Energy resolution improvement in room-temperature CZT detectors
EEnergy resolution improvement inroom-temperature CZT detectors
Y. Ramachers ∗ and D.Y. Stewart † Dept. of Physics, University of Warwick, Coventry CV4 7AL, UK
Abstract
We present methods to improve the energy resolution of single chan-nel, room-temperature Cadmium-Zinc-Telluride (CZT) detectors. A newpreamplifier design enables the acquisition of the actual transient currentfrom the crystals and straightforward data analysis methods yield unprece-dented energy resolution for our test-detectors. These consist of an eV-CAPture Plus crystal as standard and 1 cm cube Frisch collar crystalscreated in-house from low-grade coplanar grid detectors. Energy resolu-tions of 1.9% for our collar detectors and 0.8% for the eV crystal at 662keV were obtained. The latter compares favourably to the best existingenergy resolution results from pixel detectors.
The motivation to research CdZnTe (CZT) detectors originates from participationin the COBRA experiment [1], a proposed massive (several hundred kg) arrayof CZT crystals for double-beta decay research. Taking into account a typicalmass of merely a few grams for each crystal, several tens of thousands of crystalswill eventually have to be operated reliably over several years. Naturally, sucha set-up would become vastly more practical by utilising simple ways to mountand operate individual crystals in the array.Single-channel readout for each crystal as opposed to coplanar grid detectorsis considered to be an attractive option. Already the reduction of wiring closeto the crystals by a factor of two would be highly significant in this case. Frischcollar detectors [2] appeared to be the most practical way forward, so we modifiedthree existing coplanar grid crystals, low-grade from eV-Products . In order tohave a standard to compare with, we purchased one eV-CAPture Plus technologydetector which is of much higher quality [3]. ∗ E-mail: [email protected] † Email: [email protected] Purchased for the COBRA experiment and kindly provided by our collaboration partners. a r X i v : . [ phy s i c s . i n s - d e t ] S e p Experiment
Crystals and preamplifier are housed together in a standard diecast enclosurefeaturing connectors for preamplifier power, BNC output and High-voltage input(see figure 1). A single HV-power supply (Ortec 659, NIM module) delivers bothpolarities up to 5kV. Two linear DC power supplies deliver ±
5V to the preampli-fier (see figure 2). The output signal is connected directly to the data acquisitionsystem (DAQ) using a 50Ω BNC cable (RG58). The DAQ consists of a 100 MHzsampling digital oscilloscope in a 3U PXI module from National Instruments (NIPXI-5112) mounted in a PXI crate (NI PXI-1042) and is controlled by an em-bedded controller PC (NI PXI-8186) running custom-made LabView software fordigital pulse acquisition. Pulses are streamed directly to disk in binary format formaximum sampling speed when using radioactive sources for detector calibration.The dynamic range is limited to 7-bit (8-bit oscilloscope, both polarities mea-sured), hence each acquisition needs a little fine-tuning for the vertical range tocapture all important structures. We emphasize this point since this has becomean issue when calibrating with a Ba-133 source, see figure 8. The 81 keV line res-olution is limited by digitisation noise rather than our analysis method whilst theline voltage amplitude is simply too small when the system is set-up to acquirethe 356 keV line simultaneously.The core of the data acquisition system is the preamplifier design, see figure 2.It represents a rather unusual method for readout of semiconductor pulses since itis not a charge-sensitive amplification system. This preamplifier can be describedas a straightforward source-follower. The motivation to try this type initially wasto learn more about signal formation in the semiconductor, i.e. to specialise theanalysis and measurement on pulse-shape in contrast to pulse-height.The preamplifier is optimised for high-speed, consistent with the data ac-quisition bandwidth, and highest possible signal–to–noise ratio. Note that bothproperties are almost mutually exclusive, i.e. speeding up the amplifier worsensthe signal and better signal–to–noise ratio slows the amplifier. However, for ap-plications that require a higher signal gain, focussing on low-energy signals forexample, it is possible to increase the gain without losing an equal factor in speed.Our application for this set-up, however, targets rather higher energy signals ofup to several MeV, hence the circuit in figure 2 is expected to work optimally forus. The existing three coplanar grid detectors were modified to work as Frischcollar detectors. All three are cubes of volume 1cm with gold-plated electrodes,a full area cathode and a coplanar grid anode structure. All faces (excluding thecathode face) are covered with insulating paint. Since we were not interested inoperating the grid, paint covering the anode was partly removed (by dissolving itin ethanol) and the area used to contact the anode with a wire was attached by agenerous drop of silver conductive paint. The preamplifier is AC-coupled to thecrystal anode which is biased positively by the HV-power supply. The cathode2igure 1: Picture of the set up, containing two preamplifiers either for readoutof two of our in-house Frisch collar crystals (shown here) or two-channel readoutof anode and cathode signals simultaneously.is kept at ground potential. This mode was used to operate and test the crystalsas simple planar detectors.For the Frisch collar mode, each crystal was wrapped in two layers of thinteflon tape, covering the full height, leaving out anode and cathode, similar todevices fabricated