PitchD – The PhD’s Pitch

The purpose of my presentation is to exhibit my present and future work as PhD student in the field of embedded radar system prototyping.
The focus is on the design and prototyping of a collision avoidance system for Automated Guided Vehicles (AGV) based on a Time Of Flight sensor and of a Real Time Localization System (RTLS) based on Ultra Wide Band (UWB) signals. It is shown how these two systems can be used in an integrated Simultaneous Locating and Mapping (SLAM) system.
The presentation then deals with the activity I carried on for automotive radar applications and other radar systems. An overview of the different radar systems developed by the radar group is also presented.

In this seminar I will present a combined experimental and theoretical approach toward reaching new insights into the mechanisms of superconductivity through the analysis of the effects of disorder in Iron Based Superconductors (IBS). We investigate the critical temperature, penetration depth, quasiparticle conductivity and surface impedance of high-quality IBS single crystals by a planar waveguide resonator technique, in a cavity perturbation approach. The experimental method and data analysis are described, showing that this technique is reliable for the study of small crystals and, since the measurement technique is non-destructive and does not alter the crystals, the very same samples can be measured before and after irradiation, making the analysis of the effects of additional defects more reliable. The absolute values of the penetration depth are accessible by the experiment, showing a fairly good agreement with literature data.
The experimental data is compared to calculations based on the Eliashberg equations within the s± wave model, giving a remarkable agreement. This overall consistency validates the model itself, thus allowing us to estimate parameters that are missing in literature. The calculations are also able to explain in a consistent way the effects of disorder both on the critical temperature and on penetration depth, by suitably accounting for the impurity scattering due to the defects created by irradiation. This combined approach has allowed the identification of the disorder induced s± to s++ order parameter symmetry transition in Rh-doped crystals.

During my presentation I will focus on the interesting theme of human movement analysis by means of wearable magneto-inertial sensors (IMU). The aim of my research activity is to design and validate advanced algorithms of signal processing (also known as sensor fusion algorithms, e.g. Kalman filter and complementary filter) to estimate the joint kinematics starting from the inertial and magnetic data collected by IMUs. At the end of my presentation, I will introduce the main projects in which our research group is involved. In particular, we are focusing on a tele-rehabilitation project and European research in the Horizon 2020 framework to assess the digital mobility of population with and without motor impairments in real-world settings. Master thesis proposals and open research themes will be also presented.

My experience in GM
Eng. Alessandra Neri received her doctoral degree in Metrology from PoliTo with a thesis entitled ‘Development of innovative low cost sensors based on optical technologies’. She has been working in General Motors for almost 6 years, first as Hardware Electrical Architecture Development Engineer and now as Technology System Engineer, dealing with connected vehicles and big data. She will present her professional experience, providing an interesting insight on opportunities and challenges facing PhD students after the degree.

With the first detection of Gravitational Waves on September 14th, 2015 a new era of Astronomy has begun. The LIGO detectors observed the merger of two Black Holes several 10-times heavier then the sun clashing into each other at more then half the speed of light over a billion light-years and over a billion years ago. Since then the efforts by the LIGO scientific collaboration have been pushed forward and large international effort has begun to build a global network of more and better Gravitational Wave detectors and to understand what we can learn from studying Gravitational Waves.
But why are Gravitational Waves so interesting to physicists? Or what are they to begin with? What kind of events produce gravitational waves that can be detected on earth? How can they be detected? What kind of detectors are currently operating? What kind of detectors are planned?
In my talk I will discuss these questions and give a brief overview of the field of Gravitational Wave Astronomy. I will discuss the basic ideas of what are Gravitational Waves, how can they be detected, how can they be generated by Black Holes, Neutron Stars and other more exotic phenomena in the observable universe, and what can be understood from studying them.

My research activity is focused on the study of electromagnetic metamaterials and especially on their application to electromagnetic invisibility. Metamaterials are artificial periodic structures able to control and bend the propagation path of the electromagnetic waves in unprecedented ways. Among the different applications of metamaterials, cloaking is one of the most fascinating. In particular, mantle cloaking, differently from other techniques, has the advantage of suppressing the scattered field from the target object by using a 2D thin patterned metasurface, therefore reducing the cloak weight and thickness. In this seminar I will present an overview of electromagnetic metamaterials and cloaking problems, showing the design and analysis of a particular metasurface based on a width modulated microstrip line.

This presentation aims at providing an illustration of my research about an optimized deep Convolutional Neural Network (CNN) for the automatic classification of Scanning Electron Microscope (SEM) images of homogeneous nanofibers (HNF) and nonhomogeneous nanofibers (NHNF) produced by electrospinnig process. Specifically, SEM images are used as input of a Deep Learning (DL) framework consisting of a Sobel filter based pre-processing stage followed by a CNN classifier (SoCNNet). Experimental results of this study demonstrate the potential effectiveness of proposed SoCNNet in the industrial chain of nanofibers production.

Wearable electrochemical sensors play a significant role in physiology and health status monitoring since they are capable of continuously collecting biological data from human body in the field. In sports applications, it is crucial to assess muscle fatigue, mineral loss, dehydration in order to foresee muscle cramping or other physiological dysfunction. The biological compounds enabling such physiological status monitoring belong to different families of biomarkers, hence, necessitating a general purpose and multi-sensing platform. Moreover, within a multi-sensing context, efficient data processing tools are needed to predict reliably the concentration of each biomarker and to cope with interference. In this talk, electrochemical sensing platforms dedicated to wearable physiology applications are presented. The electronic front-end interfaces developed are described and characterized for endogenous metabolites, electrolytes and exogenous compounds monitoring. The programmability and versatility of the sensing front-end enables its use for broader biomedical applications. In addition, some insights on processing tools for multi-ion sensing are also presented.

The concepts of a complete plant “Internet of Things” with direct data collection from the plant, is a novel approach. Extensive plant research is available. However, study of plants in terms of electronics and electrical conduction mechanisms are not well defined. Here would like to establish an improved understanding of the electronic conduction within the plant and deploy it for sensing and communication applications. Using this new approach, we seek to establish whether a measurable electrical change, will allow detection of biological and physiological changes within the plant.

