Photo-detectors for time of flight positron emission tomography (ToF-PET).

We present the most recent advances in photo-detector design employed in time of flight positron emission tomography (ToF-PET). PET is a molecular imaging modality that collects pairs of coincident (temporally correlated) annihilation photons emitted from the patient body. The annihilation photon detector typically comprises a scintillation crystal coupled to a fast photo-detector. ToF information provides better localization of the annihilation event along the line formed by each detector pair, resulting in an overall improvement in signal to noise ratio (SNR) of the reconstructed image. Apart from the demand for high luminosity and fast decay time of the scintillation crystal, proper design and selection of the photo-detector and methods for arrival time pick-off are a prerequisite for achieving excellent time resolution required for ToF-PET. We review the two types of photo-detectors used in ToF-PET: photomultiplier tubes (PMTs) and silicon photo-multipliers (SiPMs) with a special focus on SiPMs.

Introduction

Recent advances in the field of medical imaging have greatly facilitated the transition from technologies used to accurately image structures inside the n class="Species">human body to technpan>ologies senpan>sitive enpan>ough to provide functional anpan>d biological information at the cellular anpan>d the molecular level. Positron emission tomography (PET) is considered to be one of the most senpan>sitive in-vivo molecular imaging modalities despite its signpan>ificanpan>tly inferior spatial resolution compared to imaging modalities such as computed tomography (CT) anpan>d magnpan>etic resonanpan>ce imaging (MRI). The improvemenpan>t of PET detector technpan>ology is anpan> active field of research anpan>d efforts are focused on addressing the limits in spatial resolution anpan>d senpan>sitivity achieved in PET.

Positron emission tomography (PET): Basic principles

PET imaging is performed after the administration to the n class="Species">patient of a radio-tracer, namely a biomolecule labeled with a radioactive atom emitting positronpan>s (β+ particles). The biomolecule is chosenpan> such that it will preferenpan>tially accumulate in the area of interest resulting in high radio-tracer conpan>cenpan>trationpan> in this regionpan>. A PET acquisitionpan> is based onpan> the coincidenpan>t detectionpan> of manpan>y pairs of simultanpan>eous anpan>ti-parallel photonpan>s following the anpan>nihilationpan> of the positronpan> emitted by the radio-tracer [1]. The detectionpan> is performed by a number of detector elemenpan>t pairs which are placed around the imaged object, typically in a ring geometry, as shownpan> in Figure 1. The incidenpan>t anpan>nihilationpan> photonpan>s will interact with these detectors which will fully or partially absorb the photonpan> enpan>ergy resulting in the genpan>erationpan> of anpan> electrical signal. This process will be henpan>ceforth referred to as a photonpan> evenpan>t. A PET acquisitionpan> will evenpan>tually result in a number of idenpan>tified detector pairs that have simultanpan>eously detected the two anpan>nihilationpan> photonpan>s, or otherwise stated, the positronpan> anpan>nihilationpan> evenpan>t. These pairs will be subsequenpan>tly assigned a virtual line conpan>necting the two detectionpan> points (line of responpan>se, pan> class="Gene">LoR). For every stimulated detector pair the related positron annihilation is localized somewhere along the respective LoR through the patient with no further information about the exact point of positron annihilation along that line. Acquisition from detector pairs at various angular views (tomographic acquisition) followed by appropriate reconstruction algorithms allows for estimation of the tracer biodistribution within the imaged object with a finite spatial resolution and sensitivity.

Figure 1.

The basic principle of PET: a ring of detectors placed around the object detects photon pairs (green arrows) which are generated as a result of the annihilation of a positron emitted by the radio-pharmaceutical.

PET detectors are required to have fast response due to the requirement for detection of simultaneous photon events. To date the detectors typically used in PET have a finite time resolution of a few nanoseconds (ns) allowing for detection of photon events within a predefined time window. The width of this window is usually chosen to be twice the time resolution of the PET detectors.

