This thesis presents an innovative approach to enhance the stability of floating offshore wind turbine (FOWT) platform through mooring actuation. First, an OC3- Hywind spar-buoy floating platform is modeled utilizing the Control-oriented, Reconfigurable, and Acausal Floating Turbine Simulator (CRAFTS) with a specific focus on predicting hydrodynamic and mooring line loads while intentionally excluding consideration of aerodynamic forces. The accuracy of this model is validated against the industry standard OpenFAST simulator through various test cases. The central objective of this study revolves around achieving robust stabilization of the spar buoy platform, primarily focusing on X-Z symmetric planar motions, including surge, pitch, and heave degrees of freedom (DOFs). To accomplish this, two linearization techniques are employed: one transforms the inherently complex nonlinear model from CRAFTS into a linear Mass-Spring-Damper (MSD) system, particularly targeting surge and pitch motions, while the other method involves the conversion of the nonlinear model from CRAFTS into the Functional Mockup Interface (FMI) within MATLAB/Simulink for linearization. The analysis utilizing Bode plots derived from these lin- earized models yields crucial insights into the system's response to mooring actuation. Notably, it emphasizes the inherent challenge in pitch control, characterized by lower gain compared to surge at relevant frequencies, necessitating substantial mooring actuation or cable length modifications for effective pitch stabilization. Then, a Linear Quadratic Regulator (LQR) controller is designed to mitigate surge and pitch motions. Numerical simulations conducted across diverse scenarios reveal the inherent challenge in simultaneously mitigating surge and pitch motions using the original platform configuration. To address this challenge, a control co-design strategy is proposed, leading to the development of an optimized mooring line configuration that effectively stabilizes both motions with minimal adjustments. In summary, this thesis introduces a control-oriented modeling approach and an innovative control strategy to enhance the stability of the floating wind turbine platform through mooring actuation. The results emphasize the potential for broader application of this approach to various floating platforms for FOWTs and the extension of stabilization efforts to address all six DOFs in future research, where aerodynamic loads are also incorporated.

L'objectif de ces travaux de recherche est l'étude et le développement d'un système ICM embarqué en utilisant la méthodologie de conception conjointe afin de satisfaire ses contraintes spécifiques. Il en a découlé la constitution d'un système ICM complet intégrant un système d'acquisition OpenBCI et un système de traitement à base de FPGA. Ce système pourrait être utilisé dans des contextes variés : médicale (pour les diagnostiques précoces des pathologies), technologique (informatique ubiquitaire), industriel (communication avec des robots), ludique (contrôler un joystick dans les jeux vidéo), etc. Dans notre contexte d’étude, la plateforme ICM proposée a été réalisée pour assister les personnes à mobilité réduite à commander les équipements domestiques. Nous nous sommes intéressés en particulier à l'étude et à l'implémentation des modules de filtrage adaptatif et dynamique, sous forme d'un coprocesseur codé en HDL afin de réduire son temps d'exécution car c'est le bloc le plus critique de la chaine ICM. Quant aux algorithmes d'extraction des caractéristiques et de classification, ils sont exécutés par le processeur Nios-II sous son système d'exploitation en ANSI-C. Le temps de traitement d'un trial par notre système ICM réalisé est de l'ordre de 0.4 s/trial et sa consommation ne dépasse guère 0.7 W. / The main purpose of this thesis is to study and develop an embedded brain computer interface (BCI) system using HW/SW methodology in order to satisfy the system specifications. A complete BCI system integrated in an acquisition system (OpenBCI) and a hardware platform based on the FPGA were achieved. The proposed system can be used in a variety of contexts: medical (for early diagnosis of pathologies, assisting people with severe disabilities to control home devices system through thought), technological (ubiquitous computing), industrial (communication with Robots), games (control a joystick in video games), etc. In our study, the proposed ICM platform was designed to control home devices through the thought of people with severe disabilities. A particular attention has been given to the study and implementation of the filtering module, adaptive and dynamic filtering, in the form of a coprocessor coded in HDL in order to reduce its execution time as it is the critical block in the returned ICM algorithms. For the feature extraction and classification algorithms, they are executed in the Nios-II processor using ANSI-C language. The prototype operates at 200 MHz and performs a real time classification with an execution delay of 0.4 second per trial. The power consumption of the proposed system is about 0.7 W.

