PCM-based in-memory computing: architectures, circuits and applications

Analog In-memory Computing (AIMC) has been proposed in the context of Beyond Von Neumann architectures as a valid strategy to reduce internal data transfers energy consumption and latency, and to improve compute efficiency. The aim of AIMC is to perform computations within the memory unit, typically leveraging the physical features of memory devices. Among resistive Non-volatile Memories (NVMs), Phase-change Memory (PCM) has become a promising technology due to its intrinsic capability to store multilevel data. Hence, PCM technology is currently investigated to enhance the possibilities and the applications of AIMC. This thesis aims at exploring the potential of new PCM-based architectures as in-memory computational accelerators. In a first step, a preliminar experimental characterization of PCM devices has been carried out in an AIMC perspective. PCM cells non-idealities, such as time-drift, noise, and non-linearity have been studied to develop a dedicated multilevel programming algorithm. Measurement-based simulations have been then employed to evaluate the feasibility of PCM-based operations in the fields of Deep Neural Networks (DNNs) and Structural Health Monitoring (SHM). Moreover, a first testchip has been designed and tested to evaluate the hardware implementation of Multiply-and-Accumulate (MAC) operations employing PCM cells. This prototype experimentally demonstrates the possibility to reach a 95% MAC accuracy with a circuit-level compensation of cells time drift and non-linearity. Finally, empirical circuit behavior models have been included in simulations to assess the use of this technology in specific DNN applications, and to enhance the potentiality of this innovative computation approach.

Studio molecolare dei meccanismi dell'autoincompatibilità gametofitica in pero europeo (Pyrus communis)

Self-incompatibility (SI) systems have evolved in many flowering plants to prevent self-fertilization and thus promote outbreeding. Pear and apple, as many of the species belonging to the Rosaceae, exhibit RNase-mediated gametophytic self-incompatibility, a widespread system carried also by the Solanaceae and Plantaginaceae. Pear orchards must for this reason contain at least two different cultivars that pollenize each other; to guarantee an efficient cross-pollination, they should have overlapping flowering periods and must be genetically compatible. This compatibility is determined by the S-locus, containing at least two genes encoding for a female (pistil) and a male (pollen) determinant. The female determinant in the Rosaceae, Solanaceae and Plantaginaceae system is a stylar glycoprotein with ribonuclease activity (S-RNase), that acts as a specific cytotoxin in incompatible pollen tubes degrading cellular RNAs. Since its identification, the S-RNase gene has been intensively studied and the sequences of a large number of alleles are available in online databases. On the contrary, the male determinant has been only recently identified as a pollen-expressed protein containing a F-box motif, called S-Locus F-box (abbreviated SLF or SFB). Since F-box proteins are best known for their participation to the SCF (Skp1 - Cullin - F-box) E3 ubiquitine ligase enzymatic complex, that is involved in protein degradation through the 26S proteasome pathway, the male determinant is supposed to act mediating the ubiquitination of the S-RNases, targeting them for the degradation in compatible pollen tubes. Attempts to clone SLF/SFB genes in the Pyrinae produced no results until very recently; in apple, the use of genomic libraries allowed the detection of two F-box genes linked to each S haplotype, called SFBB (S-locus F-Box Brothers). In Japanese pear, three SFBB genes linked to each haplotype were cloned from pollen cDNA. The SFBB genes exhibit S haplotype-specific sequence divergence and pollen-specific expression; their multiplicity is a feature whose interpretation is unclear: it has been hypothesized that all of them participate in the S-specific interaction with the RNase, but it is also possible that only one of them is involved in this function. Moreover, even if the S locus male and female determinants are the only responsible for the specificity of the pollen-pistil recognition, many other factors are supposed to play a role in GSI; these are not linked to the S locus and act in a S-haplotype independent manner. They can have a function in regulating the expression of S determinants (group 1 factors), modulating their activity (group 2) or acting downstream, in the accomplishment of the reaction of acceptance or rejection of the pollen tube (group 3). This study was aimed to the elucidation of the molecular mechanism of GSI in European pear (Pyrus communis) as well as in the other Pyrinae; it was divided in two parts, the first focusing on the characterization of male determinants, and the second on factors external to the S locus. The research of S locus F-box genes was primarily aimed to the identification of such genes in European pear, for which sequence data are still not available; moreover, it allowed also to investigate about the S locus structure in the Pyrinae. The analysis was carried out on a pool of varieties of the three species Pyrus communis (European pear), Pyrus pyrifolia (Japanese pear), and Malus × domestica (apple); varieties carrying S haplotypes whose RNases are highly similar were chosen, in order to check whether or not the same level of similarity is maintained also between the male determinants. A total of 82 sequences was obtained, 47 of which represent the first S-locus F-box genes sequenced from European pear. The sequence data strongly support the hypothesis that the S locus structure is conserved among the three species, and presumably among all the Pyrinae; at least five genes have homologs in the analysed S haplotypes, but the number of F-box genes surrounding the S-RNase could be even greater. The high level of sequence divergence and the similarity between alleles linked to highly conserved RNases, suggest a shared ancestral polymorphism also for the F-box genes. The F-box genes identified in European pear were mapped on a segregating population of 91 individuals from the cross 'Abbé Fétel' × 'Max Red Bartlett'. All the genes were placed on the linkage group 17, where the S locus has been placed both in pear and apple maps, and resulted strongly associated to the S-RNase gene. The linkage with the RNase was perfect for some of the F-box genes, while for others very rare single recombination events were identified. The second part of this study was focused on the research of other genes involved in the SI response in pear; it was aimed on one side to the identification of genes differentially expressed in compatible and incompatible crosses, and on the other to the cloning and characterization of the transglutaminase (TGase) gene, whose role may be crucial in pollen rejection. For the identification of differentially expressed genes, controlled pollinations were carried out in four combinations (self pollination, incompatible, half-compatible and fully compatible cross-pollination); expression profiles were compared through cDNA-AFLP. 28 fragments displaying an expression pattern related to compatibility or incompatibility were identified, cloned and sequenced; the sequence analysis allowed to assign a putative annotation to a part of them. The identified genes are involved in very different cellular processes or in defense mechanisms, suggesting a very complex change in gene expression following the pollen/pistil recognition. The pool of genes identified with this technique offers a good basis for further study toward a better understanding of how the SI response is carried out. Among the factors involved in SI response, moreover, an important role may be played by transglutaminase (TGase), an enzyme involved both in post-translational protein modification and in protein cross-linking. The TGase activity detected in pear styles was significantly higher when pollinated in incompatible combinations than in compatible ones, suggesting a role of this enzyme in the abnormal cytoskeletal reorganization observed during pollen rejection reaction. The aim of this part of the work was thus to identify and clone the pear TGase gene; the PCR amplification of fragments of this gene was achieved using primers realized on the alignment between the Arabidopsis TGase gene sequence and several apple EST fragments; the full-length coding sequence of the pear TGase gene was then cloned from cDNA, and provided a precious tool for further study of the in vitro and in vivo action of this enzyme.

