Performance of the diamond-based beam-loss monitor system of Belle II
S.Bacher, G.Bassi, L.Bosisio, G.Cautero, P.Cristaudo, M.Dorigo, A.Gabrielli, D.Giuressi, K.Hara, Y.Jin, C.La Licata, L.Lanceri, R. Manfredi, H.Nakayama, K.R.Nakamura, A.Natochii, A.Paladino, G.Rizzo, L.Vitale, H.Yin
PPerformance of the diamond-based beam-loss monitor system of Belle II
S. Bacher h , G. Bassi a,d , L. Bosisio b , G. Cautero c,b , P. Cristaudo b , M. Dorigo b , A. Gabrielli a,b , D. Giuressi c,b ,K. Hara g , Y. Jin b , C. La Licata a,b,e , L. Lanceri b , R. Manfredi a,b , H. Nakayama g , K. R. Nakamura g , A. Natochii j ,A. Paladino f , G. Rizzo f , L. Vitale a,b , H. Yin i a Dipartimento di Fisica, Universit`a di Trieste, I-34127 Trieste, Italy b INFN, Sezione di Trieste, I-34127 Trieste, Italy c Elettra Sincrotrone Trieste SCpA, AREA Science Park, I-34149 Trieste, Italy d now at: Scuola Normale Superiore, I-56126 Pisa, Italy e now at: Kavli Institute for the Physics and Mathematics of the Universe (WPI), University of Tokyo, Kashiwa 277-8583, Japan f INFN Sezione di Pisa and Dipartimento di Fisica, Universit`a di Pisa, I-56127 Pisa, Italy g High Energy Accelerator Research Organization (KEK), Tsukuba 305-0801, Japan h H. Niewodniczanski Institute of Nuclear Physics, Krakow 31-342, Poland i Institute of High Energy Physics, Vienna 1050, Austria j University of Hawaii, Honolulu, Hawaii 96822, USA
Abstract
We designed, constructed and have been operating a system based on single-crystal synthetic diamond sensors, tomonitor the beam losses at the interaction region of the SuperKEKB asymmetric-energy electron-positron collider.The system records the radiation dose-rates in positions close to the inner detectors of the Belle II experiment, andprotects both the detector and accelerator components against destructive beam losses, by participating in the beam-abort system. It also provides complementary information for the dedicated studies of beam-related backgrounds. Wedescribe the performance of the system during the commissioning of the accelerator and during the first physics datataking.
Keywords: sCVD diamond sensor, beam-loss monitoring, accelerator interlocks
1. Introduction
Beam-loss monitoring is an important componentof every particle-accelerator system. For the Su-perKEKB [1] electron-positron collider and the Belle IIexperiment [2, 3] this function is crucial, specially in theinteraction region, since the accelerator is aiming at anunprecedented design luminosity of 8 × cm − s − [1]with highly-focused intense beams. The detector has aprojected lifetime of at least a decade, to collect a verylarge sample of electron-positron annihilation events,and explore extensions of the Standard Model of fun-damental interactions.The inner Belle II detectors are silicon-based pixeland microstrip sensors with integrated front-end elec-tronics. The Belle II silicon technology is designed tosustain a dose of 10 to 20 Mrad, that is 0 . − . ff ersessential information to optimize the collider operationsand increase luminosity.To meet these challenges we designed, built, and havebeen operating a radiation monitoring system basedon single-crystal diamond sensors, grown by chemi-cal vapour deposition (sCVD sensors), and read-outby custom-designed electronics. This paper describesthe diamond system requirements, construction, and itsearly operations and performance. Preprint submitted to Nuclear Instruments and Methods A February 10, 2021 a r X i v : . [ phy s i c s . i n s - d e t ] F e b . Belle II interaction region and vertex detector The SuperKEKB layout is sketched in Figure 1 [4]with some features of the beam abort system. In-formation from hardware components such as the RF,vacuum, magnets systems and the beam loss moni-tor signals are collected at local control rooms (LCR);their abort requests, signalling abnormal accelerator andbeam conditions, are sent to the central control room(CCR) to trigger the abort kicker magnets.
Figure 1: The SuperKEKB beam loss monitor and abort system, fromreference [4].
The Belle II detector [2, 3] is a spectrometer sur-rounding the intersection of the 7 GeV “high-energy”electron ring (HER) and 4 GeV “low-energy” positronring (LER). The inner silicon-based vertex detector(VXD) has two layers of pixel detectors (PXD) directlymounted on the beam pipe, and four layers of microstripsilicon vertex detector (SVD) ladders supported by thecones and rings sketched in Figure 2. The superconduct-ing final-focus magnets (QCS) of SuperKEKB extendpartly inside the spectrometer, close to the beam-pipebellows of the intersecting rings.
