Heterojunction bipolar transistors (HBT) based on SiGeC havebeen investigated. Two high-frequency architectures have beendesigned, fabricated and characterized. Different collectordesigns were applied either by using selective epitaxial growthdoped with phosphorous or by non-selective epitaxial growthdoped with arsenic. Both designs have a non-selectivelydeposited SiGeC base doped with boron and a poly-crystallineemitter doped with phosphorous. Selective epitaxial growth of the collector layer has beendeveloped by using a reduced pressure chemical vapor deposition(RPCVD) technique. The incorporation of phosphorous and defectformation during selective deposition of these layers has beenstudied. A major problem of phosphorous-doping during selectiveepitaxy is segregation. Different methods, e.g. chemical orthermal oxidation, are shown to efficiently remove thesegregated dopants. Chemical-mechanical polishing (CMP) hasalso been used as an alternative to solve this problem. The CMPstep was successfully integrated in the HBT process flow. Epitaxial growth of Si1-x-yGexCy layers for base layerapplications in bipolar transistors has been investigated indetail. The optimization of the growth parameters has beenperformed in order to incorporate carbon substitutionally inthe SiGe matrix without increasing the defect density in theepitaxial layers. The thermal stability of npn SiGe-based heterojunctionstructures has been investigated. The influence of thediffusion of dopants in SiGe or in adjacent layers on thethermal stability of the structure has also been discussed. SiGeC-based transistors with both non-selectively depositedcollector and selectively grown collector have been fabricatedand electrically characterized. The fabricated transistorsexhibit electrostatic current gain values in the range of 1000-2000. The cut-off frequency and maximum oscillation frequencyvary from 40-80 GHz and 15-30 GHz, respectively, depending onthe lateral design. The leakage current was investigated usinga selectively deposited collector design and possible causesfor leakage has been discussed. Solutions for decreasing thejunction leakage are proposed. <b>Key words:</b>Silicon-Germanium-Carbon (SiGeC),Heterojunction bipolar transistor (HBT), chemical vapordeposition (CVD), selective epitaxy, non-selective epitaxy,collector design, high-frequency measurement, dopantsegregation, thermal stability.

<p>Heterojunction bipolar transistors (HBT) based on SiGeC havebeen investigated. Two high-frequency architectures have beendesigned, fabricated and characterized. Different collectordesigns were applied either by using selective epitaxial growthdoped with phosphorous or by non-selective epitaxial growthdoped with arsenic. Both designs have a non-selectivelydeposited SiGeC base doped with boron and a poly-crystallineemitter doped with phosphorous.</p><p>Selective epitaxial growth of the collector layer has beendeveloped by using a reduced pressure chemical vapor deposition(RPCVD) technique. The incorporation of phosphorous and defectformation during selective deposition of these layers has beenstudied. A major problem of phosphorous-doping during selectiveepitaxy is segregation. Different methods, e.g. chemical orthermal oxidation, are shown to efficiently remove thesegregated dopants. Chemical-mechanical polishing (CMP) hasalso been used as an alternative to solve this problem. The CMPstep was successfully integrated in the HBT process flow.</p><p>Epitaxial growth of Si1-x-yGexCy layers for base layerapplications in bipolar transistors has been investigated indetail. The optimization of the growth parameters has beenperformed in order to incorporate carbon substitutionally inthe SiGe matrix without increasing the defect density in theepitaxial layers.</p><p>The thermal stability of npn SiGe-based heterojunctionstructures has been investigated. The influence of thediffusion of dopants in SiGe or in adjacent layers on thethermal stability of the structure has also been discussed.</p><p>SiGeC-based transistors with both non-selectively depositedcollector and selectively grown collector have been fabricatedand electrically characterized. The fabricated transistorsexhibit electrostatic current gain values in the range of 1000-2000. The cut-off frequency and maximum oscillation frequencyvary from 40-80 GHz and 15-30 GHz, respectively, depending onthe lateral design. The leakage current was investigated usinga selectively deposited collector design and possible causesfor leakage has been discussed. Solutions for decreasing thejunction leakage are proposed.</p><p><b>Key words:</b>Silicon-Germanium-Carbon (SiGeC),Heterojunction bipolar transistor (HBT), chemical vapordeposition (CVD), selective epitaxy, non-selective epitaxy,collector design, high-frequency measurement, dopantsegregation, thermal stability.</p>