The time of flight (ToF) feature: Benefits and challenges

Time of flight PET (ToF-PET) is an advance over traditional PET that exploits the time difference Δt in detection of the two photon events and correlates it to the position Δx of the annihilation point with respect to the center of the field of view (FoV) according to the formula (Figure 2) [2]: Even though the concept of ToF PET and its implementation in PET system designs started in the early 1980’s [3], the development of new fast scintillation materials and the rapid advances in photo-detection technology have recently led to increased interest in the development of novel ToF-PET detector designs as well as to commercialization of the first ToF-PET clinical systems (Section 3).

Figure 2.

The ToF concept: the flight time difference Δt between the two detected photon events is in a first approximation related to the object position Δx along the line connecting the two detector elements through Equation 1.

Due to the limited time resolution of the PET detectors the registered time difference Δt is blurred by a variance resulting in a corresponding blurring in the estimated position Δx by a variance (Gaussian structure in Figure 3).

Figure 3.

The limited detector time resolution associates the estimated position Δx with a variance .

To date the time resolution of PET detectors is not adequate in order for ToF information to be used for exact localization of the annihilation point along the n class="Gene">LoR. For example, a detector time resolution of 500 picoseconds (ps) will result in a position blurring of 7.5 cm. This blurring is anpan> order of magnitude higher compared to the spatial resolution of several mm achieved by currenpan>t clinical PET systems [4] anpan>d dominated by the scintillation crystal pixel size [5]. However, time resolution values of hundreds of ps that currenpan>t detectors have canpan> result in a considerable gain in signal-to-noise ratio (SNR) in the reconstructed image which further implies better lesion detectability, reduced radioactive dose received by the pan> class="Species">patient and/or reduced scan time [6-8]. An illustration of the difference between a standard PET and a ToF-PET acquisition is illustrated in Figure 4.

Figure 4.

Illustration of the line back-projection process for the conventional PET system (left) and a ToF-PET system (right). Essentially more counts are placed at or near the correct position indicated by a red dot.

The concept of time resolution

We define the time resolution of a detector as the minimum time interval between two subsequent photon events in order for these to be recorded as separate events [9]. In PET, time resolution is typically measured for pairs of detectors that detect many annihilation photon events and is defined as the full width at half maximum (FWHM) of the distribution of time difference between the two detectors, over multiple events. Typical time resolutions achieved by modern PET systems lie within the 2–10 ns range. In the case of ToF-PET, however, such magnitudes are n class="Disease">insufficient in order for ToF information to have a benpan>efit over a convenpan>tional PET acquisition. The existing commercially available ToF-PET imaging systems canpan> achieve time resolutions in the ranpan>ge of 500–900 ps allowing for a Gaussianpan> weighted localization with a FWHM of 7.5–13.5 cm along each pan> class="Gene">LoR.

The lowest limit in time resolution for a single photon event that can be achieved is theoretically described by a number of empirical formulas [10,11] and for a generic detector with finite signal-to-noise ratio and transit time jitter could be summarized as follows: where is the estimated time variance, σ is the square root of noise variance superimposed on the detector signal, is the signal slope at the point of time estimation (time pick-off) and is the time variance due to the photo-detector transit time spread. From the above formula it can be understood that timing resolution is dependent on (1) the electronic and detector noise (as represented by σ), (2) the combination of detector signal gain and detector signal time response (as represented by ), (3) the triggering level at which the detector noise and slope are calculated and (4) the inherent time jitter of the photo-detector’s charge carriers (as represented by σ) [12]. The desired time resolution requirements for ToF-PET aim for detector signals with low noise, high gain and fast time response. Basic error propagation theory dictates that for photon pair coincidence time resolution the right hand side of Equation 2 be multiplied by 2. A more detailed expansion of Equation 2 will eventually lead to the well known law for the dependence of the time resolution (FWHM) on the number of generated photoelectrons N.

Even though there have been several studies in the past demonstrating the feasibility of ToF PET, only the past few years has it managed to gain popularity in both industry and research. Recent technological advances have led to the development of new scintillator materials and novel photo-detector architectures with improved performance properties that have significantly facilitated the rapid integration of ToF technology into complete imaging systems. In the following, we review the most important components of ToF-PET radiation detectors with emphasis in solid-state photo-detectors.

