Recent advancements in biomedical research and imaging applications have ignited an intense interest in single-photon detection. Along with single-photon resolution, nanosecond or sub-nanosecond timing resolution and high sensitivity of the device must be achieved at the same time. Single- Photon Avalanche Diodes (SPADs) have proved their prospectives in terms of shot-noise limited operation, excellent timing resolution and wide spec- tral range. Nonetheless, the performance of recently presented SPAD based arrays has an issue of low detection efficiency by reason of the area on the substrate occupied by additional processing electronics. This dissertation presents the design and experimental characteriza- tion of a few compact analog readout circuits for SPAD based arrays. Tar- geting the applications where the spatial resolution is the key requirement, the work is focused on the circuit compactness, that is, pixel fill factor re- finement. Consisting of only a few transistors, the proposed structures are remarkable for a small area occupation. This significant advancement has been achieved with the analog implementation of the additional circuitry instead of standard digital approach. Along with the compactness, the dis- tinguishing features of the circuits are low power consumption, low output non-linearity and pixel-to-pixel non-uniformity. In addition, experimental results on a time-gated operation have proved feasibility of a sub-nanosecond time window.

The development of Fully Depleted Monolithic Active Pixel Sensors (FD-MAPS) represents nowadays a hot-topic in the radiation detector community. The advantages in terms of production costs and easiness of manufacturing in comparison to the state-of-the-art hybrid detectors boost the research effort in the direction of developing new CMOS compatible detector technologies. To this end, the INFN ARCADIA project targeted the design of a sensor platform for the production of FD-MAPS to be employed in different scientific, medical and space applications. The sensor technology has been developed in collaboration with LFoundry on the basis of a standard 110nm CMOS production process with some modifications needed to meet the project requirements. High resistivity n-type silicon substrates have been chosen for the sensor active volume and a n-type epitaxial layer has been included at the sensor frontside to delay the onset of the punch-through current flowing between the frontside and backside p-type implants. The sensor n-type collection electrodes are surrounded by pwells, which can host the embedded analog and digital frontend electronics, and deep pwells have been included below the pwells to shield them from the sensor substrate. Three engineering runs have been submitted and the produced wafers have been delivered in 2021, 2022 and 2023, respectively. An additional p-type implant has been added in the third production run to create an embedded gain layer below the n-type collection electrodes, to enhance the signal through avalanche multiplication. A selection of the main results obtained from the TCAD simulations and of the most relevant measurements performed on the designed MAPS passive test structures will be presented and discussed in chapter 4. In an analogous way, the experimental results obtained from the characterization of an active sensor designed for brachytherapy, called COBRA, are reported in chapter 5. The calibration of the capacitance included in the internal charge injection circuit of two TJ-Monopix2 MAPS having different substrate types is reported in chapter 6. These sensors represent examples of fully functional and full scale monolithic prototypes realized in a 180nm Tower-Jazz CMOS process, that have been characterized using X-rays fluorescence techniques at the SiLab of the University of Bonn. Finally, in the Conclusions section the main results of the research activity are summarized and the possible future spin-offs of the project are presented.

