Investigation of photon counting pixel detectors for X-ray spectroscopy and imaging
Der Naturwissenschaftlichen Fakult¨at der Friedrich-Alexander-Universit¨at Erlangen-Nurnberg¨ zur Erlangung des Doktorgrades Dr.rer.nat.
vorgelegt von Patrick Takoukam Talla aus Yaounde/Kamerun Als Dissertation genehmigt von der Naturwissen- schaftlichen Fakult¨at der Universit¨at Erlangen-Nurnberg¨
Tag der mundlichen¨ Prufung:¨ 07. April 2011 Vorsitzender der Promotionskommission: Prof. Dr. Rainer Fink Erstberichterstatter: Prof. Dr. Gisela Anton Zweitberichterstatter: Prof. Dr. Valeria Rosso Dedication
To my wife Sandrine Takoukam Talla To my parents Jean Baptiste Talla and Monique Talla To my late brother Guy Bertrand Wafeu Talla
i
Contents
1 Introduction 1
2 Interaction of X-rays with Matter 3 2.1 PhotoelectricEffect...... 3 2.2 ComptonScattering ...... 4 2.3 RayleighScattering...... 5 2.4 PairProduction ...... 5 2.5 Interaction of Electrons with Matter ...... 6 2.6 Conclusion ...... 7
3 The Medipix2 Detector 9 3.1 Description and operating Modes ...... 9 3.2 CountingPrinciple ...... 11 3.3 ChargeSharing...... 12 3.4 Conclusion ...... 15
4 The Medipix3 Detector 17 4.1 Motivation ...... 17 4.2 Description ...... 18 4.3 Functionalities and operating Modes ...... 20 4.4 ChargeSummingMode ...... 20 4.5 Conclusion ...... 21
5 Monte Carlo Simulations and Energy Responses of Medipix Detectors 23 5.1 MonteCarloToolROSI ...... 23 5.2 Energy Response of the Medipix2 Detector ...... 24 5.3 Energy Response of the Medipix3 Detector ...... 27 5.4 Impact of Noise Contributions on the Energy Resolution of Medipix Detectors 29 5.5 Conclusion ...... 31
6 Characterization of the Medipix3 Detector 33 6.1 Test Pulses Measurements to determine Optimal DAC Settings ...... 33 6.2 Equalisation Procedure of the Medipix3 Detector ...... 37 6.3 Energy Calibration of the Medipix3 Detector ...... 40 6.4 Energy Resolution of the Medipix3 Detector ...... 42 6.5 Parameters Extraction for Monte Carlo Simulations ...... 43 6.6 Energy Response of the Medipix3 Detector in Charge SummingMode . . 43 6.7 Count Rate Linearity of the Medipix3 Detector ...... 45 6.8 Conclusion ...... 47
iii Contents
7 Spectrum Reconstruction with hybrid Photon counting Detectors 49 7.1 Motivation ...... 49 7.2 Theory...... 50 7.3 Methods...... 51 7.4 Reconstruction with the Medipix2 Detector ...... 57 7.5 Reconstruction with the Medipix3 Detector in Charge Summing Mode . . 67 7.6 Conclusion ...... 71
8 Determination of the kVp with the Medipix2 Detector 73 8.1 Determination of the kVp with multiple Filter Combinations ...... 74 8.2 Determination of the kVp with only one Filter Combination...... 80 8.3 Conclusion ...... 81
9 Introduction to X-ray Imaging 83 9.1 ImageQualityMetrics ...... 83 9.2 Imaging with the Medipix3 Detector ...... 85 9.3 Conclusion ...... 89
10 Spatial Resolution of the Medipix3 Detector 91 10.1Methods ...... 91 10.2 Spatial Resolution of the Medipix3 Detector in Single Pixel Mode . . . . . 93 10.3 Spatial Resolution of the Medipix3 Detector in Charge SummingMode . 96 10.4Conclusion ...... 101
11 Material Reconstruction with Photon counting Detectors 103 11.1 Theory of Material Reconstruction ...... 103 11.2 CombinationMethod...... 105 11.3 Minimization Algorithms ...... 106 11.4 Reconstruction Results with Medipix2 and Medipix3 ...... 108 11.5 Material Reconstruction of the Solder Bumps of Medipix ...... 115 11.6Conclusion ...... 118
