Position Sensitive Detectors in Particle Physics

D H Saxon

Dept of Physics and Astronomy, University of Glasgow,

Glasgow, G12 8QQ, Scotland

Invited talk given at the 5^th Int. Conf. on Position-Sensitive Detectors, London, 13-17 Sept 1999

Abstract

The experimental challenges set by nature in particle physics dictate the detector system and technology requirements. Progress in radiation-hard detector and readout systems is reviewed and new technologies for Ring Imaging Cherenkov systems are described.

1. The task

Particle Physics 'experiments' are capital facilities that are like astronomical observatories, in that a range of studies may be undertaken over a period of years using the same kit of parts. But they are unlike observatories in that each study addresses the entire data sample, taken over years in many cases, and all studies are made simultaneously. [1,2] Whereas a telescope can be pointed at a particular star or galaxy for a given study, it makes no sense to borrow someone else's LEP detector for a few days to use its vertex detector. Data wanted for a particular study must be selected out of the total sample, provided in random sequence by nature, using triggering and analysis.

Different subsystems must therefore cohabit without degrading each other's performance. For example, silicon vertex detectors (observing the decays of short lived particles) must have minimum material to reduce as far as possible multiple scattering and photon conversion. Important results are obtained by combining subsystems. Thus at LEP one can observe a few events of the type

e⁺e^- b bbar, b → J/ψ+X, J/ψ → μ⁺μ^-.

Here the μ⁺ and μ^-are identified by penetration through the outer instrumented iron. Their momenta are measured in the magnetic tracking system and the invariant mass of the pair found to indicate a J/ψ; and the vertex detector shows that these tracks do not emanate from the production point, but are produced in a b-decay after a flight-path of 1 or 2 mm.

To study high mass or high energy phenomena, high rates are unavoidable. The total cross-section for proton-proton collisions is dominated by soft processes and is roughly constant at 100 mb, but the rate for interesting point-like processes falls as 1/E² (where E is the energy characteristic of the process) on dimensional grounds. At the LHC the need to be sensitive to Higgs phenomena up to 1 TeV mass sets the luminosity target at 1-2.10³⁴ cm^-2s^-1. This gives a production rate for a Higgs of mass 500 GeV of order 1 per minute and a total event rate of 10⁹ s^-1.

The key issues are therefore detector and readout speed and radiation hardness, together with event veracity and error rates. The task of finding this Higgs amongst the 10¹¹ background events is roughly as hard as finding the proverbial needle in a haystack. One needs to search quickly, efficiently, and without errors of omission or confusion.

Event rates at certain colliding beam facilities are given in table 1. The crossing interval is the time between candidate collisions where events may have occurred.

Table 1. Event Rates

Only a few events per second can be recorded, and so candidate crossings must be selected in real time, rejecting unwanted ones and selecting (at a fraction say 10^-4 to 10^-7) a range of interesting common and rare types. The trigger system therefore needs high and known efficiency, (which implies some redundancy for internal calibration,) the ability to develop to counter unexpected backgrounds, and must undergo a learning curve during operation. From HERA onwards one needs pipelined logic, as the time needed to make a first accept/reject decision exceeds the crossing interval, and from HeraB onwards one needs a short detector memory time as backgrounds dominate each candidate crossing. The HeraB data path diagram (figure 1) is typical also of LHC detectors (except that it runs at 10 MHz not 40 MHz.) HeraB is more ambitious than ATLAS or CMS in that tracking information is used in the first level trigger in selecting rare b-events. [3]

A LEP1 event (e⁺e^- → Z⁰ → hadrons) has typically 20 tracks and is seen without background. At the LHC at 14000 GeV a typical event has about 25 tracks in the detector acceptance, A wanted high transverse momentum event might have 200 tracks. But the number of tracks seen will be

N = 200 + 24 x 25 x N_c = 800 (for N_c = 1)

where the detector sensitive time is N_c crossing intervals as there are 24 background events to be added to the interesting one. For cleanness of track reconstruction (reliable and high hit probability in each layer and clean separation from neighbouring tracks) one needs a low 'occupancy', that is the fraction of detector cells occupied, which for low occupancies equals the fraction of hits lost. For any detector

Occupancy = hit frequency x busy time/ no of cells

If one designs for 1% occupancy (0.01 = 800 x 40.10⁶ x 25.10^-9/no of cells) one finds 80000 cells per detector layer. Systems of over a million cells are therefore needed.