Energy efficiency is a key challenge to build a sustainable society. It can be declined in variety of ways: for instance, from the reduction of the environmental impact of appliances manufacturing, to the implementation of low-energy communication networks, or the management of the existing infrastructures in a smarter way. The actual direction is the integration of different energy systems with a common management scheme with the aim of harmonizing and integrating different energy systems. In this context, smart cities already envision the use of information communication technologies (ICT) to smartify objects and services, connecting people and machines. An important enabling technology for smart cities is certainly the Internet of Things (IoT). Both smart cities and IoT have been extensively investigated over the last few years, under the influence of European funded projects as well. Smart cities apply communication and networking technologies, very often using the paradigm of IoT, to address relevant issues like traffic congestion, population growth, crowding, and others, besides implementing innovative services, modernizing existing infrastructures, e.g. smart mobility. IoT greatly helps in monitoring and better managing energy consumption as well, realizing smart homes, smart buildings and smart grids. For what concern the power grid, in fact, the direction is to harness IoT technologies to improve flexibility, easiness of use and, ultimately, energy efficiency while preserving stability and safety. Today the electrical grid is facing deep changes, mostly caused by the intensive deployment of Distributed Energy Resources (DER) based on renewable sources such as photovoltaic plants or wind farms. Managing such heterogeneous active distribution networks (ADNs), represent one of the most important challenges to be faced in the future of energy systems. The integration of active elements into the grid is challenging because of both the great potential they bring in energy production and the hazard they may represent if not properly managed (e.g. violation of operational constraints). ADN implementation relies on the deployment of high-performance real-time monitoring and control systems. It is well accepted that the phasor measurement units (PMU) are one of the most promising instruments to overcome many problems in ADN management, as they support a number of applications, such as grid state estimation, topology detection, volt-var optimization and reverse power flow management. However, classic PMUs are conceived to measure synchrophasor in transmission systems, while the distribution ones have very different characteristics and, in general, different needs. Therefore, tailoring the characteristics of the new-generation PMUs to the needs of the ADNs is currently very important. This new kind of PMU must address few important design challenges: 1. improved angle measurement capabilities, to cope with the smaller angle differences that distribution grids exhibit; 2. low cost, to promote an extensive deployment in the grid. These two requirements are clearly in opposition. In this dissertation, a low-cost PMU design approach, partially influenced by IoT ideas, is presented.

Tiny machine learning (tinyML) is becoming popular in Internet of Things (IoT) systems to add intelligence to end nodes with limited resources. Nowadays, there are tens of billions of connected devices that exchange data wirelessly, making dense IoT networks. However, the data involved in interconnected devices is increasing their memory footprint, making the transmission of raw data over low-power wide-area networks a challenging and expensive step. TinyML implements in-situ data processing, ensuring the efficiency and reliability of IoT systems without overloading communication channels. Developing tinyML systems is complex because it involves the implementation of the traditional ML algorithm and then its optimization and compression to ensure successful deployment in resource-constrained devices. However, ML algorithms range from simple systems (e.g., 1-layer feed-forward neural networks), to the most complex ones, such as deep neural networks (DNNs) with tens of hidden layers and millions of parameters. The optimization of DNNs is an active research area as it presents many challenges to deploy them in IoT end-nodes efficiently. This dissertation presents a general framework aiming at developing tinyML systems. It investigates different ML algorithms and embedded platforms to validate the correct operation of the proposed framework. Furthermore, it selects different use cases to motivate and demonstrate the effectiveness of the proposed solution for developing tinyML algorithms in IoT systems. The use cases consist of real-world applications providing actual techniques and methods to implement tinyML algorithms in constrained devices successfully. Furthermore, this thesis provides clear evidence of the benefits of tinyML considering energy efficiency, reliability, and maintenance. Finally, it improves the capability of standard tinyML systems with on-device learning techniques. In this way, it is possible to obtain tinyML systems which follow the trend of the environment, learning new patterns and reducing maintenance operations.

The objective of this dissertation is to develop and test a predictive model for the passive kinematics of human joints based on the energy minimization principle. To pursue this goal, the tibio-talar joint is chosen as a reference joint, for the reduced number of bones involved and its simplicity, if compared with other sinovial joints such as the knee or the wrist. Starting from the knowledge of the articular surface shapes, the spatial trajectory of passive motion is obtained as the envelop of joint configurations that maximize the surfaces congruence. An increase in joint congruence corresponds to an improved capability of distributing an applied load, allowing the joint to attain a better strength with less material. Thus, joint congruence maximization is a simple geometric way to capture the idea of joint energy minimization. The results obtained are validated against in vitro measured trajectories. Preliminary comparison provide strong support for the predictions of the theoretical model.

