Long term aging studies of the new PMTs used for the HL-LHC ATLAS hadron calorimeter upgrade

The central hadron calorimeter (TileCal) of the ATLAS experiment at the Large Hadron Collider (LHC) is a sampling calorimeter read out by about 10,000 photomultipliers (PMTs). Earlier studies of performance showed a degradation in PMTs response as a function of the integrated anode charge. At the end of the High-Luminosity LHC (HL-LHC) program, the expected integrated anode charge for PMTs reading out the most exposed cells is 600 C, and their projected response loss is 25%. These PMTs (≈ 8% of the total TileCal PMTs) will be replaced with a newer version PMT with the same geometry. A test set-up is being used in the Pisa laboratory to study the long term response of this new PMT model and compare it to the old PMT model currently installed in TileCal. For the first time this new PMT has been tested after integrating more than 800 C of anode charge. Preliminary results obtained from data collected in Pisa over a period exceeding two years are shown in this paper. We also introduce a preliminary study aimed to disentangle the contribution to the PMT response degradation due to the loss of quantum efficiency and to the change in absolute gain.


Introduction
The Tile Calorimeter (TileCal) [1] is the central section of the hadronic calorimeter of the ATLAS detector, covering the pseudo-rapidity () range || < 1.7.TileCal provides basic information for reconstruction of hadrons, jets, and missing transverse energy.TileCal is a sampling calorimeter made up of steel as absorber and plastic scintillating tiles as active medium.It is made of wedge shaped modules in the form of a central barrel section flanked by two extended barrel sections, surrounding the Liquid Argon barrel electromagnetic and endcap hadronic calorimeters, as shown in figure 1 (left).Each barrel is segmented into 64 modules in , corresponding to a 0.1 granularity in Δ.Each module is further segmented in the radial direction into three layers.The segmentation in ,  and radial direction defines the cell structure of TileCal (see figure 1 (right)).In total, there are 5182 cells in 256 modules.Charged particles produce light in the scintillating tiles, which is collected by wavelength shifting fibers from the two sides of each plastic tile and transported to the photomultiplier tubes (PMTs).Fibers are grouped in bundles to define the unit readout cell.Each cell is readout by two PMTs Hamamatsu model R7877 (specifically produced for TileCal, characterized by a 18×18 mm 2 square cathode active area, dynode fine mesh structure type).
PMT response stability is monitored in TileCal with a laser calibration system which is part of the TileCal global calibration system [2,3].PMT signal amplitude for laser pulses is recorded during normal operation with special calibration runs.PMT response variations were observed during collision and no-collision periods.As shown in figure 2, a response loss occurs during collisions, followed by a partial response recovery in long no-collision periods (LHC machine shut-downs).The aging effects are more pronounced for the inner, most exposed cell layer A, whose readout PMTs integrate a larger amount of anode current.
Significant response variation (with loss exceeding 20%) is expected to occur for PMTs having integrated hundreds of Coulomb of anode charge, with a large spread of the response variation for different PMTs [4, chapter 13, subsection 13.9.1, figures 13-23].At the end of the High-Luminosity LHC (HL-LHC) phase [5], which is planned to start in 2029, PMTs reading out the most exposed cells are expected to integrate 600 C of anode charge.
-1 - Based on the results from laser calibration data and from a dedicated test set-up, the ATLAS TileCal Collaboration decided to replace before the start of HL-LHC about 1,000 PMTs, the ones reading out the most exposed cells, out of the total 9852 TileCal PMTs, with the latest version of the same Hamamatsu PMT, labelled as R11187 [6].

JINST 19 C03051
A dedicated experimental set-up was used in Pisa INFN Laboratory to test a recent (2019) production sample of PMTs model R11187.The main goal was to study the average PMT response stability and the spread of the response variation as a function of the integrated anode charge, with the goal of reaching the target 600 C in about 2 years.A crucial point was to understand whether replacement PMTs will perform well until the end of the HL-LHC program.

JINST 19 C03051 2 Experimental set-up
A set-up for PMT long term aging test is operating since 2020 in Pisa INFN Laboratory with the aim to study evolution of the performance of the replacement PMTs as a function of the integrated anode charge and to test them up to about 1,000 C.
Figure 3 shows a scheme of the Pisa experimental set-up.A laser source model InnoLas MosquitooX with 100 μJ pulse energy at 50 kHz,  = 532 nm, and pulse width 12 ns provides the pulsed light in this arrangement.The laser light enters an optics box where the laser beam is split into different paths.In the main path, laser light enters an integrating sphere after passing through a filter wheel with 6 locations hosting different optical density neutral filters.Two secondary, strongly attenuated laser beams, generated up-stream and down-stream the filter wheel, are sent to two monitor PMTs (M1 and M2) used to check laser stability and correct for any change in beam intensity.A DC LED source with  = 535 nm is placed on the window of the top port of the integrating sphere.The integrating sphere acts as a mixer of the pulsed and DC light.Light exiting from the output port of the integrating sphere is then sent, via a bundle of clear fibers, to 24 PMTs located in a separated black box.All PMTs are equipped with active high voltage dividers to ensure the PMT linearity over a wide range of anode currents [6]. Figure 4 shows the optical box containing the integrating sphere (in the middle).On its left it is possible to see the clear fiber bundle used to feed the individual PMTs.The laser is located outside the box (not visible) and the beam line enters the box on its bottom left side.The DAQ and control system is based on commercial VME and NIM units for control signal generation and for digitization of analog signals from the PMTs under test and from the monitor PMTs.Laser triggering, LED on-off control, PMT High Voltage Supply control, filter wheel control and DAQ cycle are managed via a host computer.