The majority of tracks are of low transverse momentum ( p_T ~ 400 MeV/c). Suppose one is looking for a Higgs particle decaying via H → ZZ → 4μ in a background of 30 events per crossing. Reconstructing and selecting tracks of p_T > 2.0 GeV/c reduces 800 tracks to about 20. Reconstructing the production point along the beam direction separates these 20 tracks into different events with a spread (σ_z ~ 7 cm) given by the proton beam bunch lengths. One finds, say, 9 tracks from the same vertex including all four muons. One is therefore confident that they come from the same event.

Figure 2 [4] illustrates schematically the radiation backgrounds as a function of distance from the beam line. One sees the inverse square law fall off of charged particles from the production point and the roughly uniform neutron flux (scaled approximately to the equivalent number of minimum ionising particles.) Close to the interaction point m.i.p.'s dominate. The third curve shows the damage to a detector of constant occupancy. This is lowest closest to the beam as the cell size is smallest here. One is motivated therefore to approach the beam as closely as possible and to learn to work in this hostile environment.

2. The systems

The LHC general purpose detectors follow the same multi-shell logic as the LEP detectors, but must be much finer-grained and faster to cope with the expected occupancies and rates. As an example figure 3 shows the ATLAS inner detector layout. [5] Working out from the production point (bottom left) one encounters first a pixel system of 140M channels measuring to accuracies of σ(rφ) = 12 μm and σ(z) = 66 μm. At a radius of 30 to 52 cm there is a system of silicon microstrips. Six million channels measure to σ(rφ) = 16 μm and σ(z) = 580 μm. At a radius of 56 to 107 cm there is a coarse-grained system of 4 mm diameter ‘straw-tube’ transition radiation tracker (TRT) elements. (0.4M channels, σ = 170 μm.) The detectors are laid out in cylindrical layers for tracks emerging perpendicular to the beam axis and in a disc format in the forward direction, with a break near 45⁰. A typical track has 3 pixel, 8 microstrip and 36 TRT hits.

Figure 4 shows a simulated event display. The eye recognises the pattern in the outer coarse-grained but radially continuous TRT. Alternative tracking algorithms carry TRT tracks inwards towards the vertex or start with the sparser inner precision layers and work outwards. [5] The most striking feature of the event display is the number of secondaries in the outer part caused by interactions within the material of the tracking system. This emphasises the importance of minimising the material.

The CMS detector follows the same layout principles as ATLAS. There are two major design choice differences. The magnetic field 4 T not 2 T. This provides a greater fanning out of jets. Also, the 36 straw hits of ATLAS with 170 μm resolution are replaced by 6 MSGC hits of 45 μm resolution. [6] (The outer tracking radius is slightly greater at 115 cm.) The total material up to the last track detection layer is below 0.4 radiation lengths for all rapidities. The pixel detectors account for about one third of this.