Nowadays, optical fibers are quite well known for their great telecommunication capabilities, but recently they have gained a lot of attention for their sensing capabilities. Small size, high sensitivity, electromagnetic compatibility and the low cost are some of their most important key features and the main reason why so much research is going on in the field. My research focuses on plasmonic based optical fiber sensors aimed to monitor the quality of potable water. The plasmonic phenomenon allows an analysis based on stable high-resolution refractive index sensing. According to UN, more than 40% of the world’s population is affected by water scarcity problems, which in a lot of cases does not mean total lack of water, but rather means having not sanitized water supply. The necessity for monitoring the quality of such water before using (either for washing or drinking), is an emergency, and in most of the cases it needs to be done actively on the spot, in order to prevent the consumption. Plasmonic optical fiber sensors seem to be a very promising solution for such an emergent problem.

While digital Integrated Circuits (IC) have been taking the advantage of CMOS technology scaling in terms of speed, power consumption, and cost, the techniques running behind the analog signal processing are still lagging behind. In other words, it is clear that overtime there have been significant improvements (and it is expected to) in digital circuits performance when compared with analog building blocks. To mitigate this, there has been an increasing trend in finding alternative IC design strategies to implement analog functions exploiting digital-in-concept design methodologies. In addition to the advantage of using the well-established fully-automated EDA digital tools, the idea of re-thinking analog functions in digital terms would make the Analog IC blocks also suitable to avail energy efficiency and the feature-size shrinking of new technologies. In this context, this presentation will cover the current status of the state-of-art digital-based analog building blocks results including circuits such as OTA, comparators and converters, as well as their impact over the emerging applications and technologies. Put differently, are you a digital guy ? Would you like to know how to design OpAmp (analog stuff) only using logical gates ? This talk is for you.

The current technological progress in the field of quantum computing poses serious problems for the security of widely adopted asymmetric cryptosystems as RSA and ECC. The asymmetric encryption requires to rely on different algorithms, such as Code-based cryptography. The topic of my research is the McEliece cryptosystem based on Error Correcting Codes, in particular a version that adopts QC-LDPC codes. The codes are structured and sparse, thus they have key of small size even if the dimensions of the inner variables are huge. I am mainly working on the hardware implementation of the decryption system in order to make the computation of the quantities faster and include the possibility to have a scalable architecture that is able to match the needs of a particular application or the limitations of a device.

Low cost, ubiquitous, tag-less, and privacy-aware indoor monitoring is essential to many applications, such as home automation and assisted living of elderly persons. Capacitive sensors can be inconspicuous, cost and power-effective solution, but afflicted by noise especially at ranges much longer than their plate diameter. Here, we explore how well different Artificial Intelligence (AI) techniques can extract location and movement information from noisy experimental data (with both high-pitch and slow drift noise) obtained from capacitive sensors operating in load mode at long ranges. For a 3 m x 3 m space equipped with four 16 cm x 16 cm capacitive sensors, we investigate first how well Machine Learning (ML) can localize a person in 16 predefined positions, then how well continuous human walking trajectories can be reconstructed by various types of Neural Network architectures. We discuss how the experimental data was obtained, preprocessed, and the outcome of the various ML and AI techniques. The key result is the ability to locate and track a person in the above setting with an average accuracy of 93% for location classification and a mean error of 28 cm for trajectory reconstruction using sensors with an estimated SNR of about 3-4 dB.

Cryogenic CMOS electronic could provide an important impact in the current conventional electronic applications, since this can offer important device performance improvements, in term of RMS noise, power consumption and speed. In the nuclear physics area, the underground astroparticle physics experiments require the usage of the cold electronics due its structure (LAr and LXe Time Projection Chamber). In this presentation, the behavior of 110 nm CMOS transistor parameters (gm, Vth, µo, rSD, q) between 300 K and 77 K and the integrated front-end electronic design for readout a sensor area of 24 cm2 are presented. The experimental result of the cryogenic electronic front-end for DarkSide20K, using a SiPM as sensor and Liquid Nitrogen, are described. This research is on behalf of DarkSide collaboration.

Power electronics-based energy conversion is making the exploitation of Renewable Energy Sources, such as sun and wind, more and more attractive and the amount of renewable energy generation is expected to rise exponentially all over the world in the next 20 years. However, this comes at a price: the power system was designed to operate with a prevalence of synchronous generators. Therefore, changes in the design of power electronics and control will be mandatory to reach a 100% renewable-based power system. In particular, electronic converters will have to guarantee the so called “ancillary services” in order to maintain the stability and the correct operation of the power grid. A promising solution is the concept of Virtual Synchronous Generator, that makes the converter behave as a synchronous machine. This talk will introduce the issues related to power generation from renewables and describe how Virtual Synchronous Generators can solve them.

The Moore’s law states the number of transistors inside a single chip doubles every eighteen months thanks to dimensions scaling. However, this trend is reaching an end due to technological limits and power density. For these reasons new technologies are being studied in order to find possible alternatives. In particular, magnetic QCA (quantum-dot cellular automata) technologies are the most promising. No current flows inside the technological elements, and information propagation is achieved through magnetic fields: they are part of the Field Coupled Nanocomputing (FCN) technologies. Technologists focus on the design of the single device and its characteristics. A circuit level exploration needs CAD (Computer Aided Design) tools in order to speed up the process. The ToPoliNano (Torino Politecnico Nanotechnologies) framework enables the design and simulation of digital circuits based on FCN technologies. The framework and the supported technologies will be described and a Live Demo of the tools will be performed.

