Group III-A nitrides (GaN, AlN, InN and alloys) are materials of considerable contemporary interest and currently enable a wide variety of optoelectronic and high-power, high-frequency electronic applications. All of these applications utilize device structures that employ a single or multiple hetero-junctions, with material compositions varying across the interface. For example, the workhorse of GaN based electronic devices is the high electron mobility transistor (HEMT) which is usually composed of an AlGaN/GaN hetero-junction, where a two-dimensional electron gas (2DEG) is formed due to differences in polarization between the two layers. In addition to such hetero-junctions in the same material family, formation of hetero-interfaces in nitrides begins right from the epitaxy of the very first layer due to the lack of native substrates for their growth. The consequences of such "dissimilar" hetero-junctions typically manifest as large defect densities at this interface which in turn gives rise to defective films. Additionally, if the substrate is also a semiconductor, the electrical properties at such dissimilar semiconductor-nitride hetero-junctions are particularly important in terms of their influence on the performance of nitride devices. Nevertheless, the large defect densities at such dissimilar 3D-3D semiconductor interfaces, which translate into more trap states, also prevents them from being used as active device layers to say nothing of reliability considerations arising because of these defects. Recently, the advent of 2D materials such as graphene and MoS2 has opened up avenues for Van der Waal’s epitaxy of these layered films with practically any other material. Such defect-free integration enables dissimilar semiconductor hetero-junctions to be used as active device layers with carrier transport across the 2D-3D hetero-interface. This thesis deals with hetero-epitaxial growth platforms for reducing defect densities, and the material and electrical properties of dissimilar hetero-junctions with the group III-A nitride material system.

Cílem této práce je vytvořit neuronovou síť která bude schopna rekonstruovat obličeje z fotografií na kterých je část obličeje překrytá maskou. Jsou prezentovány koncepty využívané při vývoji konvolučních neuronových sítí a generativních kompetitivních sítí. Dále jsou popsány koncepty používané v neuronových sítích specificky pro rekonstrukci fotografií obličejů. Je představen model generativní kompetitivní sítě využívající kombinaci hrazených konvolučních vrstev a víceškálových bloků schopný realisticky doplnit oblasti obličeje zakryté maskou.

In der vorliegenden Arbeit werden Möglichkeiten zur Maximierung der Effizienz von stromregelnden DC/DC-Wandlern für den Betrieb von Hochleistungs-LEDs in PKW-und Motorrad-Beleuchtungseinrichtungen untersucht, mit dem Ziel, das Gewicht und den Energieverbrauch der Steuergeräte zu reduzieren und so zu dem stetigen Bestreben der Minimierung der Gesamtfahrzeugemissionen beizutragen. Dafür werden verschiedene, teils sequenziell aufbauende Maßnahmen in Topologie, Bauelementen, Dimensionierung und Betriebsart betrachtet. Eine grundlegende Herausforderung für die Auslegung der Schaltung stellt dabei deren universelle Verwendbarkeit als Gleichteil in einem großen Bereich an Ausgangsstrom und -spannung in den individuellen Scheinwerfersystemen der verschiedenen Fahrzeugderivate dar. Die Grundlage für die Verringerung der Verlustleistung bildet die Vereinfachung der Schaltreglertopologie hinsichtlich des Bauteilaufwands. Dies wird durch die Versorgung der Schaltung aus dem 48 V-Energiebordnetz und die Verwendung der Topologie des Tiefsetzstellers erreicht. Elementarer Anteil dieser Arbeit ist die Untersuchung der Wirksamkeit des Einsatzes neuartiger Galliumnitrid-Leistungsschalter (GaN-HEMTs) anstelle der konventionellen Silizium-MOSFETs, was zunächst an Hand von Berechnungen und schaltungstechnischen, parasitärbehafteten und zeitvarianten Simulationen durchgeführt wird. Bereits bei herkömmlichen Schaltfrequenzen und hartgeschaltetem Betrieb können signifikante Verbesserungen des Wirkungsgrades erreicht werden. Weitergehend wird der Nutzen der durch die GaN-Transistoren ermöglichten höheren Schaltfrequenzen eruiert. Die um bis zu Faktor 20 erhöhte Schaltfrequenz macht den Einsatz einer resonanten Betriebsart (Zero-Voltage-Switching) und einer Luftspule als Hauptinduktivität notwendig. Auf Steuergeräteebene kann somit die Verlustleistung auf unter ein Drittel reduziert werden, was zudem ein deutlich einfacheres und kompakteres Gehäuse ermöglicht, wodurch das Gesamtgewicht etwa halbiert werden kann. Abschließend wird die Schaltung in einem Prototypen praktisch umgesetzt und die Funktionsfähigkeit im ZVS-Betrieb bei Schaltfrequenzen von bis zu 10 MHz verifiziert. / This thesis deals with the research of possibilities for maximising efficiency of current-regulating DC/DC-Converters for driving high-power-LEDs in passenger-car- and motorcycle-lighting-devices. The ambition is to reduce weight and energy-consumption of the electronic-control-units, to contribute to reach the continuously decreasing target-values for vehicle-emissions. Therefor different approaches in topology, components, design and operating mode are considered. A key-challenge for the circuit-design is the common-part-strategy for usage in many individual vehicle-headlamp-systems with a wide range of output-current and LED-string-voltages. Basis for the reduction of power-losses is the simplification of the converters topology in terms of quantity of components. This is achieved by using the 48 V -vehicle-electrical-system as voltage-supply and a step-down-topology. Mainpart of this research is about the potential benefits of applying novel Galliumnitride High-electron-mobility-transistors (GaN-HEMTs) instead of silicon MOS-FETs. Initially this is done by calculations and parasitic-afflicted, timevariant circuit-simulations. Already in hardswitching operation under conventional switching-frequencies significant improvements in converter-efficiency can be achieved. Furthermore the advantages of higher switching-frequencies, offered by the GaN-transistors, are investigated. Up to 20 times higher switching-frequencies necessitate a resonant operating mode of the circuit (Zero-voltage-switching) and the use of an aircoil as main-inductor. On ECU-level power-losses can be reduced down to less than one third, which enables a more simplified and compact housing-concept, so that the overall weight can drop to about the half. Finally the designed circuit is build up in a prototype and the functional capability is verified in ZVS-mode with up to 10 MHz switching-frequency.