The accuracy of reconstruction at the production point is crucial to the study of short-lived particles. The impact parameter, δ, of a track is the distance, extrapolated back to the production point, by which the track misses the nominal production point. The accuracy of a measuring system is characterised by its resolution, σ_δ . For reliable heavy flavour tagging, σ_δ << cτ, where c = velocity of light and τ is the lifetime ( cτ ~ 400 μm for b-quarks and 120 to 300 μm for charmed particles). In a toy model with a cylindrical vertex detector of two layers at radii r₁ and r₂, with resolution σ, and inner layer thickness (including the beam pipe) of t radiation lengths, one finds for a particle of momentum p GeV/c travelling normally to the beam axis

σ_δ² = σ² (r₁² + r₂²) (r₁ - r₂)^-2 + r₁² (0.015/pβ)² t sin^-3 θ

For the ATLAS design in the rφ plane, including the innermost 'b'-layer at a radius of 4 cm, one finds approximately (in units of μm)

σ_δ² = 11²+ 73² p^-2 sin^-3 θ

The key issues in detector design are seen to be: tracking complexity and pattern recognition, precision, transparency of material, speed, contribution to triggering and detector lifetime, plus the systems issues of alignment, stability and cooling.

3. Technologies for high rate, radiation hard detectors

3.1 Gaseous Microstrip Devices. Starting at the outer part of the LHC central tracker and working in, we first encounter gaseous tracking systems. These are single-wire ‘straw tube’ proportional counters with drift time readout for ATLAS, and Microstrip Gas Counters for CMS. [5,6] The MSGC is a miniaturisation of the MWPC. Electrostatic instability makes it impossible to take a wire structure much below 1 mm inter-anode spacing (0.5 mm in special cases.). The MSGC [7] represents an imaginative step in miniaturisation.

Typically a 3 mm deep gas volume is ionised by the passage of a particle to be detected. The electrons released are drifted to a plane containing alternate anode and cathode strips with a typically 200 μm cell size. These strips are laid out on a glass substrate 300 μm thick. Sauli has reviewed the problems of ageing. [8] Rapid ageing occurs if polymers are present in the chamber construction. These quickly coat the anodes. By a good choice of material (e.g. Cr electrodes on S-8900 glass and Ar-DME 90-10 gas,) these problems can be overcome. A greater problem is corrosion of anodes by discharges. A detector which is sensitive to minimum-ionising particles is vulnerable to streamer discharges caused by heavily ionising particles such as nuclear fragments which are always present at a low rate in reactions involving hadrons. A particularly dangerous condition arises if there is some gas amplification for ionisation released close to the cathode edge, which is then further multiplied by the normal anode avalanche.

For the barrel detectors CMS have addressed this by three measures (a) high resistivity plate material with 10¹⁶ Ω/square (b) narrow anodes to give high gain at low field (c) polyimide passivation of the cathode strip edges. [9] For the CMS forward MSGC, and also for the HeraB system, [10] a two-stage amplifier approach has been adopted using the GEM (Gas Electron Multiplier) foil. Sauli has taken the view that all single step devices will discharge at some gain, and so one had better include an intermediate gain stage to increase the safety. [11, see also 12]

3.2 Radiation-resistant silicon detectors. To withstand 10 MRad and 10¹⁴ neutrons cm^-2 both ATLAS and CMS began using 'n on n' technology and moved to p⁺n. [5,6] p⁺n stands higher voltages and is easier to make. As the neutron flux is increased the effective doping changes and the needed depletion voltage decreases until type inversion and then increases inexorably. 250 V or 500 V may be needed after ten years running. To withstand this, detectors have been designed with a multiple guard ring structure to prevent HV breakdown at the detector edges. The detector is 300 μm thick and the anodes are AC-coupled using polysilicon bias resistors. CMS plans to run at -10⁰C. [13] Only single-sided detectors are used because of the type inversion problem. There is the danger of thermal runaway since the leakage current at high voltage produces heat which raises the temperature and further increases the leakage current. p⁺n is now the baseline option.