12 Redesign of Charge Summing Mode of Medipix3 119 12.1 New Architecture of Charge Summing Mode of Medipix3 ...... 119 12.2 Simulation of the proposed architecture ...... 120 12.3 Impact of the new Charge Summing Mode architecture on Imaging . . . . 120
13 Summary and Outlook 123
14 Zusammenfassung und Ausblick 125
Acknowledgements 127
Literaturverzeichnis 139
iv List of Abbreviations
ASIC Application Specific Integrated Circuit CMOS Complementary Metal Oxide Semiconductor MPX Medipix PCB Printed Circuit Board DAC Digital Analog Converter CSDA Continuous Slowing Down Approximation SPM Single Pixel Mode CSM ChargeSummingMode HGM High Gain Mode LGM Low Gain Mode CM Color Mode RW Read/Write Mode SNR Signal to Noise Ratio MTF Modulation Transfer Function PSF Point Spread Function LSF Line Spread Function ESF Edge Spread Function NPS Noise Power Spectrum DQE Detective Quantum Efficiency ROI Region of Interest
1 Introduction
X-rays and gamma-rays were discovered respectively by W. C. R¨ontgen in 1895 and by Paul Villard in 1900. Since then, they are used in a wide range of applications e.g. in medicine or in industry for Non Destructive Testing. Nowadays, seventy percent of all medical inspections in imaging are carried out using X-rays [1]. They are electromagnetic waves like radiowaves or light but have a much smaller wave length. In medical radio- graphy the patient is irradiated using X-rays and the attenuated intensity is measured. This attenuation is material and energy dependent. This enables one to distinguish for example between soft tissues and bones. Films are usually used for the detection of X-rays in projective X-ray diagnostic. The rapid development in electronics in the last decades allows the digitization of the detec- tion signals and therefore a jump from integrating to photon counting pixel detectors like the Medipix2 detector. With this detector, we can count the incoming photons and gain at the same time information about their energy. Those properties of the detector enable for example the reconstruction of a polychromatic spectrum impinging on to the detector. A major drawback of the Medipix2 detector is that it suffers from charge sharing: the charge carriers produced by one photon can be distributed over several pixels. Therefore, an incoming photon can be detected by more than one pixel. As a consequence, the incoming and the measured spectrum are different. In order to suppress the influence of charge sharing the Medipix3 detector was developed. With its Charge Summing Mode it is able to correct for charge sharing in real time. The aim of this thesis is a detailed characterization of the detectors of the Medipix family. The first three chapters are about the basic interactions of radiation with matter and the introduction to photon counting detectors of the Medipix family. Chapter 5 intro- duces the Monte Carlo Simulation Tool ROSI and explains how the response functions of the Medipix detectors to monochromatic radiation are modelled. Characterization of the brand new Medipix3 detector in measurements is the main focus of chapter 6. Chapter 7 investigates the spectroscopic properties of the Medipix detectors and chapter 8 is about using the Medipix2 detector for quality assurance and constancy checks. Chapter 9 intro- duces metric quantities used to characterize an imaging system. The spatial resolution of the Medipix3 detector is presented in chapter 10. Chapter 11 focuses on energy resolved material reconstruction with Medipix detectors. The last chapter is about the redesign of Charge Summing Mode of Medipix3.