The demand for a source of renewable energy has been the driving force to achieve significant advancements in the field of photovoltaic energy generation during the past decades. III-V quantum dot (QD) solar cells hold promising potential for the realization of multiple novel photovoltaic concepts. However, one of their essential limitations arises from the weak absorption coefficient of the QD layers as well as the difficulty to grow active regions with high in-plane QD density and to stack many QD layers. Light trapping allows overcoming limitations faced by QD solar cells and Intermediate Band (IB) solar cells that generally suffer from weak QD interband and intraband photo-generation. It can be used to increase the effective optical path length of the QD layers, thereby resulting in higher photogeneration and overall efficiency. Moreover, light trapping concepts allow the use of a thin active region while maintaining constant short-circuit current, thereby minimizing volume recombination losses and enabling a higher open circuit voltage. Thin-film solar cells are associated with a high power-to-weight ratio, an essential requirement in space applications.

Quality of Experience (QoE) assessment for video games is known for being a long and expensive process, typically requiring the active involvement of several human players under different network conditions while ultimately bringing limited transferability across different games. Noteworthy, QoE is correlated with the achieved in-game score, as player frustration will arise whenever their performance is far from what is expected.
Taking inspiration from this observation, we question ourselves if it is possible to simplify the QoE assessment task by substituting human players with artificial agents trained to play the same game under varying network conditions.
To answer this question we let our artificial agents play a set of retro games with different types of interaction, showing that the observed score degradation curves can be exploited in networking devices (e.g., by prioritizing scheduling decisions), reinforcing fairness across games, and thus enhancing the overall quality of gaming experience.

Compressed Sensing (CS) is a widely researched technique employed to lower energy consumption of signal acquisition. It is especially suited for naturally sparse signals such as human body biosignals. These type of signals can be linearly mapped on a lower dimensional space through a simple vector-matrix multiplication. From this, the recovery of these signals can be partitioned in two steps: support estimation phase (i.e., the retrieval of the positions of the coefficients in the sparse vector) and coefficient estimation phase. Support estimation can be performed with an oracle based approach on a deep neural network. The output of the oracle can be then used to estimate the coefficient by solving the Least Mean Square (LMS) problem. This approach allows the definition of an encoder-decoder pair with state-of-the-art recovery capabilities when applied to biological signals such as ECG and EEG.

The power grid is assisting a huge transformation thanks to the adoption of renewable resources. which have enabled bidirectional power flows, no longer only from high and medium voltages to low voltage but also vice versa. This phenomenon, together with the fact that wind and solar energies are not easily foreseeable and programmable, leads to the need to rethink a more flexible electricity network able to store a greater electricity capacity. On the other hand, another revolution is increasingly making its way. The transport sector is destined to converge rapidly towards electricity. The combination of these two revolutions can represent a risk for the stability of the electricity grid but also a crucial opportunity not to be missed.

Recent years have observed a significant rise in popularity of music streaming services. Instead of using a physical format, music is now distributed in a completely digital way. Nevertheless, on the playback phase little has changed as it remained inherently analog. It follows that to correctly reproduce a digital audio signal, it must be converted back to its analog representation. This operation has to be performed by a Digital to Analog Converter (DAC). During the conversion phase digital signals are subject to inevitable analog errors which can deteriorate the quality of the resulting output signal. In High Fidelity audio systems, which aim to be as faithful as possible to the original source, these kind of errors represent a major obstacle to overcome. Although older analog-centric DAC chips are able to partially mitigate this issue, newer digital-centric designs struggle to reach the same level of performance, as they need to cope with error related to modern transistor production processes.This talk will focus on novel smart digital algorithms capable of concealing and mitigating issues related to digital-centric DACs, while keeping cost and power consumption low.

In the last decade, the accelerating progress in the development of more and more complex neural networks has called for performant accelerators and novel computing paradigms. Memristors, theorized to be the fourth fundamental circuit elements, have attracted the interest of the neuromorphic computing research community as promising technologies to overcome the limitations of current digital systems. Originally developed as information storage devices, memristive technologies have unvealed their true potential as basic computing elements by exploiting the Ohm’s Law when operated at very low electric field. Nonetheless, their intriguing complex physical working principles, responsible for their state-dependent highly nonlinear characteristics, open the door to very novel computing paradigms such as spiking and oscillatory neural networks. Those complex physical phenomena, which qualify memristors as perfect candidates for mimicking neuron level activities, need either very powerful computers to be simulated or application specific compact modelling techniques.

Most conventional artificial neural networks architectures are implemented on digital computers yielding order of magnitude higher energy consumption than biological brains. Neuromorphic computing aims to emulate the neural structure and operation of the brain in order to create low power computing platforms that are able to tackle complex tasks in artificial intelligence. Nonvolatile memory devices, such as memristors, stand out as promising candidates to represent synaptic weights of artificial neural networks by enabling efficient parallel computations such as matrix-vector multiplications. Despite the great success in deep learning applications, there is an increasing interest in the direct use of the intrinsic physics dynamics of these devices in approximating the complex behavior of the brain. Various short/long-term synaptic plasticities have been recently reported and the analysis of the analog neural networks dynamics has drawn particular interest among researchers. This talk will briefly introduce different applications of memristive crossbars both as accelerators for deep learning and as building blocks for analog neural networks.

Auscultation is a traditional routine screening tool in the clinical practice for cardiovascular diseases, the first world cause of death according to the World Health Organization. Phonocardiography is its digital counterpart and is growing in importance because it allows, for example, for the assessment of the electromechanical coupling of the heart. A delay in the closure of heart valves, which originates heart sounds, with respect to the ventricular depolarization in the simultaneous electrocardiogram was found to correlate with heart failure. This talk will explore how the design of a noninvasive device for the recording of phonocardiographic (PCG) and electrocardiographic (ECG) signals can help in the prevention of acute episodes of heart failure. The issues regarding the positioning of the digital stethoscope, limiting its applicability in a real-life homecare context, will be investigated along with the potentiality of multi-source phonocardiography as a possible solution.

Technology scaling has been leading the market of integrated circuits (IC) production of the last decades, to get aster and more efficient digital circuits. Meanwhile, transistor variability has increased the challenges in the design of analog blocks, especially where low area and low power embedded systems are required. This is the case for the Internet of Things (IoT) sensor nodes, where analog interfaces are needed as acting, sensing and communicating blocks in systems which need little to no maintenance, possibly harvesting energy from the environment. The strategy to re-think analog interfaces to be intrinsically mostly-digital, matching-insensitive and able to work at very low supply voltages is proving to be a promising solution. The new-in-concept Relaxation Digital-to-Analog converter (ReDAC) which is going to be presented in the talk, is targeting all these challenges as an ultra low power, bitstream based, easily scalable converter.

