Cette thèse constitue un des éléments du projet de recherche EMOCAVI (Evolution des Modèles des Composants de puissance grand gAp au cours du VIeillissement). Elle porte sur l’étude de la fiabilité des transistors de puissance en Nitrure de Gallium (GaN) récemment apparus sur le marché. Ces travaux se focalisent sur la réalisation d’une méthodologie pour paramétrer le modèle du composant GaN GIT (Gate Injection Transistor) en fonction du vieillissement auquel il a été soumis. Pour atteindre cet objectif, nous sommes passés par plusieurs étapes. La première a été consacrée à la définition, la mise en place et la validation d’un banc de vieillissement et à la caractérisation de ces composants avant et en cours de vieillissement. Un banc de test de vieillissement en court-circuit répétitif à faible puissance a été conçu et mis en oeuvre. Ce banc a permis de valider l’hypothèse du vieillissement lié à l’énergie, d’identifier son niveau déterminant d’un point de vue fiabilité du composant et enfin mettre en évidence la dégradation progressive du composant afin d’identifier les paramètres du transistor les plus sensibles au vieillissement. La deuxième étape de nos travaux a été consacrée à l’établissement d’une méthodologie de création de modèle de vieillissement du composant GaN-GIT. En reproduisant le modèle COBRA présenté dans la littérature, nous avons réussi dans nos travaux à proposer une approche novatrice permettant d’intégrer les dépendances en température et en énergie subie par le composant pendant le stress (la durée d’impulsion Tsc et le nombre de pulse subi Nsc). La dernière étape de nos travaux a été dédiée à l’analyse physique de défaillance afin de confirmer les hypothèses faites sur les mécanismes de dégradation obtenus après vieillissement du composant. Pour réaliser ces analyses, nous avons commencé par la décapsulation du composant en combinant l’ouverture laser aux attaques chimiques de la résine constituant le packaging. Une fois le défaut localisé par photoluminescence, une analyse approfondie par des vues au microscope électronique à balayage MEB puis par découpe PFIB (Plasma Fouced Ion Beam) a été réalisée afin de déterminer le mécanisme de défaillance. Il s’agissait principalement de fissures situées dans le métal d’Al au niveau du drain ainsi que la présence de cavités dans la couche métallique qui sert à réaliser le contact ohmique au niveau de la source, ce qui explique l’augmentation de la résistance RDSON. / This thesis constitute one of the elements of the EMOCAVI research project (Evolution of the Large gAp Power Component Models during the VIeillissement). It deals with the study of the reliability of Gallium Nitride (GaN) power transistors which are recently appeared on the market. This work focuses on the realization of a methodology to parameterize the model of GaN GIT component (Gate Injection Transistor) according to the aging to which it has been subjected. To achieve this goal, it will be necessary to go through several steps. The first step was dedicated to the definition, implementation and validation of an aging bench for the component and the characterization of these components before and during aging. A low power repetitive short-circuit aging test bench was designed and implemented. This bench is used to validate the energy-related aging hypothesis, to identify its determining level from a point of view of the reliability of the component and finally to highlight the progressive degradation of the component in order to identify the parameters of the transistor which are the most sensitive to aging. The second step of our work was devoted to the establishment of a methodology to create the aging model for the GaN-GIT component. By reproducing the COBRA model presented in the literature, we have succeeded in our work in proposing an innovative approach to integrate the dependencies in temperature and energy suffered by the component during stress (the Tsc pulse duration and the number of pulse suffered Nsc). The last step of our work was dedicated to the physical failure analysis in order to confirm the hypothesis made on the degradation mechanisms obtained after aging of the component. To carry out these analyzes, we started with the de-capsulation of the component by combining the laser cutting with the chemical attacks of the resin constituting the packaging. Once the defect was localized by photoluminescence, an in-depth analysis by SEM scanning and then PFIB (Plasma Focused Ion Beam) scans was performed to determine the mechanism of failure. These were mainly cracks in the Al metal at the drain and the presence of cavities in the metal layer which is used to make the Ohmic contact at the source, which explains the increase in resistance RDSON.

