Typically, a time-of-flight (TOF) PET block detector is built using application-specific integrated circuits (ASICs), since they integrate a high number of channels at a reasonable power consumption and into a small area. However, ASICs’ flexibility is limited and prototyping times are long because a semiconductor fabrication process is required in every design iteration. Alternatively, fast terminal (FT) silicon photomultipliers (SiPMs) require a simplified analog front-end in order to achieve time-of-flight (TOF) accuracy. In addition, field-programmable gate arrays (FPGAs) can allocate time-to-digital converters (TDCs) as well as complex digital readout logics. In this work, we propose building TOF-PET block detectors based on FPGAs, FT-SiPMs, and minimal amount of off-the-shelf components. In this way, TOF-PET accuracy is achieved with a full-flexible and fast prototyping solution. We evaluated the coincidence resolving time (CRT), performance degradations due to channel multiplexing, energy resolution, and scintillator pixel encoding performance of SiPM arrays utilizing the proposed approach. Experimental results show minimal timing degradations, when multiplexing FTs. Moreover, simulation results show a low reduction in the singles count rate of multiplexed channels at typical brain-PET radioactive doses.

In this paper we present a silicon photomultiplier (SiPM) readout system based on a field-programmable gate array (FPGA), which is capable of converting any commercial 16-channel analog SiPM array into a hybrid device with fullydigital readout for application in time of flight positron emission tomography (TOF-PET). In principle this hybrid SiPM array can be implemented with several leading edge discriminators (LEDs) per channel, so that multiple timestamps can be acquired per scintillation pulse allowing to estimate the time of interaction more robustly than if a single timestamp is used. These concepts were studied experimentally in two different ways. In the first approach, we utilized discrete components in combination with an array of time-to-digital converters (TDCs) implemented and validated on the FPGA. The overall resolution of each channel was 56 ps FWHM or better, while crosstalk was undetectable. In the second approach, we utilized a four-channel high-speed acquisition board operating at 5 GSPS and emulated a multiple-LED acquisition by post-processing the digitized waveforms. We present, for the first time, experimental results obtained with the best linear unbiased estimator (BLUE) to estimate the time of interaction from the multiple timestamps.