Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
Contents lists available at ScienceDirect
Nuclear Instruments and Methods in Physics Research A journal homepage: www.elsevier.com/locate/nima
Room-temperature spectroscopic performance of a very-large area silicon drift detector G. Zampa a,, R. Campana b,c,, M. Feroci b,c, A. Vacchi a, V. Bonvicini a, E. Del Monte b,c, Y. Evangelista b,c, F. Fuschino d, C. Labanti d, M. Marisaldi d, F. Muleri b, L. Pacciani b,c, M. Rapisarda b,c,f, A. Rashevsky a, A. Rubini b,c, P. Soffitta b, N. Zampa a, G. Baldazzi e, E. Costa b, I. Donnarumma b, M. Grassi g, F. Lazzarotto b, P. Malcovati g, M. Mastropietro h, E. Morelli d, L. Picolli g a
INFN – Sezione di Trieste, Padriciano 99, I-34127 Trieste, Italy INAF/IASF-Roma, Via Fosso del Cavaliere 100, I-00133 Roma, Italy INFN Sezione di Roma 2, Via della Ricerca Scientifica 1, I-00133 Roma, Italy d INAF/IASF-Bologna, Via Gobetti 101, I-40129 Bologna, Italy e Universita di Bologna, Dip. di Fisica, viale Berti Pichat 6, I-40127 Bologna, Italy f ENEA Frascati, Via Enrico Fermi 45, I-00044 Frascati, Italy g Universita di Pavia, Dip. di Ingegneria Elettrica, Via Ferrata 1, I-27100 Pavia, Italy h CNR, Istituto Metodologie Inorganiche e dei Plasmi, Via Salaria km 29.300, I-00015 Montelibretti, Italy b c
a r t i c l e in f o
abstract
Article history: Received 28 October 2010 Accepted 15 December 2010 Available online 21 December 2010
Silicon drift detectors (SDD) of small dimensions (up to 1 cm2) have been successfully employed in X-ray spectroscopy due to their small anode geometry, which allows to minimize the electronic noise due to the readout device. Many applications, however, require large sensitive areas to be covered (e.g. X-ray astronomy), so that these detectors are effectively impractical. We present the spectroscopic performance of a 53 cm2 sensitive area, multi-anode SDD, measured at room temperature using an eight-channel readout setup. The measurements, taken using 55Fe and 241Am sources, and X-ray tubes generating energies down to 2 keV, show energy resolutions in the range 290–570 eV FWHM, at 20 1C, depending on the number of anodes collecting the signal. Further developments we are carrying out could improve the detector characteristics and allow to approach the performance of small area SDDs. & 2010 Elsevier B.V. All rights reserved.
Keywords: X-ray spectroscopy Silicon drift detectors Room temperature
1. Introduction The aim of this work was to realize a multi-anode measurement setup to characterize the spectroscopic performance of a very-large area Silicon Drift Detector (SDD) in the frame of a R&D project devised to produce detectors, and electronics, to build both highenergy astrophysics instrumentation (X-ray All-sky Monitor, large-area X-ray timing, Gamma Compton Camera), and SinglePhoton-Emission Computed Tomography systems based on a Compton Camera for medical imaging. The importance of this study stems from the fundamental limitation of today spectroscopic SDDs, namely their small sensitive area of the order of 1 cm2 at most. On the contrary, the applications we are addressing, and many more that one can think of in the field of X-ray spectroscopy, require to cover sensitive
Corresponding author. Principal corresponding author at: INAF/IASF-Roma, Via Fosso del Cavaliere
100, I-00133 Roma, Italy. E-mail addresses:
[email protected] (G. Zampa),
[email protected] (R. Campana). 0168-9002/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.nima.2010.12.129
surfaces at least two orders of magnitude larger than those, providing also for a very good spatial resolution. The following Sections 2 and 3 briefly describe the SDD we are developing and the measurement setup. Section 4 provides a background of the data analysis used to obtain the results discussed in Section 5, while our conclusions are drawn in Section 6, together with a view of the prospectives to improve the detector performance.