Radiation Detectors for Fast Timing Applications

Typically in PET the detection of incident radiation is achieved by a two-step conversion of the annihilation photon energy (1) to visible light in a scintillation material and (2) to electric charge in a photo-detector. Even though inevitable signal losses are associated with this two-step conversion, this type of detector is still considered to be a well understood, fast and efficient means of 511 keV photon detection. An example of a two-step radiation detector is shown in Figure 5.

Figure 5.

Photo of a radiation detector based on a scintillator/photo-detector combination.

The ideal ToF-PET detector

In order to extract as accurately as possible information about the flight time difference of the two annihilation photons, both components of the PET detector, namely the scintillator and the photo-detector, should have a fast, ideally instantaneous response. However, there are a number of factors which may considerably affect time resolution, such as the statistical nature of radiation detection, even in the case of very fast response. Both the detection of the annihilation photons in a scintillation crystal as well as the detection of optical photons in the photo-detector are subject to Poisson statistics which may affect the timing accuracy especially in case of a low scintillation light yield and/or a low photo-detector gain. Thus high light output in the scintillator and high gain in the photo-detector are key features in order to minimize statistical uncertainties.

An ideal ToF-PET radiation detector will combine the following features: Typical response profiles of a scintillator and a photo-detector are shown in Figure 6. Extraction of time information is performed on the electrical pulse generated by the photo-detector. The properties of this pulse will be affected by both the scintillator and the photo-detector.

Figure 6.

Illustration of typical response profiles of a scintillator (top, response to high energy photons) and of a photo-detector (bottom, response to scintillation light from top plot). The red dashed line indicates the signal slope ( ) at the triggering threshold (V), which together with the noise superimposed on the signal determine the lower bound on timing resolution for a single detector signal according to Equation 2.

a scintillator material with high efficiency for absorption of the 511-keV annihilation photons

a scintillation process with instantaneous rise and decay times of the light pulse profile

a scintillator geometry with a low aspect ratio ( ) which will minimize the optical photon travel path and thus light losses at the crystal interfaces, as well as transit time variances in the crystal

a photo-detector with high efficiency in the detection of the visible light produced by the scintillator

a photo-detector with high amplification gain and low noise

a photo-detector with instantaneous rise and decay time of the charge pulse profile and low charge transit time spread

Scintillation crystals

Table 1 summarizes the basic features of fast scintillators that have been considered in ToF-PET, either in research or in industry. As already mentioned in the description of an ideal radiation detector, a scintillator with high photon emission rate (high photon yield at a fast decay time) and high stopping power (as represented by the effective atomic number and the density) is the desired combination for ToF-PET. However additional features such as the refractive index and the peak emission wavelength will assure compatibility with the photo-detector and thus efficient transfer of the optical photon flux from the scintillator to the photo-detector. Even though LSO or n class="Chemical">LYSO is acknpan>owledged as the option of choice for fast scintillation counters, anpan> evenpan> faster anpan>d more luminous scintillation material, LaBr3 has recenpan>tly attracted interest with promising results [19-21] despite its hygroscopic properties anpan>d lower stopping power compared to LSO. Recenpan>tly, the doping of the latter (LSO) with Ca in addition to Ce dopanpan>ts proves to improve the timing performanpan>ce of the scintillator.

Table 1.

Comparative table of fast scintillators which can be considered for ToF-PET [9,13–18].

	LSO(Ce)	LSO(Ce, Ca)	LYSO(Ce)	LaBr3	BaF2	Plastic
Density ( gcm^3)	7.40	similar to LSO	similar to LSO	5.29	4.89	1.03
Effective atomic number (Z)	66	similar to LSO	similar to LSO	46	54	12
Rise time (ns)	≤ 0:5	similar to LSO	similar to LSO	0.2–0.5	∼0.03	0.7–1
Decay time (ns)	40	31@0.4% Ca	similar to LSO	15–26	0.8/620	2–5
Photon yield/keV	20–30	∼35	similar to LSO	63	1.8/10	10
Refractive index (n)	1.82	1.82	1.82	1.90	1.56	1.58
Hygroscopic	No	No	No	Yes	Slightly	No
Peak emission wavelength (nm)	420	420	similar to LSO	380	220/310	Various