Investigating the role of single point mutations in the human proteome: a computational study

In the post genomic era with the massive production of biological data the understanding of factors affecting protein stability is one of the most important and challenging tasks for highlighting the role of mutations in relation to human maladies. The problem is at the basis of what is referred to as molecular medicine with the underlying idea that pathologies can be detailed at a molecular level. To this purpose scientific efforts focus on characterising mutations that hamper protein functions and by these affect biological processes at the basis of cell physiology. New techniques have been developed with the aim of detailing single nucleotide polymorphisms (SNPs) at large in all the human chromosomes and by this information in specific databases are exponentially increasing. Eventually mutations that can be found at the DNA level, when occurring in transcribed regions may then lead to mutated proteins and this can be a serious medical problem, largely affecting the phenotype. Bioinformatics tools are urgently needed to cope with the flood of genomic data stored in database and in order to analyse the role of SNPs at the protein level. In principle several experimental and theoretical observations are suggesting that protein stability in the solvent-protein space is responsible of the correct protein functioning. Then mutations that are found disease related during DNA analysis are often assumed to perturb protein stability as well. However so far no extensive analysis at the proteome level has investigated whether this is the case. Also computationally methods have been developed to infer whether a mutation is disease related and independently whether it affects protein stability. Therefore whether the perturbation of protein stability is related to what it is routinely referred to as a disease is still a big question mark. In this work we have tried for the first time to explore the relation among mutations at the protein level and their relevance to diseases with a large-scale computational study of the data from different databases. To this aim in the first part of the thesis for each mutation type we have derived two probabilistic indices (for 141 out of 150 possible SNPs): the perturbing index (Pp), which indicates the probability that a given mutation effects protein stability considering all the “in vitro” thermodynamic data available and the disease index (Pd), which indicates the probability of a mutation to be disease related, given all the mutations that have been clinically associated so far. We find with a robust statistics that the two indexes correlate with the exception of all the mutations that are somatic cancer related. By this each mutation of the 150 can be coded by two values that allow a direct comparison with data base information. Furthermore we also implement computational methods that starting from the protein structure is suited to predict the effect of a mutation on protein stability and find that overpasses a set of other predictors performing the same task. The predictor is based on support vector machines and takes as input protein tertiary structures. We show that the predicted data well correlate with the data from the databases. All our efforts therefore add to the SNP annotation process and more importantly found the relationship among protein stability perturbation and the human variome leading to the diseasome.