2. The detector The very-large area multi-anode SDD we are developing for this application was designed in the framework of the LHC-ALICE experiment at CERN [1–4], and it demonstrated very good performance in localizing the impact point of ionizing particles in a high multiplicity environment [5,6] with a spatial resolution of 30 mm in two dimensions. It features a 53 cm2 sensitive area divided in two halves, each one read out by means of an array of 256 anodes (pitch of 294 mm) placed at the detector edge. A negative HV potential supplies power to the detector from its central cathode, while several integrated voltage dividers provide both the scaled
16
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
potentials to create the constant drift field and the bias of the guard cathodes placed at the sides of the drift region. These are used to scale down the drift potentials to ground in a controlled way. The working principle of the device is the following: the photon absorption creates a cloud of electron–hole pairs that drift along the electric field lines. The holes are collected on the cathodes while the electrons are quickly focused in the middle plane of the detector, and then drift towards the anodes. Because of the diffusion mechanism, the size of the electron cloud increases with time or, equivalently, with the total drift distance as sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi k T x k T 2 s ¼ 2Dt þ s0 ¼ 2 B m þ s20 ¼ 2 B x þ s20 ð1Þ q qE mE where D is the diffusion coefficient, kB is the Boltzmann’s constant,
m is the electron mobility, T is the absolute temperature, t is the drift time and x is the drift length. The Einstein’s relation between D and m has been used to derive the final result. This phenomenon has an important effect on the energy resolution of the detector, as will be shown in Section 4, degrading the spectroscopic performance with increasing drift lengths. Due to a capacitance coupling between the anodes and the last drift cathodes, a large common mode noise has been observed on the output signal already with a single-channel setup [7,8], which nonetheless obtained an energy resolution of 390 eV FWHM at 0 1C for the Mn Ka peak. The common mode noise was reduced somewhat by decoupling these cathodes on-board the PCB hosting the detector, but better performance can be obtained removing the common mode noise from the data by measuring it with a multi-channel setup.
3. The measurement setup The spectroscopic performance of the Silicon drift detector was studied at the IASF-Roma X-ray facility [9] by means of both X-ray
tubes and 55Fe and 241Am radioactive sources, and at the INFNTrieste Silicon Detector Laboratory using a 55Fe source. The X-ray facility employs various choices of tubes and crystal polarizers that provide monochromatic 100% polarized X-ray lines by means of 451 Bragg diffraction. The photon beam can be collimated using interchangeable capillary plates and an adjustable shutter, offering a low divergence beam ( r 11). Fig. 1 shows a simplified block diagram of the measurement setup. Eight contiguous anodes of the SDD were connected to as many low gate capacitance SF-51 JFETs (CGS ¼0.4 pF) used as the input transistor of Amptek A250F-NF charge sensitive amplifiers (CSA). The feedback capacitor CF ¼ 50 fF and a reset transistor are both integrated on the JFET die, allowing to reduce the input stray capacitance for a better noise performance. Nonetheless, some stray capacitance is unavoidable due to the need to adapt the small SDD anode pitch to the much wider one of the electronics (Fig. 2). JFETs and CSAs are both powered by means of batteries in order to minimize the non-fundamental noise. To exploit the transistor reset strategy in the most profitable way, a coordination of the reset signals must be enforced. To this aim, the output ramps of all preamplifiers are compared with a threshold Vhigh, and the first comparator that fires (the one connected to the leakiest anode) starts the reset phase: all feedback capacitors are discharged to Vlow by activating a feedback loop built around the reset transistor. The frequency of the reset signal can be used to monitor the largest of the leakage currents at the anodes, which anyway are very uniform as shown by the detector characterization preliminary measurements. The preamplifiers outputs are then connected to a 16 channel NIM CAEN spectroscopy amplifier (N568B) via output buffers able to drive 50 O cables. High-pass filters and switches are placed at the input of the buffers to reduce as much as possible the large reset transients, and thus the dead time of the N568B associated with them. A CAEN VME 32 channel peak sensing ADC (V785) is used to acquire the data upon
Fig. 1. Simplified block diagram of the eight-channel readout setup.
