Time-Correlated-Single-Photon-Counting systems: challenging the limits

The analysis of optical signals by means of Single-Photon Avalanche Diodes(SPADs) has undergone a huge spread in recent years, thanks to the achieved ultimate sensitivity, which raised the interest of several applications from different fields, both scientific and industrial. For instance, systems featuring single-photon detectors are effectively employed for Fluorescence Lifetime Imaging Microscopy (FLIM) and Förster Resonance Energy Transfer (FRET) in life science, for Laser Imaging Detection and Ranging (LIDAR) in remote object sensing and for Quantum Key Distribution (QKD) in quantum cryptography. Besides providing single-photon sensitivity, SPADs can also detect the arrival time of photons with a jitter as low as few tens of picoseconds. These two features are combined by Time-Correlated Single Photon Counting (TCSPC), which is a very efficient technique for measuring weak and fast periodic signals. The technique relies on the periodic excitation of the sample of interest by means of a pulsed light source and on the recording of the arrival times of the emitted photons. The power of the light source is kept low enough so that one photon is detected per each excitation cycle at maximum. The measure is repeated and the histogram of the arrival times with respect to the optical period is built, which represents, after many measurements, the light intensity profile over time. The remarkable timing precision that is achievable made TCSPC be acknowledged as the gold standard for FLIM and spread it in the life science field in general. For instance, this technique allows to overcome some limits characteristic of intensity-based measurements that affect applications like FRET, which suffers for environmental changes when tracking the cells workings at very short distances. Despite the remarkable sensitivity and timing precision that TCSPC can achieve, this technique intrinsically suffers for a relative “long” measurement time, due to the fact that the measure has to be repeated many times to record enough photons and reconstruct the desired statistics. Aiming to overcome this limitation, many advanced TCSPC instruments have been developed, which feature several channels operated in parallel to reduce the measurement time. Nevertheless, if we look at state-of-the-art instruments, we clearly see that there is a strong trade-off. Indeed, on one hand, commercial systems feature few channels but with remarkable performance, especially in terms of timing precision, differential nonlinearity (DNL) and conversion frequency; whereas, on the other hand, research systems feature a significantly higher number of channels, even few tens of thousands, but with performance that is far away from the best in class. In multichannel systems, the huge number of channels is obtained by exploiting the standard CMOS process, which allows the integration of both the sensor and processing electronics on the same die. In this way, very dense array can be developed, but the performance is limited, because the technology is not optimized for the design of SPAD detectors and the power consumption is a constraint as well. Moreover, in the so called “smart pixel” architecture, each detector is associated to a converter, either Time-to-Digital Converter (TDC) or Time-to-Amplitude Converter (TAC), but this solution does not come with significant advantages when dense arrays are exploited. Indeed, in order to avoid distortion of the reconstructed waveform, the count rate of each pixel is kept to few percent of the laser frequency, therefore for most of the time a single converter is not working, leading to a low utilization efficiency. Furthermore, with dense arrays, the throughput that is generated is remarkable and it cannot be managed, leading to a decrease of the efficiency. I decided to follow an orthogonal approach to break the existing trade-off, by first defining the maximum throughput that the data link toward the PC can sustain (about 10 Gbps) and then tailoring the number of acquisition chains to saturate it. Since in advanced TCSPC system the ASIC and system designs are strongly connected to get the desired performance, I contributed to the development of an innovative routing architecture and a fast TAC. The purpose of the designed router is to connect the fired detectors to the lower number of acquisition chains available. Despite this solution’s being easy at a first glance, it is anything but trivial. Indeed, the easiest way would be to divide the SPAD array into sub portions and associate each one to a single converter by exploiting static multiplexers, but it would not ensure the maximum resource utilization if the illumination were sparse. Conversely, the developed smart architecture extracts the fired detectors in a random and unbiased way, by assigning variable priorities to each pixel and selecting the highest ones. In this way, the reconstructed image is not distorted due the presence of hot pixels. Finally, the router architecture is able to extract the number of pixels to be connected to the acquisition chains at each excitation cycle, i.e. the featured dead time is negligible with respect to the excitation period in order not slow down the measurement speed. The fast TAC (F-TAC) is meant for working in conjunction with the smart router and process