On-line adaptive visual tracking

Visual tracking is the problem of estimating some variables related to a target given a video sequence depicting the target. Visual tracking is key to the automation of many tasks, such as visual surveillance, robot or vehicle autonomous navigation, automatic video indexing in multimedia databases. Despite many years of research, long term tracking in real world scenarios for generic targets is still unaccomplished. The main contribution of this thesis is the definition of effective algorithms that can foster a general solution to visual tracking by letting the tracker adapt to mutating working conditions. In particular, we propose to adapt two crucial components of visual trackers: the transition model and the appearance model. The less general but widespread case of tracking from a static camera is also considered and a novel change detection algorithm robust to sudden illumination changes is proposed. Based on this, a principled adaptive framework to model the interaction between Bayesian change detection and recursive Bayesian trackers is introduced. Finally, the problem of automatic tracker initialization is considered. In particular, a novel solution for categorization of 3D data is presented. The novel category recognition algorithm is based on a novel 3D descriptors that is shown to achieve state of the art performances in several applications of surface matching.

Cache-aware development of high integrity real-time systems

Cost, performance and availability considerations are forcing even the most conservative high-integrity embedded real-time systems industry to migrate from simple hardware processors to ones equipped with caches and other acceleration features. This migration disrupts the practices and solutions that industry had developed and consolidated over the years to perform timing analysis. Industry that are confident with the efficiency/effectiveness of their verification and validation processes for old-generation processors, do not have sufficient insight on the effects of the migration to cache-equipped processors. Caches are perceived as an additional source of complexity, which has potential for shattering the guarantees of cost- and schedule-constrained qualification of their systems. The current industrial approach to timing analysis is ill-equipped to cope with the variability incurred by caches. Conversely, the application of advanced WCET analysis techniques on real-world industrial software, developed without analysability in mind, is hardly feasible. We propose a development approach aimed at minimising the cache jitters, as well as at enabling the application of advanced WCET analysis techniques to industrial systems. Our approach builds on:(i) identification of those software constructs that may impede or complicate timing analysis in industrial-scale systems; (ii) elaboration of practical means, under the model-driven engineering (MDE) paradigm, to enforce the automated generation of software that is analyzable by construction; (iii) implementation of a layout optimisation method to remove cache jitters stemming from the software layout in memory, with the intent of facilitating incremental software development, which is of high strategic interest to industry. The integration of those constituents in a structured approach to timing analysis achieves two interesting properties: the resulting software is analysable from the earliest releases onwards - as opposed to becoming so only when the system is final - and more easily amenable to advanced timing analysis by construction, regardless of the system scale and complexity.

Machine-learning methods for structure prediction of β-barrel membrane proteins

Different types of proteins exist with diverse functions that are essential for living organisms. An important class of proteins is represented by transmembrane proteins which are specifically designed to be inserted into biological membranes and devised to perform very important functions in the cell such as cell communication and active transport across the membrane. Transmembrane β-barrels (TMBBs) are a sub-class of membrane proteins largely under-represented in structure databases because of the extreme difficulty in experimental structure determination. For this reason, computational tools that are able to predict the structure of TMBBs are needed. In this thesis, two computational problems related to TMBBs were addressed: the detection of TMBBs in large datasets of proteins and the prediction of the topology of TMBB proteins. Firstly, a method for TMBB detection was presented based on a novel neural network framework for variable-length sequence classification. The proposed approach was validated on a non-redundant dataset of proteins. Furthermore, we carried-out genome-wide detection using the entire Escherichia coli proteome. In both experiments, the method significantly outperformed other existing state-of-the-art approaches, reaching very high PPV (92%) and MCC (0.82). Secondly, a method was also introduced for TMBB topology prediction. The proposed approach is based on grammatical modelling and probabilistic discriminative models for sequence data labeling. The method was evaluated using a newly generated dataset of 38 TMBB proteins obtained from high-resolution data in the PDB. Results have shown that the model is able to correctly predict topologies of 25 out of 38 protein chains in the dataset. When tested on previously released datasets, the performances of the proposed approach were measured as comparable or superior to the current state-of-the-art of TMBB topology prediction.