1
2 Interaction of X-rays with Matter
Contents 2.1 PhotoelectricEffect ...... 3 2.2 ComptonScattering...... 4 2.3 RayleighScattering ...... 5 2.4 PairProduction ...... 5 2.5 InteractionofElectronswithMatter...... 6 2.6 Conclusion ...... 7
This chapter gives an overview of the interactions which occur when X-ray photons encounter matter. Photoelectric effect, Compton scattering, coherent scattering and pair production will be presented in the first part of the chapter. In fact these processes are the basis of all current photon detection devices and thus determine the sensitivity and the efficiency of a detector. The last part of the chapter focuses on the interaction of electrons with matter.
2.1 Photoelectric Effect
In the photoelectric absorption process, a photon undergoes an interaction with an atom in which the photon completely disappears (see figure 2.1). An energetic photoelectron is released by the atom from one of its bound shells. For a photon of sufficient energy, the most probable origin of the photoelectron is the K-shell of the atoms since a free electron cannot absorb a photon and also conserve momentum. The recoil momentum is then absorbed by the nucleus. The energy of the ejected electron is given by [2]:
E − = hν Eb (2.1) e − where Eb is the binding energy of the photo electron in its original shell, h the Planck constant, ν the frequency of the incident photon. The ejection of the photoelectron leaves an ionised atom. The vacancy is filled through rearrangement of electrons from other shells of the atom. This leads to characteristic X-ray photons or to the ejection of auger electrons. In general the cross section of the photoelectric effect increases with the atomic number Z and decreases with the energy E. For photon energies below 100 keV, a rough approximation of the cross section is given by [3]:
Z4 σpe (2.2) ∝ E3
3 2 Interaction of X-rays with Matter
Incoming photon Photoelectron from an inner shell + +++
Figure 2.1 Illustration of photoelectric effect [4].
2.2 Compton Scattering
Compton scattering takes place between an incident photon and an electron of the outer atomic shell. In this process, the incoming photon is deflected through an angle θ with respect to its original direction (figure 2.2). The photon transfers a portion of its energy E to the electron. Using the equations for the conservation of energy and momentum, the energy E′ of the photon after the scattering is then [2]:
′ E E = E (2.3) 1+ 2 (1 cos θ) m·c · − θ is the scattering angle, c the speed of light, m the mass of the electron. The differential
Scattered electron Incoming photon from an outer shell
Scattered photon
+ +++
Figure 2.2 Illustration of Compton scattering [4].
dσ cross section dΩ for Compton scattering can be calculated with the Klein-Nishina formula and is given by [2]:
dσ 1 1 + cos2 θ α2(1 cos θ)2 = Zr2 1+ − dΩ 0 1+ α(1 cos θ) 2 (1 + cos2 θ)[1 + α(1 cos θ)] − −
4 2.3 Rayleigh Scattering
2 where α = hν/mc . r0 is the classical electron radius. The total cross section shows a weak energy dependence compared to the photoelectric effect.
2.3 Rayleigh Scattering
Rayleigh scattering (figure 2.3) describes photon scattering by atoms as a whole. It is also called coherent scattering as the electrons of the atom contribute to the interaction in a coherent manner. No energy is transferred to the material. This elastic process changes only the direction of the incoming photon. The cross section σR for Rayleigh scattering decreases almost quadratically with the ener- gy E of the incident photon and increases quadratically with the atomic number Z:
Z2 σR (2.4) ∝ E1.9
Incoming photon
+ +++ Elastic scattered photon
Figure 2.3 Illustration of Rayleigh scattering [4].
2.4 Pair Production
A gamma ray photon with an energy of at least 1.022 MeV can create an electron-positron pair when it is under the influence of the strong Coulomb field surrounding the nucleus (see figure 2.4). In this interaction the nucleus receives a very small amount of recoil energy to conserve momentum, but the nucleus is otherwise unchanged and the gamma ray photon disappears [2]. If the gamma ray energy exceeds 1.022 MeV, the excess energy is shared between the electron and positron as kinetic energy. Figure 2.5 depicts the dependence of photoelectric effect, Compton scattering and pair production upon photon energy for different atomic numbers Z. For silicon (dashed lines), photoelectric effect dominates for energies below 57 keV. Above this limit, Compton scattering is predominant. Coherent scattering is not taken into account since its cross section is negligible compared to the cross sections of the three other processes. In fact, only the photoelectric absorption and the Compton effect will be important in the energy range (10 keV - 160 keV) used in X-ray imaging for medical applications.