Data taking procedure
The main purpose of the study described in this paper is to perform measurements of the PMT response variation in stable and controlled conditions.To do this, events are acquired by iterating ten identical cycles every day.In each cycle, a fixed amount of anode charge is integrated by exciting each PMT with DC light produced by the green LED.The charge integrated by each PMT is computed from the measured anode current times the (fixed) integration time.Subsequently, the PMT response to laser pulses is measured for different laser light intensities by acting on the filter wheel.Measurements are done with and without superimposing the DC light to the pulsed laser light.The sequence of each individual acquisition cycle is shown in figure 5.The LED voltage is adjusted to have, on average, PMT DC anode currents of about 10 μA.In these conditions, about 1 C per day is integrated by each PMT, with some exceptions due to the non uniform light collection efficiency of the clear fibers in the bundle.No difference was observed when measuring the PMT pulse amplitude with and without superimposed DC light.report our findings based on data collected until February 2023 (after more than 2 years of almost continuous data taking).Figure 6 shows the evolution of the PMT response as a function of time (left plot) and as a function of the anode charge integrated by each PMT (right plot) for three versions of the Hamamatsu PMTs.The red crosses correspond to the three old R7877 PMTs dismounted from TileCal.The blue ones to two special R11187 produced in 2010, and finally the black crosses correspond to three out of the 19 R11187 produced in 2019.In both plots the individual PMT response is normalized to the signal of monitor M1 and to the PMT response of the first day of the observation period.There is a clear separation between the old and new PMT models.New model PMTs are clearly better in terms of response stability.-5 - • a contribution from dark current fluctuations and QDC pedestal level (pulse baseline), independent from the pulse heigth:   (dark-current) = , where  is a constant.
By combining all previous terms, the following relationship between   () and  is obtained: with:  =    ×  × .The PMT gain  can be extracted from the  parameter in equation (5.2).As a first, preliminary attempt, the absolute gain evolution for one PMT model R7877 was extracted from the above equations, using data taken at five different laser light intensities.

Conclusion
Old model R7877 and new model R11187 PMTs for the ATLAS TileCal hadron calorimeter were tested in Pisa in a dedicated experimental set-up over a long operation period at high anode current (10 μA) up to about 900 C of integrated anode charge.The time evolution of the PMT response was monitored for more than 2 years and the PMT performance of new and old models were compared.New model PMTs are better in terms of response stability and are certified for being safely operated in the harsh HL-LHC conditions.Present results confirm a previous study performed after integrating a charge of about 500 C [8].This is the first time that the PMT response variation is monitored in a controlled and reproducible environment to such high values of integrated charge, much higher than what is expected at the end of the HL-LHC phase.Old model PMTs show a fast response loss below 100 C of integrated charge followed by a smoother variation with time above 100 C.However, at HL-LHC they will be operating safely in the less exposed cells, subject to lower amounts of integrated charge.
-8 -First attempts to separately measure PMT absolute gain and quantum efficiency variations were made.Very preliminary results indicate that the  is only slightly varying with charge integration, while the absolute gain variation seems to be consistent with the PMT global response variation.
Further studies are in progress, in which more PMTs will be included in the  and absolute gain calculation, and additional methods to estimate the PMT absolute gain are considered.
In general, all results indicate that using new model PMTs for reading out the most exposed cells and old model PMTs for the less exposed cells should allow safe operation of the ATLAS TileCal readout for the full HL-LHC data taking period.

Figure 3 .
Figure 3. Experimental layout of the Pisa set-up used for PMT long term aging tests.

Figure 4 .
Figure 4. Picture of the optics box set-up.The integrating sphere can be seen in the middle, with the bundle of clear fibers exiting from the left.

Figure 5 .
Figure 5. Data acquisition sequence for one of the ten daily cycles.The overall duration of one cycle is 135 minutes.

Figure 6 .
Figure 6.PMT response normalized to the laser intensity monitor and to the first day of acquisition, as a function of time (left) and integrated charge (right).The meaning of different colors is explained in the text.The vertical dashed line indicates the integrated charge expected at the end of HL-LHC for the PMTs reading out the most exposed cells.

Figure 7 .
Figure 7. PMT response normalized to the laser intensity monitor and to the first acquisition day, as a function of time (left) and integrated charge (right), for the 19 PMTs model R11187 (2019 production).The vertical dashed line is used just to outline the integrated charge reached by most of the new PMTs under test.
Figure 10 (left) shows the   () as a function of  distribution.The quadratic expression from equation (5.2) was used to fit the data points (superimposed red curve) and to extract the absolute gain value.The gain for this PMT was measured with data taken with the long term aging set-up in several days of a two months period in spring 2023.The PMT relative global response and absolute gain, normalized to the first day of the period, are shown in figure 10 (right).One can observe that the absolute gain variation is following the global response variation.

Figure 10 .
Figure 10.Left:   () as a function of  for one R7877 PMT.A fit obtained using equation (5.2) is superimposed.Right: for the same PMT, relative global response (red cross) and absolute gain (blue circle) normalized to the first day, during a total observation time of about two months.