Efficient power conversion is essential to face the continuously increasing energy consumption of our society. GaN based vertical power field effect transistors provide excellent performance figures for power-conversion switches, due to their capability of handling high voltages and current densities with very low area consumption. This work focuses on a vertical trench gate metal oxide semiconductor field effect transistor (MOSFET) with conceptional advantages in a device fabrication preceded GaN epitaxy and enhancement mode characteristics. The functional layer stack comprises from the bottom an n+/n- drift/p body/n+ source GaN layer sequence. Special attention is paid to the Mg doping of the p-GaN body layer, which is a complex topic by itself. Hydrogen passivation of magnesium plays an essential role, since only the active (hydrogen-free) Mg concentration determines the threshold voltage of the MOSFET and the blocking capability of the body diode. Fabrication specific challenges of the concept are related to the complex integration, formation of ohmic contacts to the functional layers, the specific implementation and processing scheme of the gate trench module and the lateral edge termination. The maximum electric field, which was achieved in the pn- junction of the body diode of the MOSFET is estimated to be around 2.1 MV/cm. From double-sweep transfer measurements with relatively small hysteresis, steep subthreshold slope and a threshold voltage of 3 - 4 V a reasonably good Al2O3/GaN interface quality is indicated. In the conductive state a channel mobility of around 80 - 100 cm2/Vs is estimated. This obtained value is comparable to device with additional overgrowth of the channel. Further enhancement of the OFF-state and ON-state characteristics is expected for optimization of the device termination and the high-k/GaN interface of the vertical trench gate, respectively. From the obtained results and dependencies key figures of an area efficient and competitive device design with thick drift layer is extrapolated. Finally, an outlook is given and advancement possibilities as well as technological limits are discussed.:1 Motivation and boundary conditions 1.1 A comparison of competitive semiconductor materials 1.2 Vertical GaN device concepts 1.3 Target application for power switches 2 The vertical GaN MOSFET concept 2.1 Incomplete ionization of dopants 2.2 The pseudo-vertical approach 2.3 Considerations for the device OFF-state 2.3.1 The pn-junction in reverse operation 2.3.2 The gate trench MIS-structure in OFF-state 2.3.3 Dimensional constraints and field plates 2.4 Static ON-state and switching considerations 2.4.1 The pn-junction in forward operation 2.4.2 Resistance contributions 2.4.3 Device model and channel mobility 2.4.4 Threshold voltage and subthreshold slope 2.4.5 Interface and dielectric trap states in wide band semiconductors 2.4.6 The body bias effect 3 Fabrication and characterisation 3.1 Growth methods for GaN substrates and layers 3.2 Substrates and the desired starting material 3.2.1 Physical and micro-structural characterisation 3.2.2 Dislocations and impurities 3.3 Pseudo- and true-vertical MOSFET fabrication 3.3.1 Processing routes 3.3.2 Inductively-coupled plasma etching 3.3.3 Process flow modification 3.4 Electrical characterisation, structures and process control 3.4.1 Current voltage characterisation 3.4.2 C(V) measurements and charge carrier profiling 3.4.3 Cooperative characterisation structures 4 Properties of the functional layers 4.1 Morphology of the MOVPE grown layers 4.2 Hydrogen out-diffusion treatment 4.3 Morphology of the n+-source layer grown by MBE 4.4 N-type doping of the functional layers 4.5 P-type GaN by magnesium doping 4.6 Structural properties after the etching and gate module formation 4.7 Electrical layer characterization 4.7.1 Gate dielectric and interface evaluation 5 Pseudo- and true vertical device operation 5.1 Influences of the metal-line sheet resistance 5.2 Formation and characterisation of ohmic contacts 5.2.1 Ohmic contacts to n-type GaN 5.2.2 Ohmic contacts to p-GaN 5.3 The pn- body diode 5.4 MOSFET operation 5.4.1 ON-state and turn-ON operation 5.4.2 The body bias effect on the threshold voltage 5.4.3 Device OFF-state 6 Summary and conclusion 6.1 Device performance 6.2 Current limits of the vertical device technology 6.3 Possibilities for advancements Bibliography A Appendix A.1 Deduction: Forward diffusion current of the pn-diode A.2 Deduction: Operation regions in the EKV model Figures Tables Abbreviations Symbols Postamble and Acknowledgement