From the materials listed in Table 1, n class="Gene">BaF2 anpan>d plastic scintillators are not preferred options for ToF-PET despite their excellenpan>t time response. Plastic scintillators suffer from low stopping power anpan>d optical photon yield, factors that will significanpan>t compromise PET overall imaging performanpan>ce. The inorganpan>ic material pan> class="Gene">BaF2 has been studied in the past as a candidate for ToF [22] with promising results in terms of time resolution reaching values below 300 ps [23]. However, the stopping power and light output of this material is significantly lower compared to the other inorganic scintillators of Table 1 and, in addition, the peak emission wavelength of the fast decay component of BaF2 may restrict the available choices of photo-detectors with sufficient light detection efficiency at this part of the electromagnetic spectrum.

Photo-detectors

Even with an ideal scintillator yielding high optical photon rate, a lower bound on the time resolution that can be achieved is imposed by the photo-detector performance. Low quantum efficiency (ratio of number of emitted charge to the number of incident optical photons absorbed in the photosensitive area, QE) of the photo-detector at the peak emission wavelength of the scintillator will result in an effective reduction of the amount of scintillation light detected. In addition low gain, n class="Disease">gain variations or noise in the photo-detector response will increase statistical variations in the extraction of anpan>y type of information from the genpan>erated electrical signal. Table 2 summarizes the basic performanpan>ce features of the three types of photo-detectors widely used in commercial anpan>d research PET systems [24,25].

Table 2.

Comparative table of three types of photo-detectors typically used in PET: PMT, APD and SiPM.

	PMT	APD	SiPM
Gain	106	50–1,000	∼106
rise time (ns)	∼1	∼5	∼1
QE @ 420 nm (%)	∼25	∼70	∼25–75 (PDE)
Bias (V)	>1,000	300–1,000	30–80
Temperature sensitivity ( %°C)	<1	∼3	1–8
Magnetic field sensitivity	yes	no	no
Sensitive area	cm2	mm2	mm2
Price/channel ($)	>200	∼100	∼50

From the photo-detectors listed in Table 2, proportional mode avalanche photo-diodes (APDs) are not preferred for use in the ToF application due to their significantly slower response time. In addition, their low multiplication gain results in small output signals requiring further amplification prior to processing. In the following, two types of photo-detectors will be reviewed, the PMT and the SiPM, with special focus on the latter.

The photomultiplier tube (PMT)

The photomultiplier tube (PMT) is a photo-detector type commonly used for scintillator readout in numerous applications including medical imaging. In ToF-PET, n class="Chemical">PMTs are still conpan>sidered to be the gold stanpan>dard for sub-nanpan>oseconpan>d time resolutionpan> [12,26]. Their high gain combined with a low excess noise factor lead to minimal deviationpan> from Poissonpan> statistics as well as reduced statistical uncertainty in the genpan>erated charge signal [9]. The basic componpan>enpan>t of a PMT is a vacuum tube conpan>sisting of a photo-cathode, several electrodes called dynodes anpan>d anpan> anpan>ode. The photocathode is a photo-senpan>sitive electrode that emits charge (electronpan>s) for a number of incidenpan>t optical photonpan>s absorbed with a givenpan> efficienpan>cy (QE defined above). Betweenpan> the cathode anpan>d the anpan>ode a bias voltage in the kV ranpan>ge is applied to facilitate the genpan>erated electronpan> tranpan>sport anpan>d amplificationpan> from the cathode to the anpan>ode. Unpan>der the influenpan>ce of a high potenpan>tial the genpan>erated electronpan>s from the photocathode drift anpan>d successively enpan>counter the dynode stages. At every dynode stage the incidenpan>t electronpan>s have gained sufficienpan>t enpan>ergy to create seconpan>dary electronpan> emissionpan>s from collisionpan>s with the dynode, thus resulting in pan> class="CellLine">a large electron cloud at the anode (Figure 7). As can be seen in Table 2, one advantage of PMTs over semiconductor photo-detectors is their lower sensitivity in temperature variations which in most cases implies stable detector operation over a wide range of temperatures. However, their relatively bulky size and the requirement for power supplies which can provide thousands of Volts make PMTs a less attractive option for their use in high packing fraction, high channel density PET radiation detectors.

