Thomas Frach, Member, IEEE, Walter Ruetten, Member, IEEE, Klaus Fiedler, Gunnar Maehlum, Member, IEEE, Torsten Solf, and Andreas Thon

Transcription

1 Assessment of Photodiodes as a Light Detector for PET Scanners Thomas Frach, Member, IEEE, Walter Ruetten, Member, IEEE, Klaus Fiedler, Gunnar Maehlum, Member, IEEE, Torsten Solf, and Andreas Thon Abstract Current PET systems based on pixelated scintillator arrays coupled to photomultiplier tubes suffer from pile-up and electronics dead time at high count rates. With a pixelated readout, i.e. one-to-one coupling of a scintillator crystal to a photo detector, these effects can be strongly reduced. Recent developments of high light output scintillators like LYSO and LaBr in combination with very low noise amplifiers based on modern CMOS processes make it possible to use high quantum efficiency blue-sensitive PIN photodiodes as a light detector. To explore the potential of this approach, a model of the signal detection chain was implemented. It comprises the scintillation light pulse and its quantum noise, optical coupling, charge conversion in the diode, noise sources of the integrating amplifier, shaper circuits for the energy and timing channel, and the discriminator for the timing channel. The model is verified using off-the-shelf PIN photodiodes and a dedicated CMOS preamplifier excited by a picosecond laser as well as scintillator pulses. The model predicts that with high light output scintillators, high quantum efficiency photodiodes and optimized preamplifiers, a pixelated PET readout with very good energy resolution and sufficient timing resolution can be realized. To complement the study, an APD-based readout is also considered and the related signal to noise issues are discussed. Index Terms PET, pixelated readout, photodiode readout. I. INTRODUCTION CURRENT PET systems are usually based on a pixelated scintillator array coupled to an array of photomultiplier tubes (PMTs) via a light guide. While this technology is proven and allows for a good resolution in energy and time, the relatively high number of scintillator pixels per PMT leads to signal pile-up and electronics dead time at higher count rates. This problem can be avoided by using a pixelated readout, i.e. by coupling an individual photo detector to each scintillator crystal. Several research institutions have investigated this concept [] [] using avalanche photodiodes (APDs) or pixelated PMTs, which raises stability and price issues especially for full-body PET scanners. Considering the advances in the development of high light output scintillators (LYSO, LaBr ) and low-noise amplifiers, the approach of using PIN photodiodes (PDs) for a PET detector becomes feasible. This paper investigates the various factors Manuscript received October 9, T. Frach, W. Ruetten, K. Fiedler, T. Solf and A. Thon are with Philips Research Laboratories, Aachen, Germany. Thomas.Frach@philips.com G. Maehlum is with IDEAS ASA, Fornebu, Norway. gunnar@ideas.no influencing the performance of scintillator-photodiode-amplifier combinations. Hereby, both detailed model calculations and experimental data are presented, and the results are compared to the performance of APD detectors. As for the application, photodiodes can be used as stand-alone readout in a high count rate, high spatial resolution PET scanner, or as part of a hybrid readout system for a depth-of-interaction detector. II. SIMULATION The individual steps for the modeling of a PD-based PET detector are shown in Fig.. The calculation starts with the scintillation light pulse and takes into account the optical coupling of the scintillator to the photodiode, the conversion to electrical charge in the photodiode, the charge integration and the pulse shaping with a CR-RC shaper in the energy and timing branch of the readout electronics. Several noise sources like the photon and electron quantum noise, the photodiode dark current, the thermal noise and the /f noise of the integrator are included in the model. The equivalent noise charges from the photodiode dark current (ENC d ), the thermal noise (ENC C ) and the /f noise (ENC F ) of the first MOSFET in the integrator are given by ENC d = e τ s qi d () q ENC C = C T e q ENC F = C T e q kt ( + η) g m τ s () K f C oxwl π, () where q is the elementary charge, τ s is the time constant of the CR-RC shaper, I d is the dark current of the photodiode, C T is the total capacitance at the input node of the integrator, g m is the transconductance of the first MOSFET of the integrator, η is the ratio of bulk-channel to gate-channel transconductance, K f is the /f noise coefficient, C ox is the oxide capacitance per area and W and L are the width and length of the first MOSFET in the integrator (see also []). The equations show that a low capacitance C T at the integrator input is important to get a low noise contribution. A high transconductance g m and large width W of the first MOSFET also serve to keep the noise low //$. (C) IEEE