5 2 Interaction of X-rays with Matter
Positron Incoming photon
+ +++ Electron
Figure 2.4 Illustration of pair production [4].
Figure 2.5 Regions of predominance of photoelectric effect, Compton effect, and pair pro- duction as function of photon energy and for different atomic numbers Z. The region of predominance of photoelectric effect increases with the atomic number Z. As a result, high Z materials are indicated as sensor material for radiation detection [5].
2.5 Interaction of Electrons with Matter
The electrons that are released through interaction of primary photons with matter, can loose their energy through collisions with shell electrons. The transferred energy, when sufficient, can promote electrons from the valence band into the conduction band, resulting in electron-hole pairs in a semi conductor material. For electrons, due to their small mass, another energy loss mechanism comes into play: the emission of electromagnetic radiation (Bremsstrahlung) that may be understood as radiation arising from the acceleration of the electron by the electrical attraction of the nucleus [6]. The total energy loss of electrons Stot, therefore, is the sum of collision stopping power Scoll and radiative stopping power Srad [7]:
Stot = Scoll + Srad (2.5)
6 2.6 Conclusion with
dE Z m Scoll = ρ (2.6) dx ∝ · A · E !coll and
dE Z2 e 2 Srad = ρ E (2.7) dx ! ∝ · A · m · rad where ρ is the density, Z the charge number, A the atomic mass number of the medium. m is the electron mass, e the elementary charge and E the energy. The energy loss of electrons per collision at energies of a few MeV is comparatively small, resulting in a quasi-continuous energy loss of electrons on a track with relatively large scattering angles. They follow a zigzag course through the medium [8] (see section 3.3).
2.6 Conclusion
This chapter gave a brief overview of interactions of photons and electrons with matter. The two predominant effects are Compton and photoelectric effect in the energy range relevant in medical X-ray imaging.
7
3 The Medipix2 Detector
Contents 3.1 DescriptionandoperatingModes...... 9 3.2 CountingPrinciple...... 11 3.3 ChargeSharing...... 12 3.3.1 PathofanElectronintheSensorMaterial ...... 12 3.3.2 PhotonmultipleInteractions ...... 13 3.3.3 LateralChargeDiffusion...... 13 3.3.4 Repulsion ...... 14 3.4 Conclusion ...... 15
Nowadays, there are several types of devices to detect radiation. In X-ray radiography the film is used. With the rapid advances in microelectronics, the film is being pro- gressively replaced by digital systems. We can distinguish between two classes of digital detectors. The first class is based on the integrating principle: charge carriers that are released in the detector are added up over the frame time and digitized through analog digital converters. The second category makes use of the photon counting principle. The first photon counting semiconductor detector of the Medipix family was developed in the nineties of the last century. It was based on the idea of the active pixel detector like the one in the LHC1/omega3 [9] experiment. The first generation was the Medipix1 detector, developed in the framework of the international Medipix collaboration [10]. After the success of the Medipix1 detector, the collaboration decided to develop the Medipix2 detector with higher performance and a smaller pixel pitch for a better spatial resolution for imaging purposes. After few years, other chips were designed: Medipix2- MXR, Timepix and recently Medipix3. The latter has more functionalities and operation modes than its predecessors. This chapter will focus on the Medipix2 detector.
