction efficiency. In order to study the performance of the HGTD and optimize the detector layout, fully simulated events [6] of single muons, single electrons, single pions, jets, inelastic p-p interactions and several other physics processes have been produced. The pile-up rejection has been evaluated in different jet pT regimes and pseudorapidity intervals. It has been shown that HGTD can intro- duce an improvement of up to a factor 4 if a timing resolution per track of 30 ps is assumed [3]. Heavy-flavour tagging is another area which is critical in HL-LHC era. The b-tagging algorithms are particularly sensitive to pile-up con- tamination. The b-jets are associated with a displaced vertex so the algorithms accept tracks with a very large impact parameter (𝑧0 ) win- dow and thus increase the likeliness of using pile-up tracks. Using the Fig. 8. Normalized signal over background gain relative to ITk-only pile-up jet HGTD timing information a fraction of the pile-up tracks in this 𝑧0 suppression performance, as a function of the additional pile-up jet rejection from window is rejected and the identification of b-jets in the forward region HGTD. The solid black (dotted red) line represents the HGTD improvement from is improved by a factor 1.4 to 1.8 [3]. the Central-Forward (Forward–Forward) event topologies separately. The dotted blue The HGTD can be used to improve the identification of the leptons line shows the total improvement when the combined HGTD+ITk pile-up suppression algorithm is applied to all jets in the event. in the forward region. When working with leptons it is important to Source: Figure taken from Ref. [3]. know if they originate from the hard scatter event or if they are the result of a decay, a conversion or a fake. Ref. [3] shows how HGTD can improve the selection efficiency for electron isolation criteria by about 10%. linearity between the number of hits in the detector and the number HGTD can also help in physics analyses that can benefit from of interactions per bunch crossing is expected. improved jet and lepton reconstruction, as for example Standard Model With the HGTD a precise measurement of luminosity is possible by measurements of a VBS process and the weak mixing angle sin2 𝜃ef f , integrating all the hits in a period of 1s. A statistical precision of 0.14% and a search for VBF-produced Higgs bosons decaying invisibly. In for a run with an average number of simultaneous pp interactions, the measurement of sin2 𝜃ef f , HGTD reduces the total experimental ⟨𝜇⟩ = 1 and down to 0.01% for a run with ⟨𝜇⟩ = 200 is possible. Better uncertainty by 11% [3]. precision could also be achieved through a longer averaging time. In the search for VBF-produced Higgs bosons decaying invisibly HGTD can improve the S/B ratio, as shown in Fig. 8. This figure 6. Conclusion is normalized to ITk-only performance. For a particular pile-up jet efficiency gain from the application of HGTD relative to ITk, the dotted blue line shows the corresponding gain in S/B. For example, a 20% The HGTD will cover the forward region of the ATLAS detector in increase in forward pile-up rejection for jets above 50 GeV (without |𝜂| between 2.4 and 4.0, providing a timing resolution of 30 ps per any loss in hard-scatter efficiency), would correspond to a S/B gain of minimum-ionizing particle. This detector is proposed to be built using approximately 15% (corresponding to the 𝑦-axis when the horizontal LGAD sensors with a pixel size of 1.3 × 1.3 mm2 . The timing informa- value is 0.80). tion provided by this detector will be used to mitigate the effects of the In addition to improving the reconstruction of physics objects, the pile-up in the forward region, improving the reconstruction of physics HGTD can also be used as a luminosity measurement device. Since objects to a level similar to the one obtained in the central region. The this detector is highly granular and has a low occupancy a good HGTD will also provide luminosity measurements online and offline. 4 M.P. Casado Nuclear Inst. and Methods in Physics Research, A 1032 (2022) 166628 Declaration of competing interest [2] ATLAS Collaboration, The ATLAS experiment at the CERN large hadron collider, JINST 3 (2008) S08003, https://iopscience.iop.org/article/10.1088/1748-0221/3/ 08/S08003/pdf. The authors declare that they have no known competing finan- [3] ATLAS Collaboration, Technical Design Report: A High-Granularity Timing De- cial interests or personal relationships that could have appeared to tector for the ATLAS Phase-II Upgrade, CERN-LHCC-2020-007 ; ATLAS-TDR-031, influence the work reported in this paper. Geneva, 2021, https://cds.cern.ch/record/2719855/files/ATLAS-TDR-031.pdf. [4] C. Allaire, A High-Granularity Timing Detector (HGTD) in ATLAS: Performance at the HL-LHC, ATL-LARG-PROC-2018-003, CERN, Geneva, 2018, https://cds.cern. ch/record/2302827/files/ATL-LARG-PROC-2018-003.pdf. References [5] H.F.W. Sadrozinski, A. Seiden, N. Cartiglia, 4D tracking with ultra-fast silicon detectors, Rep. Progr. Phys. 81 (2) (2018) 026101, http://dx.doi.org/10.1088/ [1] High-Luminosity Large Hadron Collider (HL-LHC): Preliminary Design 1361-6633/aa94d3, arXiv:1704.08666. Report, CERN-2015-005, G. Apollinari and I. Béjar Alonso and O. Brüning [6] ATLAS Collaboration, The ATLAS simulation infrastructure, Eur. Phys. J. C and M. Lamont and L. Rossi, https://cds.cern.ch/record/2116337/files/CERN- 70 (2010) 823, http://dx.doi.org/10.1140/epjc/s10052-010-1429-9, arXiv:1005. 2015-005.pdf. 4568. 5