2 γ Light generation coupling photodiode integrator efficiency Fig.. discriminator Vt shaper shaper Model of the scintillator-photodiode-amplifier configuration. Energy The time dependent noise power of the quantum noise n s (t) is transfered to the shaper output according to Time n g(t) = n s(t) h (t), () where n s (t) is the quantum noise at the input of the integrator, h(t) is the combined transfer function of integrator and shaper and denotes the convolution operator. On the other hand, the signal has to be maximized using high light output scintillators, high quantum efficiency photodiodes and a good optical coupling of the two. The shaping times of the energy and timing branch can be independently chosen to provide the best signal to noise ratio at the peak signal and at trigger time respectively. If the model is applied to avalanche photodiodes, the avalanche gain M, which affects the signal and the quantum noise, is taken into account. The excess noise resulting from the stochastic nature of the avalanche process adds a further noise component to the simulation. III. SIMULATION RESULTS To explore the potential of a photodiode-based PET readout, the model was used to assess LYSO and LaBr scintillator crystals coupled to a high quantum efficiency PIN photodiode. Hereby, the photodiode parameters were chosen to correspond to commercially available devices, with the capacitance and the quantum efficiency having a major impact on the signal to noise (S/N) performance. The input stage of the preamplifier, which is the dominant noise source, was designed with the parameters of a conventional CMOS process. As an example, Fig. and show the simulated energy and coincidence timing resolution for a mm mm photodiode with pf capacitance, 9 % quantum efficiency and a preamplifier based on a. µm CMOS process. Depending on the coupling efficiency between the scintillator and the photodiode, an energy resolution of % and a coincidence timing resolution of. ns can be reached with a LaBr scintillator at optimized shaping times. The optimal shaping times in this setup are around ns for the timing branch and ns or more for the energy branch (limited by electronic pile-up). In general, it can be concluded that PD-based PET detectors making use of high light output scintillators can provide very Energy Resolution (FWHM) [%] 8 8 optical coupling =. optical coupling =. optical coupling = Shaping Time [ns] Fig.. Simulated energy resolution (FWHM) over shaping time for a photodiode with a LaBr scintillator. Coincidence Timing Res. (FWHM) [ns] optical coupling =. optical coupling =. optical coupling = Shaping Time [ns] Fig.. Simulated coincidence timing resolution (FWHM) over shaping time for a photodiode with a LaBr scintillator ( kev trigger threshold). good energy resolution and sufficient timing resolution for a standard PET scanner. The simulation results also show that the optimal shaping times differ for the energy and timing branch. To complement the study, an APD-based detector system using parameters of a commercially available APD is compared to a PD-based system. Fig. shows the expected S/N, which determines the energy resolution E/E and the coincidence timing resolution t for an APD-based detector versus the gain M. For a typical shaping time of ns, the optimum S/N in the energy channel is for a gain of M, while the PD-based system reaches an S/N of 9. The inferior S/N of the APD can be traced back to the lower quantum efficiency, the higher capacitance of the APD and the intrinsic noise of the avalanche process. The lower quantum efficiency leads to a lower signal to quantum noise ratio in the charge signal. The high capacitance results in an increased amplifier noise (see (), ()). Hence a large part of the APD gain is required to match the signal to electronic noise ratio of the photodiode. Finally, the signal to quantum noise ratio present in the APD is //$. (C) IEEE

