Mass production and characterization of 3-inch PMTs for the JUNO experiment
Chuanya Cao, Jilei Xu, Miao He, Angel Abusleme, Mathieu Bongrand, Clément Bordereau, Dominique Breton, Anatael Cabrera, Agustin Campeny, Cédric Cerna, Haoqiang Chen, Po-An Chen, Gérard Claverie, Selma Conforti Di Lorenzo, Christophe De La Taille, Frédéric Druillole, Amélie Fournier, Marco Grassi, Xiaofei Gu, Michael Haacke, Yang Han, Patrick Hellmuth, Yuekun Heng, Rafael Herrera, Yee Hsiung, Bei-Zhen Hu, Yongbo Huang, Cédric Huss, Ignacio Jeria, Xiaoping Jing, Cécile Jollet, Victor Lebrin, Frédéric Lefère, Hongwei Li, Nan Li, Hongbang Liu, Xiwen Liu, Bayarto Lubsandorzhiev, Sultim Lubsandorzhiev, Arslan Lukanov, Jihane Maalmi, Anselmo Meregaglia, Diana Navas-Nicolas, Juan Pedro Ochoa-Ricoux, Frédéric Perrot, Rebin Karaparambil Rajan, Abdel Rebii, Bed?ich Roskovec, Cayetano Santos, Mariangela Settimo, Andrey Sidorenkov, Igor Tkachev, Giancarlo Troni, Nikita Ushakov, Guillaume Van Royen, Benoit Viaud, Dmitriy Voronin, Pablo Walker, Chung-Hsiang Wang, Zhimin Wang, Diru Wu, Hangkun Xu, Meihang Xu, Chengfeng Yang, Jie Yang, Frédéric Yermia, Xuantong Zhang
MMass production and characterization of 3-inch PMTsfor the JUNO experiment
Chuanya Cao , Jilei Xu ∗ , Miao He † , Angel Abusleme , Mathieu Bongrand , Cl´ementBordereau , Dominique Breton , Anatael Cabrera , Agustin Campeny , C´edric Cerna ,Haoqiang Chen ‡ , Po-An Chen , G´erard Claverie , Selma Conforti Di Lorenzo ,Christophe De La Taille , Fr´ed´eric Druillole , Am´elie Fournier , Marco Grassi , XiaofeiGu , Michael Haacke , Yang Han , Patrick Hellmuth , Yuekun Heng , Rafael Herrera ,Yee Hsiung , Bei-Zhen Hu , Yongbo Huang , C´edric Huss , Ignacio Jeria , Xiaoping Jing ,C´ecile Jollet , Victor Lebrin , Fr´ed´eric Lef`ere , Hongwei Li , Nan Li § , Hongbang Liu ,Xiwen Liu , Bayarto Lubsandorzhiev , Sultim Lubsandorzhiev , Arslan Lukanov ,Jihane Maalmi , Anselmo Meregaglia , Diana Navas-Nicolas , Juan PedroOchoa-Ricoux , Fr´ed´eric Perrot , Rebin Karaparambil Rajan , Abdel Rebii , BedˇrichRoskovec , Cayetano Santos , Mariangela Settimo , Andrey Sidorenkov , Igor Tkachev ,Giancarlo Troni , Nikita Ushakov , Guillaume Van Royen , Benoit Viaud , DmitriyVoronin , Pablo Walker , Chung-Hsiang Wang , Zhimin Wang , Diru Wu , HangkunXu , Meihang Xu , Chengfeng Yang , Jie Yang , Fr´ed´eric Yermia , and Xuantong Zhang Institute of High Energy Physics, Beijing, China University of Chinese Academy of Sciences, Beijing 100049, China Pontificia Universidad Cat´olica de Chile, Santiago, Chile SUBATECH, Universit´e de Nantes, IMT Atlantique, CNRS-IN2P3, Nantes, France Univ. Bordeaux, CNRS, CENBG, UMR 5797, F-33170 Gradignan, France Department of Physics, National Taiwan University, Taipei IJCLab, Universit´e Paris-Saclay, CNRS/IN2P3, 91405 Orsay, France Astro-Particle Physics Laboratory, CNRS/CEA/Paris7/Observatoire de Paris, Paris,France School of Physics and Microelectronics, Zhengzhou University, Zhengzhou, China OMEGA, Ecole Polytechnique-CNRS/IN2P3, Paris, France Guangxi University, Nanning, China Hainan Zhanchuang Photonics Technology Co., Ltd, Chengmai, China Institute for Nuclear Research of the Russian Academy of Sciences, Moscow, Russia Department of Physics and Astronomy, University of California, Irvine, California92697, USA National United University, Miao-Li a r X i v : . [ phy s i c s . i n s - d e t ] F e b bstract ∼ The Jiangmen Underground Neutrino Observatory (JUNO) [1] is a multipurpose neutrino experi-ment under construction in southern China. Its main detector is located 53 km from two nuclearpower plants in a cavern with a 650 m overburden. The primary goal is to measure the neutrinomass ordering with a sensitivity better than 3 standard deviations after 6 years of data taking [2].High transparency liquid scintillator, high coverage (78%) of photomultiplier tubes (PMTs), andlow background levels are needed to achieve an energy resolution of 3%/ (cid:112) E (MeV) and an energycalibration error lower than 1%. The high coverage is achieved by closely packing ∼ ∼ ∗ [email protected] † [email protected] ‡ Now at China Electronics Technology Instruments Co., Ltd § Now at Huawei Technologies Co., Ltd ∼ XP72B20 was originally designed for KM3NeT with the curvature of the photocathode was deter-mined to be 52.4 mm [11]. The shape of the glass bulb was further optimized for both collectionefficiency and transit time spread (TTS) of photoelectrons (PEs) with simulation study in 2017 