Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–78

Contents lists available at SciVerse ScienceDirect

Nuclear Instruments and Methods inPhysics Research A

0168-90

http://d

n Corr

E-m

journal homepage: www.elsevier.com/locate/nima

Geant4 simulation of transition radiation detector based on DEPFETsilicon pixel matrices

Sergey Furletov n, Julia Furletova

University of Bonn, Bonn, Germany

a r t i c l e i n f o

Available online 14 May 2012

Keywords:

Transition radiation detector

TRD

DEPFET

Active pixel detector

Silicon detector

Geant4

02/$ - see front matter & 2012 Elsevier B.V. A

x.doi.org/10.1016/j.nima.2012.05.009

esponding author.

ail address: fourl@mail.cern.ch (S. Furletov).

a b s t r a c t

This paper presents new developments in Monte Carlo simulation for test beam measurements of a

silicon transition radiation detector based on DEPFET—a silicon active pixel detector. The test of

DEPFET with fiber radiator has been carried out at the DESY 5 GeV electron beam. Monte Carlo

simulation of the test beam setup is based on Geant4. A comparison of Geant4 simulation with test

beam data is presented.

& 2012 Elsevier B.V. All rights reserved.

1. Introduction

Transition radiation detectors (TRD) have the attractive featureof being able to separate particles by their gamma factor. In highenergy physics, TRDs are typically used for electron identificationand hadron background rejection [1]. The main problem in thedetection of transition radiation photons is to separate TR fromionization energy losses of charged particles.

The classical TRD is based on gaseous detectors filled with axenon based gas mixture to efficiently absorb transition radiationphotons. The typical absorbed TR energy is about 10–15 keVwhile the background of dE=dx from a charged particle is about2–3 keV.

Replacing the xenon-based gaseous detectors by modernsilicon detectors is complicated by the large energy losses ofcharged particles in 3002700 mm of silicon which is about 100–300 keV. A DEPFET sensor [2,3] has a good signal to noise ratiowhich allows to measure ionization from charged particles in athin layer of silicon (down to 30 mm) [9]. In this case theionization from charged particles (10–15 keV) is in order of theTR photon energy. Additionally, due to its high granularity (downto 20� 20 mm2), the DEPFET allows measuring angular distribu-tion of TR photons. The use of the angular separation between TRphotons and particle tracks is discussed in detail in Refs. [1,7,8].

The main features of DEPFET which are used in the new TRdetector concept are a fully depleted substrate and low noiseenergy measurements. Fig. 1 shows a schematic cross-section of aDEPFET sensor with a pixel size of 20� 20 mm2 and 450 mmthickness. Particles crossing the DEPFET sensor with an angle of

ll rights reserved.

incidence around 401 have a track length in one pixel of about30 mm. In this case the energy deposited in one pixel is a factor of10 lower (� 10 keV, than for a particle with an angle of incidence01 and comparable with ionization from TR photons.

About 10–30 points of dE=dx measurements, depending on theparticle angle, could be used for particle separation. However,most of the TR photons are absorbed in the first 3–7 bins (pixels)along a track.

Nowadays the DEPFET substrate can be thinned down to50 mm [15]. Fig. 2 shows an application of a thinned down DEPFETmatrix for a TR detection. The drawback of this method is that theefficiency of registering TR will be also lower. The advantage isthat the pixel size could be much larger � 502200 mm.

2. Test beam setup

The test of DEPFET with a fiber radiator has been carried out inDecember 2009 at DESY. Detailed description of the setup andresults is discussed in the previous article [4]. DESY has a pureelectron beam with an energy up to 7 GeV. For these relativelylow energy electrons, the multiple scattering in a 15 cm radiatoris about 0.2 mrad and comparable with a TR angular distribution.Readout of a DEPFET and data acquisition system is described inRef. [5]. Tests were done using a standalone DEPFET module(COCG VS H3.0.07 20� 20 mm2) with signal to noise ratio � 220for MIP, pixel size 20� 20 mm2 and 450 mm thickness. Fiberradiators with different thicknesses have been used for the tests.The radiator consists of polypropylene (PP) fibers of 20 mmdiameter, and produced as fleece layers (mats) of about 2.2 mmthickness and density 0.1 g/cm3. The layers were stacked to give atotal radiator thickness of 15.5 cm with an average density of0.083 g/cm3 [6].

dE/dX

20−30 µ

450 µ

TR photon

Radiator

5 − 15 cm

Particle track

Fig. 1. TRD with 450 mm thick silicon.

50 µ

Si

Radiator

5 − 15 cm

Particle track

TR photon

70 µ

Fig. 2. TRD with 50 mm thick silicon.

