Nuclear Instruments and Methods in Physics Research A 648 (2011) S265–S268

Contents lists available at ScienceDirect

Nuclear Instruments and Methods inPhysics Research A

0168-90

doi:10.1

� Corr

E-m

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

On the energy response function of a CdTe Medipix2 Hexa detector

Thomas Koenig a,�, Andreas Zwerger b, Marcus Zuber a, Patrick Schuenke a, Simeon Nill a, Ewald Guni c,Alex Fauler b, Michael Fiederle b, Uwe Oelfke a

a German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, 69120 Heidelberg, Germanyb Freiburg Materials Research Center (FMF), Stefan-Meier-Strasse 21, 79104 Freiburg, Germanyc Erlangen Centre for Astroparticle Physics (ECAP), Erwin-Rommel-Strasse 1, 91058 Erlangen, Germany

a r t i c l e i n f o

Available online 21 November 2010

Keywords:

Semiconductor detector

Cadmium telluride

Photon counting

Energy resolution

02/$ - see front matter & 2010 Elsevier B.V. A

016/j.nima.2010.11.071

esponding author.

ail address: t.koenig@dkfz.de (T. Koenig).

a b s t r a c t

X-ray imaging based on photon counting pixel detectors has received increased interest during the past

years. Attached to a semiconductor of choice, some of these devices enable to resolve the spectral

components of an image. This work presents the results from measuring the energy response function of a

Medipix2 MXR Hexa detector, where six individual Medipix detectors were bump bonded to a 1 mm thick

cadmium telluride sensor in order to form a 3�2 array of 4.2�2.8 cm2 size. The average FWHM of the

photo peak of an 241Am source was found to be 2.2 and 2.1 keV for single pixels and bias voltages of 200

and 350 V, respectively, across the whole Hexa detector. This corresponds to a relative energy resolution

of less than 4%. Adding up all pixel spectra of individual chips lead to an only small deterioration of energy

resolution, with line widths of 2.7 and 2.5 keV. In general, a lower detection efficiency was observed for

the lower voltage setting, along with a shift of the peak position towards lower energies.

& 2010 Elsevier B.V. All rights reserved.

1. Introduction

Conventional digital X-ray imaging relies on measuring a signalproportional to the integrated amount of energy deposited in apixel during exposure, which normally follows the conversionof X-ray photons into visible light in a scintillating material. Duringthe last decade, this approach has been supplemented by directconversion techniques based on pixelized semiconductor detectors[1]. While many of these devices are capable of counting individualphotons and thus providing spectral resolution, there is ongoingdebate as to which combinations of pixel sizes, sensor materialsand thicknesses to choose in order to achieve both high spatialand spectral resolution in combination with a high quantumefficiency. Depending on detector properties and sensor material,the recorded spectrum can significantly differ from the actual one.In order to interpret and reconstruct these spectra, a carefulmeasurement of the so-called energy response function is neces-sary. This work presents the results from measuring this function ata single photon energy of about 60 keV for a Medipix2 MXR [2]Hexa detector attached to cadmium telluride (CdTe) single crystal.With a size of 4.2�2.8 cm2, this device represents the largestMedipix detector assembled to a single sensor so far, and isintended to be used for small animal imaging at DKFZ.

In the following section, we will briefly describe the detectorand the processing steps we employed in order to obtain the results

ll rights reserved.

presented in Section 3. Section 4 will then give a summary as wellas an outlook on future experiments.

2. Materials and methods

The energy response function of pixelized semiconductors canbe severely biased by electric charges shared among neighboringpixels. In order to mitigate these effects, a pixel pitch of 165 mm waschosen rather than the 55 mm which is provided by the Medipixarchitecture. This was achieved by connecting only every ninthpixel to the sensor (3�3) by means of intentionally ohmic contacts,resulting in 258�172 pixels (44 376 in total). The sensor was a1 mm thick CdTe single-crystal manufactured by Acrorad. Thesesingle-crystals represent the state-of-the-art of high resistivity,detectorgrade material and are commercially available with adiameter of 75 mm and thicknesses of 1 and 2 mm. The crystalswere grown by the travelling heater method, which offers theadvantage of a low growth temperature that reduces the concen-tration of defects and increases homogeneity. Characterization ofthese wafers obtained a homogeneous distribution of the resistivitywith an average value of 5� 109 O cm and a variation of less than10%. Tellurium inclusions (second phase defects) were identified byinfrared microscopy with an average diameter of less than 10 mmand a concentration of about 103 cm�2. Hybridization was per-formed at FMF with low temperature solder bumps. The overallprocess temperature was kept below 130 1C, which is of crucialimportance in order to maintain high level sensor properties. All

20 30 40 50 60

0

500

1000

1500

2000

2500

3000

3500

Energy [keV]

Diff

eren

tial C

ount

s

Fig. 1. Single pixel spectra found in a 3�3 neighborhood of a randomly chosen pixel

(Vbias¼350 V). Note that these spectra were low-pass filtered using a binomial

kernel of size n¼25 (for visualization only).