3 Time [ns] TABLE I PARAMETERS OF PHOTODIODES AND APDS USED IN THE VERIFICATION SNR PD (Output) PD (Trigger) APD (Output) APD (Trigger) APD Gain Type Area Capaci- Dark Rev. nm tance current Bias [mm ] [pf] [na] [V] S7 (PD, Hamamatsu) PDCs-mu (PD, Detection Technology) S (APD, Hamamatsu) 7.. Fig.. Signal to noise ratio for PDs and APDs at peak signal for energy determination (output) and at trigger time (trigger). Shaping time τ s = ns. Fig.. Laser Attenuation Photodiode Measurement setup with picosecond laser. progressively degraded with increasing gain due to the excess noise of the avalanche process. Fig. also shows that for an optimum time stamp, a higher APD gain of 7 is needed, resulting in a S/N of at trigger time. The PD-based system still has a higher S/N of, but the advantage of the PD is smaller at the chosen trigger level of kev. It should be noted that in setting a fixed APD gain, a tradeoff between energy and timing resolution has to be made. IV. LASER MEASUREMENTS To validate the simulation model, prototype low noise amplifier ASICs have been developed and manufactured by IDEAS based on.8 µm and. µm CMOS processes. The design contains several input channels with front-end transistors of varying geometry, connected to shapers and discriminators. The measurement setup is shown in Fig.. The picosecond pulsed laser emits ps FWHM pulses at a wavelength of nm. The pulses are attenuated and focused onto the sensitive surface of the photodiode. In preparation for the crystal measurements, the packages of the photodiodes were opened to allow direct access to the active area of the die. The photodiode parameters are listed in Table I. The photodiodes and the APD were connected to the.8 µm front-end, the PDCs was also measured with the. µm front-end. Peaking times were 9 ns with the S7 photodiode, ns with the PDCs connected to the. µm ASIC and ns with the S APD. A digital sampling oscilloscope (LeCroy WavePro 9) was used to record the pulses for offline analysis. A suitable trigger signal was provided by the picosecond laser controller. Advanced Photonic Systems, head: PILG, controller: EIGD ASIC Amplitude [V] Peak Output Voltage [V]... PDC (PD).8um ASIC PDC (PD).um ASIC S7 (PD).8 um ASIC S (APD).8um ASIC Fig.. Response of (A)PD-amplifier combinations with picosecond laser pulses. Error bars indicate the RMS noise voltage. The PD-ASIC response to the laser pulses is shown in Fig.. The S APD has an effective gain of over the S7 photodiode. The RMS noise voltages are listed in Table II. Fig. 7 shows the energy resolution measured for the various photodiodes and also provides simulation data for two configurations. The agreement between simulation and measurement is good for attenuations up to to. For higher attenuations, the deviations become larger. The timing resolution (Fig. 8) shows a good agreement for the PDCs photodiode. With the S7 photodiode, the agreement is good for attenuations of up to. However, the measured timing resolution deviates from the predicted one for higher attenuations (low light intensity). The reasons for this behaviour are currently being investigated. TABLE II RMS NOISE OF (A)PD-ASIC COMBINATION FOR ATTENUATION OF (A)PD ASIC RMS noise [ mv ] PDCs.8 µm 8. PDCs. µm. S7.8 µm 9.8 S.8 µm //$. (C) IEEE

4 Energy Resolution (FWHM) [%] Simulation S7.8um ASIC Simulation PDC.8um ASIC S7 (PD).8um ASIC PDC (PD).8um ASIC PDC (PD).um ASIC S (APD).8um ASIC 8 8 FWHM =.9 % 7 Fig. 7. Measurement of the PD energy resolution with a picosecond laser. Fig.. Energy spectrum obtained with a mm mm 8 mm MLS crystal and a PDCs photodiode. Reverse bias V,. µm ASIC. Timing Resolution (FWHM) [ns] Simulation S7.8um ASIC Simulation PDC.8um ASIC S7 (PD).8um ASIC PDC (PD).8um ASIC PDC (PD).um ASIC S (APD).8um ASIC FWHM =. ns Timing Resolution [ns] Fig. 8. Measurement of the PD timing resolution with a picosecond laser. V. CRYSTAL MEASUREMENTS The photodiode-amplifier combinations were further evaluated using available scintillator crystals. Fig. 9 shows the coincidence setup used for the crystal measurements. A Na source with 7 kbq was placed between the reference PMT and the photodiode under test. The reference PMT was a Hamamatsu R9 ( mm diameter) coupled to a mm mm mm LYSO crystal. The signals from the photodiode were acquired with an oscilloscope using the PMT signal as trigger signal and taking the amplitudes of both the PMT signal and the photodiode signal into account in a qualified trigger. All scintillator crystals were wrapped in a reflective coating and coupled to the photodiode with optical grease in order to maximize the signal. Fig. shows the energy spectrum obtained with a mm Fig. 9. PMT Na Photodiode Setup for coincidence measurements. ASIC Digital Oscilloscope Fig.. Timing distribution obtained with a mm mm 8 mm MLS crystal and a