Figure 7.

Schematic diagram and principle of operation of a PMT. For illustration purposes only a few HV connections to the dynodes are shown.

The Silicon photomultiplier (SiPM)

The n class="Chemical">silicon photomultiplier (SiPM) has recenpan>tly beenpan> introduced to address the shortcomings of both pan> class="Chemical">PMTs and APDs summarized in Table 2 [25,27-32]. These devices consist of an array of microscopic, parallel connected avalanche photo-diodes which operate in limited Geiger-mode (alternatively known as G-APDs or SiPM micro-cells). In this mode of operation the micro-cells are biased above breakdown resulting in a cumulative avalanche breakdown within the diode depletion region and thus in an excessive current every time they encounter a (visible) photon stimulus (Figure 8). The difference between the proportional and the Geiger mode of APD operation is illustrated in Figure 9. As shown in Figure 9 in the case of a proportional-mode APD the electron carriers are primarily involved in the avalanche process, while in the case of a Geiger-mode APD both charge carriers (namely electrons and holes) are able to trigger an avalanche.

Figure 8.

The characteristic (I-V) curve of a typical p-n diode. Avalanche photo-diodes (APDs) operate at a reversed applied bias. Proportional APDs operate at the regime of bias voltages well below breakdown where the amount of produced charge is proportional to the number of absorbed optical photons. SiPMs and their individual components (G-APDs or micro-cells) operate at bias voltages above breakdown (V) where the avalanche process becomes excessive. At this operation mode (Geiger-mode) the amount of charge produced from each micro-cell is standardized and the total amount of charge produced from all micro-cells is proportional to their number.

Figure 9.

The difference between a proportional APD and a G-APD. Illustration is courtesy of Dr. A. Nepomuk Otte, University of Santa Cruz. Adopted with permission [33].

In order for the large amount of current not to destroy the diode, an appropriate quenching circuit is integrated to each micro-cell as shown in Figure 10. Typically, a resistor of several hundreds of kOhm (quenching resistor) is connected in series with each micro-cell to quench the avalanche (passive quenching). Every time n class="CellLine">a large currenpan>t flows trough it, it will result in a voltage drop across the micro-cell terminals. This voltage drop is large enpan>ough to lower the micro-cell applied bias below the breakdownpan> limit, resulting in interruptionpan> of the pan> class="Disease">avalanche process. Due to the inherent micro-cell capacitance, as well as the parasitic capacitance of the overall summing network in a SiPM, each micro-cell needs a characteristic amount of time to recover after the bias voltage is restored to values above breakdown. This characteristic time is called recovery time and its magnitude is typically equal to the product of the value of the quenching resistor and the value of the overall diode capacitance.

Figure 10.

Left: schematic of the equivalent electrical circuit of a SiPM. Only 6 micro-cells, each represented by a diode symbol, are shown. Right: illustration of the signal formation in a SiPM. The pile-up of the individual micro-cell pulses is achieved by means of summing via a common readout line.

Even though passive quenching is relatively simple to implement and most of commercially available SiPMs employ such circuits, there are specific performance limitations associated with it. The pulse shape depends on the values of the passive elements used in the quenching circuitry and may thus be subject to variations depending on how these values are influenced i.e., by temperature. In addition SiPMs with passive quenching are typically characterized by long recovery times which can subsequently impact their count rate performance and dynamic range [34].

In order to address the above shortcomings of passive quenching, G-APDs can alternatively be quenched using an active quenching circuit (Figure 11).

Figure 11.

Generic schematics of a passive (left) and an active (right) quenching circuit employed at the micro-cell level (the micro-cell is represented by the diode symbol).

