Reliable and Variation-Tolerant Interconnection Network for Low Power MPSOCS

Multi-Processor SoC (MPSOC) design brings to the foreground a large number of challenges, one of the most prominent of which is the design of the chip interconnection. With a number of on-chip blocks presently ranging in the tens, and quickly approaching the hundreds, the novel issue of how to best provide on-chip communication resources is clearly felt. Scaling down of process technologies has increased process and dynamic variations as well as transistor wearout. Because of this, delay variations increase and impact the performance of the MPSoCs. The interconnect architecture inMPSoCs becomes a single point of failure as it connects all other components of the system together. A faulty processing element may be shut down entirely, but the interconnect architecture must be able to tolerate partial failure and variations and operate with performance, power or latency overhead. This dissertation focuses on techniques at different levels of abstraction to face with the reliability and variability issues in on-chip interconnection networks. By showing the test results of a GALS NoC testchip this dissertation motivates the need for techniques to detect and work around manufacturing faults and process variations in MPSoCs’ interconnection infrastructure. As a physical design technique, we propose the bundle routing framework as an effective way to route the Network on Chips’ global links. For architecture-level design, two cases are addressed: (I) Intra-cluster communication where we propose a low-latency interconnect with variability robustness (ii) Inter-cluster communication where an online functional testing with a reliable NoC configuration are proposed. We also propose dualVdd as an orthogonal way of compensating variability at the post-fabrication stage. This is an alternative strategy with respect to the design techniques, since it enforces the compensation at post silicon stage.

Detection and Modeling of Boundary Layer Height

This thesis tackles the problem of the automated detection of the atmospheric boundary layer (BL) height, h, from aerosol lidar/ceilometer observations. A new method, the Bayesian Selective Method (BSM), is presented. It implements a Bayesian statistical inference procedure which combines in an statistically optimal way different sources of information. Firstly atmospheric stratification boundaries are located from discontinuities in the ceilometer back-scattered signal. The BSM then identifies the discontinuity edge that has the highest probability to effectively mark the BL height. Information from the contemporaneus physical boundary layer model simulations and a climatological dataset of BL height evolution are combined in the assimilation framework to assist this choice. The BSM algorithm has been tested for four months of continuous ceilometer measurements collected during the BASE:ALFA project and is shown to realistically diagnose the BL depth evolution in many different weather conditions. Then the BASE:ALFA dataset is used to investigate the boundary layer structure in stable conditions. Functions from the Obukhov similarity theory are used as regression curves to fit observed velocity and temperature profiles in the lower half of the stable boundary layer. Surface fluxes of heat and momentum are best-fitting parameters in this exercise and are compared with what measured by a sonic anemometer. The comparison shows remarkable discrepancies, more evident in cases for which the bulk Richardson number turns out to be quite large. This analysis supports earlier results, that surface turbulent fluxes are not the appropriate scaling parameters for profiles of mean quantities in very stable conditions. One of the practical consequences is that boundary layer height diagnostic formulations which mainly rely on surface fluxes are in disagreement to what obtained by inspecting co-located radiosounding profiles.

Photoactive (supra)molecular devices in double layer membranes

Biological systems are complex and highly organized architectures governed by non-covalent interactions responsible for the regulation of essential tasks in all living organisms. These systems are a constant source of inspiration for supramolecular chemists aiming to design multicomponent molecular assemblies able to perform elaborated tasks, thanks to the role and action of the components that constitute them. Artificial supramolecular systems exploit non-covalent interactions to mimic naturally occurring events. In this context, stimuli-responsive supramolecular systems have attracted attention due to the possibility to control macroscopic effects through modifications at the nanoscale. This thesis is divided in three experimental chapters, characterized by a progressive increase in molecular complexity. Initially, the preparation and studies of liposomes functionalized with a photoactive guest such as azobenzene in the bilayer were tackled, in order to evaluate the effect of such photochrome on the vesicle properties. Subsequently, the synthesis and studies of thread-like molecules comprising an azobenzene functionality was reported. Such molecules were conceived to be intercalated in the bilayer membrane of liposomes with the aim to be used as components for photoresponsive transmembrane molecular pumps. Finally, a [3]rotaxane was developed and studied in solution. This system is composed of two crown ether rings interlocked with an axle containing three recognition sites for the macrocycles, i.e. two pH-switchable ammonium stations and a permanent triazolium station. Such molecule was designed to achieve a change in the ratio between the recognition sites and the crown ethers as a consequence of acid-base inputs. This leads to the formation of rotaxanes containing a number of recognition sites respectively larger, equal or lower than the number of interlocked rings and connected by a network of acid-base reactions.