Meanwhile there are new approaches to enhance radiation resistance. [14] We briefly consider these: freezing out the traps, [15] defect engineering and the so-called 3D-detectors. [16,17]

One of the effects of radiation damage is to create lattice vacancies and interstitial atoms. These produce isolated energy-levels in the band gap which can trap electrons and holes, leading to a loss of charge collection efficiency during electron or hole drift from the place where the ionisation was caused to the electrodes as the radiation dose increases. The RD39 collaboration have shown that if an irradiated Si detector is quickly cooled to around 130K these traps cease to operate. One surmises that they become filled with e and h that lack the thermal energy to escape. They are thereby neutralised and high charge collection efficiency is regained at modest bias voltages. This unexpected revival of a dead detector by cooling has gained the name of the ‘Lazarus effect’. Under reverse bias the effect wears off over minutes (see figure 5) as, presumably, the trapped carriers evaporate. If the detector is illuminated by an LED which injects e-h pairs, the recovery is maintained. Intriguingly, one can also use the detector forward-biased. The resistivity is so high at 130K that it is quite usable.

An alternative approach by the RD48 (ROSE) collaboration [14] engineers the lattice defects differently. If oxygen is diffused into the silicon wafer, (by 16 hours in oxygen atmosphere at 1150⁰C,) it combines with vacancies (V) to form clusters such as VO and V₂O. The electronegative oxygen reduces the effect of the vacancy. One finds a reduction of order 3 in the depletion voltage needed after type inversion. (Using carbon instead of oxygen makes things worse rather than better.) There is sufficient gain in performance using oxygen diffusion that this may become a standard technique.

Following ideas of Sherwood Parker, two groups have investigated a 3D structure in Si and GaAs. [16,17] Instead of the customary planar structure, anodes and cathodes are drilled through the medium on 25 μm spacing. This has the advantage of shorter collection distances and times, and hence lower depletion voltages, but the disadvantage of complex processing. First results suggest worthwhile detectors can indeed be constructed.

3.3 Pixel detectors. Moving in closest to the vertex one uses pixel detectors. [5,18] If such detectors are to be read out with a crossing rate of 40 MHz one cannot afford the time to scan through 140M ATLAS channels. Each channel needs on-chip discrimination and intelligence to sparsify the readout. This is accomplished by bump-bonding a readout chip to the detector medium, so that each detector pixel has its own readout electronics - so called ‘smart pixels’.

The WA97 experiment have shown first examples of the use of such a pixel telescope in 33.3 TeV Pb-Pb collisions, [19] and a series of readout chips (the W series) have been designed by M Campbell et al. The Ω3 chip has 128 x 16 pixels each 50 x 50 μm² with individual addressing, a global threshold and an ENC of 100 e^- RMS. The Medipix chip (64 x 64, 170 x 170 μm²) has individual thresholds and runs in a shutter mode. There are similar examples from other groups.

The choice of technology for radiation-hard readout is also changing with time. The Harris and Dmill/Temic processes are baseline strategies at the LHC, These are specialist and therefore expensive processes. It has proved non-trivial to achieve 40 MHz operation and low noise ( < 1000 e^- with S/N ~ 20). A new approach has arisen through deep sub-micron (0.25 μm) technology. It seems that the small features and enclosed gate geometry to prevent transistor leakage may hold the key. Campbell et al. have built electronics that suffers only minor loss of signal and noise gain beyond 10 MRad. [20]

The ultimate spatial precision, (pixels say 22 x 22 μm² not 50 x 500 μm²) is obtained using CCDs. [21] The smaller pixel size allows one to get in closer, providing dramatically smaller extrapolation errors and less lever arm for scattering to degrade the resolution. Detectors have significantly less material, (a three layer system has 2.1% X₀,) and a thinner active depth (20 μm not 300 μm) means that there is no loss of precision on steeply angled tracks. The blue-riband system at the SLC has 307M pixels and achieves σ_x = σ_y = 4 μm 3D points, running at 190K. But readout is slow (about 10⁴ times slower than smart LHC pixels). This slow rate is not a problem at proposed future e⁺e^- colliders (see table 1). Damerell and co-workers plan a design report by 2001/2 on a detector scheme for such a collider. They plan a system that is thinner still (from 0.4% to 0.12% X₀ per layer) and is read out at 50 MHz, ten times faster than was achieved in 1996. [22] There is a gain in efficiency and background rejection in tagging b and c quarks, as compared to smart pixels.