Esta pesquisa aborda como problema principal, a falta de participação do usuário de Habitação de Interesse Social (HIS) na definição da sua moradia. O modelo atual do processo de provisão da habitação é baseado, principalmente, no sistema de produção em massa, resultando em unidades habitacionais padronizadas que não atendem às necessidades de seus usuários. Dentro deste cenário, a Customização em Massa (CM), tem sido investigada e apontada como uma estratégia promissora para solucionar este problema. A CM é definida como uma síntese de dois sistemas, aparentemente concorrentes: o da produção em massa com a customização individual de bens ou serviços. A participação do cliente na configuração do seu produto é uma condição essencial no conceito da CM, para o atendimento de suas necessidades e desejos. Para viabilizar essa participação, é necessária a utilização de ferramentas que permitam aos clientes configurarem seus produtos, a partir de opções disponibilizadas pelo fabricante. Especificamente no setor da construção civil, estas ferramentas ainda precisam ser desenvolvidas e experimentadas. Neste contexto, o objetivo principal da pesquisa é desenvolver um instrumento funcional que apoie os usuários, nas decisões da customização da sua habitação, em processos participativos de projetos de HIS. A pesquisa utiliza a abordagem da Design Science Research, a qual orienta para que, após a compreensão da questão central, seja proposto um artefato aplicável em soluções práticas. Os resultados obtidos com a aplicação do artefato desenvolvido demonstraram que o mesmo pode ser utilizado, não apenas na CM, mas também em outros processos de projetos participativos. A principal contribuição desta pesquisa está na originalidade das características paramétricas do modelo físico proposto. / This research approaches, as a main problem, the lack of user participation in Social Housing regarding the definition of its dwelling. The current model of housing provision process is based, mainly, on the mass production system, resulting in standardized housing units that do not satisfy the needs of users. Within this background, the Mass Customization (MC), has been investigated and identified as a promising strategy to solve this problem. The MC is defined as a synthesis of two systems, seemingly competitive: the mass production with the individual customization of goods or services. The customer participation in his product configuration is an essential condition in the concept of MC to fulfill his needs and desires. In order to enable this participation, the use of tools is required to enable customers to configure their products, from options, provided by the manufacturer. Specifically, in the civil construction sector these tools still need to be developed and experienced. In this context, the main objective of this research is the development of a functional device that can support users in their decisions of housing customization in participatory processes of Social Housing projects. The research uses the Design Science Research approach, which after understanding the central issue, recommends the development of an artifact and its application in practical solutions. The results obtained by applying the developed artifact demonstrated that the latter can be used, not only in MC, but also in other participatory design processes. The main contribution of this research is the originality of the parametric features of the proposed physical model.