Nowadays, computer architectures are widely studied in the literature. The von Neumann scheme, which is essentially composed of CPU and memory, is the most adopted structure. Both elements have made significant improvements in terms of performance: CPUs are becoming faster and more efficient over the years, achieving high clock frequencies and multitasking capabilities. However, memory cannot follow the trend of the CPU in terms of computational speed, so the communication between these two elements represents a bottleneck: a lot of power, energy, and execution time are wasted only to wait for the data from memory. This phenomenon is called the von Neumann bottleneck. What could we do to reduce this drawback? Logic-in-Memory (LiM) could be a solution: it merges storage and computation, so instead of moving data back and forth in the memory hierarchy, calculations are performed directly inside the memory array, reducing the data traffic and improving the performance.

Brain-computer interfaces (BCI) are communication channels alternative to natural peripheral nerves and muscles. Indeed, these were born for people with motor impairments, but they opened up new possibilities also in fields like gaming, entertainment, education, and industry. As in many other domains, what yesterday seemed just “magic”, today is becoming a technological reality. However, BCI applications outside the laboratory are still limited, and many research groups are trying to reach higher performances and user-friendliness. The present talk focuses on the development of non-invasive and wearable BCIs to be easily used in daily life. Two well-known paradigms are considered, namely “steady-state visually evoked potentials” and “motor imagery”, and the whole BCI loop will be briefly described, i.e. signal acquisition, signal processing, and possible applications. The overall aim of the R&D described in this talk is to contribute in spreading the BCI technology over the world.

The possibility of controlling transport properties of semiconductors by means of an external electric field goes back to the invention of the field-effect transistor (FET) in 1960 and in particular to its modern implementation, the metal-oxide-semiconductor field-effect transistor (MOSFET). However, field-effect doping can also be employed for modifying the conductivity of thin films of noble metals or tuning strongly correlated phases of matter (such as superconductivity, ferromagnetism, insulator-to-metal transitions, charge-density waves, etc.). In order to describe these situations and try to compare the experimental outcomes to theoretical predictions, it is important to understand how FET geometry affects the electronic properties of materials, e.g. using density functional theory (DFT). In this presentation I will show you how field-effect doping is treated in the framework of plane-waves pseudopotential DFT codes and what we can learn from the electronic band-structure using two case studies: niobium nitride (NbN) and hydrogenated diamond thin films.

Coronary artery disease – the narrowing or blockage of coronary arteries – is one of the leading causes of mortality worldwide. The most common surgical treatment in this case is bypass grafting, where some arteries are used to bypass the blockage. Unfortunately, up to 60% of grafts fail some years after surgery, and the reasons behind such failures are still largely unknown. A huge help in this sense comes from engineers and mathematicians, which can use advanced computational models to simulate how blood flows in the affected arteries, sometimes borrowing modeling techniques from circuit applications. Such models provide information on a patient’s unique physiology, and may help clinicians in performing a better diagnosis and treatment planning. This talk will provide an overview of the use of computational models for cardiovascular applications and the ongoing work to overcome one of its main challenges, the correct choice of boundary conditions.

The photovoltaic (PV) technology, blending cheapness and sustainability, represents a viable alternative to the fossil-fuel system responsible of climate change and global warming. Nowadays, the terrestrial PV market is dominated by c-Si single-junction solar cell achieving an efficiency close to its theoretical limit. However, in recent years, a good solution to overcome this limit is found in Perovskite-Silicon tandem solar cells thanks to the low manufacturing costs as well as the strong solar absorption of the perovskite (PVK) material. Aiming at increasing the PVK/Si solar efficiency and overcoming several constraints of the series connected double junction cell, such as the current matching and the need of tunnel junctions, it is studied an alternative architecture: the three-terminal bipolar transistor (3T-HBT) solar cell. To evaluate the performance potential and design optimization of this novel PVK/Si 3T-HBTsc, the classical Hovel model is extended to deal with the 3T-HBT structure demonstrating efficiencies up to 28.6%.

Fully depleted monolithic active pixel sensors (FD-MAPS) are the future of particle and radiation detectors. Full depletion improves charge collection timing and signal to noise ratio versus power dissipated. Furthermore, a smaller material budget and less interconnections push down production costs and time. The approach envisioned by the ARCADIA (Advanced Readout CMOS Architectures with Depleted Integrated sensor Arrays) collaboration consists in implementing the front-end electronics in a deep p-well embedded in an n-doped epitaxial layer, decreasing the sensor capacitance and effectively decoupling the substrate from the electronics. A patterned backside, with appropriate guard rings, allows the necessary high voltage to uniformly deplete the full wafer thickness. Despite the backside implants, the sensor fabrication is fully compatible with a standard CMOS production line. The talk will present the current R&D progress and results of tests on small-scale matrices and test structures.

Alternative internet congestion control strategies have been proposed since the TCP’s birth, mainly because of the variance of conditions and assumptions in different network scenarios, e.g., cellular versus data center networks. However, years of research on transport protocols have not solved the tussle between in-network and end-to-end congestion control. Recently, the community has proposed a few transport protocols driven by machine learning, nonetheless limited to end-to-end approaches. This talk will introduce a transport protocol based on reinforcement learning, whose goal is to select the proper congestion window learning from end-to-end features and network signals, when available. A solution like the one presented has been proved to converge to a fair resource allocation after the learning overhead. Finally, it can provide higher throughput and lower delays compared to benchmark solutions based on end-to-end or in-network congestion control.