atthe Xi’an Institute of Optics and Precision Mechanics of the Chinese Academy of Sciences at therequest of JUNO. At a given voltage of 265 V which was calculated from gain 3 × between thephotocathode and the first dynode, the electric field distribution was simulated, and the maximumdifference of the transit time of PEs emitted at 6 positions with the polar angle from 0 ◦ to 50 ◦ wasfound to be 1.4 ns. A new glass bulb was then designed with a combination of two curvatures:54.9 mm and 42.6 mm, as shown in Fig. 1. The maximum transit time difference was reduced to0.5 ns.The simulation also indicated that the collection of the multiplied PEs between the first andthe second dynode played a significant role in reducing the TTS. The resistor ratio (high voltageratio) of the first 3 dynodes was originally set to 3:1:1 in an early study of JUNO [12]. In order toimprove the TTS, a dedicated study was done with different resistor ratios. A ratio of 3:2:1 wasfinally selected, which gave a 25% improvement of the TTS, from 5.0 ns to 3.7 ns in terms of fullwidth at half maximum (FWHM) for single PEs. Although the ratio 3:3:1 gave a slightly betterTTS, an additional ∼
50 V (4%) would be required to compensate for the decrease of the gain andthe single PE resolution was found to be reduced relatively 5%.As a low-background experiment, the radioactivity of each detector component of JUNO hasto be carefully controlled. The requirement on the radioactivity of the glass bulb for the smallPMT in JUNO is 400 ppb (4.94 Bq/kg), 400 ppb (1.63 Bq/kg) and 200 ppb (52.47 Bq/kg), for U, Th and K, respectively, based on an investigation of the glass manufacture [13] and thesimulation of the background event rate in the detector [2]. The major composition of the glassbulb is quartz sand and 3 different sand samples were obtained from the market and measured bya High Purity Germanium detector. The results are shown in Table 1. The normal sand has muchhigher
Th than the requirement.
U and
Th were reduced by a factor of 3 and 20 afteracid pickling, resulting in a small cost increase. The high-purity sand yielded another factor of 3reduction on
U and
Th, while K was found to be increased significantly probably due to thecontamination in the purification procedure. Taking into account the radioactivity and the price,3igure 1: Left: Engineering drawing of PMT XP72B22. Right: Typical electronic field simulation.The dimensions are given in millimeters (left) and the potential in Volts (right).Table 1: Different raw meterial radioactivities and glass bulb radioactivity requirementsRaw Material (Bq/kg) U Th KNormal quartz sand 2.95 ± ± ± ± ± ± ± ± ± < < < .To evaluate the radon contribution, 29 SPMT glass bulbs were placed into a 700 L large chamber instainless steel filled with nitrogen to accumulate radon till secular equilibrium was reached. Part ofthe gas was then pumped into an electrostatic radon detector to measure the alpha particles emittedby radon daughters, especially Po. An introduction to this facility can be found in Refs. [14, 15].This measurement gave an emanation rate of <
350 atoms of
Rn/day/m , corresponding to atotal contribution from the 25,600 SPMTs of < in the JUNO water pool, which isnegligible compared to the requirement.The production line of HZC was imported from PHOTONIS France in 2011 with a full pro-duction capacity of 250,000 tubes per year. The high degree of automation in both the production4ine and the performance testing largely ensures the stability of the product quality and reducesthe need for human labor and required skills. The quality management system is based on ISO9001:2005 standards. A dedicated production team was organized and quality control strategieswere applied for JUNO. For example, 6 additional steps were implemented for the componentinspection. Weekly meetings were organized to analyze product quality issues. In 2017, a pilotproduction of several hundreds of qualified PMTs was reviewed by JUNO. The quality of thesetubes was satisfying and thus the mass production was approved to start at the beginning of 2018.There was no major issue in the entire production period of two years, and the PMTs were suppliedto JUNO continuously every three months. The ratio of PMTs that passed the outgoing qualitycontrol before delivering to JUNO, defined as the good products yield, was below 50% in 2017, thenincreased to 77.5% in 2018 and 87.8% in 2019. The average yield was 80.5%, with the two