As the world has become more connected meetings have moved online. However, since few have access to studio lighting and uses the embedded webcam the video quality can be far from good. Hence, there is an interest in using a software solution to enhance the video quality in real time. This thesis investigates the feasibility to train a machine learning model to automatically enhance the quality of images. The model must learn without using paired images, since it is difficult to capture images with the exact same content but different quality. Furthermore, the model has to process at least 30 images per second which is a common frequency for videos. Therefore, this thesis investigates the possibility to train a model without paired images and whether such a model can be used in real-time. To answer these questions several sizes of the same model was trained. These were evaluated using six different measures during in order to determine if training without paired data is possible. The models image enhancement capabilities and inference speed were investigated followed by attempts at improving the speed. Finally, different combinations of datasets were investigated to test how well the model generalised to new data. The results show that it is possible to train models for image enhancement without paired data. However, to use such a model in real time a graphics card is needed to reach above 30 images per second.

Eyewitness identification plays an important role during criminal proceedings and may lead to prosecution and conviction of a suspect. One of the methods of eyewitness identification is a police photo lineup when a collection of photographs is presented to the witness in order to identify the perpetrator of the crime. In the lineup, there is typically at most one photograph (typically exactly one) of the suspect and the remaining photographs are the so-called fillers, i.e. photographs of innocent people. Positive identification of the suspect by the witness may result in charge or conviction of the suspect. Assembly of the lineup is a challenging and tedious problem, because the wrong selection of the fillers may end up in a biased lineup, where the suspect will stand out from the fillers and would be easily identifiable even by a highly uncertain witness. The reason why it is tedious is due to the fact that this process is still done manually or only semi-automatically. This thesis tries to solve both issues by proposing a model that will be capable of generating synthetic data, together with an application that will allow users to obtain the fillers for a given suspect's photograph. 1

Les composants de puissance à base de GaN sont connus par l’instabilité de leurs caractéristiques électriques, en particulier la tension de seuil et la résistance à l'état passant. Cela est dû aux effets des mécanismes de piégeage/de-piégeage des charges dans la structure. Le travail présenté dans ce mémoire se compose de deux grandes parties. Dans un premier temps, nous mettons en évidence l’effet d’un certain nombre de paramètres de commutation sur l'évolution de la résistance dynamique avec des cycles de commutation successifs. Nous analysons, en particulier, l’effet de la tension au blocage, la fréquence de commutation et la température sur l'évolution de la résistance dynamique. Dans un deuxième temps, nous présentons les résultats des essais de cyclage de puissance effectués en utilisant 80K de variation de température de jonction sur des puces de puissance MOS-HEMT Al2O3/AlGaN/GaN Normally-On. Ensuite, nous réalisons des caractérisations de pièges, basées sur des analyses de mesures de courants transitoires pendant le processus de vieillissement. Les résultats montrent qu’une dégradation irréversible affecte la tension de seuil avec une dérive vers des valeurs négatives. Ces dérives sont principalement attribuables au piégeage cumulatif avec des cycles de puissance, probablement induits par des électrons chauds, d’une manière progressive et non récupérable. / Power components based on GaN are known by the instability of their electrical characteristics, in particular the threshold voltage and the on-state resistance. This is due to the effects of trapping/de-trapping mechanisms in the structure. The work presented in this memoir consists of two main parts. At the first step, we highlight the effect of a number of switching parameters on the evolution of the dynamic resistance with successive switching cycles. In particular, we analyze the effect of blocking voltage, switching frequency and temperature on the evolution of dynamic resistance. In a second step, we present the results of power cycling tests performed using 80K of junction temperature swing on Normally-ON Al2O3/AlGaN/GaN MOS-HEMTs. Then, we perform trap characterizations, based on the analyses of transient current measurements, during the aging process. The results show that irreversible degradation affects threshold voltage with drift to negative values. These drifts were mainly attributed to cumulative trapping with power cycles, probably induced by hot electrons, in a progressive and non-recoverable way.