3. The diamond-based beam-loss monitor system
The commissioning and data taking phases of Su-perKEKB and Belle II are summarized in Table 1.Four prototype diamond detectors and their readoutwere installed and operated during the initial acceler-ator commissioning Phase 1 in 2016, as part of theBEAST [5] project for initial beam-background studies;eight detectors were installed on the final beam pipe forthe Phase 2 commissioning run, with first collisions anda provisional inner detector installed, in 2018. The finalsystem consists of 28 diamond detectors, mounted on
Table 1: The three running phases of SuperKEKB and Belle II.
Phase / Year Collisions / Detector / / BEAST II [5]2 / / BEAST II & Belle II, no VXD3 / − Collisions / Full Belle II with VXD the beam pipe (BP) near the interaction point (IP), onthe SVD support cones, and on the beam-pipe bellows,close to the QCS magnets (Figures 2 and 3). We in-stalled it in 2018 and have been operating it since then,for the physics data taking (Phase 3) with the completeBelle II apparatus.
The KEKB peak luminosity was exceeded and a newworld record of 2 . × cm − s − was establishedby SuperKEKB during Phase 3 in June 2020. It wasachieved with a product of beam currents that was lessthan 25% that of KEKB, and by the initial implemen-tation of the “nano-beam scheme” and the “crab waistscheme” [6, 7]: the vertical height of the beams at the in-teraction point was squeezed to about 220 nanometers.To approach the design luminosity, almost a factor 40higher, the beam height at the IP will have to decrease toapproximately 50 nanometers, and the product of beamcurrents increase to four times that of KEKB [1].This luminosity enhancement is associated withharsher background conditions. The main beam-background sources are Touschek scattering, radiativeBhabha scattering, electron-positron pair production inphoton-photon scattering, and o ff -momentum particlesfrom beam-gas interactions [5]. The intensities of thesebackgrounds are strongly dependent on the beam op-tics. At the design luminosity, QED backgrounds willbecome relevant.We based the initial requirements of the monitoringsystem on background simulations and on the extrapo-lation of the experience of KEKB [8], the BaBar exper-iment at PEP-II (SLAC) [9, 10], CDF at the Tevatron(FermiLab) [11], ATLAS [12] and CMS [13] at LHC.To design the read-out electronics, we assumed a sen-sor current to dose-rate conversion factor in the range0 . − / s) / nA; after calibrations, our sCVD sen-sors turned out to have sensitivities larger by an order ofmagnitude, of about 35 (mrad / s) / nA on average.We expected to amplify and digitize diamond cur-rents from pA up to several mA, to be sensitive toboth very low beam losses with radiation levels below2 igure 2: Sketch of the location of the 28 diamond detectors on the beam pipe, the SVD support cones, and the beam-pipe bellows, as shown bythe on-line display of dose rates, available to Belle II and SuperKEKB operators.Figure 3: Photographs of diamond detectors (red dashed boxes) mounted on the beam pipe (top), on the backward SVD support cone (bottom left),and on the bellows close to the backward QCS (bottom right). / s, and large spiky losses due to noisy beam in-jections or machine operation accidents, of the order of10 krad / s or more.The time response, range, and precision of the asso-ciated read-out electronics were initially constrained asfollows. • The time scale relevant for beam abort is the 10 µ srevolution time of electrons and positrons in theSuperKEKB rings: the sampling and comparisonof signals with abort thresholds must match thecorresponding frequency of 100 kHz, or better. • Localised radiation damage in SVD microstrip de-tectors may occur when a radiation dose in ex-cess of O(1 rad) is delivered in a time interval ofO(1 ms) [10]. To produce “fast” abort requests, thedigitised diamond current is integrated (averaged)in moving sums, updated and compared with anabort threshold every 10 µ s. With an integrationtime of about 1 ms, a precision at the level of about10 nA is required on a 5 mA measurement rangefor such aborts. The integration time and thresh-olds must be programmable, to satisfy evolving re-quirements. • Finally a recording rate of 10 Hz is chosen for doserate monitoring, archiving and integration. Thesensitivity obtained by averaging the dose rates onthese time scales is also available for “slow” abortrequests, that could be delivered to the acceleratorby the slow control software rather than by the di-amond electronics directly.
