Design of CMOS Digital Silicon Photomultipliers with ToF for Positron Emission Tomography

Resum

This thesis presents a contribution to the design of single-photon detectors for medical imaging. Specifically, the focus has been on the development of a pixel capable of single-photon counting in CMOS technology, and the associated sensor thereof. These sensors can work under low light conditions and provide timing information to determine the time-stamp of the incoming photons. For instance, this is particularly attractive for applications that rely either on time-of-flight measurements or on exponential decay determination of the light source, like positron emission tomography or fluorescence-lifetime imaging, respectively. This thesis proposes the study of the pixel architecture to optimize its performance in terms of sensitivity, linearity and signal to noise ratio. The design of the pixel has followed a bottom-up approach, taking care of the smallest building block and studying how the different architecture choices affect performance. Among the various building blocks needed, special emphasis has been placed on the following: • the Single-Photon Avalanche Diode (SPAD), a photodiode able to detect photons one by one; • the front-end circuitry of this diode, commonly called quenching and recharge circuit; • the Time-to-Digital Converter (TDC), which determines the timing performance of the pixel. The proposed architectural exploration provides a comprehensive insight into the design space of the pixel, allowing to determine the optimum design points in terms of sensor sensitivity, linearity or signal to noise ratio, thus helping designers to navigate through non-straightforward trade-offs. The proposed TDC is based on a voltage-controlled ring oscillator, since this architecture provides moderate time resolutions while keeping the footprint, the power, and conversion time relatively small. Two pseudo-differential delay stages have been studied, one with cross-coupled PMOS transistors and the other with cross-coupled inverters. Analytical studies and simulations have shown that cross-coupled inverters are the most appropriate to implement the TDC because they achieve better time resolution with smaller energy per conversion than cross-coupled PMOS transistor stages. A 1.3×1.3 mm2 pixel has been implemented in an 110 nm CMOS image sensor technology, to have the benefits of sub-micron technologies along with the cleanliness of CMOS image sensor technologies. The fabricated chips have been used to characterize the single-photon avalanche diodes. The results agree with expectations: a maximum photon detection probability of 46 % and a median dark count rate of 0.4 Hz/µm2 with an excess voltage of 3 V. Furthermore, the characterization of the TDC shows that the time resolution is below 100 ps, which agrees with post-layout simulations. The differential non-linearity is ±0.4LSB, and the integral non-linearity is ±6.1LSB. Photoemission occurs during characterization - an indication that the avalanches are not quenched properly. The cause of this has been identified to be in the design of the SPAD and the quenching circuit. SPADs are sensitive devices which maximum reverse current must be well defined and limited by the quenching circuit, otherwise unwanted effects like excessive cross-talk, noise, and power consumption may happen. Although this issue limits the operation of the implemented pixel, the information obtained during the characterization will help to avoid mistakes in future implementations.