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
17
signal[i] = ADC[i] - offset[i] [ADC units]
100 80 60 40 energy threshold
20
event baseline
-20
7 channels
0 Fig. 2. Close-up of the front-end board used to connect eight anodes to the JFETs and to adapt the anode pitch to that of the electronics.
the activation of a 250 ns gate signal generated by discriminating one of the very fast outputs of the N568B (rise-time of 25 ns). The detector and the front-end and biasing electronics are placed in an aluminum box which is fixed to the sample holder of the X-ray tube facility. This provides a six axis micro-positioning motorized system that allows to control the relative position of the detector with respect to the collimated photon beam, so that it is possible to study the detector performance in great detail. The X-ray tube facility is situated inside a lead shielded room, which is kept at an almost constant temperature of 17 1C by means of an air conditioner. Since the detector is placed into the box, its temperature is few degrees higher than that of the room (an estimation of 3 1C can be made based on the variation of the detector bias current with temperature). All of the setup parameters and the micro-positioning stage motors are controlled by means of a computer, which is also responsible of the data acquisition. During the data taking only seven channels where operative because of a defective JFET that was not replaced in order to avoid possible damaging of the other channels. 4. Data analysis The data analysis is aimed at the characterization of both the spectroscopic and the imaging performance of the detector with respect to X-ray photons in the energy range from 2 to 60 keV. In this article we report on the spectroscopic related issues, while the imaging considerations and results are detailed on a companion paper [10]. Since the detector presents a non-negligible common mode noise, and the readout channels are few, event simulations were carried out both to optimize the analysis software and to evaluate the intrinsic performance of the system. To be conservative, the noise parameters used in the simulations (ENC ¼29 e RMS, common mode noise of 125 e RMS) are slightly higher than the actual ones, these latter being estimated by the measurements. Fig. 3 shows a 4 keV event, and highlights the problems related to the limitation on the number of readout channels (only seven here), that is discussed in the next section. 4.1. Cluster analysis One of the most important tasks performed by the analysis is to determine the level of the common mode noise on an event-by-event basis. This can be viewed as a time-varying baseline: in fact, the output voltage of a channel can be expressed as Vi ¼ Pi þCM þ Ni þ Si
ð2Þ
calculated baseline
0
5
10
15
20
25
30
Channel number Fig. 3. 32-Channel simulation of a 4 keV event. Noise parameters (ENC ¼29 e RMS, baseline noise of 125 e RMS, or 450 eV RMS) are slightly higher than the actual ones. The simulation shows the data-analysis difficulties arising from the small number of readout channels.
where Pi is the pedestal, Ni is the uncorrelated channel noise, and Si is the signal of the i-th channel, while CM is the common mode noise baseline. The pedestal Pi itself is composed of two terms, the selectable offset voltage of the spectroscopy amplifier used to adjust its outputs to the ADC input range, and an offset proportional to the average input voltage of the N568B due to the differentiator circuit of the shaper. Since Ni and CM can be considered to have zero average, it is straightforward to determine the pedestals when no signal is present. This can be accomplished by switching off the tube, shielding the radioactive source, or using an oscillator trigger in anti-coincidence with the signal. Subtracting the pedestals from the data, the common mode noise can be estimated using the channels where no signal charge was integrated: CM ¼
M 1X ðVi Pi Þ M i
ð3Þ
where i spans the whole set of channels excluding the triggering one and any other channel above a given threshold, M being its number. A cluster finding algorithm is used to distinguish between baseline channels and signal channels, thus we refer to this technique as cluster analysis. The standard deviation of the distribution of the baseline error depends both on the noise present on the channels, sNi , and also on the distribution of the signal residual tail on the baseline channels (i.e. the signal charge collected by anodes whose output voltage falls below threshold): vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u M u 1 X s2 : ð4Þ sCM ¼ t 2 s2Ni þ tail M2 M i The first term is the variance of the mean of M values, each one having a standard deviation sNi , while the second term is the variance of the signal tail distribution, whose amount is divided by M when performing the average calculation. Note that stail is correlated to the noise of the channels where the signal falls below threshold, and this must be considered when calculating this contribution. Also, very noisy channels, if they exist, should be excluded from the event baseline calculation, because they could significantly bias its value. The spectrum of the source is then obtained accumulating an histogram of the photon energy calculated summing up the contribution of all the remaining channels, after the baseline has been subtracted from each one of them. Clearly, the energy resolution of the SDD is not constant with respect to the position where the photon has been absorbed, simply because the number