Diamond sensors, with a bias voltage applied on twoopposite electrodes, operate functionally as solid-stateionisation chambers, delivering a current induced by thedrift of electrons and holes produced by ionisation in thediamond bulk. Their compact size and radiation hard-ness are well-suited for radiation monitoring in the in-teraction region; the temperature-independent responseallows operation without additional temperature moni-toring and related corrections.The assembly, test, and calibration of the 28 sCVDdiamond detectors are detailed in ref. [14].The electronic grade sCVD sensors were grown byElement Six [15] in the standard size 4 . × . × . ;two (Ti + Pt + Au), (100 + + Figure 4: Diamond sensor packaged into a detector unit. The sensoris glued on a Rogers printed-circuit board; electrical contact is estab-lished between the two electrodes (front / back) and the inner conduc-tors of two miniature coaxial cables; outer shielding is also provided. cover. The inner conductors of two thin coaxial cableswere soldered to two pads, connected to the sensor elec-trodes by conductive glue (back side) and gold wire ballbonding respectively. The outer conductors of the ca-bles were connected together and with the external de-tector shield, as shown in Figure 4.Each detector was characterized by measuring [14] • the current I as a function of the bias voltage V , inthe dark and in the presence of ionization by a β radioactive source; • the stability of the current signal in time; • the transport properties of charge carriers, elec-trons and holes, using a 5 kBq Am source ofmonochromatic α -particles. • a current to dose-rate calibration factor k , using analmost point like 3 MBq Sr radioactive β sourcelocated at varying distance.The dose rate, expressed in mrad / s, is proportional tothe measured current I m (nA), dDdt = m dEdt = m I m G E eh q e = FG I m = kI m (1) F = E eh mq e , (2)where m is the sensor mass, E eh the average ioniza-tion energy per electron-hole pair, q e the elementarycharge. The unit conversion factor F is computed as F = . / s) / nA. A bias voltage of 100 V is cho-sen for full charge-collection e ffi ciency. For a typicalmeasured current of 1 nA, the dose rate is 34 . / sif G = G describe possible devia-tions of each diamond sensor from the ideal behaviourin stationary conditions: 100% e ffi ciency in the collec-tion of charge carriers from the assumed active volume,4ero net trapping-detrapping rate, and absence of chargeinjection from the electrodes. They are obtained fromthe ratios of the measured current to the correspondingenergy deposit per unit time and sensor mass, estimatedby a detailed simulation of the source and of the set-up.Similar measurements and simulations on a referencesilicon diode are used to account for the source activityand to reduce the uncertainties of the set-up model. Themeasured G values cluster around G (cid:39)
1, with maxi-mum variations of about 50%.The uncertainty in the determination of the calibra-tion factors k is dominated by systematic e ff ects; we es-timate an accuracy of 8% [14]. The linear behaviour(equation 1) is confirmed by measurements performedon a sub-sample of diamond detectors with a Co γ source at dose rates and currents larger by two orders ofmagnitude. The 28 diamond detectors are controlled by seven di-amond control units (DCU). Each DCU pilots four di-amond detectors, as sketched in the block diagram ofFigure 5 .
Figure 5: Diamond control unit (DCU) block diagram. Arrows on theleft side indicate currents from diamond sensors and arrows on thedown side indicate high voltage applied on the diamond sensors.
The digital core is a board hosting a Cyclone V FPGAby Intel [18], which receives commands via an Ether-net interface, drives four HV modules independentlythrough a DAC, and accepts input data from an ana-log module with amplifiers and ADC conversion, assketched in Figure 5.The DCU is also able to deliver VXD abort requestsseparately for the electron higher-energy ring (HER)and the positron lower-energy ring (LER), and receivesthe SuperKEKB abort signals. The diamond currents are amplified by trans-impedance amplifiers, digitised by a 16-bit ADC [19] at50 Msamples / s, and processed by FPGAs in the DCUs,as shown in the simplified block diagram of Figure 6. Figure 6: Diamond control unit (DCU): simplified block diagram ofthe FPGA firmware.
Three amplifier gain values can be selected by resis-tors in the feedback loop of the front-end operationalamplifier [20], to provide three di ff erent current mea-surement ranges, indexed as 0, 1, and 2 in the follow-ing. The analogue bandwidth at the lowest gain (range2) is about 10 MHz, matched to large and fast signals;it is reduced to the order of 10 kHz at the highest gain(range 0) used for monitoring smaller signals at 10 Hz.The oversampling 16-bit ADC followed by digitalintegration performs similarly to slower ADCs with alarger number of bits. At the design stage, this solutionwas preferred for its flexibility in choosing the level ofdigital integration.The sum of 125 samples is obtained every 2.5 µ s, at400 kHz. These sums, called in the following “400 kHzdata”, are written in a 4 Gbit DDR circular bu ff er mem-ory. These numbers refer to an upgraded version of thefirmware, in use since January 2020; in the previous ver-sion, sums of 500 samples were obtained at 100 kHz,every 10 µ s.Two “moving sums” of “400 kHz data” are updatedfor each diamond sensor at each memory cycle, by sub-tracting the oldest added value of “400 kHz data” andadding the newest one. Corresponding to integrateddoses over programmable time intervals, they are com-pared with programmable thresholds, initially intendedto provide abort requests on two di ff erent time scales:(1) “fast” or “acute” aborts for sudden large spikes inbeam losses, typically in the millisecond time range,5 able 2: Options for current-measurement ranges of diamond controlunits, with the corresponding rms noise values, measured in 100 kHzdata (third column) and in 10 Hz data (fourth column). Range Current Rms noise Rms noiseindex range @100 kHz @10 Hz0 36 nA 0 .