No description available.

With increasing usage of credit card payments, credit card fraud has also been increasing. Therefore a fast and accurate fraud detection system is vital for the banks. To solve the problem of fraud detection, different machine learning classifiers have been designed and trained on a credit card transaction dataset. However, the dataset is heavily imbalanced which poses a problem for the performance of the algorithms. To resolve this issue, the generative methods Generative Adversarial Network (GAN), Variational Autoencoders (VAE) and Synthetic Minority Oversampling Technique (SMOTE) have been used to generate synthetic samples for the minority class in order to achieve a more balanced dataset. The main purpose of this study is to evaluate the generative methods and investigate the impact of their generated minority samples on the classifiers. The results from this study indicated that GAN does not outperform the other classifiers as the generated samples from VAE were most effective in three out of five classifiers. Also the validation and histogram of the generated samples indicate that the VAE samples have captured the distribution of the data better than SMOTE and GAN. A suggestion to improve on this work is to perform data engineering on the dataset. For instance, using correlation analysis for the features and analysing which features have the greatest impact on the classification and subsequently dropping the less important features and train the generative methods and classifiers with the trimmed down samples. / Med ökande användning av kreditkort som betalningsmetod i världen, har även kreditkort bedrägeri ökat. Därför finns det behov av ett snabbt och tillförligt system för att upptäcka bedrägliga transkationer. För att lösa problemet med att detektera kreditkort bedrägerier, har olika maskininlärnings klassifiseringsmetoder designats och tränats med ett dataset som innehåller kreditkortstransaktioner. Dock är dessa dataset väldigt obalanserade och innehåller mest normala transaktioner, vilket är problematiskt för systemets noggranhet vid klassificering. Därför har generativa metoderna Generative adversarial networks, Variational autoencoder och Synthetic minority oversampling technique använs för att skapa syntetisk data av minoritetsklassen för att balansera datasetet och uppnå bättre noggranhet. Det centrala målet med denna studie var därmed att evaluera dessa generativa metoder och invetigera påverkan av de syntetiska datapunkterna på klassifiseringsmetoderna. Resultatet av denna studie visade att den generativa metoden generative adversarial networks inte överträffade de andra generativa metoderna då syntetisk data från variational autoencoders var mest effektiv i tre av de fem klassifisieringsmetoderna som testades i denna studie. Dessutom visar valideringsmetoden att variational autoencoder lyckades bäst med att lära sig distributionen av orginal datat bättre än de andra generativa metoderna. Ett förslag för vidare utveckling av denna studie är att jobba med data behandling på datasetet innan datasetet används för träning av algoritmerna. Till exempel kan man använda korrelationsanalys för att analysera vilka features i datasetet har störst påverkan på klassificeringen och därmed radera de minst viktiga och sedan träna algortimerna med data som innehåller färre features.

Fluorescence microscopy is an invaluable technique for studying the intricate process of organism development. The acquisition process, however, is associated with the fundamental trade-off between the quality and reliability of the acquired data. On one hand, the goal of capturing the development in its entirety, often times across multiple spatial and temporal scales, requires extended acquisition periods. On the other hand, high doses of light required for such experiments are harmful for living samples and can introduce non-physiological artifacts in the normal course of development. Conventionally, a single set of acquisition parameters is chosen in the beginning of the acquisition and constitutes the experimenter’s best guess of the overall optimal configuration within the aforementioned trade-off. In the paradigm of adaptive microscopy, in turn, one aims at achieving more efficient photon budget distribution by dynamically adjusting the acquisition parameters to the changing properties of the sample. In this thesis, I explore the principles of adaptive microscopy and propose a range of improvements for two real imaging scenarios. Chapter 2 summarizes the design and implementation of an adaptive pipeline for efficient observation of the asymmetrically dividing neurogenic progenitors in Zebrafish retina. In the described approach the fast and expensive acquisition mode is automatically activated only when the mitotic cells are present in the field of view. The method illustrates the benefits of the adaptive acquisition in the common scenario of the individual events of interest being sparsely distributed throughout the duration of the acquisition. Chapter 3 focuses on computational aspects of segmentation-based adaptive schemes for efficient acquisition of the developing Drosophila pupal wing. Fast sample segmentation is shown to provide a valuable output for the accurate evaluation of the sample morphology and dynamics in real time. This knowledge proves instrumental for adjusting the acquisition parameters to the current properties of the sample and reducing the required photon budget with minimal effects to the quality of the acquired data. Chapter 4 addresses the generation of synthetic training data for learning-based methods in bioimage analysis, making them more practical and accessible for smart microscopy pipelines. State-of-the-art deep learning models trained exclusively on the generated synthetic data are shown to yield powerful predictions when applied to the real microscopy images. In the end, in-depth evaluation of the segmentation quality of both real and synthetic data-based models illustrates the important practical aspects of the approach and outlines the directions for further research.