Energy, keV0 5 10 15 20 25 30 35 40 45 50

dN d

E, n

orm

aliz

ed

0

500

1000

1500

2000

2500

3000

3500

4000

4500 incoming photonsabsorbed photonsescaped photons

Fig. 3. A Monte Carlo spectrum of generated, absorbed and escaped transition

radiation photons for 600 mm thick silicon and 15.5 cm thick radiator.

S. Furletov, J. Furletova / Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–7874

3. Monte Carlo simulation

For initial Monte Carlo simulation the ATLSIM package [14]based on GEANT3 is used together with a transition radiationsimulation program, which has been developed for the simulationof the ATLAS TRT [16].

The absorbed transition radiation energy in � 500 mm thicksilicon detector with 15 cm of radiator, placed at a distance of � 2cm from the detector surface is in good agreement with ameasured spectrum [4]. However, the TR model of this programdoes not simulate the angular distribution of TR photons, which isvery important in this case. For this reason, we had to switch toGeant4 where the TR angular distribution is implemented. Thetest beam setup was implemented in Geant4.9.4.p01. Unfortu-nately, the default version also does not describe the experimen-tal data in the part of TR angular distribution.

Investigation of this problem shows that the agreement is notgood due to lack of experimental data on angular distribution ofTR and some theoretical uncertainties [11]. After tuning simula-tion parameters which are related to the angular distribution of

TR photons, a good agreement between Geant4 and experiment isachieved. Using the recent experimental data allows to improvethe simulation of TR photon angle. Tuned version of the code willbe available in the next release of Geant4.

The simulation of pixel detector response includes the noise ofreadout electronics, the charge diffusion in pixels and the ADCdiscretization. Geant4 offers several options for simulation ofionization losses and transition radiation processes. The particledE/dx in 450 mm silicon is simulated according to the photoabsorption ionization (PAI) model [12,13].

The current Geant4 transition radiation model includes severalclasses. The two most important ones are G4RegularXTRadiator fordescription of radiator with fixed foil and gas gap thicknessand G4GammaXTRadiator for description of irregular radiatorwith gamma distributed foil and gas gap thickness [10]. Regularradiator is described in terms of foil and gas gap thickness anddensity. The number of foils is calculated from the radiator length.For irregular radiators there are two more parameters describingthe thickness fluctuation of foils and the gas gap: AlphaPlate andAlphaGas. G4GammaXTRadiator was selected for simulating thefiber radiator which was used in the test beam. A foil thickness

parameter of 16 mm was chosen for 20 mm fibers as average pathin the fiber. The gas gap is a free parameter and could vary from100 to 1000 mm; the value of 300 mm was found well tuned todescribe experimental data. Variation of fluctuation parametersAlphaPlate and AlphaGas does not show significant sensitivity ofthe result to these parameters and the default value of 100 wasused for both parameters. The Geant4 data processing flow wasthe same as for the experimental data. The energy calibration inMonte Carlo is done in the same way as for the experimental datasimulating the data from a gamma source of 59.5 keV (241Am).

4. Comparison between the experimental data and the Geant4Monte Carlo

Silicon as detector for TR photons shows a good efficiency. Themaximum possible path of TR photons in 450 mm DEPFET rotatedat 401 is about 600 mm. In this case, the DEPFET has almost a 100%efficiency for TR up to 12 keV (Fig. 3).

4.1. Cluster reconstruction

Due to charge sharing, a deposited charge in the silicondetector usually spreads over several pixels, typically within a3�3 pixel area. Reconstruction of a cluster in an event is done by

columns25 30 35 40 45 50

row

s

70

75

80

85

90

95

7950

8000

8050

8100

8150

8200 Run:2537 Event: 9866

columns35 40 45 50 55 60

row

s

170

175

180

185

190

195

7950

8000

8050

8100

8150

8200

8250

8300

8350

8400 Run:8532 Event: 1602

Fig. 4. Event display of 5 GeV electrons, pixels 20� 20 mm2, radiator thickness 15.5 cm. Left: DESY test beam, right: Geant4 MC. The arrow on the event display shows the

center of TR cluster found by the analysis software.

cluster length, pixels

0 5 10 15 20 25

entri

es

0

5

10

15

20

25

30

35

40×103

With RadiatorNO Radiator

DATA

cluster length, pixels0 5 10 15 20 25

entri

es

0

1

2

3

4

5

6

7

×103

With RadiatorNO Radiator

MC

Fig. 5. Cluster length distribution, measured in DEPFET sensor with a pixel pitch

of 20� 20 mm2 (top: test beam and bottom: Geant4).

S. Furletov, J. Furletova / Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–78 75

identifying a seed pixel as the pixel with the largest signal over athreshold, which is chosen to be five times larger than the noise(5snoise � 870e�). The adjacent pixels in a defined area around theseed pixel are added to the cluster, if their signal is larger than thesecond level threshold. This second threshold is chosen to betwice as large as the noise (2snoise � 350e�).