4. Progress in Ring Imaging Cherenkov Counters

This is a different area where the last year or two has seen major progress, Recall that magnetic systems measure particle momenta but not velocities and therefore not particle masses. One can identify particle species (in the case of LHCb over the range 2-150 GeV/c) by the Cherenkov radiation emitted in a cone of angle θ = cos^-1 (1/βn) where β = v/c and n is the refractive index of the medium being traversed. Thus HERMES chose two radiators, aerogel of n = 1.03 and C₄F₁₀ gas of n = 1.0014, to identify π,K,p by the radiation cone angle up to 20 GeV/c. The recent progress has been in the readout systems for the optical and near UV photons using photomultipliers, hybrid photodiodes and CsI photocathodes.

Figure 7 illustrates the principles. It is an exercise for the reader to convince himself that a cone of light emitted at a constant angle along the path of a particle is focussed by a concave mirror into a ring. (Second exercise: calculate the ring radius.) One then covers the focal plane with light-detecting pixels. HERMES [23] use a hexagonal close-packed array of photomultipliers at 23.3 mm centres. The photocathodes cover only 38% of the focal plane area. A simple system of soft steel funnels with aluminium foil inserts increases this coverage to 92%. They expect about 10 photoelectrons for aerogel and 25 for CF₄.

HeraB use smaller pixels. Multianode photomultipliers (each with 4 or 16 pixels) are laid out in a square array. [24] Each one has a field lens and collector lens system 140 mm deep designed to image a 35.3 x 35.3 mm² area on to the 18 x 18 mm² sensitive face of the photomultiplier. First HeraB event displays show excellent clear rings with 35 photons for β = 1. The detector is so robust one can even use it to measure momentum and mass without using tracking information, assuming only a production point beyond a spectrometer magnet! [25]

LHCb progress has been reported by Websdale. [26] One approach uses 26 x 26 mm² multianode photomultipliers, but now with 64 anodes in an 18 x 18 mm² area and again with a simple lens system. They are also looking at hybrid photodiode systems. [27] A great attraction of these is that there is no avalanche gain, but rather each electron is accelerated through 15 kV and then deposits all its energy in a silicon detector. Peaks for 1,2,3,4 photoelectrons are clearly resolved.

Finally, one should mention the HMPID (High Momentum Particle Identifier) system of the LHC Heavy Ion experiment ALICE. [28] (see figure 8). This is a compact system designed to work in a magnet using 10-14 mm depth of liquid C₆F₁₄ radiator of n = 1.28 at a wavelength of 200 nm. The Cherenkov light is proximity focussed through a quartz window into a ring on a CsI photocathode plane. The pulses are amplified using an MWPC plane and cathode-pads read out the signals. The system has a quantum efficiency (including window losses) which peaks at 12% at 185 nm. Clean rings with 15 hits are seen.

5. Conclusions

Particle Physics continues to set exacting demands on detector technology and readout: in speed, position resolution, double hit resolution, material thickness, radiation hardness, data flow, data compression and event triggering. After 10 years of LHC R&D, (since the Barcelona ECFA workshop in 1989, [29]) new ideas are still coming through for enhanced radiation hardness of detector and readout. 0.25 μm technology helps pixel size and radiation hardness, and there are robust new options in single photon detection (pixels, RICH). Watch this space!

6. Acknowledgements

People have been generous in helping me prepare this article. I have not had space to do justice to all the good things they showed me. Let me thank especially R Bellazzini, N H Brook, C J S Damerell, S Hayward, R Horisberger, P Krizan, L Latronico, F Muheim, V O’Shea, A Parker, F Piuz, R Schwitters, K M Smith, G Tonelli, O Villalobos-Baillie, T Virdee and J J Ward.