Les communications sans fil sont de plus en plus présentes dans notre quotidien, et avec elles, les ondes électromagnétiques qui leurs sont associées. Cela a créé un besoin de produits permettant de contrôler la portée du réseau, pour des raisons de sécurité de données, d'augmentation de débit en cas d'interférences avec d'autres réseaux, ou bien pour des raisons sanitaires. A l'inverse, dans d'autres cas, l'amélioration de la transmission de ces ondes est recherchée. Par exemple, dans les bâtiments de nouvelle construction ou rénovées, l'installation de vitrages à isolation renforcée permet d'avoir des bâtiments plus performantes au niveau thermique. Malheureusement, ces vitrages bloquent également les radiofréquences, dont notamment les signaux de téléphonie mobile. L'objectif de la thèse est la conception et le développement de structures filtrantes imprimées sur substrat souple. Dans un premier temps, des solutions de filtrage sélectif coupe bande sont proposées afin d'empêcher autant que possible la transmission d'ondes électromagnétiques, par exemple celles du WiFi entre les pièces d'un bâtiment. Dans un deuxième temps, une solution permettant d'améliorer la transmission des ondes électromagnétiques au travers des vitrages thermiques est proposée. Outre ces études, un système de caractérisation permettant de caractériser finement ces structures FSS a été développé lors de ces travaux de thèse. / Nowadays wireless communication systems are more and more present in our lives, and with them, electromagnetic waves associated to them. A need of products capable to control the range of the network has appeared, for data security reasons, for data rate increase in case of network interferences or for health reasons. On the opposite, in other cases, improving the transmission of these waves is desired. For instance, in new construction or renovated buildings, energy saving windows allows having more efficient buildings thermal speaking. Unfortunately, those windows also block radiofrequencies, including mobile phone signals. The aim of this thesis is the design and development of filtering structures printed on flexible substrate. First, stop band selective filtering solutions are proposed in order to block as much as possible transmission waves, for instance, WiFi in between different rooms of the same building. Secondly, a solution to improve electromagnetic wave transmission through energy saving windows is proposed. In addition to these studies, a characterization system for testing these FSS structures has been developed during this thesis work.

<b>Background:</b> Systemic chemotherapy and radiotherapy play an important role in the treatment of colorectal cancer. Tumour perfusion and oxygenation is known to influence radiosensitivity and chemosensitivity. In this thesis, I propose that the evaluation of changes in tumour perfusion using perfusion CT (pCT) and dynamic contrast-enhanced (Dce) MRI can guide the rational sequencing of drugs and radiation. <b>Methods:</b> Dce-MRI and pCT scans were incorporated into a clinical trial of hypofractionated pelvic radiotherapy and nelfinavir in 10 patients with rectal cancer. Toxicity and tissue biomarkers (tumour cell density, microvessel density, CAIX, HIF1-alpha, phospho-Akt and phospho-PRAS40) were evaluated. pCT liver scans were incorporated into an imaging study in patients with colorectal liver metastases randomised to receive either oxaliplatin/ 5FU chemotherapy or oxaliplatin/ 5FU chemotherapy plus selective internal radiotherapy. <b>Results:</b> After 7 days of nelfinavir concurrent with hypo-fractionated pelvic radiotherapy, there was a mean 42&percnt; increase in median K^trans (P=0.03, paired t test) on Dce-MRI and a median 30&percnt; increase in mean blood flow on pCT (P=0.028, Wilcoxon Rank Sum), although no statistically significant changes in perfusion parameters were demonstrated after 7 days of nelfinavir prior to radiotherapy. The feasibility of evaluating tumour cell density in rectal biopsies before and after radiotherapy and a radiosensitising drug as an early endpoint of response was demonstrated. In patients with colorectal liver metastases who received oxaliplatin and modified de Gramont chemotherapy alone, after 4 cycles of chemotherapy, a 28&percnt; decrease in the mean hepatic arterial fraction was observed (P=0.018, paired t test). Between pCT scans 2 days before SIRT and 39-47 days following SIRT and continued 2-weekly chemotherapy, there was a mean 62&percnt; (P=0.009) reduction in Blood Flow and 61&percnt; (P=0.006) reduction in Blood Volume (paired t test). <b>Conclusions</b> This research does not support the hypothesis that nelfinavir before radiotherapy improves blood flow to human rectal cancer. Increases in rectal tumour perfusion during radiotherapy and concurrent nelfinavir are likely to be primarily explained by the acute biological effects of radiation. Four or more cycles of oxaliplatin and modified de Gramont chemotherapy may result in changes in tumour perfusion of colorectal liver metastases which would be detrimental to subsequent radiotherapy. Selective internal radiotherapy resulted in substantial reductions in tumour perfusion 39-47 days after the treatment. Perfusion imaging can be used to detect changes in tumour perfusion in response to radiotherapy and systemic therapy which have implications for the sequencing of therapies.