majorsources of disqualification by HZC being low gain and high dark count rate. A further acceptancetest by JUNO was done based on the good PMTs, which will be introduced in Sec. 4. A waterproof seal will be applied to all 26,000 PMTs together with the HV divider and the cableby HZC. Therefore, an acceptance test by JUNO to ensure the quality of the PMTs was necessarybefore the sealing. Considering the large number of PMTs, as well as the fact that each of themhas 15 parameters (table 3) to be characterized, and in order to reduce the cost, manpower, andrisks associated with PMT transportation back and forth, JUNO adopted an onsite sampling teststrategy by sending a team to HZC roughly every three months during the production but usingthe test facilities and the manpower of HZC. This strategy also allowed to inspect the PMTs’performance at an early stage, ensuring good quality control of the production.As part of the incoming material inspection, the diameters of the glass bulbs were first measuredto ensure they fell into the (78, 82) mm range. The produced PMTs were measured in four main teststations, which were built or improved before the mass production started, and their performancewas reviewed and monitored through the production period.1. Static station: testing quantum efficiency (QE) and high voltage (HV) at a nominal gain(3 × ).2. Single photoelectron (SPE) station: testing SPE resolution, peak to valley (PV) ratio, darkcount rate (DCR).3. Transit time spread (TTS) station: testing TTS, pre-pulse, and after-pulse.4. Scanning station: testing QE non-uniformity and the effective diameter of the cathode.The first two stations were used by HZC as a standard procedure to test the basic parameters(QE, HV, SPE resolution, PV ratio, DCR) for all PMTs. Only tubes that were qualified duringthis procedure were given over to JUNO for further testing. All four stations were used by JUNOfor the sampling tests. 5 .1 Static station The static station (Fig. 2) was used to measure the quantum efficiency (QE) and the high voltage(HV) at nominal gain (3 × ). Experimentally, QE is defined as the ratio between the photoelec-trons produced by photocathode and then collected by the first dynode and the photons emittinginto photocathode. However, it is hard to measure the absolute incident photons precisely, so weused a standard PMT to be the reference. For the QE measurement, the light from a quartztungsten lamp passed through a 400 nm bandpass filter (BPF) and directly hit the cathode withan aperture diameter of 70 mm. The first-dynode current I k was read out and compared withthe current of a reference PMT I kc whose QE c was calibrated by a 10 mm ×
20 mm referencephotodiode S2744 [16] with the method of Ref. [17] with the relative uncertainty of reference PMTQE was estimated about 0.5%. The QE of the measured PMT was obtained from equation
PMTHV
I Aperture ABPF Light Source
Figure 2: Diagram of the static station to measure QE and HV. The system was in the darkroom.A light spot with 400 nm wavelength and a diameter of 70 mm was provided by a quartz tungstenlamp passed through a band pass filter (BPF) and an aperture. An optical attenuator (A) wasadded between BPF and light source when measured the anode current.QE = I k I kc QE c , (1)For the HV measurement at such a high gain, an optical attenuator (with attenuation factorA) was added to reduce the anode current I a into the range of the ampere meter, and the gain ( G )was extracted as G = I a I k A, (2)where I k was measured without attenuation. At nominal gain G nom , the corresponding nominalanode current I noma was calculated using Eq. (2) and the HV was tuned till I a was close to I noma .There were three light filters at HZC with wavelengths of 320 nm, 400 nm, and 550 nm, while theQE requirement by JUNO was defined at 420 nm. Therefore, the QE of five XP72B22 PMTs wasscanned from 300 nm to 700 nm by JUNO [10], and the average QE at 420 nm was found to be 6.8%lower than that at 400 nm. A correction factor 0.932 was thus applied to HZC’s result at 400 nmand delivered to JUNO. The other two filters were used for the spectral response measurement.6hree XP72B22 PMTs were measured every day to monitor the working stability of the stationduring the whole production. As shown in Fig. 3 (left), the QE measurements were stable over thefull production period. A few exceptional data points were attributed to the accidental measurementerror for a single monitor PMT. The cumulative statistics of QE over the production period is shownin Fig. 3 (right), and their average fluctuation of 0.2%, corresponding to a relative uncertainty 0.8%. Q E / % @ n m PMTID 80010PMTID 80061PMTID 80064 22 24 26 28 30 32QE / % @ 420 nm0255075100125150175 C o un t s PMTID 80010