3.1 Description and operating Modes
The Medipix2 detector is a pixelated, direct converting photon counting semiconductor ñ detector. It has 256 256 pixel with a pixel pitch of 55 ñm. The use of 0.25 m CMOS × (Complementary Metal Oxide Semiconductor) technology enables the implementation of more than 33 millions of transistors over the active area (1.982 cm2) of the chip. The detector is designed in a hybrid technology. Hybrid means that sensor material and
readout electronics (ASIC) are produced separately. An advantage is that different sensor
ñ ñ materials (Si, GaAs, CdTe) and sensor thicknesses (300 ñm, 700 m, 1000 m) can be used depending on the specific application: for low-energy X-ray, silicon sensors are used whereas the other materials mentioned are appropriate for higher energies due to their
9 3 The Medipix2 Detector higher quantum efficiency. Sensor and ASIC are connected via solder bumps made of Ag/Sn or Pb/Sn as depicted in figure 3.1. Figure 3.2 shows a block diagram of a Medipix2 pixel cell. The diagram exhibits an analog and a digital part: the analog part comprises a charge sensitive preamplifier, two discriminators (a Low threshold THL (DiscL) and a high threshold THH (DiscH) ) and 2 Digital to Analog Converters (DAC) for threshold equalization. The digital part consists of a pulse processing circuitry (Double Disc Logic DDL) and a Shift Register that acts as a 13 bit counter during data acquisition. The digital output of the discriminators are given as input to the DDL that enables two operating modes: photons with an energy above a defined THL can be registered (Single Threshold Mode) or only photons with an energy in a defined THL-THH window (Energy Window Mode) are counted. Table 3.1 summerizes the characteristics of the Medipix2 detector.
Figure 3.1 Schematic view of a Medipix assembly [11]. The sensor is connected to the ASIC via bump bonds.
Table 3.1 Characteristics of the Medipix2 detector [5]. ñ Pixel size 55 ñm 55 m × Number of pixels 256 256 × Sensitive area 1.98 cm2 (87 % of the total area) ∼ ENC (Electronic Noise Charge) 140 e− ∼ Pixel count rate up to 1 MHz Threshold linearity < 3 % to 100k e− Threshold spread (not adjusted) 500 e− Threshold spread (adjusted) 100 e−
10 3.2 Counting Principle
Figure 3.2 Electronic circuitry of a Medipix2 pixel cell. The analog part consists of a preamplifier and two discriminators. The digital part receives the output of the discriminators and decides in the DDL (Double Disc Logic), if the counter will be incremented. This circuitry is also responsible for the readout of the data [12].
3.2 Counting Principle
An incoming X-ray photon impinging on the detector interacts with the sensor material through Compton or photoelectric effect. The electron that is released will loose its energy gradually in the sensor material. The energy loss, when sufficient, can promote electrons from the valence band into the conduction band: as a result electron-hole pairs are created. These charge carriers are separated and drifted towards the pixel electrodes using an external applied electrical potential difference. The voltage is typically 150 V
(for a 300 ñm Si sensor). On the electrodes’ side, mirror charges are created due to the motion of the carriers. As a consequence an electrical current is registered in the electrodes. The magnitude of the current can be determined using the so-called weighting potential (Ramo Theorem) [13]. The temporal integral over the current in a pixel electrode corresponds to the amount of charge collected by this electrode. For Medipix2, this current is preamplified. After that the charge to voltage conversion is performed in the preamplifier integrating capacitance and compared directly with the energy threshold. If the pulse is above the threshold value, a counter is incremented. The Medipix2 detector
11 3 The Medipix2 Detector can therefore count photons with energies above a given threshold or within a defined energy window.
3.3 Charge Sharing
Most of pixelated detectors based on planar technology suffer from the effect called charge sharing: this implies that a distribution of free charge carriers released by an electron in the sensor material can be collected by more than one pixel. Many effects can cause charge sharing: on one hand we can have multiple interactions of incoming photons, diffusion and repulsion of generated charge carriers, on the other hand the track of a primary electron can be extended over several pixels.