In the thesis the development of two equipment for preparation of ultrathin films under ultrahign vacuum conditions (UHV) is discussed. Here, additionally to a brief description of theoretical principles, more details on the design of these units are given. In the first part the design of a thermal source of oxygen or hydrogen atomic beams is discussed. Further, a design and construction of an ion–atomic beam source for ion-beam assisted deposition of thin films is detailed. The source combines the principles of an efusion cell and electron-impact ion beam source generating ions of (30 – 100) eV energy. The source has been successfully applied for the growth of GaN on the Si(111) 7x7 substrate under room temperature.

Data augmentation is a technique that can be performed in various ways to improve the training of discriminative models. The recent developments in deep generative models offer new ways of augmenting existing data sets. In this thesis, a framework for augmenting annotated data sets with deep generative models is proposed together with a method for quantitatively evaluating the quality of the generated data sets. Using this framework, two data sets for pupil localization was generated with different generative models, including both well-established models and a novel model proposed for this purpose. The unique model was shown both qualitatively and quantitatively to generate the best data sets. A set of smaller experiments on standard data sets also revealed cases where this generative model could improve the performance of an existing discriminative model. The results indicate that generative models can be used to augment or replace existing data sets when training discriminative models. / Dataaugmentering är en teknik som kan utföras på flera sätt för att förbättra träningen av diskriminativa modeller. De senaste framgångarna inom djupa generativa modeller har öppnat upp nya sätt att augmentera existerande dataset. I detta arbete har ett ramverk för augmentering av annoterade dataset med hjälp av djupa generativa modeller föreslagits. Utöver detta så har en metod för kvantitativ evaulering av kvaliteten hos genererade data set tagits fram. Med hjälp av detta ramverk har två dataset för pupillokalisering genererats med olika generativa modeller. Både väletablerade modeller och en ny modell utvecklad för detta syfte har testats. Den unika modellen visades både kvalitativt och kvantitativt att den genererade de bästa dataseten. Ett antal mindre experiment på standardiserade dataset visade exempel på fall där denna generativa modell kunde förbättra prestandan hos en existerande diskriminativ modell. Resultaten indikerar att generativa modeller kan användas för att augmentera eller ersätta existerande dataset vid träning av diskriminativa modeller.

No description available.