Mean =26.8
RMS =0.2PMTID 80061
Mean =25.6
RMS =0.3PMTID 80064
Mean =26.2
RMS =0.2
Figure 3: Left: QE monitoring of three PMTs as a function of time. Right: QE distribution foreach of the three monitor PMTs.The HV monitoring data of the same three PMTs are shown in Fig. 4. There were largefluctuations up to ±
20 V before August 2018. An investigation of the test station suggested someinterference between the power supply and the signal readout since they were in the same crate.The power supply was then moved out and the grounding of the readout electronics was improved.As a result, fluctuations were reduced by a factor of three. The three monitor PMTs give similarresults, and the overall uncertainty of the HV measurement was estimated as 0.6%. H V / V @ g a i n × PMTID 80010PMTID 80061PMTID 80064 1000 1100 1200 1300 1400HV / V @ gain 3×10 C o un t s PMTID 80010
Mean =1113.9
RMS =6.7PMTID 80061
Mean =1191.5
RMS =7.4PMTID 80064
Mean =1049.4
RMS =7.1
Figure 4: Left: HV monitoring of three PMTs as a function of time. Right: HV distribution foreach of the three monitor PMTs.
The SPE station (Fig. 5) was used to measure the SPE spectrum-related parameters (SPE res-olution, PV ratio) and DCRs with 0.25 PE and 3.0 PE threshold, respectively. A LED with anappropriate driving voltage provided single 420 nm photons with a distance to PMT of about15 cm, which fully covered the PMT cathode. The PMT signal was amplified sequentially by twoamplifiers and then fed into a 512-channel multichannel analyzer to get the SPE spectrum. Posi-tions of the peak and valley, as well as the FWHM, could be extracted automatically. The SPE7esolution and the PV ratio could be calculated accordingly. The LED light was turned off whenmeasuring DCR. The DCRs were measured at two thresholds 0.25 PE and 3.0 PE, while the latterwas required specifically by JUNO, trying to identify PMTs with large spontaneous light emission.PMTs were kept in the dark box for at least 4 hours before measuring. PMT PA MCAAmp.HV
LED
Figure 5: Diagram of SPE station to measure SPE resolution, PV ratio, and DCR. PMT signal wasamplified by a preamplifier (PA, CANBERRA Model 2005), a main-amplifier (Amp., CANBERRAModel 2022), and fed into a multichannel analyzer (MCA). The PMT and the LED were in a darkbox, while the rest parts were in a room with weak light from the computers’ screen and someindicator lamps. The room temperature was controlled by air conditioner at 20 ◦ C.There was one PMT selected randomly from the early production to monitor the SPE resolutionmeasurement, as JUNO’s requirement. The monitor data of the SPE resolution is shown in Fig. 6.There was no time-dependent variation but only random fluctuations, showing good stability of theSPE measurement. The relative uncertainty (RMS / Mean) is about 4%. In the factory’s standardprocedure, another PMT was used just to monitor possible light leakage in the dark box. The DCRmonitoring data in Fig. 7 shows a slow decrease at 0.25 PE threshold in the first several monthsfollowed by a period of stability after the PMT was in operation for a longer time. The relativestandard deviation 33% was used to characterize the uncertainty of the DCR measurement. S P E r e s o l u t i o n () / % PMTID 70226 25.0 27.5 30.0 32.5 35.0 37.5 40.0 42.5 45.0SPE resolution ( ) / %020406080100 C o un t s PMTID 70226
Mean =32.9
RMS =1.3
Figure 6: Left: SPE resolution monitoring of the monitor PMT as a function of time. Each pointrepresents one measurement result on working day. Right: Distribution of SPE resolution for themonitor PMT. Defined as FWHM/peak/2.36 in this paper. Some PMT factories and papers use FWHM/peak. For physics study,the σ from Gaussian error is more widely used, which is about 2.36 times smaller than FWHM in mathematics. /18 9/18 1/19 6/19 9/19 12/19Date020040060080010001200 D a r k c o un t i n g r a t e / H z @ . P E PMTID 472 0 200 400 600 800 1000Dark counting rate / Hz @ 0.25 PE0510152025303540 C o un t s PMTID 472
Mean =320.0
RMS =105.5
Figure 7: Left: DCR monitoring of the monitor PMT as a function of time. Each point was themeasured result in each working day. Right: Distribution of DCR for the monitor PMT.