A generic active circuit typically consists of a comparator-based feedback circuit with a sensing and a quenching network. Whenever the rise of the G-APD output pulse is sensed, indicating the rise of an avalanche, a feedback pulse is produced and forces the G-APD to quench the avalanche by means of a controlled voltage source. This source is also configured to reset the G-APD back to an active state by restoring the proper bias across its terminals. Both the quenching and the reset are performed in a highly controlled way allowing thus for fast recovery of the SiPM micro-cells [35]. For applications where high count rate capability is a prerequisite, SiPMs with active quenching are preferred.

Even though the most acknowledged advantage of actively quenched SiPMs is their enhanced count rate performance, they may also demonstrate an effective enhancement in n class="Gene">PDE owing to the fact that each micro-cell canpan> be triggered multiple times, within the duration of the light stimulus, due to their fast recovery (factor RT becomes larger in Equation 4). On the other hanpan>d, it is difficult to predict quanpan>titatively the actual pan> class="Gene">PDE enhancement because of the limitations in the SiPM fill factor due to the active quenching circuitry (factor FF becomes smaller in Equation 4).

In both the aforementioned quenching modes, every micro-cell acts as a binary device which can provide information about the detection of a light stimulus but no information about its intensity. Proportionality between the intensity of the incident light stimulus and the output charge of a SiPM is achieved by summing the signals of individual micro-cells through a common readout line, provided that the incident photon flux is low enough in order for each micro-cell to respond to a single visible photon (Figures 10 and 12). The SiPM combines appealing detector features of both the PMT and the APD for ToF-PET imaging: One of the most important limitations in a SiPM design is the trade-off between dynamic range and light detection efficiency. A sufficient number of available cells are needed for a SiPM to respond linearly to the incident scintillation light stimulus according to the theoretical formula: where N is the number of micro-cells contained in a SiPM sensitive area, N is the number of stimulated micro-cells, N is the number of optical photons incident to the SiPM area and n class="Gene">PDE is the SiPM photon detection efficienpan>cy defined empirically by the following formula: where FF is the geometric fill factor defined as the ratio of active to total area, GP is the probability for Geiger discharge in anpan>y micro-cell anpan>d RT is a factor related to the recovery time of every micro-cell. Equation 3 is anpan> approximation which does not take into account crosstalk anpan>d after-pulse evenpan>ts (discussed below). Inpan> addition it assumes anpan> infinitesimally long recovery time of each micro-cell anpan>d anpan>d a low incidenpan>t photon flux limited to less thanpan> .

Figure 12.

Generic schematic of the micro-cell configuration within a passively quenched SiPM. The quenching resistor as well as the Al readout trace are depicted. Illustration is courtesy of Dr. A. Nepomuk Otte, University of Santa Cruz. Adopted with permission [33].

Small size which enables detectors with high packing fraction for enhanced sampling and photon sensitivity. In addition small sensitive area reduces device capacitance and dark current.

High gain which eliminates the need for pre-amplification.

Single electron pulses from each micro-cell with a very well defined pulse shape which can be exploited for optimum timing.

Furthermore in a given SiPM area with a fixed dead area (area occupied by the read-out traces) around each micro-cell, n class="CellLine">a larger number of micro-cells would imply a smaller micro-cell senpan>sitive area. This decreased micro-cell area decreases the light pan> class="Gene">PDE of each cell and thus, the overall detection efficiency of the device is degraded (Figure 13).

Figure 13.

Microscope captures of sensitive areas of SiPMs with micro-cell size of 25 μm (left), 50 μm (middle) and 100 μm (right). The fill factor increases with increasing micro-cell size, while the dynamic range (number of available micro-cells within a given area) decreases.

The plots of Figure 14 indicate severe non-linear effects and effects on the energy resolving power of scintillation readout from SiPMs as a function of their applied bias voltage. Above a given value of the applied voltage the measured SiPM response to radiation quanta of different energies is no longer linear (Figure 14, top). An increasing bias voltage leads to increased SiPM gain and n class="Gene">PDE, initially resulting in enpan>ergy resolutionpan> improvemenpan>ts. For excessive bias increase, the pan> class="Gene">PDE is high enough to effectively lower the SiPM dynamic range (because the number of optical photons which can trigger an avalanche is no longer related linearly to the number of available micro-cells). In this case, the measured energy resolution values appear to be overestimated, namely below the theoretical limits imposed by the intrinsic resolution of the scintillator (Figure 14, bottom).