18
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
of anodes that collect the corresponding charge varies due to the diffusion of the cloud of electrons as the drift length increases. vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi uNM M uX E ðNMÞ2 X N2 t sE ¼ s2Ni þ s2Nj þ 2 s2tail þ F EI E: ð5Þ 2 N M M j¼1 E Etail i ¼ 1 M The resolution is expressed by Eq. (5) which includes the contribution of the noise of the N M signal channels (first term, N being the total number of readout anodes), the one of the baseline error (second and third terms), and that of the signal statistics (last term, where F 0:11 and EI ¼3.6 eV are the Fano factor and the Ionization Energy of Silicon). E is the true energy of the photon, and Etail is the average value of the energy distribution of the signal tail, and all energy and noise terms are expressed in eV units. The factor that multiplies the square root is an energy renormalization term due to the shift of the spectrum to lower values caused by the signal tail on the baseline channels, and it represents a further contribution to the resolution worsening (Fig. 4a). Two issues remain to be further addressed: the gain calibration of the various channels needed to normalize the system response along the anodes, and the selection of the threshold used to determine the presence of a signal. The former can be done taking spectra of known radioactive sources. Looking at the correlation of the various outputs in pedestal events, it is also possible to monitor gain drifts during long acquisition runs. This calibration method is based on the linearity of the system, and exploits the presence of a large common mode noise. The selection of the threshold is a critical task, since it involves a compromise between: i. the need to minimize the signal tail on the baseline channels that would require to maximize the number of channels that integrate the signal (low threshold) and ii. the need to minimize the number of channel measuring the signal in order to obtain the best possible energy resolution (high threshold). The threshold value selection is based on the probability that the noise of the channel under evaluation is above threshold when there is no signal (noise pollution), and the average amount of signal that falls on the tail associated with this threshold. This tail can be determined by a Monte Carlo simulation, using a Gaussian distribution for the noise, and a uniform one for the signal. As an example for a single channel, using thresholds of 0, s, 2s and 3s, one obtains average energy tails of 0:626s, 0:888s, 1:24s and 1:67s, and noise pollution probabilities of 50%, 15.87%, 2.28%, and 0.135%, respectively, in the hypothesis that the baseline is known
exactly. The total energy that is not measured depends on the number of signal channels that fall below threshold. 4.2. Charge distribution reconstruction analysis Despite the cluster analysis is commonly used in multi-channel environments, it presents a strong limitation in the case of this SDD. Contrary to (micro-)strip Silicon detectors where the charge diffusion is negligible, as the photon conversion point moves out from the anodes along the drift direction, an increasing number of channels begin to fall below threshold giving rise to a larger amount of signal assigned to baseline channels. This is evidenced by a shift of the spectrum towards lower energies, which at first sight could be ascribed to energy loss in the detector. For this reason, we used this approach only during calibration, to select single-channel events when the source is placed nearby the anodes. For all the other aspects of the analysis we opted for a different solution which reconstructs the distribution of the signal charge at the anodes, hence helping the baseline determination by exploiting the information contained in the size of the event. The cloud of electrons generated by the photon conversion has a Gaussian distribution, having a width proportional to the square root of the drift time, which is sampled by the anodes. If Qtot is the total charge, then the photon energy can be estimated, after the pedestal subtraction, by fitting the event data with the function: Qtot i þ 0:5X0 i0:5X0 pffiffiffi pffiffiffi erf FðiÞ ¼ K þ erf ð6Þ 2 s 2 s 2 where K is a constant representing the common baseline, X0 and s are the position and the standard deviation of the charge distribution, and i is the position of the center of the anode. We use a scale where the pitch has been normalized to one. K, Qtot, X0 and s are the free parameters of the fit. The energy resolution of the detector is still given by an expression similar to Eq. (5), with the exception that now the signal tail is negligible as illustrated by Fig. 4b where it is possible to see that the signal peaks of the simulations, representing photons absorbed at different drift distances, are now aligned indicating that the fit allows to recover the signal in the channels below threshold. A residual offset of the spectrum close to the anodes, however, can still be seen, but in this case the shift is in the opposite direction with respect to a signal tail that would give a lower total energy, and is not yet clearly understood. Note that this approach does not require a threshold to distinguish between signal and noise, but conversely it has a limitation on the minimum signal that can be measured, since a Gaussian distribution has to be discernible against the baseline noise.