In recent years, the reduction of cost of Next Generation Sequencing (NGS) technologies has enabled the widespread of omics data: genomics (DNA), transcriptomics (RNA), proteomics (proteins), and many others. However, omics data are characterized by complexity and high dimensionality, which makes the biological interpretation a tricky process. Nowadays, AI contributed to the major technological advances in many fields, including biology. My work focuses on two areas in cancer analysis: gene fusion and multi-omics data integration. Gene fusions consist of a biological event in which two genes break and their portions are juxtaposed together, potentially causing cancer. Therefore, deep learning models have been exploited to predict the probability for a gene fusion to be involved in cancer genesis and progression. Since cancer is a complex phenomenon, one omic alone provides only partial information on the biological processes. Therefore, multi-omics data integration is crucial in the cancer domain and AI techniques allow a more efficient prediction.

Air quality, especially particulate matter, has attracted growing attention from governments, industry, and academia in recent years, motivating the use of denser air quality monitoring networks based on low-cost sensing strategies. However, low-cost sensors are frequently sensitive to aging, environmental conditions, and pollutant cross-sensitivities. These issues have been only partially addressed, limiting their usage.The talk would describe the development of a low-cost particulate matter monitoring system, deployed for monitoring air quality on both stationary and mobile sensor platforms. We explore the influence of all model variables and the quality of different calibration strategies. Tests of statically immovable stations include an analysis of accuracy and sensors’ reliability made by comparing our results with more accurate and expensive standard β-radiation sensors. Tests on mobile stations have been designed to analyze the reactivity of our system to unexpected and abrupt events. These experiments embrace traffic analysis, pollution investigation using different means of transport and pollution analysis during peculiar events.

A brain stroke is a widespread disorder that affects around one in six people in their life, top-ranking worldwide as causes of death, disability, and dementia in otherwise healthy adults. It is a medical emergency caused by the interruption of the regular supply of oxygen-rich blood to the brain, leading to the loss of millions of brain cells per minute. Thus, requiring prompt treatment. Stroke care widely relies on brain imaging technologies, identifying the specific pathophysiologic conditions for tailored treatment and enhancing effectiveness. The most well-established solutions are computerized X-ray tomography (CT) and magnetic resonance imaging (MRI). Moreover, in recent years microwave imaging (MWI) has emerged as a complementary technology that allows early diagnosis and bed-side follow-up. MWI relies on the electric contrast between the healthy brain tissues and the pathologies to retrieve essential diagnostic information according to their typology and their physiopathological status.

The term “Industry 4.0” is the current automation and data exchange trend in the manufacturing environment. It includes cyber-physical systems, the Internet of things, and cloud computing, which depict factories as intelligent environments. Their composing elements (e.g., machines, storage systems, edge and system devices, etc.) are seen as smart components, able to communicate with each other and take decisions autonomously. These benefits come when reliable communication is established between the shop floor and the factory’s top floor. Depending on the application, Wireless Sensor Network (WSN) and Internet of Things (IoT) help communicate devices and people with each other, both within the factory and outside it, reaching customers, suppliers, and broader governing bodies. Nowadays, Wireless communication plays a significant role in Industry 4.0. WSNs help factories to gather data and send control commands to the factory shop floor. WSN and IoT relevant challenges guarantee reliable communication in the factory where constrained communication is often required. The talk discusses and explains methods and algorithms to increase performance indicators in Industry 4.0, to have reliable communication between the shop floor and business applications in the factory.

Microwave Imaging (MI) in medical applications is a technique to observe the internal structure of an object by using electromagnetic fields at microwave frequencies. One of the challenges in MI is the high execution time of the image reconstruction algorithms making it difficult to be used in real-time systems. Our aim is to increase the speed of the recurrent algorithms used in MI by means of hardware acceleration techniques. These algorithms include compute-intensive parts which we call “kernels”. In this presentation, I will introduce you some of the most commonly-used kernels in MI and the methodology to design specific hardware accelerators in FPGAs for these kernels by means of High Level Synthesis (HLS) tool. These kernels include a forward electromagnetic solver called FDTD and some Machine Learning classifiers. The performance is compared with CPU and GPU-based designs in terms of processing time, power consumption and resource usage.

Over the last decade, many studies have focused on Virtual Reality (VR) frameworks for remotely controlling robotic systems. Although VR systems have been used to teleoperate robots in simple scenarios, their effectiveness in terms of accuracy, speed, and usability has not been rigorously evaluated for complex tasks that require accurate trajectories. In this talk, an Enhanced Virtual Reality (EVR) framework for robotic teleoperation is evaluated to assess if it can be efficiently used in complex tasks that require accurate control of the robotic end-effector. The environment and the employed robot are captured using RGB-D cameras, while the remote user controls the motion of the robot with VR controllers. The captured data are transmitted and reconstructed in 3D so as to allow the remote user to monitor the task execution progress in real time, using a VR headset. The EVR system is compared with two other interface alternatives: i) teleoperation in pure VR (the model of the robot is rendered with respect to its real joint states), and ii) teleoperation in EVRR (the model of the robot is superimposed on the real robot).

The goal of quantum sensing is to reach an advantage over classical benchmarks using quantum resources. In the protocol of quantum reading this advantage is obtained for the task of information recovery from a classical digital memory. In the work presented we showed, both theoretically and experimentally, how this advantage can be achieved using an entangled two-mode squeezed vacuum source paired with a photon counting measurement and a maximum likelihood decision. This quantum strategy is able to outperform any classical strategy for the same number of input photons. Our experimental results prove how quantum entanglement and simple optics are able to enhance the redout of digital data, paving the way to real application of quantum reading as well as any other model based on the binary discrimination of bosonic loss.

Point clouds are becoming increasingly popular due to the availability of instruments such as LiDARs and the interest in exploiting geometric representation in many challenging applications such as autonomous driving, medical imaging or virtual reality. However, the acquisition methods are imperfect and insert a non-negligible noise. In general, a point cloud is a set of points in the 3D space to which a specific location is associated, the insertion of the noise changes their position and the collected shape is different from the original one. Therefore, in case of safety-critical applications, it is important to perform denoising of point clouds, to recover the original shape from corrupted data. This talk will be focused on the challenges of dealing with point clouds, the limitations of the existing methods in the context of point cloud denoising and the presentation of a neural network based on graph convolutions as a possible solution.