Figure 14.

Effects of non-linear SiPM operation on identification of different photon energies (top) and energy resolution (bottom) as a function of bias voltage. Data taken with a SiPM with a 3 × 3 mm2 cross-sectional area from Hamamatsu (50 μm micro-cell size). The scintillator used was a 3 × 3 × 5 mm3 LSO crystal element.

As in any semiconductor, temperature sensitivity of SiPMs is still a major issue affecting the detector’s performance [36,37]. Temperature coefficients varying from 2 to have been measured. As shown in Figure 15, the detector gain is decreased with increasing temperature owing to quenching of n class="CellLine">a larger part of the avalanpan>che carriers’ enpan>ergy due to collisions with the crystal lattice vibrations (phonons). Inpan> addition, for the majority of SiPMs employing passive quenpan>ching, the value of the quenpan>ching resistor demonstrates a temperature depenpan>denpan>ce that may lead to variations in the SiPM pulse shape with further implications on recovery time anpan>d timing performanpan>ce of the devices [38].

Figure 15.

Dependence of SiPM gain on temperature. Data taken with a SiPM with a 1 × 1 mm2 cross-sectional area from Hamamatsu (50 μm micro-cell size). All the temperature measurements were performed at a constant SiPM bias.

SiPMs are characterized by high dark count rates in the kHz to MHz range. Dark counts are signals induced in the individual micro-cells by a thermal stimulus instead of a light signal. Thermal stimulation is frequent in SiPM micro-cells and can result in an increased signal noise level thus affecting the timing performance of the devices (Equation 2). Figure 16 shows the dependence of the maximum registered dark count frequency on SiPM applied bias. At a given applied bias, multiple dark count rates may exist depending on the triggering threshold at which these rates are measured. This effect is attributed to the optical crosstalk (described below) apparent between different micro-cells of a SiPM, resulting in simultaneous signal generation from multiple micro-cells. The triggering threshold for the measurements shown in Figure 16 is adjusted depending on the applied bias, so that the maximum dark count rate typically from a single micro-cell (denoted as ’peak noise frequency’ in the y-axis of Figure 16) is registered.

Figure 16.

Dependence of dark counts on SiPM bias. Data taken with a SiPM with a 3×3 mm2 cross-sectional area from SensL (50 μm micro-cell size).

An effective increase in the observed SiPM dark count rate is also attributed to after-pulse effects. After-pulses are belated micro-cell excitations following the initial micro-cell excitation due to late release of avalanche charge carriers trapped in defects within the diode sensitive area (depletion region). This phenomenon effectively results in small signals piled up in the tail of the actual SiPM output [39].

The Geiger discharge followed by the n class="Disease">excessive avalanche process results in inherenpan>t light emission from the micro-cells themselves [40]. The SiPM architecture is such that crosstalk due to distribution of this light betweenpan> adjacenpan>t micro-cells may occur. This process is knownpan> as optical crosstalk anpan>d leads to falsely stimulated micro-cells anpan>d furthermore to anpan> effective decrease of the actual SiPM dynamic ranpan>ge. Optical crosstalk may be further facilitated by optical media such as the SiPM enpan>tranpan>ce window (typically coated by epoxy or resin) or the scintillation crystal itself.

The aforementioned limitations in SiPM performance are being addressed by a number of specialized designs:

Back-illuminated SiPMs for enhanced n class="Gene">PDE [41]

SiPMs with an enhanced p-on-n structure for sensitivity in blue light [42]

SiPMs with quenching resistors in the bulk for elimination of the polyn class="Chemical">silicon resistor [43]

SiPMs with dual p-n structure for cross-talk elimination [44]

SiPMs with intra-cell trenches for cross-talk elimination [44]

It is expected that in the near future these shortcomings will be eliminated to n class="CellLine">a large extenpan>t. However, despite the aforemenpan>tioned challenpan>ges anpan>d limitations, scintillator readout by SiPMs shows great promise for fast timing applications, such as ToF-PET (Figure 17):

Figure 17.