No description available.

Development of deep learning methods is dependent majorly on availability of large amounts of high quality data. To tackle the problem of data scarcity one of the workarounds is to generate synthetic data using deep learning methods. Especially, when dealing with trajectory data there are added challenges that come in to the picture such as high dependencies of the spatial and temporal component, geographical context sensitivity, privacy laws that protect an individual from being traced back to them based on their mobility patterns etc. This project is an attempt to overcome these challenges by exploring the capabilities of Generative Adversarial Networks (GANs) to generate synthetic trajectories which have characteristics close to the original trajectories. A naive model is designed as a baseline in comparison with a Long Short Term Memorys (LSTMs) based GAN. GANs are generally associated with image data and that is why Convolutional Neural Network (CNN) based GANs are very popular in recent studies. However, in this project an LSTM-based GAN was chosen to work with in order to explore its capabilities and strength of handling long-term dependencies sequential data well. The methods are evaluated using qualitative metrics of visually inspecting the trajectories on a real-world map as well as quantitative metrics by calculating the statistical distance between the underlying data distributions of the original and synthetic trajectories. Results indicate that the baseline method implemented performed better than the GAN model. The baseline model generated trajectories that had feasible spatial and temporal components, whereas the GAN model was able to learn the spatial component of the data well and not the temporal component. Conditional map information could be added as part of training the networks and this can be a research question for future work. / Utveckling av metoder för djupinlärning är till stor del beroende av tillgången på stora mängder data av hög kvalitet. För att ta itu med problemet med databrist är en av lösningarna att generera syntetisk data med hjälp av djupinlärning. Speciellt när man hanterar bana data finns det ytterligare utmaningar som kommer in i bilden såsom starka beroenden av den rumsliga och tidsmässiga komponenten, geografiska känsliga sammanhang, samt integritetslagar som skyddar en individ från att spåras tillbaka till dem baserat på deras mobilitetsmönster etc. Detta projekt är ett försök att överkomma dessa utmaningar genom att utforska kapaciteten hos generativa motståndsnätverk (GAN) för att generera syntetiska banor som har egenskaper nära de ursprungliga banorna. En naiv modell är utformad som en baslinje i jämförelse med en LSTM-baserad GAN. GAN:er är i allmänhet förknippade med bilddata och det är därför som CNN-baserade GAN:er är mycket populära i nya studier. I det här projektet valdes dock en LSTM-baserad GAN att arbeta med för att utforska dess förmåga och styrka att hantera långsiktiga beroenden och sekventiella data på ett bra sätt. Metoderna utvärderas med hjälp av kvalitativa mått för att visuellt inspektera banorna på en verklig världskarta samt kvantitativa mått genom att beräkna det statistiska avståndet mellan de underliggande datafördelningarna för de ursprungliga och syntetiska banorna. Resultaten indikerar att den implementerade baslinjemetoden fungerade bättre än GAN-modellen. Baslinjemodellen genererade banor som hade genomförbara rumsliga och tidsmässiga komponenter, medan GAN-modellen kunde lära sig den rumsliga komponenten av data väl men inte den tidsmässiga komponenten. Villkorskarta skulle kunna läggas till som en del av träningen av nätverken och detta kan vara en forskningsfråga för framtida arbete.