The TTS station shown in Fig. 8 is a replication of another setup of JUNO [10], which was able tomeasure not only TTS but also the pre-pulses and the after-pulses. A picosecond laser (405 nm)was used as a light source. The light was reflected and went through a shutter, then into a shortplastic optical fiber. There was a divergence angle when the light went out of the fiber into the airand hit the PMT cathode randomly in diameter of ∼ ∼ σ to express the TTS, which is equal to FWHM/2.36 for a Gaussian distribution. Forthe pre/after-pulse measurement, the average light level was ∼
100 PE. Integration of the waveformin the (-90, -10) ns, (-10, 15) ns and (0.05, 20) µ s windows with respect to the peak of the mainpulse gave the charge of the pre-pulse Q pre , main pulse Q main and after-pulse Q after , respectively.The ratio of the pre/after-pulse to the main pulse was calculated as Q pre / Q main and Q after / Q main . PMT ps-laser &controller HV shutter ReflectorOptical FiberCoupler Osc.
Figure 8: Diagram of the TTS station.There were two PMTs to monitor the long-time stability of the TTS system, as shown in Fig. 10.Intervals of approximately 3 months can be seen in the plots, corresponding to the onsite testing9 / ndf c – – – – - A m p lit ud e / m V / ndf c – – – – Figure 9: An example of the single PE waveform from one measured PMT, fitting with the Landaufunction added with a constant baseline. The typical amplitude is between 2 and 3 mV, with theelectronics noise smaller than 1 mV. P0 represents the most probable value which is used for timing,p1 and p2 the scale parameters, p3 the baseline.periods of JUNO. The TTS measurement was very stable with an uncertainty estimated as 10%based on the standard deviation of all data points. The after-pulse showed a slow decrease inparticular for PMT ID 75395, which is a suspect of a continuous ionization of the residual gasmolecules in the glass bulb. TT S () / n s PMTID 74406PMTID 75395 7/18 10/18 2/19 5/19 8/19 12/19Date0.02.55.07.510.012.515.017.520.0 A f t e r - p u l s e / % PMTID 74406PMTID 75395
Figure 10: TTS and after-pulse monitoring of two PMTs as a function of time. The data pointsare grouped reflecting the JUNO onsite testing periods.
The scanning station shown in Fig. 11 was required by JUNO to measure the non-uniformity of QEand the effective diameter of the photocathode. A quartz tungsten lamp served as a light source,provided a ∼ ×
100 mm squarewith 2 mm step size, and thus realized QE scanning in 2,500 pixels covering the photocathodeduring a testing process. An example of the scanning result of the anode current value is shown inFig. 12, which portrays the relative changes of QE along the PMT surface by showing the measuredanode current ( I a ). The two-dimensional projection of the photocathode is clearly demonstrated.10here is a ring with higher QE at the edge of the photocathode due to an effect of the glass bulbgeometry. The inner area ( φ
60 mm) was used to calculate the QE non-uniformity, expressedas the ratio of the standard deviation to the average. The effective photocathode diameter wasdetermined as an average of diameters determined along the main axes used in the scan. The edgesof the photocathode were set at the pixels, for which the anode current drops below 50% of theaverage of the inner area. T r a n s f e r b a r B P F Amp.HV
PMT S e r v o m o t o r A p er t u re Figure 11: Diagram of the scanning station.
X-pixel Y - p i x e l - - Position / mm A m C u rr e n t / Figure 12: Example of one photocathode scanning. Left: The anode current ( I a ) in µ A determinedfor each pixel. Right: The average anode current in the range 20 < Y pixels <
30 as a functionof X. The current near the edge of PMT is larger than the central area because of larger incidentangle of the light near the edge of the cathode ball and the reflection of the light at the inner sideof the lower hemisphere where has an aluminum coating.