1.25 mm 11.25 mm 21.25 mm
6000
5000 Counts
5000 Counts
1.25 mm 11.25 mm 21.25 mm 31.25 mm
6000
4000 3000
4000 3000
2000
2000
1000
1000
0 0
100 200 300 400 500 600 700 Energy [ADC units]
0
0
100 200 300 400 500 600 700 Energy [ADC units]
Fig. 4. Simulations of a Titanium X-ray tube with a PET crystal generating lines at energies multiple of 2.006 keV read out by a seven-channel setup. The spectra were obtained using different analysis algorithms to determine the event photon energy. The two plots show the dramatic improvement of performance it is possible to achieve using the solution that aims at reconstructing the charge distribution at the anodes (right panel), versus the cluster finding algorithm (left panel). See the discussion in Sections 4.1 and 4.2. (a) Cluster analysis and (b) charge distribution Gaussian fit.
2000 1800 1600 1400 1200 1000 800 600 400 200 0 260 280 300 320 340 360 380 400 420
19
40 35 RMS [e- ENC]
Counts
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
Measured Cin= 1.0 pF Cin= 1.5 pF Cin= 2.0 pF Cin= 2.5 pF Cin= 3.0 pF
30 25 20 15 0
2
4
Energy [ADC units]
6
8
10
Shaping time [µs]
Fig. 5. Noise performance of the system. Left panel: 55Fe spectrum taken at 20 1C using a shaping time constant of 3 ms. Single-anode events were selected using the cluster analysis (threshold of smean ). The energy resolution is 282 eV FWHM for the Mn Ka line. Right panel: the ENC was simulated for several shaping time constants and input capacitances, while the leakage current was set to 8 pA. (a) 55Fe spectra from a single channel and (b) equivalent noise charge.
900 6000
Energy Resolution [eV FWHM]
1.25 mm 11.25 mm
5000
21.25 mm
Counts
31.25 mm
4000 3000 2000 1000 0
0
100
200
300
400
500
600
700
Energy [ADC units]
800 700 600 500
Sim. 4.012 keV, 7 ch. Sim. 4.012 keV, 32 ch. Measurements 4.012 keV Sim. 10.03 keV, 7 ch. Sim. 10.03 keV, 32 ch. Measurements 10.03 keV
400 300
0
5
10
15
20
25
30
35
Drift distance [mm] Fig. 6. Spectra of the X-ray tube with PET crystal obtained at 20 1C from a scan of the drift distance.
Although this analysis method is more computationally intensive, it provides information on the position of the photon conversion by means of the X0 and s parameters, allowing to obtain a very good imaging performance along the anode coordinate, and even a coarse one along the drift [10].
5. Results and discussion A scan of the shaping time was performed to characterize the noise of the detector/front-end system. Non-collimated 55Fe spectra were measured at a detector temperature of 20 1C using shaping time constants tsh ¼ 1, 3, and 6 ms, and the data were analyzed with the cluster method to select the events where the signal was collected by a single anode (threshold of one smean ). The maximum leakage current recorded during the measurements was C 10 pA. Fig. 5a shows a spectrum for the case tsh ¼ 3 ms. The Mn Ka and Mn Kb peaks were fitted with a Gaussian distribution having an exponential tail at the left to account for the reduced-signal events due to the threshold, and are superimposed to a step-like function which is used to represent the effect of multiple Compton interaction events. The energy resolution calculated for all the channels ranges from 270 to 300 eV. Fig. 5b shows the corresponding measured equivalent noise charge (ENC, circles), and SPICE simulations of a channel circuit (curves). The latter take into account only the principal contributors to the noise (the detector leakage current, the noise originated
Fig. 7. Simulated and measured energy resolution at various drift distances for the 4.012 and 10.03 keV lines.