Fattori eziologici della Sindrome del Tunnel Carpale: una revisione sistematica con meta-analisi degli studi analitici

OBIETTIVO: sintetizzare le evidenze disponibili sulla relazione tra i fattori di rischio (personali e lavorativi) e l’insorgenza della Sindrome del Tunnel Carpale (STC). METODI: è stata condotta una revisione sistematica della letteratura su database elettronici considerando gli studi caso-controllo e di coorte. Abbiamo valutato la qualità del reporting degli studi con la checklist STROBE. Le stime studio-specifiche sono state espresse come OR (IC95%) e combinate con una meta-analisi condotta con un modello a effetti casuali. La presenza di eventuali bias di pubblicazione è stata valutata osservando l’asimmetria del funnel plot e con il test di Egger. RISULTATI: Sono stati selezionati 29 studi di cui 19 inseriti nella meta-analisi: 13 studi caso-controllo e 6 di coorte. La meta-analisi ha mostrato un aumento significativo di casi di STC tra i soggetti obesi sia negli studi caso-controllo [OR 2,4 (1,9-3,1); I(2)=70,7%] che in quelli di coorte [OR 2,0 (1,6-2,7); I(2)=0%]. L'eterogeneità totale era significativa (I(2)=59,6%). Risultati simili si sono ottenuti per i diabetici e soggetti affetti da malattie della tiroide. L’esposizione al fumo non era associata alla STC sia negli studi caso-controllo [OR 0,7 (0,4-1,1); I(2)=83,2%] che di coorte [OR 0,8 (0,6-1,2); I(2)=45,8%]. A causa delle molteplici modalità di valutazione non è stato possibile calcolare una stima combinata delle esposizioni professionali con tecniche meta-analitiche. Dalla revisione, è risultato che STC è associata con: esposizione a vibrazioni, movimenti ripetitivi e posture incongrue di mano-polso. CONCLUSIONI: I risultati della revisione sistematica confermano le evidenze dell'esistenza di un'associazione tra fattori di rischio personali e STC. Nonostante la diversa qualità dei dati sull'esposizione e le differenze degli effetti dei disegni di studio, i nostri risultati indicano elementi di prova sufficienti di un legame tra fattori di rischio professionali e STC. La misurazione dell'esposizione soprattutto per i fattori di rischio professionali, è un obiettivo necessario per studi futuri.

Evaluation of alpha-synuclein RT-QuIC for an early and differential diagnosis of Lewy body disease

Synucleinopathies are a group of neurodegenerative diseases characterized by tissue deposition of insoluble aggregates of the protein α-synuclein. Currently, the clinical diagnosis of these diseases, including Parkinson’s disease (PD), dementia with Lewy bodies (DLB), and multiple system atrophy (MSA), is very challenging, especially at an early disease stage, due to the heterogeneous and often non-specific clinical manifestations. Therefore, identifying specific biomarkers to aid the diagnosis and improve the clinical management of patients with these disorders represents a primary goal in the field. Pursuing this aim, we applied the α-Syn Real-Time Quaking-Induced Conversion (RT-QuIC), an ultrasensitive technique able to detect minute amounts of amyloidogenic proteins, to a large cohort of 953 CSF samples from clinically well-characterized (“clinical” group), or neuropathologically verified (“NP” group) patients with parkinsonism or dementia. Of significance, we also studied patients with prodromal synucleinopathies (“prodromal” group), such as pure autonomic failure (PAF) (n = 28), isolated REM sleep behavior disorder (iRBD) (n = 18), and mild cognitive impairment due to probable Lewy body (LB) disease (MCI-LB) (n = 81). Our findings show that α-syn RT-QuIC can accurately detect α-Syn seeding activity across the whole spectrum of LB-related disorders (LBD), exhibiting a mean sensitivity of 95.2% in the “clinical” and “NP” group, while ranging between 89.3% (PAF) and 100% (RBD) in the “prodromal group”. Moreover, we observed 95.1% sensitivity and 96.6% specificity in the distinction between MCI-LB patients and cognitively unimpaired controls, demonstrating the solid diagnostic potential of α-Syn RT-QuIC in the early phase of the disease. Finally, 13.3% of MCI-AD patients also had a positive test, suggesting an underlying LB co-pathology. This work demonstrated that α-Syn RT-QuIC is an efficient assay for accurate and early diagnosis of LBD, which should be implemented for clinical management and recruitment for clinical trials in memory clinics.