in [2]. The teflon layer was wrapped in a metal foil (alu-minium kitchen foil worked best for us) and the foil attached via a small ’lip’to the cathode (using silver conductive paint). The metal foil height determinesthe performance of the Frisch collar detector [2]. We achieved best operatingperformance with 9mm - 9.5mm foil height. Any higher shield results in stabil-ity problems when biasing the anode since the grounded shield appears to betoo close, particularly at the cube corners. Finally, prepared crystals can bemounted for operation on a ground plane. We used a small copper-clad piece ofprinted-circuit board connected to the preamplifier ground, see figure 1.The eV-Products CAPture Plus technology crystal is essentially a Frisch collardetector using a resistively coupled shield instead of a capacitively one as out-lined above. It has, however, one important feature in addition, a small anode,utilising the ’small pixel’ effect [6]. The ’best of both worlds’ has been combined3igure 2: The pre-amplifier circuit diagram used in this work. The JFET J201can in principle be replaced with any FET input-stage, however low intrinsic noiseis the crucial selection criterium. The 3kV capacitor at the input represents theanode AC-coupling.with this type of detector in order to achieve an electron-signal-only operationwith a single channel readout. The surface area of the rectangular crystal is 1cm and its height is quoted as 5mm. It arrived factory–certified with an energy reso-lution of 2.20% at the Cs-137 line at 662 keV and 4.40% at the Co-57 line at 122keV. According to [3], an energy resolution of 1.5% at 662 keV has been achievedunder favourable conditions. The crystal is contacted and biased identically toour modified crystals, except that this crystal can withstand higher bias than theFrisch collar crystals (recommended bias is 1.5 kV). Note that an alternative de-sign, utilising the capacitively coupled shield and a small anode has been realisedin [4] with excellent results.In case anode as well as cathode signals are required to be measured with twopreamplifiers, it is important to raise the ground plane (PCB) off the diecast boxfloor to minimise capacitance for the cathode signals. A 1cm insulating spacershould be sufficient. The two-channel readout serves to gain further insight intothe signal formation by picking up anode and cathode signal simultaneously fora single event. The ratio of these signals is expected to yield depth information[5] and its sum could improve energy resolution by boosting the total amount ofcharge collected. So far, we utilised this type of readout only for the eV crystal.Further research into this mode of operation is in progress.4 Results
Optimising energy resolution for Frisch collar detectors is fully documented in[2]. Performance of crystal operation in this mode was successfully re-produced.However, in this process we noticed that our absolute energy resolutions weresurprisingly good compared to the literature. Note that our crystals are re-markably large for CZT crystals operating as Frisch collar detectors and thecrystal quality was expected to be poor compared to spectroscopy-grade crys-tals. Additionally, our readout was custom-made for pulse-shape analysis andnot specialised for spectroscopy operation. Obtaining energy information fromintegrated digitised pulses is always considered to be inferior to analogue oper-ations (current integration on a charge-sensitive preamplifier and spectroscopyshaping). This ’common-knowledge’ can lead to interesting preamplifier designscombining charge and current sensitive readout [7]. Taking all of this into con-sideration, we were surprised to see our initial energy resolutions comparable topublished results. In order to achieve further improvements the following dataanalysis method has been developed.Frisch collar detector fabrication inevitably leads to high input capacitancesrelative to planar crystals. The metal shield wrapped around the bulk of thecrystal functions by collecting induced charges from any current flowing insidethe capacitively coupled crystal. Thereby, it prevents the majority of the signalappearing on the signal electrodes until the charge cloud comes close to the anode,hence achieving close to electron-only signal formation.A high input capacitance coupled to the source-follower preamplifier effec-tively smears or low-pass–filters the image of the current flowing in the detectorcrystal (interpreted here as a current source). Therefore our signal (see figure 3)is not a true transient current measurement initially. However, a CR-RC filterapplied to the digitised pulse can enhance the fast components in the signal byeffectively transforming the pulse to its rate-of-change image. After shaping, thepulse more closely represents the actual transient current flowing in the crystal.We found that for the given preamplifier settings and the data acquisition band-width, time-constants of 200 ns for the high-pass and 4 µ s for the low-pass yieldsatisfactory results for all our cystals. The result of such a shaping is shown infigure 4 for a typical pulse.Naturally, this shaping process enhances the high-frequency noise, so we de-cided to apply a second step to this two-stage software filtering in the form of amoving average filter (width 200 ns). Note that a moving average filter belongs tothe class of optimal filters, where the moving average filter represents the optimalmethod to preserve any steep, fast-changing feature in a pulse while smoothinghigh-frequency noise . Figure 5 shows its effect on the pulse from figure 4.The final analysis step involves calculating basic pulse parameters which serve All digital filter algorithms as described in the text have been taken from [8].