23 nA 0 . µ A 3 nA 70 pA2 4 . . µ A 40 nAand (2) “slow” or “chronic” aborts for moderate, but in-creasing beam losses. Operations experience showedthat abort requests are delivered mainly on quickly ris-ing beam losses. Integration-time windows were there-fore shortened, producing “very fast” and “fast” aborts,as described in Section 5.If a moving sum exceeds the corresponding pro-grammed dose threshold, a logical signal is generated;in total, eight signals are available per DCU. Individ-ual masks can be applied to exclude noisy channels,if needed. “Abort request” signals are generated sep-arately for LER and HER then sent to SuperKEKB,when a programmed minimum number of unmaskedsignals above threshold is reached. After activating thebeam abort kicker magnets, the accelerator control sys-tem broadcasts “SuperKEKB Abort” signals, which areinput to DCUs.Incoming “SuperKEKB LER Abort” and “Su-perKEKB HER Abort” signals from SuperKEKB stopthe memory writing. The 400 kHz data can then be readout and written to a file for “post-abort” analysis of thebeam losses preceding the abort.Data at 400 kHz are further added up in groups of40000, to provide sums of 5000000 ADC values thatcan be read out at 10 Hz (“10 Hz data”).Prototypes were extensively characterized duringcommissioning: Table 2 shows the measured charac-teristics of the three options for current-measurementranges, selected by changing the amplifier gain, as im-plemented in the final production version of the DCUs.The range 0 (36 nA) allows precise monitoring of rel-atively small beam losses, while the range 2 (4 . The slow control handles the DCU configuration andthe monitoring data in EPICS 3.14 [21] code. Processvariables (PV) are employed to regularly query 10 Hzdata. After subtraction of pedestals, which are ADCvalues averaged over a prior no-beam time, and mul-tiplication with calibration factors, values of dose-ratesare obtained. All dose rate PVs are archived.A finite-state machine controls the various phases,with transitions between configurations, 10 Hz datareadout, and bu ff er memory readout for post-abort anal-ysis. This last state transition is triggered by the ar-rival of “SuperKEKB Abort” signals, which stop theadvancement of the bu ff er memory pointers. After com-pletion of bu ff er memory readout, the system resumesthe 10 Hz data readout.Time-dependent dose rates are displayed in both theBelle II and the SuperKEKB main control rooms (Fig-ure 2). Also the pre-abort history of dose rates with2.5 µ s time resolution is available to operators for post-abort analysis of the beam conditions, as discussed inSection 5. Individual diamond detectors are labelled asa b c according to their position; a = BP, SVD, QCSdescribes the location on beam pipe (BP), SVD supportcones, or QCS bellows; b = FW, BW denotes the for-ward or backward position with respect to the interac-tion point, in the electron beam direction; c representsthe azimuthal angle in degrees, in the standard Belle IIlaboratory reference frame.
4. Dose rate measurements
The dose rates from the 28 diamond detectors, mea-sured and archived at 10 Hz, provide on-line monitor-ing of beam losses and accelerator-related backgrounds,and are correlated with the data of Belle II inner detec-tors. We describe here the main features and results ofthese measurements. O ff sets (pedestals) and noise of each readout channelare obtained by repeated measurements without circu-lating beams. A few minutes are su ffi cient to preciselydetermine the central value and noise value for the cho-sen current-measurement range, from histograms simi-lar to the examples in Figure 7. Table 3 gives a summaryof measured noise for a typical diamond sensor calibra-tion factor k =
30 (mrad / s) / nA.The measured noise includes a random componentthat scales with the data averages from 100 kHz to 10Hz, and common-mode components, picked up from6 − − − C oun t s / . m r ad / s BP_BW_215
Range 0, 100 kHz = 8.6 mrad/s σ − − − − − × Dose rate [mrad/s]050010001500200025003000 C oun t s / m r ad / s BP_BW_35
Range 2, 100 kHz = 6.4 rad/s σ r ad / s µ C oun t s / . BP_BW_215
Range 0, 10 Hz = 0.038 mrad/s σ − × Dose rate [mrad/s]050100150200250300350400 C oun t s / m r ad / s BP_BW_35
Range 2, 10 Hz = 1.4 rad/s σ Figure 7: Examples of noise measurements performed without circulating beams for diamond detectors installed on beam pipe, expressed in mrad / s:(top left) range 0, 100 kHz; (top right) range 2, 100 kHz; (bottom left) range 0, 10 Hz; (bottom right) range 2, 10 Hz. Entries in the top plots areread from the 100 kHz bu ff er memory, spanning 1 s (100000 entries), pedestal-subtracted and converted from ADC to dose-rate units; the slightasymmetry in these distributions is due to an intrinsic cuto ff of the largest negative fluctuations in the almost-unipolar front-end range. Entries inthe bottom plots are 10 Hz data collected in 17 minutes ( ∼ Time − − − D o s e r a t e [ m r ad / s ] Pedestal over 3 daysBP_FW_35BP_BW_145BP_FW_145BP_BW_35
DCU1 (Range 2)
Time − − − − − − D o s e r a t e [ m r ad / s ] Pedestal over 3 daysQCS_BW_45QCS_BW_135QCS_FW_45QCS_FW_135
DCU5 (Range 0)
Figure 8: Example of time evolution of pedestals and noise (mrad / s)over three days without beams. The central values represent pedestals(averages of 10 Hz data) evaluated in short time intervals of 10 min-utes every three hours; error bars show the standard deviations ofthe corresponding approximately Gaussian distributions, representa-tive of noise, for measurements performed in range 2 (top) and inrange 0 (bottom). Table 3: Typical measured rms-noise values expressed in dose rateunits for the three measurement ranges; the approximate noise for100 kHz data in range 2 is relevant for the “very fast abort” thresh-old with 10 µ s integration time (Section 5.2). Range Dose rate Rms noise Rms noiseindex range @100 kHz @ 10 Hz0 1 . / s 6 . / s 0 .