There are numerous challenges derived from the personalized health paradigm and novel diagnostic and therapeutic approaches are required. Interest in impedance-based assays is rising due to their remarkable advantages, including label-free, low cost, non-invasive, non-destructive, quantitative and real-time. In this application-oriented talk, first, a custom-made impedance measuring system based on electric cell-substrate impedance sensing (ECIS) will be introduced. Emphasis will be given to its potential in cancer treatment decision and early detection of chemoresistance. Then, jumping from cell analysis to more complex systems such as 3D biological constructs, an overview of a microfluidic device under construction, for impedance-based measurements of biopsies, will be presented.

To allow effective storage and transmission of images/videos, compression algorithms that limit perceptual quality degradation are needed. This requires effective Quality Prediction Algorithms (QPAs). QPAs aims at predicting, for a given image/video, the Mean of the Opinion Scores (MOS) that a group of human observers would give if they were asked to watch it and express their opinion on how visible the artifacts are, using a numerical scale. There is still a large room for improvement of state-of-the-art QPAs. This talk will introduce you to how machine learning models and advanced statistical methods can be used to improve the quality prediction of these algorithms while considering the individual user’s expectation which is typically disregarded when predicting the MOS. In particular, it will be shown how an Artificial Neural Network can be trained to mimic a human subject in terms of quality perception, yielding what is called an AI Observers (AIO).

Synchromodality is a recent paradigm for long-haul transportation to decrease costs, emissions, and delivery times in modern supply chains. These goals are achieved through stronger stakeholders’ cooperation based on integrating their businesses to build one flexible and reliable system. This system allows synchronizing resource utilization and goods flow to react to disruptions with real-time practices like re-routing, rescheduling, and modal shift. However, implementing synchromodality requires improving critical success factors by using innovative technology solutions. This talk aims to overview synchromodality, focusing on peculiarities, critical success factors, and enabling technologies. In particular, we discuss three different projects. First, we present SYNCHRO-NET, an integration platform offering technology tools to various stakeholders. Second, we focus on optimizing a tactical problem concerning handling contracts with terminals and managing shipment flows. Lastly, we introduce a simulation/optimization approach to deal with real-time re-planning.

EU aims to be climate-neutral by 2050 – an economy with net-zero green gas emission. The geometric progression in the sales of electric vehicles from 2010 till 2019 reinforces the hope of reaching this goal. One major foreseen challenge that may limit the adoption of electric and hybrid vehicle is the dreadful thermal runaway. A good understanding of the major cause of this phenomenon can help to ensure that the battery operates within the risk-free zone. The chemical industry is not mature enough to produce cells with uniform ageing characteristics. State of health (SOH) is a useful index for measuring battery ageing. It is essential to adopt algorithms that can accurately estimate the varying state of health (SOH) of all the cells in the battery pack. An approach for enhancing the generalization of SOH estimation using a parallel layer extreme learning machine (PL-ELM) algorithm is presented here.

Complexity of modern neural network has grown dramatically in recent years, requiring hardware-acceleration-based solutions to deal with the demanded computational complexity. The reliability evaluation of neural networks is still premature and requires platforms to measure the safety standards required by safety-critical applications such as automotive and avionic. Software-level simulation approaches have been consolidated as the most common method to assess the reliability of neural network systems through simulation/emulation, often ignoring the underlying hardware. In this talk, the limitation of an analysis based only on a software-level approach is discussed. A new methodology for evaluating the resiliency of neural networks by using programmable hardware is presented as an alternative.

Graphics Processing Units (GPUs) are crucial devices that boost the execution of complex algorithms in the scientific and artificial intelligence domains. Moreover, GPUs are also relevant components now included in several safety-critical applications (e.g., automotive, and autonomous machines), where reliability and functional safety are essential requirements. This speech focuses on the exploration and development of new in-field techniques aiming to test and mitigate faults affecting GPUs. For this purpose, a functional testing mechanism (Software-Based Self-Test or SBST) is exploited to propose or adopt several test strategies for GPUs. Finally, hybrid mechanisms are proposed and evaluated targeting the in-field mitigation of fault in GPUs. According to experimental results, both in- field testing and mitigation mechanisms can be considered fault-tolerance solutions for GPU devices.

In the last few decades, the growing complexity of emerging computing systems has called for enhanced computing paradigms. Among all the existing possibilities, artificial intelligence (AI) based solutions and, specifically, brain-inspired computing models, have gained large interest in the industry and academia for their outstanding computational capabilities. Nowadays, the usage of electronic devices running applications based on Artificial Neural Networks (ANNs) is spreading in our everyday life. In fact, ANNs are now considered attractive solutions for different areas including safety-critical applications such as self-driving cars, radars, flight control, robots, and space applications. Often, ANNs are considered intrinsically robust and fault tolerant for being brain-inspired and redundant computing models. However, when they are deployed on resource-constrained hardware devices, single physical faults might jeopardize the activity of multiple neurons, leading to unwanted outcomes. Therefore, to use them safely in human contexts, there is a compelling need for assessing their reliability and tolerance to faults.

Climate change is probably the biggest challenge humanity has ever faced. As it is widely known, factors causing it have mainly anthropic origins. Among these factors, one of the most impactful is agricultural activity. It is responsible for about 30% of greenhouse gases emission and 70% of freshwater withdrawals. Agricultural’s environmental impact is destined to increase in the following decades since the world population is expected to grow dramatically. Therefore, producing enough food for the entire population while reducing production footprint will be a crucial goal that must be achieved. Smart agriculture aims at merging together both engineering and farming knowledges to increase harvests and, at the same time, reducing their impact.