Since the DEPFET module has been tilted about 401 withrespect to the beam, the total length of the clusters from thebeam particles is about 20 pixels. Fig. 4 shows an event display forthe test beam (left) and Geant4 (right). Due to the low energy ofelectron beam (5 GeV), the TR photons are clearly visible andseparated from particle track by a few pixels. The arrow on theevent display shows the center of TR cluster found by the analysissoftware.

4.2. Cluster length and energy distributions

All found clusters were evaluated using four parameters:cluster length, cluster width, number of pixels in cluster andcluster energy. Runs with a radiator have a large number of smallsize clusters compared to the runs without radiator (see Fig. 5).These small size clusters are classified as TR photons.

Cluster energy distributions for two cases—with and withoutradiator, are shown in Fig. 6 for test beam (left) and Geant4(right). The most probable value (MPV) of the dE=dx energydeposit by the ionizing particle is about 170 keV.

In the runs with radiator (solid line), a shift of the MPV ofdE=dx spectra to higher values comes from the additional con-tribution of transition radiation photons which were not sepa-rated from the particle track. In addition low energy clusterswhich belong to transition radiation photons were found.

4.3. Angular distribution of TR photons

The measured angular distribution of transition radiationphotons is affected by two effects, the angular distribution oftransition radiation photons itself and a multiple scattering of5 GeV electrons in 15 cm of a radiator (� 2%X0). A distancebetween clusters from TR photons and clusters from the particletrack in Y-direction is shown on Fig. 8 for test beam data andGeant4 MC. On the plot only photons separated from the particletrack are included, therefore the inefficiency in the middle isexpected. A distance between clusters from TR photons andclusters from the particle track is shown on Fig. 7 for experimental

data with radiator. On the plot only photons separated from theparticle track are included, therefore the inefficiency in the middleof the circle is expected. A Y -projection of this distribution isshown on Fig. 8 for experimental data and for Monte Carlo.

Energy, keV0 50 100 150 200 250 300 350 400

dN/d

E, n

orm

aliz

ed

0

2

4

6

8

10

12

14

16

18

20×103

With RadiatorNO Radiator

DATA

Energy, keV0 50 100 150 200 250 300 350 400

dN/d

E, n

orm

aliz

ed

0

0.5

1

1.5

2

2.5

3

3.5

4

×103

With RadiatorNO Radiator

MC

Fig. 6. Cluster energy spectrum measured by DEPFET. Particle clusters peaks

at 170 keV. Top: test beam data and bottom: Geant4.

X dist., pixels-20

Y di

st.,

pixe

ls

-15

-10

-5

0

5

10

15

With Radiator

-10 0 10 20

Fig. 7. Distance between TR photons and a particle track measured in a DEPFET

sensor with a pixel pitch 20� 20 mm2 with radiator.

distance, pixels-30

norm

. ent

ries

0

10

20

30

40

50

×10-3

ExperimentGeant4

-20 -10 0 10 20 30

Fig. 8. Y -projection of distance between TR photons and particle track measured

in a DEPFET sensor with a pixel pitch 20� 20 mm2.

Number of found clusters0 1 2 3 4 5

0

Pro

babi

lity

0.2

0.4

0.6

0.8

1 With RadiatorNO Radiator

DATA

Number of found clusters

0 1 2 3 4 5

Pro

babi

lity

0

0.2

0.4

0.6

0.8

1With RadiatorNO Radiator

MC

Fig. 9. Number of photon clusters (a clusters with length o5 pixels, Fig. 5), which

were found near the particle track (o30 pixels). Top: test beam data and bottom:

Geant4.

S. Furletov, J. Furletova / Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–7876

The flat background in the experimental data is supposed to comefrom the other particles: the probability to have two and moreparticles in the matrix is about 40%.

4.4. Transition radiation clusters efficiency

The probability to find a TR photon cluster near a cluster from a5 GeV electron is � 50% with radiator and � 5% without radiator(Fig. 9). The inefficiency includes the inefficiency of the cluster searchalgorithm, which could be improved. The overlapping of TR clusterswith the cluster from the particle track also contributes to theinefficiency. Clusters from transition radiation photons which were

S. Furletov, J. Furletova / Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–78 77

located too far (430 pixels) from the particle track were also nottaken into account. The TR photons which were not stopped in thesilicon due to too high energy and also the events where no TRphotons were emitted by the charged particle give an additionalcontribution to the inefficiency. This could be improved using an

[col]10

[row

]

120

125

130

135

140

145

150

155

7950

8000

8050

8100

8150

8200

8250

8300

8350

8400XY RAW (Mod6)

dE/dX area

TR area

15 20 25 30 35 40 45

Fig. 10. Event display with particle area and TR area highlighted.