In the era in which the robots have started to live and work everywhere and in close contact with humans, they should accurately know their own location at any time to be able to move and perform safely. In particular, large and crowded indoor environments are challenging scenarios for robots' accurate and robust localization. The theory and the results presented in this dissertation intend to address the crucial issue of wheeled robots indoor localization by proposing some novel solutions in three complementary ways, i.e. improving robots self-localization through data fusion, adopting collaborative localization (e.g. using the position information from other robots) and finally optimizing the placement of landmarks in the environment once the detection range of the chosen sensors is known. As far as the first subject is concerned, a robot should be able to localize itself in a given reference frame. This problem is studied in detail to achieve a proper and affordable technique for self-localization, regardless of specific environmental features. The proposed solution relies on the integration of relative and absolute position measurements. The former are based on odometry and on an inertial measurement unit. The absolute position and heading data instead are measured sporadically anytime some landmark spread in the environment is detected. Due to the event-based nature of such measurement data, the robot can work autonomously most of time, even if accuracy degrades. Of course, in order to keep positioning uncertainty bounded, it is important that absolute and relative position data are fused properly. For this reason, four different techniques are analyzed and compared in the dissertation. Once the local kinematic state of each robot is estimated, a group of them moving in the same environment and able to detect and communicate with one another can also collaborate to share their position information to refine self-localization results. In the dissertation, it will be shown that this approach can provide some benefits, although performances strongly depend on the metrological features of the adopted sensors as well as on the communication range. Finally, as far as the problem optimal landmark placement is concerned, this is addressed by suggesting a novel and easy-to-use geometrical criterion to maximize the distance between the landmarks deployed over a triangular lattice grid, while ensuring that the absolute position measurement sensors can always detect at least one landmark.

The identification of the networks connecting different brain areas, as well as the understanding of their role in executing complex behavioral tasks, are crucial issues in cognitive neurosciences. In this context, several time series analysis approaches are available for the investigation of brain connectivity from non-invasive electroencephalographic (EEG) and magnetoencephalographic (MEG) recordings. Among them, multivariate autoregressive (MVAR) models, studied in the frequency domain, allow quantitative assessment of connectivity separately for each specific brain rhythm. In spite of its widespread utilization and great potential, MVAR-based brain connectivity analysis is complicated by a number of theoretical and practical aspects. An important issue is that the MVAR model, commonly applied to neurophysiological time series, accounts only for lagged effects among the series, forsaking instantaneous (i.e., not lagged) effects. Despite this, instantaneous correlations among EEG/MEG signals are largely expected, mainly as a consequence of volume conduction, and the impact of their exclusion on frequency-domain connectivity measures has not been investigated yet. The aim of the present thesis was to introduce and validate a new methodological framework for the frequency-domain evaluation of brain connectivity during visuo-motor integration processes. To this end, we provided first a comprehensive description of the most common MVAR-based connectivity measures, enhancing their theoretical interpretation. Then, we introduced an extended MVAR (eMVAR) model representation explicitly accounting for instantaneous effects. Accordingly, new frequency-domain connectivity measures were defined, and procedures for improving model identification and significance assessment were given. The proposed approach was validated on theoretical illustrative examples, and then applied to EEG and MEG multichannel data recorded from subjects performing a visuo-motor task combining precise grip motor commands with sensory visual feedback. The theoretical validation showed that, in the presence of significant instantaneous correlations, the traditional MVAR formulation may yield misleading connectivity patterns, while the correct patterns can be detected from the new measures based on eMVAR model identification. The practical application showed that instantaneous correlations are non negligible in the considered neurophysiological recordings, strongly suggesting the necessity of using the proposed eMVAR model in place of the traditional one. Results showed that execution of the visuo-motor task evokes the activation of a specific network subserving sensorimotor integration, which involves occipito-parietal and precentral cortices. The new connectivity measures revealed connections which were peculiar of different brain rhythms. Specifically, in the alpha frequency band (8-13 Hz) we documented an enhanced driving role of the visual cortex on the left motor cortex, suggesting a relation between this rhythm and the lateralization of the visuo-motor task. In the beta band (13-30 Hz), task-induced connectivity changes were bilateral, suggesting an involvement of both hemispheres. In both alpha and beta bands, the new connectivity measures suggested an important role for the parietal cortex in mediating the information flow from visual to motor areas, confirming previous evidences from invasive studies based on intra-cranical recordings, TMS or PET examinations. This thesis was produced in collaboration with the Department of Physics of the University of Trento.