To verify the test facilities at HZC, three parameters that are most important to JUNO wereinvestigated before the mass production: QE, HV, and SPE resolution at a gain of 3 × . FivePMTs were selected randomly and measured by an independent system at the Institute of HighEnergy Physics [10], and compared with the results by HZC. They were found to be consistentwithin the uncertainty as shown in Table 2. 11able 2: Comparison of QE, HV and SPE resolution measurements between JUNO and HZC usingthe average of 5 PMTs.Parameters QE HV SPE Res. ( σ )/ % @ 420 nm / V @ Gain 3 × / %JUNO 24.9 ± ± ± ± ± ± All 26,000 3-inch PMTs have been produced, and the 6 parameters from the static station and theSPE station measured by HZC for each PMT. Only PMTs with all of these parameters meetingthe requirements were delivered to JUNO. The measured parameters for those (called the vendordata) are shown in Fig. 13, where has a cutoff at 900 V and 1,300 V at the HV distribution (900,1,300) and < There were 7 parameters contained in class A: the diameter of the glass bulb, QE, HV, SPEresolution, PV ratio, DCRs at 0.25 PE, and 3.0 PE threshold. 150 PMTs were defined as a sub-batch since 75 PMTs were packaged in one box. 10% of them were randomly selected by the JUNOshifter. The diameter was examined first by two rings with inner diameters of 78 mm and 82 mm.After that, the sampled PMTs were delivered to the HZC worker to test at the static station andthe SPE station, and the results were sent back to the JUNO shifter. If any parameter was found12able 3: Summary of the 3-inch PMTs acceptance criteria and test results for different parameters.Results for class A parameters were from 26,000 PMT mean value of vendor data after acceptancemeasurement introduced in section 4.2, and other results were from acceptance measurement only.Unless specified, all of the parameters were measured at 3 × gain.Parameters Class Requirement Test fraction Tolerance Results(limit) (mean) HZC JUNO of diff. (mean)Φ (glass bulb) A (78, 82) mm - 100% 10% - OKQE@420 nm A > >
24% 100% 10% <
5% 24.9%High Voltage A (900,1300) V - 100% 10% <
3% 1113 VSPE resolution A < <
35% 100% 10% <
15% 33.2%PV ratio A > > < < <
30 Hz - 100% 10% - 7.2 HzTTS ( σ ) B < < < < <
10% - 3% - 3.9%QE non-uniformity B <
11% - - 3% - 5%Φ (eff. cathode) B >
74 mm - - 3% - 77.2 mmQE@320 nm C >
5% - - 1% - 10.2%QE@550 nm C >
5% - - 1% - 8.6%Aging D >
200 nA · years - - 3 PMTs - OKto exceed the limitation, this PMT was measured again. If the second test gave the same result,this PMT was rejected and replaced with a new one. Among all 2,600 PMTs selected for class Aparameter acceptance measurements, only 3 were rejected at this step, one with HV lower than900 V, one with DCR at 0.25 PE larger than 1.8 kHz, and one with DCR at 3.0 PE larger than30 Hz. The sampling test results are compared with the vendor data in Fig. 13, obtaining goodconsistency.For QE, HV, and SPE resolution, the difference between the sampling test results and the vendordata was required to be smaller than a tolerance, defined as 5%, 3%, and 15% for each single PMT,respectively, based on the 2-year stability results of the test stations reported in Sec. 3. Thesetolerances corresponded to 4-6 σ to allow the normal fluctuation to be accepted. Only exceptions,such as a sudden change of the test system performance, an unstable PMT, or a human mistake wasexpected to be caught. Once a big difference was found, a second test was done for the problematicPMT. If the second result was consistent with the vendor data, this PMT would be accepted. Ifthe two rounds of sampling test agreed with each other but were far from the vendor data, thisPMT would be also accepted but the vendor data would be changed to the new one. In the worstcase that all of these 3 tests were very different, this PMT would be rejected. Only one PMTwas rejected at this step because of unstable QE. The comparison of the first sampling test resultwith the vendor data for the same 2,600 PMTs is shown in Fig. 14, with the tolerances range13 C o un t s Vendor
Mean =24.9
RMS =0.8JUNO
Mean =24.9
RMS =0.8 900 1000 1100 1200 1300 1400HV / V @ gain 3×10 C o un t s Vendor
Mean =1113.1
RMS =88.4JUNO
Mean =1107.3
RMS =89.925.0 27.5 30.0 32.5 35.0 37.5 40.0 42.5SPE resolution ( ) / %050100150200250 C o un t s Vendor
Mean =33.2
RMS =2.3JUNO
Mean =33.2
RMS =2.2 2.0 2.5 3.0 3.5 4.0 4.5PV Ratio050100150200250300 C o un t s Vendor
Mean =3.2
RMS =0.4JUNO
Mean =3.2
RMS =0.40 250 500 750 1000 1250 1500 1750 2000Dark counting rate / Hz @ 0.25 PE050100150200250 C o un t s Vendor
Mean =512.2
RMS =433.7JUNO
Mean =488.3
RMS =400.2 0 5 10 15 20 25 30 35Dark counting rate / Hz @ 3 PE0100200300400 C o un t s Vendor
Mean =7.2
RMS =3.3JUNO
Mean =7.0
RMS =3.5
Figure 13: The PMT sampling test results (2,600 PMTs) for class A parameters and comparisonwith vendor data (26,000 PMTs) after normalization.depicted. The fractions of PMTs out of tolerances were 1.6%, 2.7%, and 2.4% for QE, HV, andSPE, respectively, and the majority agreed with the vendor data after re-testing.