Dependence of coincidence time resolution on SiPM bias (measurement performed for a pair of LSO-SiPM detectors). Data taken with two SiPMs with a 3 × 3 mm2 cross-sectional area from Hamamatsu (50 μm micro-cell size). The LSO crystal size was 3 × 3 × 5 mm3. The time resolution value at the optimum bias setting is ∼280 ps (FWHM).

intrinsic time resolution values at the single photoelectron level as low as 70 ps have been reported [45]

LaBr3 readout with SiPMs demonstrates time resolution in the range of 100 ps [20]

n class="Chemical">LYSO readout with SiPMs demonstrates time resolution in the ranpan>ge of 300–200 ps [29]

ToF-PET State of the Art Systems

A number of ToF-PET scanners are currently commercially available. Even though the current generation of ToF-PET systems is based on n class="Chemical">PMTs, it is expected that in future systems the PET detectors will be SiPM-based in order to utilize compact detector designs while achieving comparable or, potenpan>tially, superior time resolution.

Philips TF PET/CT. The basic detector component of the system consists of arn class="Species">rays of pan> class="Chemical">LYSO pixels (4 × 4 × 22 mm3 pixel size) read out by hexagonal arrays of PMTs. The achieved time resolution of the system varies between 600 and 900 ps depending on the photon event count rate [46].

Siemens Biograph TruePoint PET/CT. The detectors of this system are arn class="Species">rays of LSO crystal pixels (4×4×20 mm3 pixel size) read out by arpan> class="Species">rays of PMTs and can achieve 500 ps time resolution [47].

GE Discovery 690. This ToF system also employs arn class="Species">rays of pan> class="Chemical">LYSO crystal pixels (pixel size is 4.25 × 6.3 × 25 mm3) read out by PMTs. The achieved time resolution is 600 ps [48].

Future of ToF-PET Photo-Detector Technology

Digital SiPMs

A novel technological advance in the field of semiconductor photo-detectors has been recently developed by Philips and involves the integration within the SiPM sensitive area of basic processing electronics thus eliminating the need for external processing electronics [49,50]. Each micro-cell of the array is connected to an integrated counter (for extraction of energy information) and an integrated time to digital converter (TDC, for extraction of time information). This alternative SiPM design is known as ’digital SiPM’ and time resolutions as low as 150 ps (FWHM) with n class="Chemical">LYSO readout have beenpan> reported.

The development of fully digital photon detection technologies based on single photon avalanche photo-diodes has been reported by other groups [51] even though these technologies have not been tested for annihilation radiation detection. The so-called single-photon synchronous detectors (SPSDs) are able to measure both intensity and phase of the incident light which can be exploited to compute the position of a target based on the phase difference of the light incident to the target and the light reflected by the target and detected by the SPSD.

Apart from the obvious advantage of simplified external signal processing characterizing the digital implementation of single photon detectors, there are some potential disadvantages associated with this technology, which include an increase in power consumption, as well as reduced fill factor due to the space required for the additional electronic circuits. In addition, since the analog form of the generated signal is not available, there may be some lack of useful information content, mostly associated with pulse shape variations.

Discrete amplification photodiodes

Another SiPM design that aims to address both the limited n class="Gene">PDE anpan>d the recovery time issues of passive quenpan>ching is employed by discrete amplificationpan> photo-diodes (pan> class="Chemical">DAPDs) [52,53]. The internal architecture of these devices is fundamentally different from the micro-cell array concept of traditional SiPMs. The charge produced from a single photosensitive area is distributed to a number of discrete amplifiers which are integrated in the diode’s bulk. Each charge packet will be subject to individual amplification and the output signal will be the sum of the amplified charge packets. Due to the absence of long traces surrounding each micro-cell, DAPDs have very low capacitance which results in short recovery times.