024 mrad / s1 270 rad / s 90 mrad / s 2 . / s2 140 krad / s 7 rad / s 1 . / sthe environment, which are only partially averaged out.Common-mode noise depends on the selected rangeand shows a pattern correlated with the detector ca-bling: substantial attenuation is obtained by placing fer-rite bead filters on the DCU input signal cables.The measured average values (pedestals) are mostlystable (Figure 8), drifting typically less than one ortwo times the standard deviations of beam-o ff pedestalsduring the time interval between programmed acceler-ator beam-o ff periods: a weekly or bi-weekly updateof pedestals is su ffi cient for most purposes, includingthe pedestal subtraction for monitoring of instantaneous7ose rates. For o ff -line computations requiring a moreaccurate subtraction, we use the data archived duringshort accidental beam-o ff periods to determine smallerpedestal shifts more frequently. Diamond current signals, converted to dose rate unitsafter pedestal subtraction, are continuously monitoredin all accelerator conditions. They are particularly rele-vant as indicators of beam quality during injection andcollimator tuning, and for beam optics studies.
The integrated radiation doses are approximatelycomputed on-line, using the 10 Hz monitoring data, andmade available on a daily basis. Small pedestal shiftsare then determined (Section 4.1) and used o ff -line for amore precise estimate of the integrated dose, as reportedin Figure 9 for the most exposed diamond detectors, lo-cated on the beam pipe (BP) and the QCS bellows. Thediamond detectors mounted on SVD cones received ra-diation doses lower by almost an order of magnitude,concentrated mostly in the horizontal plane of the circu-lating beams.The most exposed diamond detector QCS FW 225received an estimated total dose of about 960 krad inPhase 3 until July 2020.The pre-amplifier of diamond detector BP FW 325was damaged during a beam incident with very largelosses in May 2020, as shown by the dose for that chan-nel in Figure 9(a), no longer increasing after the inci-dent. The pre-amplifier was repaired and the diamondread-out channel recovered for the next run period. During Phase 2 several studies were dedicated to in-vestigate the origin of beam-related backgrounds, bycorrelating detector outputs with the accelerator condi-tions. Diamond detectors contributed to these studies,as shown in the following examples [22].The left panels in Figure 10 show the dose rates, mea-sured by the diamond detector on the beam pipe show-ing the highest signal, as functions of the circulatingbeam current, separately for LER and HER, at di ff erentdates.Single beam backgrounds, shown here, are dom-inated by two types of contributions. Beam lossesdue to beam interactions with the residual gas, the socalled beam-gas backgrounds, scale with the product ofthe beam current and the residual gas pressure. Tou-schek background, due to intra-bunch scattering, scalesquadratically with the beam current. (a) Beam pipe(b) QCS backward(c) QCS forwardFigure 9: Integrated dose during Phase 3 (March 13, 2019 to July 1,2020), in diamond detectors located on the beam pipe (a) and close tothe final-focusing superconducting magnets in the backward (b) andforward (c) regions.