T. Koenig et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) S265–S268S266

measurements described below were performed using the MedipixUSB interface [5] and the Pixelman software [6].

As every other Medipix2 detector, the Hexa detector introducedhere offers two modes of operation: The single threshold mode,where every event above a predefined energy threshold is counted,and the dual threshold or window mode, which introduces anadditional threshold in order to obtain an energy window.

The response of every pixel can be calibrated with two sets ofthree adjustment bits to bring each of the two thresholds more inline with their neighbors. The process determining the optimalvalues of these bits is called threshold equalization and wasperformed according to standard procedures. In detail, the Kafluorescence of a silver foil was used to equalize the lowerthresholds, while the upper thresholds were equalized employingthe switching point between single and dual threshold modes asdescribed by Tlustos et al. [3].

Energy calibration was performed using the Ka lines of molyb-denum (17.4 keV) and silver (22.2 keV) as well as the photo peak ofan 241Am source (59.6 keV), which is described below. Here, onehas to keep in mind that charge sharing shifts the maximumposition of the peak to slightly lower energies. This shift wasdetermined using a Monte Carlo simulation as developed by Durstet al. [4]. The corrected energy values used for the three sourceswere 16.4, 21.4 and 59.0 keV for a bias voltage of 350 V.

In principle, the window mode can be used to obtain spectro-scopic information about the incident X-rays. However, the energywindow must not be chosen too small as manufacturing toleranceslead to a slight variation of each pixel’s response to radiation evenafter threshold equalization, also called residual threshold disper-sion. The remaining small differences become more and moredominant as the energy window gets thinner, and so the effectivewindow size cannot be set below a few keV, which in turn leads to ablurring of the spectra recorded. While this behavior does notrepresent a drawback in most imaging applications, it prevents theprecise characterization of the detector under study. For thisreason, the energy response function is usually measured differ-ently by scanning in single threshold mode across the energy rangeof interest, yielding an integrated spectrum. This spectrum is thendifferentiated to obtain the actual photon energy spectrum.

In order to determine the energy response function, i.e. thedetector’s response to monochromatic radiation, we employed thesame 241Am source that we already used for energy calibration. Itemits X-ray photons with an energy of about 59.6 keV and istherefore well suited for the investigation of the energy responsefunction at an X-ray energy typical for medical diagnostics, whenother monochromatic sources such as synchrotron light are notavailable.

The 241Am used here has a nominal activity of about 1.1 GBq andwas placed in front of the detector at a distance of about 5 cm. Themeasurement was performed at a bias voltage of 350 V for anenergy range of 10–60 keV and took about 18 h in order to acquireenough photons to perform the peak fitting without any significantlow-pass filtering, which would have biased the width of the photopeak. Another measurement of just the photo peak was performedat 200 V to investigate the dependance on the bias voltage.

The full width at half maximum (FWHM) of the photo peak wasdetermined by fitting a Gaussian to individual pixels as well as thesum over each of the six chips’ pixel spectra. A binomial filter ofsize three was applied to the spectra prior to fitting in order tomake the results more stable (which broadens the peak by lessthan 0.05 keV). The medians and the standard deviations obtainedfrom the fits on the single pixel level can then be considered ameasure for the inter-pixel dissimilarity, while the fits on the sumspectra give an impression of how much the residual dispersion ofthe lower thresholds’ responses affect the energy resolution of awhole chip.

In order to demonstrate how the energy response functionaffects the acquisition of broad band X-ray spectra, the tubespectrum from a medical Siemens Powerphos X-ray tube wasmeasured at a voltage of 110 kVp and a current of 1 mA. The tubefeatures a built-in aluminium filter, which is normally used toreduce the skin dose in patients and blocks almost every photonbelow 20 keV from exiting. The detector was placed about 1.2 mfrom the tube to prevent both signal pile-up in the detectorelectronics as well as polarization of the CdTe sensor caused bythe low hole mobility. Again, the spectrum was recorded by a scanin single threshold mode between energies of 10 and 110 keV usingexposure times of 4.5 s per value of the internal Medipix channel(about 0.2 keV).