Physics-based device simulations represent an ideal environment to accurately model the behaviour of active devices in RF/microwave circuits, as they keep trace of technological and physical parameters. The frequency domain analysis of electron devices in highly nonlinear conditions has proved to be a fairly manageable task in TCAD simulators, especially using the Harmonic Balance technique for mixed-mode simulations. To be successfully used for circuit analysis, physics-based models must be able to predict the sensitivity of a nonlinear stage towards the variations of: temperature, physical/technological parameters and the embedding circuit. In this talk, the problem of temperature variations, especially relevant in the scenario of power devices (e.g. GaAs or GaN based HEMTs) and nanoscale devices (e.g. FinFETs), is addressed. In particular, an efficient approach to the temperature-dependent physics-based variability analysis of a FinFET-based power amplifier in Large Signal (LS) nonlinear conditions is presented. The method extends, with negligible numerical overhead, the Green’s Function approach and allows to calculate the LS device response to the temperature variation from a nominal”cold” condition with concurrent variations of the external load or technological parameters.

Packet measurements at scale are essential for several applications, such as cyber-security, accounting and troubleshooting. They, however, threaten users’ privacy by exposing sensitive information. Anonymization has been the answer to this challenge, however, it comes with some challenges and drawbacks. First, it reduces the value of data. Second, it requires to consider diverse protocols because information may leak from non-encrypted fields. Third, it must be performed at high speeds directly at the monitor, to prevent data leakage. We present α-Mon, a flexible tool for privacy-preserving packet monitoring. It replicates input packet streams to different consumers while anonymizing according to flexible policies covering all protocol layers. α-Mon supports α-anonymization, a novel solution to obfuscate rare values: it works on a streaming fashion, with zero delay, operating at high-speed links on a packet-by-packet basis. We quantify the impact of α-anonymization on traffic measurements, finding that it introduces minimal error when it comes to finding heavy-hitter services. We evaluate α-Mon performance using packet traces collected from an ISP network on a Commercial Off-the Shelf server.

Among the technologies proposed for the beyond-CMOS scenario, the molecular Field-Coupled Nanocomputing (FCN) promises very high device density and possible high-speed operations at room temperature. The binary information is encoded in the charge distribution of molecules and propagated through intermolecular electrostatic interactions. No current transport participates in the information propagation, significantly reducing power dissipation. In this scenario, ab initio tools are essential for evaluating the fundamental properties of molecules. Nevertheless, the calculation on a single molecule requires several CPU hours, making it unfeasible for simulating and designing digital devices with thousands of molecules. The SCERPA tool, presented in this talk, is a novel algorithm that considers molecules as digital devices, with input and output, whose transcharacteristics are characterized with ab initio calculation. SCERPA evaluates the propagation and the elaboration of the information in molecular circuits with a minimal computational cost, permitting the analysis, simulation, and design of molecular FCN architectures.

Since the first developments, for both the main media in the eXtended Reality (XR) family, i.e., Augmented Reality (AR) and Virtual Reality (VR), training represented the application attracting most of the interest. This statement is even more true nowadays that XR Training Systems (XRTSs) are moving from laboratory settings to the industry, being more and more frequently integrated into the companies’ training programs. Although the efficacy of XRTSs as a fancy alternative to traditional learning material used by trainers in their lectures has already been proved, their effectiveness especially in the form of self-learning tools not requiring human instructors interventions is still controversial. During this speech, will be discussed the findings of research carried out to validate design guidelines and tools to devise effectively such learning experiences, both from trainees’ and trainers’ perspectives. Will be also given a sneak peek into novel pedagogical models and opportunities that this kind of media are enabling.

Instrumented movement analysis is central for evaluating the level of mobility in populations with and without motor impairments. Recently, video-based marker less systems have been presented as a promising alternative to marker-based systems, considered the gold standard for instrumented human movement analysis, due to their affordability and the simple setup of the subject under analysis. For selected clinical applications, the output of marker less methods may be sufficiently accurate, making the use of marker-based systems unnecessary. In the field of marker less systems, RGB-D technology has been presented as a promising tool for a new generation of low-cost movement analysis systems. This talk will present a clinical marker less gait analysis protocol using a single RGB-D camera to estimate the lower limb joint kinematics for follow-up and early diagnoses purposes.

Sleep plays a fundamental role in health and provides a restorative function to the organism. There is growing evidence that poor sleep quality contributes to the onset of neurodegenerative processes. Therefore, sleep disturbances are commonly acknowledged among the earliest manifestations of neurological impairment. In particular, Rapid-Eye Movement Sleep Behaviour Disorder (RBD) is considered a precursor of alpha-synucleinopathies, with high conversion rates to Parkinson’s Disease. In its idiopathic stage, RBD offers a window for disease-modifying interventions. Therefore, early detection is pivotal for developing prevention strategies and personalised follow-up procedures. However, the diagnosis is a challenging task, and most cases frequently remain undiagnosed. This talk will present a lightweight framework for the early detection of RBD from electromyographic (EMG) recordings and introduce a strategy to assess and monitor the disease progression, thus leading to personalised outcomes that may improve the subjects’ quality of life.

Network virtualization introduced higher flexibility and dynamicity, but at the same time it led to new threats and challenges. The traditional approach of a manual configuration of Network Security Functions (NSFs) such as firewalls and VPN gateways is not feasible anymore, since it is not adequate to the ever-changing nature of modern networks and it is prone to human errors. To overcome this problem, the native flexibility provided by virtualization could be exploited to automate security configuration. However, achieving a high level of automation while providing formal assurance that the security configuration fulfills some security properties is still an open research challenge. Therefore, this speech presents a novel approach combining automation, formal verification and optimization for the allocation and configuration of NSFs in virtualized networks. Specifically, this approach pursues “security by construction”, avoiding a traditional a-posteriori formal verification, and fulfills optimality criteria to improve the efficiency of the security operations.

Optical network traffic continues to increase due to innovations and growth in end-user applications such as media streaming, cloud services, and the upcoming wide deployment of 5G networking. As a result, network operators wish to increase network capacities in a financially sustainable and costefficient manner, preferably by maximally exploiting already-installed hardware. To achieve this goal, software-defined networking (SDN) implementations have been formulated, which provide optical infrastructures with a greater degree of flexibility, exploiting capacity dynamically and efficiently, and permitting high levels of automation in network functions such as path computation and failure recovery. Simultaneously, the concept of open and disaggregated networking is gaining momentum, enabling operators to go beyond single-vendor frameworks and branch out into multi-vendor implementations. In this talk the advantages and challenges in open and disaggregated networking are presented, focusing on the application of analytical methods and artificial intelligence algorithms for quality of transmission estimation.