in the JFET and its drain bias resistor, and that of the output buffer), and the filtering functions (the high-pass filter and the second order pulse shaping of the N568B). The total input capacitance and the leakage current were varied in the simulations to estimate their actual values. The measured data are in agreement with a Cin ¼3 pF and a leakage current ILeak ¼8 pA, which is compatible with the measured period of the reset signal. The optimum shaping time constant is close to 3 ms, so we used this value for the detector performance characterization, resulting in an ENC of about 24 e RMS. The event baseline (i.e. the common mode noise) has a distribution with an RMS of 350 eV, more than three times larger than the channel noise standard deviation. Note that the optimization of the input stray capacitance, e.g. using an integrated frontend placed as close as possible to the anodes, allows a considerable improvement of the electronic noise if the input transistor transconductance can be made equal to that of the JFET. A Titanium X-ray tube with a PET crystal, providing Braggdiffracted photons with energies multiple of 2.006 keV, was used to investigate the energy resolution of the detector. The photon beam was collimated to provide a spot of about 2.5 2.5 mm2 on the detector, obtaining in this way a relatively high output flux. The measurements, taken at 20 1C, were performed at distances from the anodes of 0, 1, 2, and 3 cm (Fig. 6) to study the dependence of the energy resolution on the charge diffusion. As shown by Fig. 7, the seven-channel simulations are in agreement with the measurements, providing an energy resolution slightly worse than the actual one (remember that the noise in the
20
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
Table 1 FWHM energy resolution in eV units of the peaks of Fig. 6 along with the results obtained from 32-channel simulations. 2.006 keV
6.018 keV
8.024 keV
10.03 keV
Meas.
Sim.
Meas.
Sim.
Meas.
Sim.
Meas.
Sim.
Meas.
Sim.
372.3 7 1.6 483 7 6 – –
368.9 7 1.3 443.9 7 1.4 510.17 1.6 560.87 2.0
409.2 71.3 538.4 71.6 641.8 71.7 752 73
385.9 7 1.4 466.2 7 1.4 521.6 7 1.7 556.0 7 1.9
425.5 7 1.8 553.3 7 2.1 6407 3 773 7 6
396.2 71.4 480.0 71.4 518.3 71.6 558.5 71.7
443.3 7 1.9 574.6 7 1.9 664.9 7 2.6 801 7 5
402.8 7 1.4 487.8 7 1.5 520.9 7 1.6 563.1 7 1.8
468 7 4 593 7 4 688 7 5 8407 9
409.1 7 1.4 492.5 7 1.5 526.1 7 1.6 564.6 7 1.7
Photon energy [keV]
0 1 2 3
4.012 keV
60
Integral non linearity [%]
Position (cm)
50 40 30 20 10 0
0
500 1000 1500 2000 2500 3000 3500 Measured energy [ADC units]
10 5 0 -5 -10
0
10
20
30
40
50
60
Photon energy [keV]
Fig. 8. Linearity at a drift length of 3 cm (worst energy resolution). The plots were made using both X-ray tube and non-collimated 241Am source data (lines at 10.551, 11.87, 13.95, 26.35, and 59.54 keV). The circles represent the energy in ADC units (left panel) and the errors from the linear fit in percent units (right panel), while the error bars show the FWHM energy resolutions. (a) Linearity of the energy measurement and (b) error from the fit.
8000 Pedestal 7000 6000 Counts
5000 4000 3000 2000 1000 0
1
2
3
4 5 6 Energy [keV]
7
8
9
Fig. 9. Spectra of the X-ray Titanium tube with PET crystal obtained by analog subtraction of the common mode noise from the triggering signal. The spectrum corresponds to single-channel events selected with the cluster analysis.
simulations is higher). Furthermore, the energy resolution obtained from 32-channel simulations are representative of the intrinsic resolution one can obtain with the SDD, being entirely due to the fundamental noises, the statistics of the signal, and the number of channels integrating the charge. Table 1 provides the values of the energy resolution for the measurements and the 32-channel simulations. It is worth to point out that the energy resolution here demonstrated can be significantly improved both by cooling the detector to reduce the leakage current and by using an integrated front-end electronics, which allows to reduce the input stray capacitance while maintaining a fairly high transconductance of the input transistor at a lower power consumption. To this end it is possible to use a CMOS technology since very low 1/f noise processes are already available (e.g. the 0:35 mm C35B4M3 from AustriaMicroSystems).