To finally give an impression of the overall quality of our Hexadetector, we acquired two images of a plastic lighter in windowmode at two different photon energies (30–35 and 65–70 keV).Furthermore, we discuss the influence of the two bias voltages onthe resulting image quality using the example of the 241Am scan.

3. Results and discussion

The energy response function of nine pixels measured at 350 Vis depicted in Fig. 1. The spectra in this figure have been aligned bymeans of a cross-correlation to enhance the visibility of the peaks. Itdemonstrates the occurrence of not only the charge sharing back-ground, but also of the Ka lines of cadmium (23.2 keV) andtellurium (27.5 keV) plus their associated escape peaks (36.3 and32.0 keV). It follows that the pixel pitch of 165 mm is not largeenough to reduce the number of fluorescence photons that escape apixel down to a level where their influence would be neglectable.

The medians and the standard deviations of the FWHMs for thesingle pixel spectra are shown in Table 1 for the two voltagesinvestigated. It can be seen that increasing the bias voltage from200 to 350 V only amounts to a decrease of the peak width by0.1 keV. However, this goes along with a further movement of thepeak position to lower energies by about 0.6 keV on average, whichis due to charge sharing (data not shown).

The same conclusion can be drawn for the spectra summed overthe whole chip, as listed in Table 2. Furthermore, the increase due tothe summation is found to be only about 20%, a result that iscertainly depending on the quality of the adjustment bits obtainedfrom the lower threshold equalization procedure. Fig. 2 shows thesame spectra as Fig. 1 without alignment to give a visual impressionof the residual variations in the threshold equalization which areresponsible for this behavior.

Table 1Medians and standard deviations of the FWHMs obtained for the single pixel

spectra, averaged over the whole sensor or individual chips.

FWHM (200 V) (keV) FWHM (350 V) (keV)

All chips 2.270.4 2.170.3

Chip 0 2.270.6 2.270.3

Chip 1 2.570.4 2.270.3

Chip 2 2.470.3 2.270.3

Chip 3 2.170.3 2.170.3

Chip 4 2.170.2 2.170.2

Chip 5 2.370.3 2.170.2

Table 2Medians and standard deviations of the FWHMs obtained when summing up all

single pixel spectra over entire chips. The first quantity (‘‘all chips’’) is obtained as

the median and the standard deviation of the other measurements.

FWHM (200 V) (keV) FWHM (350 V) (keV)

All chips 2.770.3 2.570.3

Chip 0 NAa 3.0

Chip 1 2.7 2.6

Chip 2 3.2 2.9

Chip 3 2.5 2.4

Chip 4 2.3 2.1

Chip 5 2.8 2.3

a Unreliable fit, as the spectra added up to two distinct Gaussians.

20 30 40 50 60

0

500

1000

1500

2000

2500

3000

3500

Energy [keV]

Diff

eren

tial C

ount

s

Fig. 2. As in Fig. 2, but without alignment of the spectra in order to show the residual

threshold dispersion.

20 40 60 80 100

0

0.5·106

1.0·106

1.5·106

Energy [keV]

Diff

eren

tial C

ount

s

Fig. 3. Recorded spectrum of an X-ray tube operated at 110 kVp ( noisy line: original

data after differentiation, smooth line: low-pass filtered version (n¼101);

Vbias¼350 V).

Fig. 4. Two flat-field corrected images of a plastic lighter acquired in window mode

at different X-ray energies (left: 30–35 keV, right: 65–70 keV). The salt-and-pepper

noise is despite the flat-field correction and occurs because the bias voltage (200 V)

had not been switched on long enough before the measurements took place ð � 1 hÞ.

Fig. 5. Example of how assembly defects can influence the resulting images at bias

voltages of 200 V (left) and 350 V (right). Blue: low number, red: high number of

counts.

T. Koenig et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) S265–S268 S267

It is worthy to note that the energy resolution does not varysignificantly across the sensor for the 350 V setting, with 2.1 keVobtained for the best chips and 2.2 keV for the worst, which is wellbelow one standard deviation.

The situation changes when one considers the chip spectrawhere all the contributions from individual pixels have beensummed up and peak widths exceeding 3 keV can be observed.Whether this behavior can be changed with a different thresholdequalization or not will have to be investigated in further studies.