50 100 150 200 250Current [mA]00.511.522.53 D o s e r a t e [ m r ad / s ]
19 April24 April01 May
Dose rate vs LER current (comparison) − × P r e ss u r e [ P a ]
19 April24 April01 May
Pressure vs LER current (comparison) D o s e r a t e [ m r ad / s ]
24 April01 May
Dose rate vs HER current (comparison) − × P r e ss u r e [ P a ]
24 April01 May
Pressure vs HER current (comparison)
Figure 10: (left) Dose rate of the diamond detector with highest signal as a function of beam currents, separately for LER and HER, during single-beam measurements at di ff erent dates in Phase 2; (right) residual gas pressure increase with the beam current, during the same measurements. Two features are evident in these plots: a progres-sive lowering of dose rates with increasing date and aquadratic dependence of dose rate on beam current.The former feature is explained by several factors.In particular, the accelerator vacuum was improved by“beam scrubbing” the vacuum pipes, with consequentdecrease of the beam losses due to beam interactionswith the residual gas, and collimator tuning contributedto limit the beam losses.The quadratic dependence from beam current, validfor Touschek losses, is qualitatively explained also forthe beam-gas background, since the residual gas pres-sure is increasing with the beam current, as shown inthe right panels of Figure 10.A quantitative analysis separates the contributions ofbeam-gas and Touschek backgrounds, using data fromdedicated studies with single beams, as functions ofbeam size and number of bunches at fixed beam current I [5]. Beam-gas backgrounds remain constant in theseconditions, while Touschek intra-bunch scattering is ex-pected to increase when the density of particles in eachbunch increases, by decreasing the beam size or low-ering the number of bunches at the same total current.This is qualitatively shown in Figure 11 and Figure 12and quantitatively expressed in the parameterization [5] dDdt P e I = S bg Z e + S T IP e σ y , (3)where P e is the residual gas average e ff ective pressure, Z e the e ff ective atomic number of the gas mixture, σ y the vertical beam size, S bg and S T coe ffi cients relatedto the relative contributions of beam-gas and Touschekbackgrounds respectively. × mPa] µ P [A/ y σ I/ × D o s e r a t e /I P [ m r ad / s AP a ]
789 bunches1576 bunches
LER beam size study (12 June)
Figure 11: Dose rate of one diamond detector as a function of thereciprocal of the LER beam size. Both variables are normalized usingthe constant LER beam current and the e ff ective vacuum pressure.The linear fit extracts the beam-gas (constant) and Touschek (linearlyincreasing) contributions, as shown in equation 3. Sharing the samebeam current in a larger number of bunches (1576 instead of 789)decreases the Touschek contribution, as expected. × mPa] µ P [A/ y σ I/ × D o s e r a t e /I P [ m r ad / s AP a ] =150mA in I =75mA in I LER beam size study (24 May)
Figure 12: Dose rate in one diamond detector as a function of thereciprocal of the LER beam size, similar to Figure 11, for two di ff er-ent LER beam currents. The two sets of measurements overlap, asexpected for the dose rate normalized on the beam current. ff the vacuum-chamber ma-terial. Simulations of beam-related backgrounds are neededfor two main purposes. The space and time distributionof the radiation field in the interaction region correlatesthe radiation doses measured by diamond sensors withthe radiation dose received by the neighboring BelleII inner detectors, PXD and SVD. Moreover, compar-isons between simulations and measurements validatethe simulations, which can be then used to identify andestimate the background sources and to extrapolate theire ff ects in the future accelerator conditions.Single-beam background simulation studies startwith the simulation of a single circulating beam and itsinteractions (Touschek, Bremsstrahlung, and Coulombscattering) based on the SAD code [23], followed bytracking of lost particles and secondaries in the beampipe and the surrounding material with the GEANT4simulation of Belle II [24]. The geometry and materialsof the diamond detectors are included in the simulation.The energy deposited per unit time by particles cross-ing each diamond detector is converted into a dose rate,using the measured calibration constants.Simulations are repeated for the specific beam opticsand collimator settings of successive data taking peri-ods and accelerator tuning studies. Figure 13 shows anexample from LER single-beam background studies inJune 2020. After initial injection, the beam is kept forsome minutes at constant current with continuous injec-tion, and then the current decays without injection (redline in Figure 13(top)). This pattern is repeated threetimes, with di ff erent number of bunches n b filled into L E R bea m c u rr en t [ m A ] = 783 b n = 1565 b n = 393 b n D o s e r a t e [ m r ad / s ] + D b n Y s I + T I I + B Dose Rate = B 5.32e-04 – (cid:9) = 3.42e-03 B 1.50e-06 – (cid:9) = 4.32e-05 B 2.09e-02 – T (cid:9) = 1.92e+00 4.30e-02 – D (cid:9) = 2.23e-01 non-Fitted DataFitted DataFit Result - - - - -
10 110 D a t a / M C LERHER