Selective laser melting (SLM) of metallic powders is an additive manufacturing technique that is widely employed to produce 3D components, and is fast becoming an important method for manufacturing near-net shape and complex metallic parts. In this thesis, a comprehensive investigation on the effect of SLM on the mechanical and corrosion properties of the Al-12Si (AS), 316L stainless steel (SS), and 18(Ni)-300 grade managing steel (MS) is investigated, with particular emphasis on the developing (micro- as well as mesa-)structure -property correlations. Detailed microstructural characterization combined with quasi-static tensile, fracture toughness, fatigue crack growth, and unmatched fatigue tests were conducted. The effect of post-SLM heat treatment as well as the scanning strategy (linear vs. checker board hatch style) was examined and the results are compared with those of conventionally manufactured (CM) counterparts. The SLM alloys exhibit a mesostructured, in addition to the fine cellular structure along the boundaries. In a case of SLM-AS, the fine cellular structure imparts higher strength at the cost of ductility, while the mesostructured, which arises due to the laser track hatching, causes the crack path to be tortuous, and in turn leads to substantial increase in fracture toughness. This imparts significant anisotropy to the toughness while tensile properties are nearly-isotropic. The experimental results of SLM-SS also show that higher tensile strengths properties with a marked reduction ductility. In spite of these, the fracture toughness, which ranges between 63 and 87 MPa.m0.5, of the SLM-SS is good, which is a result of the mesostructured induced crack tortuousity.Both tensile and toughness properties of SLM-SS were found to be anisotropic in nature. Upon aging SLM-MS, nanoscale precipitation of intermetallic compounds occurs within the cells that, in turn, lead in marked improvements in tensile strengths properties, but substantial reductions in ductility and fracture toughness. Overall, the mechanical performance, except ductility, of the SLM-MS after aging is found to be similar to that of CM-MS. Importantly, the lack of ductility does not lead to a reduction in toughness. Although the SLM-MS alloy possesses a mesostructured, no significant anisotropy in the mechanical behaviour is observed. The unnoticed studies on SLM-AS, -SS, and -MS reveal that the tensile residual stresses, gas-pores, and unmelted powder particles, can degrade the unmatched highest fatigue properties considerably and hence need be eliminated for high fatigue strength. Room temperature, electrochemical corrosion resistances (CRs) of SLM-AS, -SS and -MS in 0.1M NaCl solution were also evaluated and compared with those CM counterparts. While SLM improves CRs of AS and SS, it degrades that of MS. The results are discussed in terms of microstructural refinement and porosity that are common in SLM alloys.