22 23 24 25 26 27 28
Vendor QE / % @ 420 nm M ea s u r e d Q E / % @ n m
900 950 1000 1050 1100 1150 1200 1250 1300 · Vendor HV / V @ gain 3 · M ea s u r e d HV / V @ g a i n
25 30 35 40 45 ) / % s Vendor SPE resolution ( ) / % s M ea s u r e d SP E r e s o l u ti on ( Figure 14: A comparison of the sampling test result with vendor data for the same 2,600 PMTs,for QE, HV and SPE resolution, respectively. The red lines are the proportional function. Theblue dash lines are the maximum and minimum tolerance ranges.In JUNO, groups of sixteen 3-inch PMTs will be powered with one single HV channel. Therefore,14he working HV measurement was required to be more reliable than other parameters to ensurethat the gains of all PMTs in each group are as close as possible. Once a PMT was rejectedbecause of HV, we re-sampled another 15 PMTs (10%) in the same sub-batch and repeated theabove procedure. The full test procedure is shown in Fig. 15, using HV as the most complicatedexample.
10% PMTs sampled
Match vendor? RemeasuresuspiciousPMTs
Match vendor? Qualified?
Match the first test?Fix vendor dataPMTs accepted HV diff > 3%? PMTs rejected
Resampling another 10%
PMTs
NoYes No YesYesNo Yes NoNoYes
Figure 15: The flow chart of sampling test for HV parameter. The procedure was operated in eachsub-batch PMTs (150 PMTs).Because of the large fluctuation of the HV measurement before August 2018, as indicated inFig. 4, the first 10,000 PMTs produced and tested in that period were tested again in 2020 with anew system, whose principle was the same as the static station (Fig. 2), but with better control ofthe noise. One JUNO PMT was randomly selected and tested in this system every working day tomonitor the stability, as shown in Fig. 16. The variations were found to be within ± The ratio of PMTs tested for parameter classes B and C were 3% and 1%, respectively, bothsampled by the JUNO shifter from those 15 PMTs (10%) which were tested in class A acceptancetest. Similarly, the class C sample was fully contained in the class B sample, resulting in 1% of allPMTs being fully characterized. All of these parameters were required to be within the limits or asecond test was done. PMTs with two failed tests were rejected and a re-sampling of 3% or 1% inthis sub-batch was required.In the TTS station, TTS, pre-pulse, and after-pulse were tested, with the results shown inFig. 17. At the HV ratio 3:2:1. The TTS distribution was very stable. The average was 1.6 ns andthe relative deviation was 10.5%. No PMTs were rejected because of TTS or pre-pulse. However, 11PMTs were found to be unqualified due to the after-pulse being larger than 15%, which represented1.3% of all of the tested PMTs including those from re-sampling. Considering that the primaryfunction in JUNO of the 3-inch PMTs is single-photon detection, we concluded that this ratio wasacceptable. 15
Date H V / V @ g a i n × PMTID 80010
Figure 16: HV monitoring during re-testing of the first 10,000 PMTs. Each point shows themeasured result of a randomly selected PMT (ID 80010) in each working day.
TTS
Entries 710Mean 1.561Std Dev 0.1646 · ) / ns @ Gain 3 s TTS ( C oun t s TTS
Entries 710Mean 1.561Std Dev 0.1646
Pre-pulse
Entries 852Mean 0.4852Std Dev 0.3592
Pre-pulse ratio / % C oun t s Pre-pulse
Entries 852Mean 0.4852Std Dev 0.3592
After-pulse
Entries 852Mean 3.895Std Dev 3.282
After-pulse ratio / % C oun t s After-pulse
Entries 852Mean 3.895Std Dev 3.282
Figure 17: Distribution of TTS, pre-pulse and after-pulse from the sampling test. The number ofentries of the TTS plot is less than the other two because the first tens of PMTs were measuredat a HV ratio of 3:1:1 and those data were not used. In addition some statistics was added to thepre/after-pulse results due to the resampling after negative test results.In the scanning station, the non-uniformity of the QE and the effective photocathode diameterwere evaluated. The results are shown in Fig. 18. There were 7 PMTs with the QE non-uniformitylower than 2%. We did an investigation and found they belonged to one batch and were tested inthree consecutive days. The 2D scanning map indicated a little light leakage of the test box duringthat period, which caused a larger mean current value for all pixels and get lower non-uniformitypercent value. We concluded they were still qualified. The effective photocathode diameters of asmall fraction of PMTs were measured to be larger than 82 mm due to the 2 mm scanning steplength. No PMT was rejected at this step.To verify the range of the spectrum response, JUNO required the QE at 320 nm and 550 nmlarger than 5%. The measurement was done also in the static station but with different light filters.The results are shown in Fig. 19. All of the sampled PMTs met the requirement.16 on-uniformity
Entries 790Mean 5.048Std Dev 1.264
QE non-uniformity / % C oun t s Non-uniformity
Entries 790Mean 5.048Std Dev 1.264
Effective Dia.