Figure 13: Comparison between dose rate data and simulations: anexample from a study of backgrounds from single LER beam. (Top)LER beam current I as a function of time; red line: beam-currentdecay period; n b : number of bunches. (Middle) Corresponding doserate from beam losses recorded by QCS FW 135; black squares: non-fitted data, red circles: fitted data; blue line: fit results for the study.(Bottom) Ratio of recorded and simulated data for eight QCS andthree BP diamond detectors, as a function of their positions; red cir-cles: LER; blue squares: HER. I , number of bunches,vertical beam size σ Y , beam-gas ( B and B ) and Tou-schek T background components, as shown in the insertof the figure. The data agree with simulations within anorder of magnitude (Figure 13(bottom)); the compari-son helps understanding the e ff ect of collimators, align-ments and other accelerator parameters in the simula-tion. ff ects In the continuous injection mode of SuperKEKB, theintensities of the circulating beam bunches are toppedup repeatedly over short time intervals, at a frequencyof up to 25 Hz. Immediately after each injection, beamlosses in the interaction region grow until the beam os-cillations due to the injection perturbation are dumpedaway. As a result, beam losses and diamond detectorsignals vary in time; a significant part of the radiationdose is delivered during short time intervals, adding upto a few milliseconds per second, depending on the spe-cific injection pattern and frequency.The dose-rate monitor at 10 Hz and the total dosecomputation are based on digital sums of the 50 MHzADC data over 100 ms time intervals (Section 3.3, Sec-tion 3.4). If saturation of the current signals occurs inrange 0 over much shorter time intervals, it becomespartially hidden by averaging in 10 Hz data, leading toan underestimate of radiation dose rates and integrateddoses.These e ff ects were studied in November 2019 byrepeatedly dumping the DCU bu ff er memories duringcontinuous injection, to observe the beam loss patternsin 100 kHz data with 10 µ s resolution, for both range0 (36 nA) and range 1 (9 µ A). The example in Fig-ure 14 shows that saturation of range 0 could be ob-served in some BP and QCS diamond detectors, ex-posed to higher beam losses. Saturation typically lastedfor one or a few 100 kHz samples (10-100 µ s). This ob-servation suggests to switch to range 1 for some DCUsin future operations, to avoid saturation at higher beamcurrents; an upgrade of the electronics toward expand-ing the dynamic range is also considered in the longterm. The occupancy of the inner layer of the SVD double-sided microstrip detector of Belle II is a critical param-eter for track reconstruction. The performance of the
Time [s] D o s e r a t e [ m r ad / s ] Time [s] D o s e r a t e [ m r ad / s ] Figure 14: (Top) An example of saturation in range 0, observed in aone-second time interval. The 100 kHz data are read during continu-ous injection at 25 Hz for diamond detector BP BW 325. Ten out of25 injection-related peaks have a saturated value of about 1 . / s, atleast for one 10 µ s sampling each; (bottom) one of the above peakssaturates the measurement range during 40 µ s. D o s e r a t e [ m r ad / s ] Entries 590Mean x 0.03433Mean y 2.547Entries 590Mean x 0.03433Mean y 2.547
Figure 15: (Top) Correlation between the dose rate in diamond de-tector BP BW 215 and the occupancy of the inner SVD microstripdetector layer, measured during single-beam background studies withchanging beam currents. (Bottom) the beam-pipe diamond detectors’dose rates measured at 10 Hz give an estimated occupancy (blue),compared with the SVD occupancy (red), measured at 1 Hz. Their ra-tio (pink) is close to unity and stable in time. On the 3rd of May 2020,the estimated occupancy exceeded the limit and triggered a diamond“injection inhibit” (green).
5. Beam aborts
24 diamond detectors were dedicated to monitoring,and the corresponding DCUs preset on the most sensi-tive current measurement range 0 (section 3.3). Fourdetectors mounted on the beam pipe were dedicated forthe abort function, with range 2 selected in the corre-sponding DCU.
Minimizing the time delays in the abort system is es-sential for the protection of Belle II and of acceleratorcomponents in the interaction region: during the interimbetween the initial detection of abnormal beam lossesand the complete dumping of the beams, radiation lev-els may increase up to damaging values [25], as shownin the example of Figure 16.
10 µs m s i n t eg r a t eddo s e [ m r ad ] Th r es ho l d c r o sse d High dose detectedAbort in progress
Normal condition Beams aborted
Time I n t eg r a t ed do s e [ m r ad ] Figure 16: An example of large beam loss, which occurred beforethe SuperKEKB reduction of abort delays and the DCU firmware im-provement described in the text. The integrated dose in a movingwindow of 1 ms, updated every 10 µ s, is shown as a function of time:a dose of 6 rad is integrated in the almost 50 µ s interim between abortthreshold crossing and beam dump completion. On the accelerator end, optical fibres are used forlong-distance propagation of abort signals. Their prop-agation delay was minimized by optimizing the signalpaths in the accelerator central control system; in 2020,a second “abort gap” [26] was also introduced in the cir-culating bunch trains, to e ff ectively reduce the waitingtime for the activation of abort kicker magnets.The generation of abort-request signals in thediamond-based system was initially cycling every 10 µ s,to compare the moving sums of data stored at 100 kHzwith the abort thresholds. At the beginning of 2020, themodified DCU firmware (section 3.3) contributed to thedelay-minimization e ff ort, by storing data in the bu ff ermemory at 400 kHz and performing the abort-thresholdcomparison every 2 . µ s. As a result, the internal de-lay of the DCU between a large step-like fast-rising in-put signal and the delivery of the output abort requestis halved to about 6 µ s. This time includes both thee ff ect of the analog front-end bandwidth and the over-head of an additional confirmation cycle after threshold-crossing. The diamond abort thresholds were adjusted accord-ing to the evolving accelerator conditions. In Phase2, several quenches of QCS