Figure captions

1. HeraB data flow

2. Sketch of the radiation damage to silicon due to (i) minimum ionising particles (ii) neutrons (no calorimeter shielding) (iii) relative damage per detector element assuming constant occupancy.

3. A cross-section of the ATLAS inner detector engineering layout through the beam axis. The complete detector is formed by rotating this section around the beam axis and reflecting it in the plane at the left edge of the picture.

4. A simulated event display of the process H → ZZ → μ⁺μ^-e⁺e^- in the barrel part of the ATLAS inner detector.

5. The Lazarus effect. Depletion as a function of voltage (forward and reverse) in a detector irradiated to 4x10¹⁴ protons cm^-2 and then held at 130K. Different curves correspond to different times after cooling.

6. ROSE Collaboration data - depletion voltage plotted against proton fluence. top curve - carbon-doped, middle curve - no doping, lowest curve - oxygen doped.

7. LHCb RICH prototype. Schematic diagram showing cones of light emitted from aerogel and gas being imaged into large and small rings in focal plane.

8. ALICE experiment: schematic view of a proximity RICH detector with liquid radiator and CsI photocathode.

References

[1] D H Saxon, Nucl. Instr. Meth A367 (1995) 41

[2] D H Saxon, 'The Tools: Detectors', in 'The Particle Century', G Fraser (ed), (Inst. of Phys. Publ., 1998), 183

[3] T Lohse et al, HeraB, an experiment to study CP violation in the B system using an internal target at the HERA proton ring, DESY-PRC 94/02 (1994)

[4] D H Saxon, Nucl. Instr. Meth A315 (1992) 1

[5] ATLAS Collaboration, 'ATLAS inner detector technical design report', CERN/LHCC/97-16 (1997)

[6] CMS Collaboration, 'CMS inner detector technical design report', CERN/LHCC/98-6 (1998)

[7] A Oed, Nucl. Instr. Meth A263 (1988) 351

[8] F Sauli, these proceedings

[9] The CMS MSGC Collaboration, Proc Beauty99, Bled-Slovenia, (1999) submitted to Nucl. Instr. Meth.

[10] T Hott, these proceedings and M Medinnis, Proc. Lepton-Photon 99, (SLAC, 1999), to be published.

[11] F Sauli, Nucl. Instr. Meth A386 (1997) 531

[12] I Giomataris, these proceedings

[13] G Tonelli, Proc. Vertex 98 (to be published)

[14] S Watts, these proceedings

[15] L Casagrande, these proceedings. V G Palmieri et al., Nucl. Instr. Meth A413 (1999) 475

[16} C Kenney, these proceedings

[17] R Bates, these proceedings

[18] D Kotlinski, these proceedings

[19] G Alexeev et al., Nucl. Phys. A590 (1995) 139c, E H M Heijne et al., Nucl. Instr. Meth. A344 (1994) 138

[20] M Campbell et al., IEEE Trans. Nucl. Sci. NS-46 (1999) 156

[21] K Abe et al., Nucl, Instr. Meth A400 (1997) 287

[22] see http://hep.ph.liv.ac.uk/^~green/lcfi/lcfihome.html

[23] D Ryckbosch. Nucl. Instr. Meth A433 (1999) 98

[24] S Korpar et al., Nucl. Instr. Meth A433 (1999) 128, D R Broemmelsiek, ibid p 136

[25] R. Schwitters, private communication

[26] D Websdale, these proceedings

[27] E Albrecht et al., Nucl. Instr. Meth A411 (1998) 249, ibid., A433 (1999) 159

[28] ALICE Collaboration. 'Detector for High Momentum PID' CERN/LHCC/98-19 (1998). A Di Mauro et al CERN-EP/99-52 (1999), F Piuz et al CERN-EP/99-33 (1999) and CERN-EP/99-55 (1999),

[29] D H Saxon, Proc. ECFA study week on Instrumentation Technology for High-Luminosity Hadron Colliders, CERN 89-10, ECFA 89-124 (1989), 53