This thesis aims to investigate the use of additive manufacturing (AM) as a novel manufacturing process for the production of milli-scale chemical reaction systems. Five well developed additive manufacturing techniques; stereolithography (SL), selective laser melting (SLM), fused deposition modelling (FDM), ultrasonic additive manufacture (UAM) and selective laser sintering (SLS) were used to manufacture a number of miniaturised flow devices which were tested using a range of organic and inorganic reactions. SL was used to manufacture a range of functioning milli-scale flow devices from Accura 60 photoresin, with both simple and complex internal channel networks. These devices were used to perform a range of organic and inorganic reactions, including aldehyde and ketone functional group interconversions. Conversion of products within these reactors, were shown to be comparable to commercially available milli-scale coil reactors. More complex designs, which allowed SL parts to be integrated to existing flow and analytical instrumentation, allowed us to develop an automated reaction analysis and optimisation platform. This platform allowed precise control over the reaction conditions, including flow rate, temperature and reagent composition. We also designed a simplex type reaction optimisation software package that could input data in the form of reaction conversions, peak intensities, and thermocouple data, and generate a new set of optimal reaction conditions. SL parts which incorporated embedded analytical components were also manufactured, which allowed us to perform inline reaction analysis as a feedback method for input into the optimisation platform. Stereolithography was shown to be a highly versatile manufacturing method for designing and producing these flow devices, however the process was shown to be still limited by the range of processable materials currently commercially available. SLM was also used to manufacture a number of functioning milli-scale flow devices from stainless steel and titanium, which had simplistic internal channel designs of diameters ranging from 1 to 3 mm. Again, SLM parts were manufactured which incorporated embedded analytical components, which could be integrated into an automated reaction platform. These devices, unlike parts produced via SL, could be attached to heating platforms to allow us to perform high temperature reactions. This control over the reaction temperature formed an essential part of the reaction optimisation platform. These parts were again used to perform a ketone functional group interconversion. Internal structures of these SLM parts were also visualised via micro computed tomography (μCT or microCT) scanning as well as optical microscopy. FDM was used throughout the project as an inexpensive method of prototyping parts which were to be manufactured via more expensive manufacturing processes. This prototyping allowed the optimisation of intricate design features, such as the manufacture of an inline spectroscopic flow cell for integration with a commercially available LC system. FDM was also proposed as a customisable approach to designing and manufacturing flow devices with embedded components, however the current limitations in build resolution and materials choices severely limited the use of FDM for this application. UAM was also proposed as a novel manufacturing process whereby the build process would allow discrete components to be embedded directly into a flow channel. This was demonstrated by embedding a type-k thermocouple across a 2 mm channel. The data from this thermocouple was monitored during a heated reaction, and used as a method of determining the exact reaction conditions the reaction medium was being exposed to. SLS was also proposed as a possible manufacturing method for milli-scale flow devices, however it proved difficult to remove un-sintered powder from parts with internal channel diameters as high as 5 mm. It was shown that this powder was forming a dense semi solid, due to the large degree of shrinkage upon cooling of the SLS parts, which was compressing the powder. More research into optimum processing conditions is required before SLS could be used for the production of intricate channel networks.

The thesis deals with optimization of the manufacturing process of the muzzle device designed for assault rifle. The most common titanium alloy named Ti-6Al-4V was chosen for this task. The introduction summarizes previously existing types of muzzle devices and further describes the SLM technology with a special focus on titanium alloys processing. The optimization methods and their follow-up testing were designed according to theoretical knowledge that is summarized in the theoretical part of this work. Firstly, the aim was to describe the optimization of the manufacturing process with attention to preserving the relative density of the parts. Secondly, the mechanical properties of the parts that underwent different heat treatment were tested.The obtained data were then used to design and manufacture a muzzle device that underwent further testing in real condition afterwards.

This diploma thesis describes the design of filling element with trabecular structure for revision implant of total knee arthroplasty. Design of the filling element is created by the digital data of the patient tissue. Production of a functional sample is performed using additive technology Selective Laser Melting. This work also include analyze of the accuracy of this technology for the production of titanium alloy Ti6Al4V.

Thermal radiation is a fundamental heat transfer process, with certain basic aspects still not fully understood. Furthermore, tailoring its properties has potential to affect a wide range of applications, particularly thermophotovoltaics (TPV) and radiative cooling. TPV converts heat into electricity using thermal radiation to illuminate a photovoltaic diode, with no moving parts. With its realistic efficiency limit up to 50% (heat source at 1200 ^oC), TPV has garnered substantial interest. However, state-of-the-art TPV demonstrations are still well below theoretical limits, because of losses from generating and efficiently converting or recycling thermal radiation. In this thesis, tailored integrated photonic crystal structures are numerically simulated to enhance the efficiency of solar TPV. Next, a high-temperature thin-film Si-based selective absorber and emitter is designed, fabricated and experimentally characterized. It exhibits great potential to open up new applications, as it lends itself to large-scale production with substantial mechanical flexibility and excellent spectral selectivity for extended time periods, even when operating under high operating temperatures (600 ^oC) for up to 6 hours, with partial degradation after 24 hours. To perform this high-temperature characterization, an emittance measurement setup has been built; its performance agrees well with numerical simulations. Second, a unique passive cooling mechanism known as radiative cooling is developed to reduce the operating temperature of the photovoltaic diode. The significant effect of radiative cooling as a complement for an all-passive-cooling TPV system is proposed and numerically analyzed under a range of conditions. Furthermore, an outdoor experiment has been performed to demonstrate the effect of radiative cooling on a concentrating photovoltaic system, which can potentially be applied to the thermal management of a TPV system. In summary, this work paves the way towards the development of reliable, quiet, lightweight, and sustainable TPV and radiatively cooled power sources for outdoor applications.