Despite enormous efforts and investments, the efficiency of InGaN-based green and yellow-green light emitters remains relatively low, and that limits progress in developing full color display, laser diodes, and bright light sources for general lighting. The low efficiency of light emitting devices in the green-to-yellow spectral range, also known as the “Green Gap”, is considered a global concern in the LED industry. The polar c-plane orientation of GaN, which is the mainstay in the LED industry, suffers from polarization-induced separation of electrons and hole wavefunctions (also known as the “quantum confined Stark effect”) and low indium incorporation efficiency that are the two main factors that contribute to the Green Gap phenomenon. One possible approach that holds promise for a new generation of green and yellow light emitting devices with higher efficiency is the deployment of nonpolar and semi-polar crystallographic orientations of GaN to eliminate or mitigate polarization fields. In theory, the use of other GaN planes for light emitters could also enhance the efficiency of indium incorporation compared to c-plane. In this thesis, I present a systematic exploration of the suitable GaN orientation for future lighting technologies. First, in order to lay the groundwork for further studies, it is important to discuss the analysis of processes limiting LED efficiency and some novel designs of active regions to overcome these limitations. Afterwards, the choice of nonpolar orientations as an alternative is discussed. For nonpolar orientation, the (1-100)-oriented (m-plane) structures on patterned Si (112) and freestanding m-GaN are studied. The semi-polar orientations having substantially reduced polarization field are found to be more promising for light-emitting diodes (LEDs) owing to high indium incorporation efficiency predicted by theoretical studies. Thus, the semi-polar orientations are given close attention as alternatives for future LED technology. One of the obstacles impeding the development of this technology is the lack of suitable substrates for high quality materials having semi-polar and nonpolar orientations. Even though the growth of free-standing GaN substrates (homoepitaxy) could produce material of reasonable quality, the native nonpolar and semi-polar substrates are very expensive and small in size. On the other hand, GaN growth of semi-polar and nonpolar orientations on inexpensive, large-size foreign substrates (heteroepitaxy), including silicon (Si) and sapphire (Al2O3), usually leads to high density of extended defects (dislocations and stacking faults). Therefore, it is imperative to explore approaches that allow the reduction of defect density in the semi-polar GaN layers grown on foreign substrates. In the presented work, I develop a cost-effective preparation technique of high performance light emitting structures (GaN-on-Si, and GaN-on-Sapphire technologies). Based on theoretical calculations predicting the maximum indium incorporation efficiency at θ ~ 62º (θ being the tilt angle of the orientation with respect to c-plane), I investigate (11-22) and (1-101) semi-polar orientations featured by θ = 58º and θ = 62º, respectively, as promising candidates for green emitters. The (11-22)-oriented GaN layers are grown on planar m-plane sapphire, while the semi-polar (1-101) GaN are grown on patterned Si (001). The in-situ epitaxial lateral overgrowth techniques using SiNx nanoporous interlayers are utilized to improve the crystal quality of the layers. The data indicates the improvement of photoluminescence intensity by a factor of 5, as well as the improvement carrier lifetime by up to 85% by employing the in-situ ELO technique. The electronic and optoelectronic properties of these nonpolar and semi-polar planes include excitonic recombination dynamics, optical anisotropy, exciton localization, indium incorporation efficiency, defect-related optical activities, and some challenges associated with these new technologies are discussed. A polarized emission from GaN quantum wells (with a degree of polarization close to 58%) with low non-radiative components is demonstrated for semi-polar (1-101) structure grown on patterned Si (001). We also demonstrated that indium incorporation efficiency is around 20% higher for the semi-polar (11-22) InGaN quantum wells compared to its c-plane counterpart. The spatially resolved cathodoluminescence spectroscopy demonstrates the uniform distribution of indium in the growth plane. The uniformity of indium is also supported by the relatively low exciton localization energy of Eloc = 7meV at 15 K for these semi-polar (11-22) InGaN quantum wells compared to several other literature reports on c-plane. The excitons are observed to undergo radiative recombination in the quantum wells in basal-plane stacking faults at room temperature. The wurtzite/zincblende electronic band-alignment of BSFs is proven to be of type II using the time-resolved differential transmission (TRDT) method. The knowledge of band alignment and degree of carrier localization in BSFs are extremely important for evaluating their effects on device properties. Future research for better understanding and potential developments of the semi-polar LEDs is pointed out at the end.

Moving parts cause vibrations and tend to wear out. In applications where maintenance is complicated, solutions without moving parts are therefore advantageous. Electrohydrodynamic pumps are such a solution. Instead of mechanical propulsion, they use strong electric fields to induce movement in a dielectric cooling liquid. These pumps require very little power, but to generate sufficiently strong electric fields, they need to be fed with very high voltage.  This project explored various methods for designing DC/DC-converters which fulfil the demands of an electrohydrodynamic pump. This was done by altering and combining existing topologies that were deemed to be relevant. The main method for testing and evaluation was by simulating in LTspice. The project also briefly investigated methods of overcurrent protection. This was relevant because gas bubbles in the cooling fluid can cause electric arcs which damage the pumps. Three converter topologies were chosen for further evaluation. First, a conventional resonant Royer-based converter that has previously been used by APR Technologies which was altered by the inclusion of a feedback loop. Second, a high-frequency resonant Royer-based converter with a planar air-core transformer. Third, a transformerless converter with a switched boost converter IC. All circuits included a Cockroft-Walton voltage multiplier bridge. The two resonant Royer-based converters fulfilled all requirements except the one on efficiency, while the transformerless converter fulfilled all requirements except the one on cost, set by APR. The more expensive transformerless converter had a significantly higher efficiency and a wider range of acceptable input voltages. Furthermore three general conclusions were drawn. The first was that planar air-core transformers are not beneficial compared to conventional transformers in these type of applications. The second was that a discrete voltage regulator controlled by feedback from the output is more effective than using a voltage regulator without feedback, as it also eliminates temperature and load variations. The third conclusion was that to protect the circuits from overcurrent, a large series resistor is needed, which causes significantly lowered efficiency.