Role of non-coding RNA alterations in melanoma

Cutaneous melanoma (CM) is a potentially lethal form of skin cancer and its most important histopathologic factor for staging is Breslow thickness (BT). Its correct determination is fundamental for pathologists. A deeper understanding of the molecular processes guiding CM pathogenesis could improve diagnosis, treatment and prognosis. MicroRNAs (miRNAs) play a key role in CM biology. The firs aim was to investigate miRNA expression in reference to BT assessment. We found that the combined miRNA expression of miR-21-5p and miR-146a-5p above or below 1.5 was significantly associated with overall survival and successfully identified all superficially spreading melanoma (SSM) patients with relapsing suggesting that the combined assessment of these miRNAs expression could aid in SSM staging. Secondly, we focus on multiple primary melanoma (MPM) patients, which develop multiple primary melanomas in their lifetime, and represent a model of high-risk CM occurrence. We explored the miRNome of single CM and MPM: CM and MPM present several dysregulated miRNAs, including key miRNAs involved in epithelial-mesenchymal transition. A different miRNA profile was observed between 1st and 2nd melanoma from the same patient. MiRNA target analysis revealed a more differentiated and less invasive status of MPMs compared to CMs. This characterization of the miRNA regulatory network of MPMs highlights molecular features differentiating this subtype from CM. Recently, NGS experiments revealed the existence of miRNA variants (isomiRs) with different length and sequence. We identified a shorter 3’isoform as tenfold over-represented compared to the canonical form of miR-125a-5p. Target analysis revealed that miRNA shortening could change the pattern of target gene regulation. Finally, we study miRNA and isomiR dysregulation in benign nevi (BN) and CM and in CM and melanoma metastasis. The reported non-random dysregulation of specific isomiRs contributes to the understanding of the complex melanoma pathogenesis and serves as the basis for further functional studies.

Automated service provisioning in programmable network infrastructures

Modern networks are undergoing a fast and drastic evolution, with software taking a more predominant role. Virtualization and cloud-like approaches are replacing physical network appliances, reducing the management burden of the operators. Furthermore, networks now expose programmable interfaces for fast and dynamic control over traffic forwarding. This evolution is backed by standard organizations such as ETSI, 3GPP, and IETF. This thesis will describe which are the main trends in this evolution. Then, it will present solutions developed during the three years of Ph.D. to exploit the capabilities these new technologies offer and to study their possible limitations to push further the state-of-the-art. Namely, it will deal with programmable network infrastructure, introducing the concept of Service Function Chaining (SFC) and presenting two possible solutions, one with Openstack and OpenFlow and the other using Segment Routing and IPv6. Then, it will continue with network service provisioning, presenting concepts from Network Function Virtualization (NFV) and Multi-access Edge Computing (MEC). These concepts will be applied to network slicing for mission-critical communications and Industrial IoT (IIoT). Finally, it will deal with network abstraction, with a focus on Intent Based Networking (IBN). To summarize, the thesis will include solutions for data plane programming with evaluation on well-known platforms, performance metrics on virtual resource allocations, novel practical application of network slicing on mission-critical communications, an architectural proposal and its implementation for edge technologies in Industrial IoT scenarios, and a formal definition of intent using a category theory approach.