The tube spectrum is shown in Fig. 3 and has been obtained bysumming up all pixel spectra for chip 4 in order to increase thesignal-to-noise ratio. It clearly shows that the charge sharingbackground and the CdTe fluorescences distort the actual spectrumpredominantly at lower energies, where almost no X-rays happento pass the X-ray filter. The tungsten Ka line can be found just below60 keV, whereas the Kb line is barely visible at about 67 keV. Theescape peaks encountered in Fig. 1 are now a continuum due to thebroad X-ray spectrum.

Fig. 4 shows the two images of the plastic lighter acquired inwindow mode. As expected, the contrast between metal and plasticparts diminishes at higher energies. The images also show somedefective bump bonds at the bottom and on the left. Apart from thatthe occurrence of assembly defects seems to be well controlled.

The influence of the bias voltage on the resulting images isshown in Fig. 5. Striking is the generally higher number of countsobserved for the higher bias voltage. This behavior is probably due

T. Koenig et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) S265–S268S268

to charge sharing, but may even be a result of an insufficient chargecollection efficiency. Further investigations of this behavior areplanned.

It can also be observed that a higher bias voltage affects the sizeof the non-working detector areas negatively. Furthermore, thepixels in the periphery of these defects show higher counts in thissingle threshold measurement than the areas free of (visible)defects, which indicates that charges generated near the defectsget deflected towards the sides. Note that while the border pixelswith higher counts can be dealt with by applying a flat-fieldcorrection, those in the interior show zero counts and require aninterpolation during image reconstruction.

Although the reduction of dead pixels is generally favorableespecially when it comes to computed tomography, the lowerdetection efficiency along with the movement of the peak positionis probably more serious, and using the higher voltage setting isadvised.

4. Conclusions and outlook

The results presented in the previous section show that thequality of CdTe single-crystals available on the market is now goodenough to facilitate the production and the use of larger monolithicsensors ð4:2� 2:8 cm2Þ than previously possible. We have demon-strated that Medipix2 detectors with a pixel pitch of 165 mm and a1 mm thick CdTe sensor can reach energy resolutions of about2.1 keV under near optimal conditions, and that an average of2.2 keV is a realistic value for assemblies like the Hexa detectorcharacterized in this study. In conjunction with bigger pixelsthis leaves room for increasing the sensor thickness to improveabsorption.

We have also shown that a value of 200 V for the bias voltage isnot appropriate to operate this combination of detector and sensor.Although the decrease in spectral resolution amounts to only0.1 keV and the influence of assembly defects on the images getssignificantly smaller, a lower number of counts was observed forthe lower bias voltage, and so the use of this setting is discouraged.It will be the subject of further studies to investigate the optimaltrade-off between spectral resolution, dead areas and photondetection efficiency for imaging applications.

Acknowledgements

The authors would like to thank Gisela Anton and Thilo Michel(both ECAP Erlangen) for helpful discussions and the provision ofthe Am source at their institute.

References

[1] J. Schlomka, E. Roessl, R. Dorscheid, S. Dill, G. Martens, T. Istel, C. Bumer,C. Herrmann, R. Steadman, G. Zeitler, A. Livne, R. Proksa, Experimental feasibilityof multi-energy photon-counting k-edge imaging in pre-clinical computedtomography, Phys. Med. Biol. 53 (15) (2008) 4031 doi: S0031-9155(08)75296-5.

[2] X. Llopart, M. Campbell, R. Dinapoli, D.S. Segundo, E. Pernigotti, IEEE Trans. Nucl.Sci. NS-49 (5) (2002) 2279.

[3] L. Tlustos, R. Ballabriga, M. Campbell, E. Heijne, K. Kincade, X. Llopart, P. Stejskal,IEEE Trans. Nucl. Sci. NS-53 (1–2) (2006) 367.

[4] J. Durst, P. Bartl, B. Kreisler, T. Michel, G. Anton, Monte Carlo simulation ofpixelated photon counting X-ray detectors like the Medipix2 and the Medipix3using high-Z sensor materials, in: IEEE Nuclear Science Symposium ConferenceRecord, 2008, 2008, pp. 4990–4994.

[5] Z. Vykydal, J. Jakubek, S. Pospisil, Nucl. Instr. and Meth. 563 (1) (2006) 112.[6] T. Holy, J. Jakubek, S. Pospisil, J. Uher, D. Vavrik, Z. Vykydal, Nucl. Instr. and Meth.

A 563 (1) (2006) 254.