Ces dernières années, le domaine de l’apprentissage profond a connu des progrès énormes dans des applications allant de la génération d’images, détection d’objets, modélisation du langage à la réponse aux questions visuelles. Les approches classiques telles que l’apprentissage supervisé nécessitent de grandes quantités de données étiquetées et spécifiques à la tâches. Cependant, celles-ci sont parfois coûteuses, peu pratiques, ou trop longues à collecter. La modélisation efficace en données, qui comprend des techniques comme l’apprentissage few-shot (à partir de peu d’exemples) et l’apprentissage self-supervised (auto-supervisé), tentent de remédier au manque de données spécifiques à la tâche en exploitant de grandes quantités de données plus “générales”. Les progrès de l’apprentissage profond, et en particulier de l’apprentissage few-shot, s’appuient sur les benchmarks (suites d’évaluation), les métriques d’évaluation et les jeux de données, car ceux-ci sont utilisés pour tester et départager différentes méthodes sur des tâches précises, et identifier l’état de l’art. Cependant, du fait qu’il s’agit de versions idéalisées de la tâche à résoudre, les benchmarks sont rarement équivalents à la tâche originelle, et peuvent avoir plusieurs limitations qui entravent leur rôle de sélection des directions de recherche les plus prometteuses. De plus, la définition de métriques d’évaluation pertinentes peut être difficile, en particulier dans le cas de sorties structurées et en haute dimension, telles que des images, de l’audio, de la parole ou encore du texte. Cette thèse discute des limites et des perspectives des benchmarks existants, des fonctions de coût (training losses) et des métriques d’évaluation (evaluation metrics), en mettant l’accent sur la modélisation générative - les Réseaux Antagonistes Génératifs (GANs) en particulier - et la modélisation efficace des données, qui comprend l’apprentissage few-shot et self-supervised. La première contribution est une discussion de la tâche de modélisation générative, suivie d’une exploration des propriétés théoriques et empiriques des fonctions de coût des GANs. La deuxième contribution est une discussion sur la limitation des few-shot classification benchmarks, certains ne nécessitant pas de généralisation à de nouvelles sémantiques de classe pour être résolus, et la proposition d’une méthode de base pour les résoudre sans étiquettes en phase de testing. La troisième contribution est une revue sur les méthodes few-shot et self-supervised de détection d’objets , qui souligne les limites et directions de recherche prometteuses. Enfin, la quatrième contribution est une méthode efficace en données pour la description de vidéo qui exploite des jeux de données texte et vidéo non supervisés. / In recent years, the field of deep learning has seen tremendous progress for applications ranging from image generation, object detection, language modeling, to visual question answering. Classic approaches such as supervised learning require large amounts of task-specific and labeled data, which may be too expensive, time-consuming, or impractical to collect. Data-efficient methods, such as few-shot and self-supervised learning, attempt to deal with the limited availability of task-specific data by leveraging large amounts of general data. Progress in deep learning, and in particular, few-shot learning, is largely driven by the relevant benchmarks, evaluation metrics, and datasets. They are used to test and compare different methods on a given task, and determine the state-of-the-art. However, due to being idealized versions of the task to solve, benchmarks are rarely equivalent to the original task, and can have several limitations which hinder their role of identifying the most promising research directions. Moreover, defining meaningful evaluation metrics can be challenging, especially in the case of high-dimensional and structured outputs, such as images, audio, speech, or text. This thesis discusses the limitations and perspectives of existing benchmarks, training losses, and evaluation metrics, with a focus on generative modeling—Generative Adversarial Networks (GANs) in particular—and data-efficient modeling, which includes few-shot and self-supervised learning. The first contribution is a discussion of the generative modeling task, followed by an exploration of theoretical and empirical properties of the GAN loss. The second contribution is a discussion of a limitation of few-shot classification benchmarks, which is that they may not require class semantic generalization to be solved, and the proposal of a baseline method for solving them without test-time labels. The third contribution is a survey of few-shot and self-supervised object detection, which points out the limitations and promising future research for the field. Finally, the fourth contribution is a data-efficient method for video captioning, which leverages unsupervised text and video datasets, and explores several multimodal pretraining strategies.