Exploiting the environment for structural health monitoring: Strategies, algorithms, and technologies

A densely built environment is a complex system of infrastructure, nature, and people closely interconnected and interacting. Vehicles, public transport, weather action, and sports activities constitute a manifold set of excitation and degradation sources for civil structures. In this context, operators should consider different factors in a holistic approach for assessing the structural health state. Vibration-based structural health monitoring (SHM) has demonstrated great potential as a decision-supporting tool to schedule maintenance interventions. However, most excitation sources are considered an issue for practical SHM applications since traditional methods are typically based on strict assumptions on input stationarity. Last-generation low-cost sensors present limitations related to a modest sensitivity and high noise floor compared to traditional instrumentation. If these devices are used for SHM in urban scenarios, short vibration recordings collected during high-intensity events and vehicle passage may be the only available datasets with a sufficient signal-to-noise ratio. While researchers have spent efforts to mitigate the effects of short-term phenomena in vibration-based SHM, the ultimate goal of this thesis is to exploit them and obtain valuable information on the structural health state. First, this thesis proposes strategies and algorithms for smart sensors operating individually or in a distributed computing framework to identify damage-sensitive features based on instantaneous modal parameters and influence lines. Ordinary traffic and people activities become essential sources of excitation, while human-powered vehicles, instrumented with smartphones, take the role of roving sensors in crowdsourced monitoring strategies. The technical and computational apparatus is optimized using in-memory computing technologies. Moreover, identifying additional local features can be particularly useful to support the damage assessment of complex structures. Thereby, smart coatings are studied to enable the self-sensing properties of ordinary structural elements. In this context, a machine-learning-aided tomography method is proposed to interpret the data provided by a nanocomposite paint interrogated electrically.

Flexible Computing Systems For AI Acceleration At The Extreme Edge Of The IoT

Embedding intelligence in extreme edge devices allows distilling raw data acquired from sensors into actionable information, directly on IoT end-nodes. This computing paradigm, in which end-nodes no longer depend entirely on the Cloud, offers undeniable benefits, driving a large research area (TinyML) to deploy leading Machine Learning (ML) algorithms on micro-controller class of devices. To fit the limited memory storage capability of these tiny platforms, full-precision Deep Neural Networks (DNNs) are compressed by representing their data down to byte and sub-byte formats, in the integer domain. However, the current generation of micro-controller systems can barely cope with the computing requirements of QNNs. This thesis tackles the challenge from many perspectives, presenting solutions both at software and hardware levels, exploiting parallelism, heterogeneity and software programmability to guarantee high flexibility and high energy-performance proportionality. The first contribution, PULP-NN, is an optimized software computing library for QNN inference on parallel ultra-low-power (PULP) clusters of RISC-V processors, showing one order of magnitude improvements in performance and energy efficiency, compared to current State-of-the-Art (SoA) STM32 micro-controller systems (MCUs) based on ARM Cortex-M cores. The second contribution is XpulpNN, a set of RISC-V domain specific instruction set architecture (ISA) extensions to deal with sub-byte integer arithmetic computation. The solution, including the ISA extensions and the micro-architecture to support them, achieves energy efficiency comparable with dedicated DNN accelerators and surpasses the efficiency of SoA ARM Cortex-M based MCUs, such as the low-end STM32M4 and the high-end STM32H7 devices, by up to three orders of magnitude. To overcome the Von Neumann bottleneck while guaranteeing the highest flexibility, the final contribution integrates an Analog In-Memory Computing accelerator into the PULP cluster, creating a fully programmable heterogeneous fabric that demonstrates end-to-end inference capabilities of SoA MobileNetV2 models, showing two orders of magnitude performance improvements over current SoA analog/digital solutions.