In order to evaluate the linearity of the system on the whole range of energies required by the applications we are considering, the spectrum of a 241Am source was measured. The emission lines with energies of 10.55, 11.87, 13.95, 26.35, and 59.54 keV were used, along with the X-ray tube data, to determine the integral nonlinearity (INL), Fig. 8. By fitting the data with a straight line crossing the origin of the coordinate axis, we determined an INL better than 2%, which is negligible in comparison to the energy resolution. Notice that, because of the very poor noise filtering on the fast output of the N568B used to generate the trigger, a high discriminator threshold is needed to reject the false triggers, hence the 2.006 keV line is suppressed when moving the beam away from the anodes. A way to extend downward the energy range consists in the analog subtraction of the common mode noise from the triggering anode, using one or more of the channels placed at a suitable distance from it. Filtering this signal with a shaping time constant close to the optimum one, and using a constant fraction discriminator to adjust the timing, we were able to obtain the spectra of Fig. 9. The 1.49 keV fluorescence peak due to the aluminum box containing the detector can be easily distinguished in the single-channel spectrum, as well as the pedestal. The latter illustrates the good noise performance of the detector even at room temperature.
6. Conclusions and outlook The room-temperature (20 1C) spectroscopy performance of a 53 cm2 Silicon drift detector has been presented. Within the limitations of a seven-channel readout setup, and the large common mode noise observed, the SDD demonstrated an energy resolution in the range between 270 and 843 eV FWHM, for energies up to 10 keV, depending on the number of anodes collecting the signal charge. Simulations of a 32-channel setup show that the energy resolution can be improved by measuring with more precision the baseline (which is time-dependent due to the common mode noise), resulting in a resolution better than 570 eV FWHM. Further improvement can
G. Zampa et al. / Nuclear Instruments and Methods in Physics Research A 633 (2011) 15–21
be made minimizing the input stray capacitance by using an integrated front-end. The SDD is able to detect photons with energies lower than 2 keV, the actual limitation being the high noise on the triggering signal. Already in its present layout, the SDD can be profitably used for applications requiring modest spectroscopy resolutions, but verylarge sensitive area. Of course, the performance of the SDD can be improved revising both the detector itself and the electronics. We are working on the detector layout to remove the coupling of the integrated divider noise to the anodes. This will be accomplished disconnecting the last cathodes from the divider, and biasing them from outside. We are also increasing the SDD thickness to improve its sensitivity at higher energies, and this will also decrease the anode capacitance. Finally, we have designed and produced a first 32-channel frontend ASIC prototype: using this device it will be possible to reduce
21
the stray capacitance improving the overall performance of the system. The design and performance of the ASIC will be reported in a future article.
References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
A. Vacchi, et al., Nucl. Instr. and Meth. A 306 (1991) 187. A. Rashevsky, et al., Nucl. Instr. and Meth. A 485 (2002) 54. C. Piemonte, A. Rashevsky, A. Vacchi, Microelectron. J. 37 (2006) 12. P. Burger, C. Piemonte, A. Rashevsky, A. Roncastri, A. Vacchi, INFN/TC-02/07. S. Beole´, et al., Nucl. Instr. and Meth. A 582 (2007) 733. E. Crescio, et al., Nucl. Instr. and Meth. A 539 (2005) 250. G. Zampa, A. Rashevsky, A. Vacchi, IEEE Nucl. Sci. Symp. Conf. Rec. 6 (2006) 3790. G. Zampa, A. Rashevsky, A. Vacchi, IEEE Trans. Nucl. Sci. NS-56 (2009) 832. F. Muleri, P. Soffitta, R. Bellazzini, et al., Proc. SPIE 7011 (2008) 84. R. Campana, et al., Nucl. Instr. and Meth. A, in press, doi:10.1016/j.nima.2010. 12.237.