PDCs photodiode. Reverse bias V,. µm ASIC, kev trigger threshold. mm 8 mm MLS scintillator, the PDCs photodiode and the. µm ASIC. The FWHM energy resolution for this setup is.9%. The single channel FWHM timing resolution for this setup is. ns (Fig. ). The delay of the PD timing signal relative to the PMT trigger is due to the shaping in the ASIC. Fig. compares the energy resolution of a mm mm 8 mm MLS scintillator and a mm mm mm LYSO crystal coupled to an S7 photodiode and the.8 µm ASIC. The shorter geometry and better light yield of the MLS crystal compared to the LYSO crystal result in a slightly better energy resolution of % for MLS compared to % for LYSO. Measurements with an S APD coupled to mm mm mm LYSO crystal are given in Figs. and. The FWHM energy resolution reaches.8% after drift correction, the single channel FWHM timing resolution is. ns. The ASIC settings were the same as used in the photodiode measurements. The output signal from the test setup showed that the optical coupling between the scintillator crystals and the photodiodes //$. (C) IEEE

5 Probability [%] LYSO xx mm MLS xx8 mm 7 8 Fig.. Energy spectra of a mm mm mm LYSO crystal and a mm mm 8 mm MLS crystal coupled to an S7 PD. Reverse bias V,.8 µm ASIC. FWHM =.8 % 7 Fig.. Energy spectrum obtained with a mm mm mm LYSO crystal and an S APD. Reverse bias V,.8 µm ASIC. FWHM =. ns Timing Resolution [ns] Fig.. Timing distribution obtained with a mm mm mm LYSO crystal and an S APD. Reverse bias V,.8 µm ASIC. was.. A kev energy deposition resulted in a signal comparable to a laser pulse with an attenuation of. A LaBr scintillator should give approximately twice the signal. The measurements with scintillator crystals and off-the-shelf photodiodes show that energy and timing resolution need to be improved further. As can be inferred from Fig., the timing resolution deteriorates fast with low optical coupling. VI. CONCLUSION A model for a photodiode based PET front-end was developed to predict the achievable energy and timing resolution. It includes optical quantum noise, electronic noise contributions and, in the case of APDs, excess noise from the avalanche process. Based on the simulations with high light output scintillators and high quantum efficiency photodiodes, the requirements for a readout ASIC were derived and prototype ASICs in two different CMOS processed were designed. The simulation model predicts that energy and timing resolutions suitable for a PET scanner are achievable. The simulation model was verified against a test setup using low noise amplifier ASICs together with photodiodes excited by a picosecond pulse laser. For low capacitance photodiodes, a good agreement was found between laser measurements and simulations. First measurements with medium light output scintillators (LYSO, MLS) and off-the-shelf photodiodes show that an optimization of the whole signal chain, especially the light yield, quantum efficiency, optical coupling and photodiode capacitance, is important. With further improvements in the optical coupling, a photodiode readout suitable for PET applications can be realized. ACKNOWLEDGMENT The authors would like to thank Sindre Mikkelsen, Dag Mattis Pettersen, Jahanzad Talebi and Petter Øya of IDEAS ASA for their work on the ASIC and PCB design and Jussi Koskinen and Tuomas Holma of Detection Technology Inc. for samples of the PDC photodiode. REFERENCES [] R. Lecomte, J. Cadorette, S. Rodrigue, D. Lapointe, D. Rouleau, M. Bentourkia, R. Yao, and P. Msaki, Initial results from the Sherbrooke avalanche photodiode positron tomograph, IEEE Trans. Nucl. Sci., vol., no., pp. 9 97, June 99. [] B. Pichler, F. Bernecker, G. Boning, M. Rafecas, W. Pimpl, M. Schwaiger, E. Lorenz, and S. Ziegler, A 8 APD array, consisting of two monolithic silicon wafers, coupled to a -channel LSO matrix for high-resolution PET, IEEE Trans. Nucl. Sci., vol. 8, no., pp. 9 9, Aug.. [] B. Pichler, W. Pimpl, W. Buttler, L. Kotoulas, G. Boning, M. Rafecas, E. Lorenz, and S. Ziegler, Integrated low-noise low-power fast chargesensitive preamplifier for avalanche photodiodes in JFET-CMOS technology, IEEE Trans. Nucl. Sci., vol. 8, no., pp. 7 7, Dec.. [] M. Kapusta, P. Crespo, M. Moszynski, W. Enghardt, M. Szawlowski, B. Zhou, and D. Wolski, Evaluation of LAAPD arrays for high-resolution scintillator matrices readout, IEEE Trans. Nucl. Sci., vol. 9, no., pp. 9 98, Aug.. [] E. Nygard, P. Aspell, P. Jarron, P. Weilhammer, and K. Yoshioka, CMOS low noise amplifier for micro strip readout Design and results, Nucl. Instr. and Meth., vol. A, pp., //$. (C) IEEE