In the last decade, the machine learning universe embraced deep neural networks (DNNs) wholeheartedly with the advent of neural architectures such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), transformers, etc. These models have empowered many applications, such as ChatGPT, Imagen, etc., and have achieved state-of-the-art (SOTA) performance on many vision, speech, and language modeling tasks. However, SOTA performance comes with various issues, such as large model size, compute-intensive training, increased inference latency, higher working memory, etc. This thesis aims at improving the resource efficiency of neural architectures, i.e., significantly reducing the computational, storage, and energy consumption of a DNN without any significant loss in performance. Towards this goal, we explore novel neural architectures as well as training algorithms that allow low-capacity models to achieve near SOTA performance. We divide this thesis into two dimensions: \textit{Efficient Low Complexity Models}, and \textit{Input Hardness Adaptive Models}. Along the first dimension, i.e., \textit{Efficient Low Complexity Models}, we improve DNN performance by addressing instabilities in the existing architectures and training methods. We propose novel neural architectures inspired by ordinary differential equations (ODEs) to reinforce input signals and attend to salient feature regions. In addition, we show that carefully designed training schemes improve the performance of existing neural networks. We divide this exploration into two parts: \textsc{(a) Efficient Low Complexity RNNs.} We improve RNN resource efficiency by addressing poor gradients, noise amplifications, and BPTT training issues. First, we improve RNNs by solving ODEs that eliminate vanishing and exploding gradients during the training. To do so, we present Incremental Recurrent Neural Networks (iRNNs) that keep track of increments in the equilibrium surface. Next, we propose Time Adaptive RNNs that mitigate the noise propagation issue in RNNs by modulating the time constants in the ODE-based transition function. We empirically demonstrate the superiority of ODE-based neural architectures over existing RNNs. Finally, we propose Forward Propagation Through Time (FPTT) algorithm for training RNNs. We show that FPTT yields significant gains compared to the more conventional Backward Propagation Through Time (BPTT) scheme. \textsc{(b) Efficient Low Complexity CNNs.} Next, we improve CNN architectures by reducing their resource usage. They require greater depth to generate high-level features, resulting in computationally expensive models. We design a novel residual block, the Global layer, that constrains the input and output features by approximately solving partial differential equations (PDEs). It yields better receptive fields than traditional convolutional blocks and thus results in shallower networks. Further, we reduce the model footprint by enforcing a novel inductive bias that formulates the output of a residual block as a spatial interpolation between high-compute anchor pixels and low-compute cheaper pixels. This results in spatially interpolated convolutional blocks (SI-CNNs) that have better compute and performance trade-offs. Finally, we propose an algorithm that enforces various distributional constraints during training in order to achieve better generalization. We refer to this scheme as distributionally constrained learning (DCL). In the second dimension, i.e., \textit{Input Hardness Adaptive Models}, we introduce the notion of the hardness of any input relative to any architecture. In the first dimension, a neural network allocates the same resources, such as compute, storage, and working memory, for all the inputs. It inherently assumes that all examples are equally hard for a model. In this dimension, we challenge this assumption using input hardness as our reasoning that some inputs are relatively easy for a network to predict compared to others. Input hardness enables us to create selective classifiers wherein a low-capacity network handles simple inputs while abstaining from a prediction on the complex inputs. Next, we create hybrid models that route the hard inputs from the low-capacity abstaining network to a high-capacity expert model. We design various architectures that adhere to this hybrid inference style. Further, input hardness enables us to selectively distill the knowledge of a high-capacity model into a low-capacity model by cleverly discarding hard inputs during the distillation procedure. Finally, we conclude this thesis by sketching out various interesting future research directions that emerge as an extension of different ideas explored in this work.