Driven by the increasing and greedy Internet data traffic request, optical network operators are working to satisfy this need, improving the already installed resources, or updating them thanks to the introduction of new technological discoveries. In this context, the most relevant support for service capacity increase and system management is conferred by optical network automation due to standardization and the consequent implementation of software-defined networks. Another important characteristic for an efficient usage of optical networks is the capability of the infrastructure to be agnostic with respect to the adopted vendor equipment, also favoring a more rapid deployment and integration of new functions. Definitely, this is allowed by hardware and software disaggregation, pushing in the direction of cognitive optical networks. Starting from the last decade, cognition has been introduced and theorized as an emerging feature of the next generation of optical networks. The response to the increasing complexity of the infrastructure is given by the possibility to probe the condition of the network through monitoring devices and to efficiently analyze the extracted information using flexible software modules. In this scenario, telemetry and monitoring devices cover a fundamental role, since they make it possible to retrieve information from the field to address different tasks and operations.

In the last few years, Neural Networks (NNets) have been used to beat the world champion in the game of Go and to achieve super-human accuracy in complex tasks, such as image classification and automatic language translation. The eagerness to adopt these powerful algorithms in a wide variety of applications with high user privacy, low latency, and low cost has recently required moving the NNets inference process from the cloud, powered by near-infinite resources, to “the edge”, that is, on lightweight resource-constrained embedded systems.
However, such a paradigm shift creates a tremendous technical challenge: filling the gap between the computational and memory requirements of modern NNets and the limited hardware and energy resources of embedded systems.
This talk will describe a vertical and automated approach to make NNets smaller, faster, and more energy-efficient.
Specifically, I will present several novel optimization techniques that push further the boundary of accurate NNets that can be deployed on tiny embedded devices.

The fast growth in global optical network traffic and the advancement in the communication technologies, such as coherent optical transmission, elastic optical networks, and the introduction of software-defined optical networks, introduce many tunable parameters, making the design and operation of optical networks more complex. In this framework, exact system modeling using closed-form formulations is very challenging, and typically, a ‘margin’ is needed while adopting the analytical models. The deployment of such margins leads to the underutilization of resources and eventually enlarges the network operational cost. To cater this limitation of analytical models, the telecom industry is strongly pushing the optical community to move towards intelligent optical networks that can perform autonomous and flexible network management and optimize network resource utilization. This talk presents the basic framework of software-defined optical networks assisted by artificial intelligence in the direction of cognitive optical networks and demonstrates some used case applications.

Movement disorders, such as Parkinson’s disease, involve a wide variety of motor symptoms. Wearable motion sensors represent minimally invasive solutions to accurately record human movement. When combined with advanced artificial intelligence techniques such as machine and deep learning, they have the potential to revolutionize the disease management paradigm. This talk will discuss the current challenges and opportunities of wearable technology and machine learning for monitoring Parkinson’s disease. Furthermore, computer programs capable of automatically assessing motor performance will be presented and described. Finally, the current limitations of such technology will be discussed.

This talk will dive into the exciting field of haptic technology in eXtended Reality (XR) and its potential to revolutionize the way we interact with digital experiences. From wearable devices to VR/AR applications, haptics is poised to bring a new level of immersion and realism to the virtual world. This presentation will provide an overview of some recent works in the field of haptic technology in XR, explore the different types of haptic feedback available, and examine some of the challenges and opportunities ahead. With its ability to deliver tangible touch sensations, haptics is shaping up to be the next big thing in XR, and this talk will provide a comprehensive overview of this emerging field.

Blockchain technology is gaining the interest of the academy, companies, and institutions. Nonetheless, the path toward blockchain adoption is not straightforward, as blockchain is a complex technology that requires revisiting the standard way of addressing problems and tackling them from a decentralized perspective. Thus, decision-makers adopt blockchain technology for the wrong reasons or mistakenly prefer it to more suitable ones. This speech introduces blockchain technology from a high-level perspective and discusses a decision-making framework for assessing blockchain’s suitability. The framework abstracts the complexity of blockchain technology and focuses on the relevant decision drivers. The framework helps decision-makers understand when blockchain is applicable, valuable, and preferable to other solutions. The framework can be used with no prior blockchain knowledge, becoming an effective tool for decision-makers that do not have the skills or the time to learn the complex blockchain technology.

Large amounts of data, such as that produced by sensors, especially cameras, can be used to reveal hidden links between the same data. To do this, visualization is often the most important tool. The main objective of the research is therefore to analyse, visualize and propose to the user data from different types of sensors.

The integration of Internet of Things (IoT) systems into our homes, workplaces, and everyday lives has greatly expanded the surface area of cyber attacks. To mitigate the issue of having compromised devices involved in massive distributed attacks, the Internet Engineering Task Force (IETF) has introduced the Manufacturer Usage Description (MUD). This standard enables manufacturers to specify the allowed endpoints for their devices.
Considering that even developers play a critical role in ensuring the cybersecurity of IoT systems, this presentation will explore an extended version of the MUD architecture that focuses on smart home gateways extendible through plug-ins. This solution allows developers to specify the required endpoints for their plug-ins, resulting in a consolidated gateway-level MUD file. This ensures that even non-MUD-enabled devices, integrated through plug-ins, are protected by the MUD standard.

Everyday, as we go about our business in a city, we carry around several devices such as smartphones, tablets or even laptops, most of them with an active WiFi interface. This interface “leaks” wireless traces, or footprints, in the form of beacon or probe packets that can be used to identify the presence of people in certain areas. In particular, the analysis of device footprints allows the detection, tracking and monitoring of people in indoor and outdoor scenarios. In this paper, we focus on the probe request messages broadcast by wireless devices and we analyze the behaviour and the characteristics of these messages from different devices, coming from various vendors, with different operating systems and features, also considering the user interaction with them. In particular, we provide a detailed picture of the adoption of MAC address randomization techniques, and on the variety of fields present within the probe request messages.