Optimizing AI at the Edge: from network topology design to MCU deployment

The first topic analyzed in the thesis will be Neural Architecture Search (NAS). I will focus on two different tools that I developed, one to optimize the architecture of Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged, and one to optimize the data precision of tensors inside CNNs. The first NAS proposed explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive field, and the number of features in each layer. Note that this is the first NAS that explicitly targets these networks. The second NAS proposed instead focuses on finding the most efficient data format for a target CNN, with the granularity of the layer filter. Note that applying these two NASes in sequence allows an "application designer" to minimize the structure of the neural network employed, minimizing the number of operations or the memory usage of the network. After that, the second topic described is the optimization of neural network deployment on edge devices. Importantly, exploiting edge platforms' scarce resources is critical for NN efficient execution on MCUs. To do so, I will introduce DORY (Deployment Oriented to memoRY) -- an automatic tool to deploy CNNs on low-cost MCUs. DORY, in different steps, can manage different levels of memory inside the MCU automatically, offload the computation workload (i.e., the different layers of a neural network) to dedicated hardware accelerators, and automatically generates ANSI C code that orchestrates off- and on-chip transfers with the computation phases. On top of this, I will introduce two optimized computation libraries that DORY can exploit to deploy TCNs and Transformers on edge efficiently. I conclude the thesis with two different applications on bio-signal analysis, i.e., heart rate tracking and sEMG-based gesture recognition.

Characterisation of RNase Y in H. pylori: a non-essential enzyme involved in the processing of cag-PAI non coding RNA 1 (CncR1) sRNA

The importance of Helicobacter pylori as a human pathogen is underlined by the plethora of diseases it is responsible for. The capacity of H. pylori to adapt to the restricted host-associated environment andto evade the host immune response largely depends on a streamlined signalling network. The peculiar H. pylori small genome size combined with its paucity of transcriptional regulators highlights the relevance of post-transcriptional regulatory mechanisms as small non-coding RNAs (sRNAs). However, among the 8 RNases represented in H. pylori genome, a regulator guiding sRNAs metabolism is still not well studied. We investigated for the first time the physiological role in H. pylori G27 strain of the RNase Y enzyme. In the first line of research we provide a comprehensive characterization of the RNase Y activity by analysing its genomic organization and the factors that orchestrate its expression. Then, based on bioinformatic prediction models, we depict the most relevant determinants of RNase Y function, demonstrating a correlation of both structure and domain organization with orthologues represented in Gram-positive bacteria. To unveil the post-transcriptional regulatory effect exerted by the RNase Y, we compared the transcriptome of an RNase Y knock-out mutant to the parental wild type strain by RNA-seq approach. In the second line of research we characterized the activity of this single strand specific endoribonuclease on cag-PAI non coding RNA 1 (CncR1) sRNA. We found that deletion or inactivation of RNase Y led to the accumulation of a 3’-extended CncR1 (CncR1-L) transcript over time. Moreover, beneath its increased half-life, CncR1-L resembled a CncR1 inactive phenotype. Finally, we focused on the characterization of the in vivo interactome of CncR1. We set up a preliminary MS2-affinity purification coupled with RNA-sequencing (MAPS) approach and we evaluated the enrichment of specific targets, demonstrating the suitability of the technique in the H. pylori G27 strain.

Physical modeling and numerical simulations of degradation mechanisms in devices and insulators for power applications