A new readout and data analysis method for single channel, room-temperaturesemiconductor detectors is introduced which significantly improves energy resolu-tion despite its simplicity. The new preamplifer circuit and data analysis methodshave been discussed in detail. The application to three potentially rather poorCZT crystals operated as Frisch collar detectors yields energy resolutions com-parable to or better than existing values from literature either for Frisch collardetectors or coplanar grid detectors of similar size (but often, where mentioned,far superior crystal quality). Our chosen test standard for comparison, an eV-Products CAPture Plus detector, surpasses all expectations and shows energyresolutions, to the best of our knowledge, better than any measured with a simi-lar device so far. Energy resolutions for such a large volume CZT detector of wellunder 1% at 662 keV without subsequent corrections, i.e. efficiency losses, haveso far only been reported for much more complicated (in terms of fabrication and9igure 9: Example of a rise-time–pulse height plot (pulse height corresponds tocharge amplitude, see text) for one of the Frisch collar detectors at 800 V biaswith a 9.5mm collar, irradiated with a Cs-137 source. The energy resolutionamounts to 1.9% at 662 keV. As can be seen, a bi-parametric correction likedemonstrated in [11] would be beneficial but could lead to conflicts for our mainapplication, see text.Figure 10: Calibration spectrum taken with a Cs-137 source at 1.0kV bias withthe Frisch collar detector
References [1] K. Zuber,
Phys. Lett.
B519 (2001) 1; T. Bloxham et al., COBRA collabo-ration,
Phys. Rev.
C76 (2007) 025501[2] D.S. McGregor et al.,
Appl. Phys. Lett. (1998) 792; W.J. McNeil et al., Appl. Phys. Lett. (2004) 1988; G. Montemont et al., IEEE Trans. Nucl.Sci. (2001) 278; A. Kargar et al., Nucl. Instr. Meth.
A558 (2006) 497; A.Kargar et al.,
Nucl. Instr. Meth.
A562 (2006) 262; M. Harrison, A. Kargarand D.S. McGregor,
Nucl. Instr. Meth.
A579 (2007) 134[3] D.S. Bale and C. Szeles,
Proc. SPIE
Vol. 6319 (2006); C.Szeles et al., ibid. (2006) 191; see also White Papers at [4] L. Verger et al.,
IEEE Trans. Nucl. Sci. (2005) 1733[5] Zhong He et al., Nucl. Instr. Meth.
A380 (1996) 228;
Nucl. Instr. Meth.
A388 (1997) 180 116] H. Barrett, J. Eskin and H. Barber,
Phys. Rev. Lett. (1995) 156[7] A. Hamrita et al., Nucl. Instr. Meth.
A531 (2004) 607[8] St. W. Smith,
The Scientist and Engineer’s Guide to Digital Signal Process-ing , Cal. Tech. Publ. (1997) USA; see also [9] Gunnink and Arlt,
Nucl. Instr. Meth.
A458 (2001) 196[10] R.M. Keyser,
IEEE Nuclear Science Symposium Conf. Record (2001) 315[11] L. Verger et al.,
Nucl. Instr. Meth.