Gamma-ray and Fluorescence Lifetime Spectroscopies are driving the development of non-imaging silicon photon sensors and, in this context, Silicon Photo-Multipliers (SiPM)s are leading the starring role. They are 2D array of optical diodes called Single Photon Avalanche Diodes (SPAD)s, and are normally fabricated with a dedicated silicon process. SPADs amplify the charge produced by the single absorbed photon in a way that recalls the avalanche amplification exploited in Photo-Multiplier Tubes (PMT)s. Recently 2D arrays of SPADs have been realized also in standard CMOS technology, paving the way to the realization of completely custom sensors that can host ancillary electronic and digital logic on-chip. The designs of scientific apparatus have been influenced for years by the bulky PMT-based detectors. An overwhelming interest in both SiPMs and CMOS SPADs lies in the possibility of displacing these small sensors realizing new detectors geometries. This thesis examines the potential deployment of SiPM-based detector in an apparatus built for the study of the Time-Of-Flight (TOF) of Positronium (Ps) and the displacement of 2D array of CMOS SPADs in a lab-on-chip apparatus for Fluorescence Lifetime Spectroscopy. The two design procedures are performed using Monte-Carlo simulations. Characterizations of the two sensor have been carried out, allowing for a performance evaluation and a validation of the two design procedures.

The goal of this thesis is the discovery of a bioinformatics solution for network-based predictive analysis of NGS data, in which network structures can substitute gene lists as a more rich and complex signature of disease. I have focused on methods for network stability, network inference and network comparison, as additional components of the pipeline and as methods to detects outliers in high-throughput datasets. Besides a first work on GEO datasets, the main application of my pipeline has been on original data from the FDA SEQC (Sequencing Quality Control)project. Here I will report some initial findings to which I have contributed with methods and analysis: as the corresponding papers are being submitted. My goal is to provide a comprehensive tool for network reconstruction and network comparison as an R package and user-friendly web service interface available on-line at https://renette.fbk.eu The goal of this thesis is the discovery of a bioinformatics solution for network-based predictive analysis of NGS data, in which network structures can substitute gene lists as a more rich and complex signature of disease. I have focused on methods for network stability, network inference and network comparison, as additional components of the pipeline and as methods to detects outliers in high-throughput datasets. Besides a first work on GEO datasets, the main application of my pipeline has been on original data from the FDA SEQC (Sequencing Quality Control)project. Here I will report some initial findings to which I have contributed with methods and analysis: as the corresponding papers are being submitted. My goal is to provide a comprehensive tool for network reconstruction and network comparison as an R package and user-friendly web service interface available on-line at https://renette.fbk.eu.

3D Si radiation sensors came along with extreme radiation hard properties, primarily owing to the geometrical advantages over planar sensors where electrodes are formed penetrating through the active substrate volume. Among them: reduction of the inter-electrode distance, lower depletion voltage requirement, inter-columnar high electric field distribution, lower trapping probability, faster charge collection capability, lower power dissipation, and lower inter-pitch charge sharing. Since several years, FBK has developed 3D sensors with a double-sided technology, that have also been installed in the ATLAS Insertable B-Layer at LHC. However, the future High-Luminosity LHC (HL-LHC) upgrades, aimed to be operational by 2024, impose a complete swap of current 3D detectors with more radiation hard sensor design, able to withstand very large particle fluences up to 2×1016 cm-2 1-MeV equivalent neutrons. The extreme luminosity conditions and related issues in occupancy and radiation hardness lead to very dense pixel granularity (50×50 or 25×100 μm2), thinner active region (~100 μm), narrower columnar electrodes (~5μm diameter) with reduced inter-electrode spacing (~30 μm), and very slim edges (~100 μm) into the 3D pixel sensor design. This thesis includes the development of this new generation of small-pitch and thin 3D radiation sensors aimed at the foreseen Inner Tracker (ITk) upgrades at HL-LHC.