Energy in TR area0 100 200 300 400 500 600 700 800 900

Pro

babi

lity

0

0.2

0.4

0.6

0.8

1 With RadiatorNO Radiator

DATA

Energy in TR area0 100 200 300 400 500 600 700 800 900

Pro

babi

lity

0

0.2

0.4

0.6

0.8

1 With RadiatorNO Radiator

MC

Fig. 11. Energy deposition in TR area with radiator and without radiator (top: test

beam and bottom: Geant4).

additional layer of silicon detector or using a thicker detector and alsoby increasing the radiator length in front of the sensor.

4.5. Identification with external track

If a particle track has been reconstructed, it is possible to skipcluster finding and to use another algorithm, calculating energydeposition in two areas (Fig. 10). The first is the particle area, hereone finds mainly the particle dE/dx and also TR photons over-lapped with particle track. The second is the TR area, where aremainly the TR photons, delta electrons and other particles in caseof high occupancy. Energy in TR cluster area could be used foridentification. Fig. 11 shows that this algorithm has a comparableefficiency with the cluster counting method.

4.6. Energy distribution along a particle track

Transition radiation photons might also be discriminated onthe top of ionization losses from charged particles using dE/dx

measurements along the particles track. Fig. 12 shows the averagedE=dx for each of 20 pixels along the beam particle track, wherethe solid line represents data taken with radiator, the dashed linebelong to the data taken without radiator and the filled area isadditional ionization from transition radiation photons. Theenergy loss along the particle track without radiator is non-uniform due to diffusion and varies depending on pixel size and

pixel number

0 5 10 15 20

mea

n dE

/dX

, AD

C u

nits

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9×103

With RadiatorNO RadiatorTransition Radiation

DATA

pixel number0 5 10 15 20

mea

n dE

/dX

, AD

C u

nits

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9×103

With RadiatorNO RadiatorTransition Radiation

MC

Fig. 12. Average energy loss along a particle track (top: test beam and bottom:

Geant4).

S. Furletov, J. Furletova / Nuclear Instruments and Methods in Physics Research A 706 (2013) 73–7878

geometry of DEPFET. Since the energy in each pixel is comparedindependently, the bin uniformity is not important.

5. Conclusions

The first tests of a DEPFET sensor as a transition radiationdetector have shown promising results. Geant4 based Monte Carloin good agreement with experimental data and can be used forperformance calculation of silicon transition radiation detectors. Incomparison to traditional gaseous wire chamber detectors, DEPFEThas advantages: its high granularity allows a combined tracker andidentificator to be built which is especially important for spacemissions. In the momentum range from 1 to 10 GeV, it is possibleto use the angular separation between TR photons and the particletrack as standalone, or in combination with the discrimination ofTR on top of ionization losses.

Acknowledgment

We are grateful to N. Wermes and H. Kruger for their interestand support of this work. We wish to thank V. Grichine for hishelp in tuning the Geant4 transition radiation package. Also manythanks to members of the DEPFET Collaboration especially HLL ofMax-Planck-Institut in Munchen for DEPFET matrices.

References

[1] B. Dolgoshein, Nuclear Instruments and Methods in Physics Research Section

A 326 (1993) 434.[2] J. Kemmer, G. Lutz, Nuclear Instruments and Methods in Physics Research

Section A 253 (1987) 365.[3] N. Wermes, Nuclear Instruments and Methods in Physics Research Section A

604 (2009) 370.[4] J. Furletova, S. Furletov, Nuclear Instruments and Methods in Physics

Research Section A 628 (2011) 309.[5] S. Furletov, Nuclear Instruments and Methods in Physics Research Section A

628 (2011) 221.[6] R.D. APPUHN, et al., Nuclear Instruments and Methods in Physics Research

Section A 263 (1988) 309.[7] V. Grishin, et al., Lebedev Phys . Institute Preprint No. 201, 1977.[8] A.I. Alikhanian, et al., Nuclear Instruments and Methods in Physics Research

Section A 158 (1979) 137.[9] M. Porro, et al., IEEE Transactions on Nuclear Science NS-2 (2004) 724.

[10] V.M. Grichine, S.S. Sadilov, Nuclear Instruments and Methods in Physics

Research Section A 563 (2006) 299.[11] V. Grichine, Private communication.[12] V.S. Asoskov, et al., Lebedev Institute Annual Report, v. 140, 1982, p. 3.[13] J. Apostolakis, et al., Nuclear Instruments and Methods in Physics Research

Section A 453 (2000) 597.[14] P. Nevski, et al., ATL-SOFT-95-014, CERN, 1995.[15] L. Andricek, et al., Nuclear Instruments and Methods in Physics Research

Section A 565 (2006) 165.[16] The ATLAS TRT collaboration, et al., Jounral of Instrumentation 3 (2008)

P0201.