3.3.1 Path of an Electron in the Sensor Material
A released electron will deposit its energy progressively in the sensor material, since it is deviated (because of its charge) for example in the Coulomb field of the nucleus of some atoms. As a consequence, the path of the electron is not a straight line and can be extended over several pixels. Depending on the threshold, the counter in more than one pixel will be incremented. Figure 3.3 shows the path of a 60 keV electron in silicon
calculated with the CSDA (Continuous Slowing down Approximation) [14]. It springs ñ from the figure that the track of the electron is around 20 ñm for a pixel pitch of 55 m. Thus electrons with higher energies will have their track extended over more than one pixel causing charge sharing. This fact can be deduced from figure 3.4 which depicts the path length as function of the electron energy in silicon.
6 m
µ 4 0
z in 2 −5 0 −10 −15 5 −20 µ 0 x in m y in µm
Figure 3.3 Illustration of the path of an electron in silicon. The extension of that path over several pixel leads to charge sharing.
12 3.3 Charge Sharing
160
140
120
100
80
CSDA range in µm 60
40
20
0 15 30 45 60 75 90 105 120 135 150 Energy in keV
Figure 3.4 CSDA range for electrons in silicon versus electron energy [15].
3.3.2 Photon multiple Interactions In some cases, an incoming ionizing particle can interact with the sensor material and release more than one electron. Figure 3.5 shows how a photon that impinges on the sensor at (0), interacts firstly through Compton effect (1) at a certain position in the sensor and then, the Compton scattered photon interacts through photoelectric effect (2) at another place. Each time a charge carrier distribution is created. The dashed circles represent the position of the collection of the charge carriers after the drift process. These diameters are bigger than the initial ones (solid lines) due to diffusion. In this example, depending on the threshold, the incoming photon will be detected by 0, 1, 2 or 3 pixel.
3.3.3 Lateral Charge Diffusion A released charge carrier distribution in the sensor material will experience a lateral diffusion while drifting towards the pixel electrodes. The solution of the diffusion equation for an initial point-like distribution is a Gaussian distribution as illustrated in figure 3.6.
The spread of this distribution in ñm is given by [17]:
σRadius = √2 Dt (3.1) where D is the diffusion coefficient of the charge carriers and t is the respective drift time. The diffusion coefficient is related to the temperature T and the mobility µ of the charge carriers through the following Einstein formula:
k T D = µ B (3.2) e An estimate of the upperlimit for the drift time of charge carriers is given by [18]:
d dsensor d tc(d)= · = (3.3) µ VBias µ E
13 3 The Medipix2 Detector
0
2 γein
γf
1
j-1
j
j+1
i-1 i i+1
Figure 3.5 Illustration of one effect leading to charge sharing [16]. An incoming photon (0) interacts with the sensor and creates at (1) through Compton and at (2) through photo effect a free electron. The bold dashed lines represent the path of the electron during the energy loss process. The dashed circles represent the end diameter of a charge cloud at the end of the drift process. The diameters due to diffusion are bigger than the initial diameters (solid circles) at the beginning of the drift process. dsensor and VBias denote the sensor thickness and the bias voltage, respectively. d is the drift distance. E is the electric field between the sensor and the electrodes and is given in V/cm. Taking into account equations 3.2 and 3.3, the spread due to diffusion (see equation 3.1) can be rewritten as follows:
kB T d σRadius = 2 r e · E
At room temperature, for drift distance in ñm and E in V/cm, the equation above be- comes: d σRadius = 23 [µm] (3.4) rE Obviously, σRadius that denotes the spread of the charge cloud during the drift, increases with the sensor thickness. The thicker the sensor, the more probable is charge sharing at constant field E.
3.3.4 Repulsion The released electron-hole pairs are separated through the applied electric field. Electrons within a charge cloud push each other away (repulsion). The charge distribution due to
14 3.4 Conclusion d Sensor d Drift direction Sensor thickness Distance