In this thesis, a TCAD approach for the investigation of charge transport in amorphous silicon dioxide is presented for the first time. The proposed approach is used to investigate high-voltage silicon oxide thick TEOS capacitors embedded in the back-end inter-level dielectric layers for galvanic insulation applications. In the first part of this thesis, a detailed review of the main physical and chemical properties of silicon dioxide and the main physical models for the description of charge transport in insulators are presented. In the second part, the characterization of high-voltage MIM structures at different high-field stress conditions up to the breakdown is presented. The main physical mechanisms responsible of the observed results are then discussed in details. The third part is dedicated to the implementation of a TCAD approach capable of describing charge transport in silicon dioxide layers in order to gain insight into the microscopic physical mechanisms responsible of the leakage current in MIM structures. In particular, I investigated and modeled the role of charge injection at contacts and charge build-up due to trapping and de-trapping mechanisms in the oxide layer to the purpose of understanding its behavior under DC and AC stress conditions. In addition, oxide breakdown due to impact-ionization of carriers has been taken into account in order to have a complete representation of the oxide behavior at very high fields. Numerical simulations have been compared against experiments to quantitatively validate the proposed approach. In the last part of the thesis, the proposed approach has been applied to simulate the breakdown in realistic structures under different stress conditions. The TCAD tool has been used to carry out a detailed analysis of the most relevant physical quantities, in order to gain a detailed understanding on the main mechanisms responsible for breakdown and guide design optimization.

An architecture for network acceleration as a service in the cloud continuum

The pervasive availability of connected devices in any industrial and societal sector is pushing for an evolution of the well-established cloud computing model. The emerging paradigm of the cloud continuum embraces this decentralization trend and envisions virtualized computing resources physically located between traditional datacenters and data sources. By totally or partially executing closer to the network edge, applications can have quicker reactions to events, thus enabling advanced forms of automation and intelligence. However, these applications also induce new data-intensive workloads with low-latency constraints that require the adoption of specialized resources, such as high-performance communication options (e.g., RDMA, DPDK, XDP, etc.). Unfortunately, cloud providers still struggle to integrate these options into their infrastructures. That risks undermining the principle of generality that underlies the cloud computing scale economy by forcing developers to tailor their code to low-level APIs, non-standard programming models, and static execution environments. This thesis proposes a novel system architecture to empower cloud platforms across the whole cloud continuum with Network Acceleration as a Service (NAaaS). To provide commodity yet efficient access to acceleration, this architecture defines a layer of agnostic high-performance I/O APIs, exposed to applications and clearly separated from the heterogeneous protocols, interfaces, and hardware devices that implement it. A novel system component embodies this decoupling by offering a set of agnostic OS features to applications: memory management for zero-copy transfers, asynchronous I/O processing, and efficient packet scheduling. This thesis also explores the design space of the possible implementations of this architecture by proposing two reference middleware systems and by adopting them to support interactive use cases in the cloud continuum: a serverless platform and an Industry 4.0 scenario. A detailed discussion and a thorough performance evaluation demonstrate that the proposed architecture is suitable to enable the easy-to-use, flexible integration of modern network acceleration into next-generation cloud platforms.

Grant-free protocols for massive multiple access

In next generation Internet-of-Things, the overhead introduced by grant-based multiple access protocols may engulf the access network as a consequence of the proliferation of connected devices. Grant-free access protocols are therefore gaining an increasing interest to support massive multiple access. In addition to scalability requirements, new demands have emerged for massive multiple access, including latency and reliability. The challenges envisaged for future wireless communication networks, particularly in the context of massive access, include: i) a very large population size of low power devices transmitting short packets; ii) an ever-increasing scalability requirement; iii) a mild fixed maximum latency requirement; iv) a non-trivial requirement on reliability. To this aim, we suggest the joint utilization of grant-free access protocols, massive MIMO at the base station side, framed schemes to let the contention start and end within a frame, and succesive interference cancellation techniques at the base station side. In essence, this approach is encapsulated in the concept of coded random access with massive MIMO processing. These schemes can be explored from various angles, spanning the protocol stack from the physical (PHY) to the medium access control (MAC) layer. In this thesis, we delve into both of these layers, examining topics ranging from symbol-level signal processing to succesive interference cancellation-based scheduling strategies. In parallel with proposing new schemes, our work includes a theoretical analysis aimed at providing valuable system design guidelines. As a main theoretical outcome, we propose a novel joint PHY and MAC layer design based on density evolution on sparse graphs.