superconducting magnets12 able 4: Abort thresholds in Phase 2 (2018) and Phase 3 (startingin 2019), adopted after initial tests and the adjustments preventingQCS quenches. Range 2 and multiplicity “at least one sum abovethreshold” have been adopted, except for the first period (range 1,multiplicity “at least two”). The abort cycle initially had a frequencyof 100 kHz, modified to 400 kHz since 18 Feb 2020 (section 3.3). date label integr. thresholdtime28 May 2018 fast 1 ms 20 mradslow 1 s 400 mrad5 Jun 2019 fast 1 ms 140 mradslow 1 s 4 . µ s 8 mrad28 Oct 2019 fast 1 ms 80 mradvery fast 40 µ s 80 mrad18 Feb 2020 fast 1 ms 80 mradvery fast 10 µ s 8 mradwere accompanied by SuperKEKB aborts, coming toolate to prevent large energy depositions in the supercon-ducting coils by the rapidly increasing beam losses. Astudy of the recorded diamond signals in these eventsshowed that in most cases a lower diamond abort-threshold could have triggered an earlier abort of thebeams, reducing in this way the energy deposition bybeam losses and preventing the corresponding QCSquenches. As a result of this study, the diamond abortthresholds were lowered, still remaining well above thenoise levels (Table 4, first line).We also observed that shorter integration time-intervals for the moving sums (section 3.3) allowed toadvance the abort request in case of rapidly rising beamlosses, with the advantage of minimizing the receivedradiation dose. Moreover, the reduction of the abort-cycle time in the DCU firmware from 10 to 2 . µ s con-tributed to the e ff ort of shortening the delays in all thecomponents of the SuperKEKB abort system. Table 4shows the evolution of diamond beam abort parametersettings during Phases 2 and 3. A total of 305 abort requests were delivered by thediamond-based system in 2019, 655 in 2020 spring runs.Most aborts were determined by the “fast” threshold ini-tially, and by the “very fast” threshold after its imple-mentation. The latest “very fast” threshold setting, witha time window of 10 µ s for the moving sum (four times (a) − − − µ Time from diamond abort trigger [01000200030004000500060007000 × D o s e r a t e [ m r ad / s ] BP_FW_35BP_BW_35BP_FW_145BP_BW_145 integrated dose (10us+50us)BP_FW_35 25.3 mrad BP_BW_35 10.7 mrad BP_FW_145 9.8 mrad BP_BW_145 4.5 mrad threshold(10us) 8 mrad, firedby BP_BW_35 with 9.4 mrad (b) “Very fast” abort − − − µ Time from diamond abort trigger [050010001500200025003000350040004500 × D o s e r a t e [ m r ad / s ] BP_FW_35BP_BW_35BP_FW_145BP_BW_145 integrated dose (1ms+50us)BP_FW_35 96.3 mrad BP_BW_35 14.1 mrad BP_FW_145 16.8 mrad BP_BW_145 11.7 mrad threshold( 1ms) 80 mrad, firedby BP_FW_35 with 80.2 mrad (c) “Fast” abortFigure 17: (a) Histogram of the number of aborts per day (scale on theleft) in February-June 2020, and cumulative number of diamond abortrequests (continuous line, scale on the right). (b, c) Two examplesof post-abort bu ff er memory dump: dose rate as a function of time,whose zero corresponds to the reconstructed threshold-crossing time. . µ s), corresponds to an average dose rate of 0 . / sin the time interval, or to a single 3 . / s spike con-centrated in one 2 . µ s sample.Figure 17(a) shows the number of aborts per day in2020. The higher peaks correspond to accelerator tun-ing. The decreasing slope of the cumulative distribu-tion indicates a decrease in the average number of abortsper day after the initial period, correlated with improve-ments in accelerator conditions, namely tighter collima-tors and stable injections.Figure 17(b,c) shows two examples of post-abortmemory dump for four aborting diamond detectors, af-ter pedestal subtraction and conversion to dose units.The first event corresponds to a very sharp rise of thedose rate, which reaches a large value of about 5 krad / simmediately after crossing the abort threshold and be-fore the beam abort process is completed. The secondevent shows a much less frequent occurrence of moreslowly rising dose rate, whose integral over 1 ms crossesthe “fast” threshold of 80 mrad.
6. Conclusions and outlook
The Belle II diamond-based beam-loss monitoringand beam-abort system, designed and assembled at theINFN Trieste laboratories, was installed and operatedsuccessfully in the interaction region of the SuperKEKBelectron-positron collider. It has been reliably provid-ing dose-rate monitoring data to keep the radiation bud-get of Belle II inner detectors under control, and hasgiven valuable feedback for accelerator studies and tun-ing. The participation in the beam-abort system, withthe delivery of a substantial fraction of the abort re-quests, has ensured an e ff ective protection of the Belle IIinner detectors and of the SuperKEKB superconductingfinal-focus magnets.To meet the challenge of SuperKEKB long-term op-eration, with the projected future large increase in lumi-nosity, possible improvements of the diamond systemare considered. In particular, the performance of thebeam-loss monitoring would be improved by electron-ics with a wider dynamic range and by a more flexibleaccess to dose-rate data with high time resolution in thering bu ff er memories. Acknowledgements
The construction of the diamond-based beam lossmonitor and beam abort system was funded by IstitutoNazionale di Fisica Nucleare (INFN) in the frameworkof the Belle II experiment. The electronics was de-signed by the Instrumentation and Detectors Laboratory of Elettra Sincrotrone Trieste ScpA. We thank our col-leagues of the SVD, PXD, and BEAST groups withinthe Belle II collaboration for fruitful discussions andcomments, and the Machine-Detector Interface group(MDI) of Belle II and SuperKEKB, an essential discus-sion forum for the operation of the system.