the received timing signals without blocking the data flow. The core of this device is a single-channel Time-to-Amplitude Converter that is parallelized 16 times to make the overall dead time negligible with respect to the laser operating frequency (typically up to 80 MHz). Even if the converter has been designed to be connected to the router, it can be also effectively employed with single detectors, because the negligible dead time leads to an increase of the recording efficiency and therefore to a reduction of the measurement time. When operating together, the router and the F-TAC generate a data throughput of several Gbps that has to be properly managed in order not to lose information and impair the measurement speed. In particular, data should be managed both upstream, at the system level, and downstream, at the PC level, to avoid bottlenecks during the acquisition. As for the upstream management, I designed and developed hardware and firmware solutions to implement on-board histogram feature and time-tag mode. Regarding the downstream management, there are two aspects that have to be carefully considered: the physical communication with the PC and the software processing. To this aim, I developed a USB 3.0 connection, since it is wide spread, and an Ethernet 10G link, for the most demanding applications, like High Content Screening (HCS), to deal with the data transfer and sustain the generated throughput. As for the software processing, it is a crucial part of the system, because when running at full speed, the 10-Gbps throughput that is generated cannot be stored onto an external SSD, otherwise the storing capacity would limit the measurement duration. Conversely, it is of utmost importance that the software receives the data and directly extracts the desired information. To this aim, a collaboration has started with another research group at Politecnico di Milano, to develop a custom software solution able to process the recorded photons without impairing the throughput. Besides exploiting several detectors in parallel to reduce the measurement time, advanced TCSPC systems can interface with microscopes to perform a scan of the sample. Indeed, even if modern instruments can feature thousands of detectors, the number of pixels is not high enough for some FLIM applications. To this aim, I developed a general-purpose firmware interface to manage the synchronization signals and divide the acquisition into pixels and frames. The proposed solution can either receive the synchronization signals from the microscope or provide it with the properly phased signals. I implemented some of the developed solutions on a 32-channel complete TCSPC system to test and validate them. The instrument, which features state-of-the-art performance, has been employed on a research project to distinguish various stages of aggregation of alpha synuclein (aSyn) in cells, which is a small, natively unstructured protein that can aggregate into insoluble structures that are toxic to neurons, a phenomenon closely linked to the pathology of Parkinson’s disease. The system should enable the direct measurement of the efficacy of aggregation-inhibiting drugs. The results obtained from the testing of the 32-channel system were state of the art and showed that a break of the trade-off was effectively feasible. These premises pushed me toward the design of advanced TCSPC systems that could feature a higher and higher number of channels. As a first technological step on this way, I worked on a 32 32 instrument whose main core is a SPAD array developed exploiting a custom technology. The main goal of this work was to analyze the main issues that affect the design of large TCSPC instruments in order to find possible solutions. In the presented system, the detector array is directly connected to the smart router, which selects five fired detectors at each excitation cycle and connects them to five acquisition chains containing the fast TACs. The communication with the PC is handled through the implemented USB 3.0 connection and the Ethernet 10G link, whereas the presented custom interface manages the communication with the microscope. This system is meant to fully exploit the bandwidth of the data link toward the PC and maximize the utilization of the acquisition chains. As already mentioned, the designed instrument is only a technological step toward the development of large multichannel TCSPC systems; indeed, when a faster data link toward the PC is enabled, also larger systems will be developed. This thesis is organized as follows: in chapter 1, the principles of Time-Correlated Single Photon Counting are presented, together with a state of the art for the available instruments; in chapter 2, the building blocks of TCSPC devices are described and a detailed description of the contribution that I gave to the development of the router and the FTAC is given; in chapter 3, the main issues that affect the system-level design of TCSPC instruments are outlined and the solution that I developed are presented; the 32-channel system that I partially re-designed to implement the solutions detailed in the previous chapter is presented in chapter 4, together with the obtained results; in chapter 5 the design of the new 32 32 system is reported; finally, in chapter 6, the future developments are sketched and the conclusions of this work are drawn.