High-speed, high-fidelity solutions for single photon counting and timing applications

Single photon photodetectors are used in a wide variety of applications, from remote sensing to biological applications, from quantum key distribution to non-invasive medical imaging. In satellite Light Detection and Ranging (LiDAR) applications, for example, photodetectors are used to extract important parameters concerning the constitution of the atmosphere, in particular of clouds and aerosols. They are used to evaluate the yield of large agricultural areas as well as for the control of forest canopy height. Some LiDAR systems have even proved capable of allowing the study of particular phenomena for which they had not been developed, such as, for example, the study of the complex mechanisms governing the atmosphere-ocean interaction. A LiDAR system consists of a laser emitter and a photodetector receiver. Periodically the system emits a laser pulse towards a target and the number of photons hitting the receiver with respect to a time reference is recorded. There are different types of laser systems, but here I will focus on ‘backscatter’ or ‘full-waveform’ LiDARs. These particular systems only require counting the number of photons that are backscattered. Knowing the number of photons that are backscattered by the target instant by instant, it is possible to i) locate the target, and ii) allow to retrive the physical-chemical properties of the target. The recent satellite LiDAR systems, such as for example the Cloud-Aerosol LiDAR with Orthogonal Polarization (CALIOP), how- ever, are characterized by an acquisition system with a limited bandwidth, prohibiting for example the possibility to vertically resolve extinction within the cloud. In addition, the laser beam hits tar- gets which return a very variable number of photons: for example, the return signal from the aerosol is ∼ 106 weaker than that of a cloud. This scenario requires a system characterized by a high dynamic range. Moreover, it is further aggravated when is necessary to study the first meters below the ocean surface. To obtain a high dynamic range it is possible to proceed in different ways. First of all, it is possible to reduce the noise of the detector. In this context, the use of photodetectors based on Single Photon Avalanche Diode (SPAD) provides a valid solution, especially if the detector is made with a custom manufacturing process. A custom manufacturing approach gives the designer every degree of freedom to customize and optimize the detector to meet application requirements. A custom SPAD can therefore be optimized in order to have a low dark count rate and ensure, for example, a high quantum efficiency at the wavelength of interest, which is particularly important for collecting information from weaker signals. In order to further optimize the dynamic range, it is possible to use SPADs characterized by a small active area and consequently characterized by a lower dark count rate compared to an equivalent SPAD with a larger active area. Yet using a SPAD that is too small makes it more optically difficult to channel the light, complicating the optical setup. To overcome this problem it is possible to use bidimensional arrays of SPADs, which facilitate the light focus on the detector. Finally, to reach the high dynamic range required, it is necessary to drive the SPAD quickly to reduce its dead time. To do this, a special quenching circuit is needed, optimized to be fast. In this regard, I have been involved in optimizing an active quenching circuit in order to reduce the dead time of the detector, without worsening other metrics of interest such as occupied space and dissipated power. To effectively connect a bidimensional array of custom SPADs to the respective quench circuits, it is necessary to carry out a 3D stacking of the two chips, superimposing them on each other. In this regard, I was responsible for laying the foundations for the floorplanning of the first 8x8 bidimensional array of 3D quenching circuits stacked under a same-size array of SPADs. SPADs, in addition to being able to detect the single photon, are able to provide the arrival time of the photon with exceptional precision. This peculiarity makes them the perfect candidates for Time-Correlated Single Photon Counting (TCSPC) applications. Nowadays, TCSPC represents a key measurement technique in many scientific and industrial applications demanding for the acquisition of extremely fast and faint luminous signals with picosecond resolution. In a TCSPC experiment the intensity of the light signal from a sample under investigation is recorded as a function of time. To accomplish this task, the sample is excited with very short light pulses and the time delays between the excitation pulse and the detection of a single photon is recorded. By repeating this procedure many times the probability distribution for the emission of a single photon is obtained. The result is the generation of a time histogram in which each time bin contains the information on how many photons have been recorded in that time frame. To avoid the distortion of the recorded curve caused by the so- called pile-up effect, the average number of detected photons per period is typically limited between 1% and 5%. Furthermore, many acquisitions are required to correctly build the histogram. Consequently, TCSPC is acknowledged as a relatively slow technique com- pared to other imaging techniques such as the time-gated. One solution to speed up a TCSPC measurement consists in parallelizing the number of acquisition channels, moving from a single channel to a multichannel architecture. While there are multichannel modules available in the market that are based on discrete components and offer superior resolution and linearity, they have certain drawbacks. Specifically, the high power dissipation and large space requirement per channel have restricted the level of parallelism achievable to only 4 or 8 channels at present. On the other hand, the advancements of CMOS technologies have enabled the development of multichannel systems comprised of linear or bidimensional arrays of SPADs, and the associated electronics. However, the existing proposals for large arrays with detectors and electronics combined on a single chip face a dilemma between channel count and performance. This is because incorporating both detectors and conversion electronics within the same pixel area places strict limitations on the power dissipation and area occupied by the electronics, thereby compromising timing performance in terms of linearity and precision. Additionally, using standard CMOS technologies restrict the designers from achieving the optimal detector performance in terms of photon detection efficiency, dark count rate, and afterpulsing probability. Moreover, large multichannel systems suffer from a bottleneck in data transfer, which significantly restricts the measurement speed that can be achieved. Large dense SPAD arrays can theoretically produce large amount of data, which can easily reach 100 Gbit/s. Unfortunately, such high- speed real-time handling necessitates a huge bus bandwidth to the external processor, a high number of I/O pads, as well as significant system complexity. As a result, bandwidth saturation is currently one of the primary constraints on the speed of TCSPC measurements. Among the various readout architectures proposed in the literature to manage the amount of generated data, the one proposed by Cominelli et al. shows promising results, allowing the best use of available resources and ensuring the best use of the transfer band- width. This architecture, called router-based, starting from the transfer bandwidth, and from the conditions of a typical TCSPC experiment, determines the number of converters needed to get the most of it. The result is an architecture which consists in associating a large number of pixels to a limited set of external time converters, guaranteeing the absence of bias in the association phase. The limited number of time converters allows for extreme optimization, over- coming the constraints of occupied space and dissipated power that typically afflict other readout architectures. Starting from already designed implementations, I dealt with their updating and development, overcoming in particular some limitations of these architectures and proposing new solutions. For the first time, the difficult issue of extracting and sending time information from pixels to time converters has been tackled. After having experimentally tested different circuit blocks constituting the architecture, I started to lay the foundations for the design of a first 64-channel implementation and to realize the design of an 8-channel architecture.