Entries 790Mean 77.21Std Dev 1.148
74 75 76 77 78 79 80 81 82 83 84
Effective diameter of Cathode / mm C oun t s Effective Dia.
Entries 790Mean 77.21Std Dev 1.148
Figure 18: Distribution of QE non-uniformity and effective photocathode diameter from the sam-pling test. The value in QE non-uniformity plot is the relative percentage of QE by measured theanode current of PMT. Some PMT’s effective diameter was larger than 82 mm, because of theuncertainty caused by the 2 mm step length.
QE@320nm
Entries 234Mean 10.19Std Dev 2.482
QE @ 320 nm / % C oun t s QE@320nm
Entries 234Mean 10.19Std Dev 2.482
QE@550nm
Entries 234Mean 8.59Std Dev 1.02
QE @ 550 nm / % C oun t s QE@550nm
Entries 234Mean 8.59Std Dev 1.02
Figure 19: Distribution of QE at light wavelength 320 nm and 550 nm. These two results wereused to verify the spectrum response range.
The PMT gain was expected to decrease as the charge accumulates at the anode. Since JUNOwas designed to operate for 20 - 30 years, considering the normal light level in the JUNO detector,the gain decrease was required to be smaller than 50% with 6.1 coulombs (C) accumulated anodecharge, which was calculated from Q = R noise × e × G × T (3)Where Q is the charge; R noise is the PMT noise, set 2000 Hz here as the maximum noise; e iselectron charge, 1 . × − C; G is the PMT gain, set 3 × as the maximum gain JUNO usedin future; T is the time length of PMT working, 20 years.Before mass production, three PMTs were selected for the aging test and exposed to high-intensity light of 10 µ A for 8 days and then 100 µ A for another 8 days continuously, which equals76 C, about 10 times the JUNO requirement. Their gains were set to 3 × in the beginning,and in the end decreased by 8%, 20%, and 33% (Fig. 20), respectively, while the QE of each PMTessentially did not change. This meets greatly JUNO requirements.17
10 20 30 40 50 60 70 80Charge / C2.02.22.42.62.83.03.23.4 G a i n / PMTID 70275PMTID 70280PMTID 70284
Figure 20: Destructive test to evaluate PMT life time. High intensity light hit the PMT photo-cathode for several days to simulate the condition of PMTs in JUNO’s dark detector with mostlydark noise for 20 years.
During the PMT mass production, the radioactivity of the glass bulb was continuously monitored.The glass bulbs were produced roughly every three months as a batch, and a sample of each batchwas sent to JUNO for the radioactivity measurement. There were 7 batches in total and the resultsare shown in Fig. 21. The first two batches were received in the middle of 2018, and
Th wasfound to exceed the acceptance criteria by 50%-60%. Considering that the overall backgroundcontribution from 3-inch PMTs is very small, these two batches were still accepted. On the otherhand, an investigation of the glass bulb factory was done, where the production environment andthe procedures were carefully reviewed. In the end, the production was moved to another furnace,and a new stainless steel container was used for the mixing and storage of the raw material (quartzsand, borax, boric acid, aluminum hydroxide, and other minor components) to reduce the dustcontamination from the environment. The new sample from the following batch was received onemonth later and both
U and
Th were reduced by a factor of 2. After that, later batchesshowed good stability below the acceptance criteria in Table 1 for all of the three elements.18 R a d i o ac ti v it y / B q / kg U U Req. Th Th Req. K K Req. Figure 21: Radioactivity measurements for the glass bulb sample. The dash lines represent theacceptance upper limits. ∼ Acknowledgments
We thank the JUNO low-background working group for radioactivity measurements for the glassbulb. This work was supported by the National Natural Science Foundation of China No. 11975258and 12005044, the Strategic Priority Research Program of the Chinese Academy of Sciences, GrantNo. XDA10011200, the CAS Center for Excellence in Particle Physics, the Funds for Major Scienceand Technology Programs of Hainan Province (project number: ZDKJ2017011), the Hainan Scienceand Technology Department, the Special Fund of Science and Technology Innovation Strategy ofGuangdong Province, the MOST and MOE in Taiwan, the National Research and DevelopmentAgency of Chile, Institut National de Physique Nucl´eaire et de Physique de Particules (IN2P3) inFrance, and the University of California at Irvine.