Esta pesquisa aborda como problema principal, a falta de participação do usuário de Habitação de Interesse Social (HIS) na definição da sua moradia. O modelo atual do processo de provisão da habitação é baseado, principalmente, no sistema de produção em massa, resultando em unidades habitacionais padronizadas que não atendem às necessidades de seus usuários. Dentro deste cenário, a Customização em Massa (CM), tem sido investigada e apontada como uma estratégia promissora para solucionar este problema. A CM é definida como uma síntese de dois sistemas, aparentemente concorrentes: o da produção em massa com a customização individual de bens ou serviços. A participação do cliente na configuração do seu produto é uma condição essencial no conceito da CM, para o atendimento de suas necessidades e desejos. Para viabilizar essa participação, é necessária a utilização de ferramentas que permitam aos clientes configurarem seus produtos, a partir de opções disponibilizadas pelo fabricante. Especificamente no setor da construção civil, estas ferramentas ainda precisam ser desenvolvidas e experimentadas. Neste contexto, o objetivo principal da pesquisa é desenvolver um instrumento funcional que apoie os usuários, nas decisões da customização da sua habitação, em processos participativos de projetos de HIS. A pesquisa utiliza a abordagem da Design Science Research, a qual orienta para que, após a compreensão da questão central, seja proposto um artefato aplicável em soluções práticas. Os resultados obtidos com a aplicação do artefato desenvolvido demonstraram que o mesmo pode ser utilizado, não apenas na CM, mas também em outros processos de projetos participativos. A principal contribuição desta pesquisa está na originalidade das características paramétricas do modelo físico proposto. / This research approaches, as a main problem, the lack of user participation in Social Housing regarding the definition of its dwelling. The current model of housing provision process is based, mainly, on the mass production system, resulting in standardized housing units that do not satisfy the needs of users. Within this background, the Mass Customization (MC), has been investigated and identified as a promising strategy to solve this problem. The MC is defined as a synthesis of two systems, seemingly competitive: the mass production with the individual customization of goods or services. The customer participation in his product configuration is an essential condition in the concept of MC to fulfill his needs and desires. In order to enable this participation, the use of tools is required to enable customers to configure their products, from options, provided by the manufacturer. Specifically, in the civil construction sector these tools still need to be developed and experienced. In this context, the main objective of this research is the development of a functional device that can support users in their decisions of housing customization in participatory processes of Social Housing projects. The research uses the Design Science Research approach, which after understanding the central issue, recommends the development of an artifact and its application in practical solutions. The results obtained by applying the developed artifact demonstrated that the latter can be used, not only in MC, but also in other participatory design processes. The main contribution of this research is the originality of the parametric features of the proposed physical model.

Les produits d'ameublement domestique sont soumis à l'accélération de leur cycle de renouvellement. Ce phénomène n'est pas sans conséquences écologiques et sanitaires. Cette recherche analyse les pratiques de consommation de ces produits par les femmes, le genre le plus intéressé par l'aménagement et la décoration intérieure, dans le but d'étudier les facteurs qui sont à l'origine du désir de changement, notamment du sentiment de lassitude, les habitudes et les fréquences de renouvellement ainsi que leurs intentions en matière de préoccupation écologique. La mise en œuvre des méthodes de recherche et de traitement, issues pour l'essentiel des sciences humaines et sociales, met en lumière une catégorie de consommatrices qui résistent à la consommation des produits industrialisés et choisissent des pratiques de consommation citoyenne qui se basent essentiellement sur Je bricolage et le recyclage des produits usagés. Ces «créatifs culturels» sont à l'origine de l'émergence d'une culture éthique qui est en train de se confirmer et qui va sans doute modifier les valeurs de la société du futur. Les designers sont plus que jamais invités à se pencher de plus près sur ce nouveau profil de consommatrices dans l'idée de la production participative, du co-design. / Home furnishings are subject to an acceleration of their renewal cycle. This phenomenon has ecological and sanitary consequences. This research analyzes the consumption practices of these products by women, the gender that is most interested in interior design and decoration. It studies the factors that drive the desire for change, such as the feeling of weariness, the habits and frequencies of renewal, and women's concerns towards ecological issues. The implementation of research and treatment methods, ail coming from social sciences, highlights a category of consumers who resist to the consumption of industrialized products and choose consumption practices that are mainly based on DIY and recycling. These "cultural creatives" individuals are at the origin of the emergence of an ethical culture that is growing and that will undoubtedly modify the values that will shape the society of the future. The designers are more than ever invited to look more closely at this new profile of consumers in the idea of participative production : the co-design.

L’équipe Physique des Galaxies du Laboratoire d’Astrophysique de Marseille a développé un nouveau type d'interféromètre de Fabry-Perot, équipé de trois actionneurs piézoélectriques amplifiés et de trois capteurs capacitifs permettant le contrôle de l'espacement et du parallélisme des lames de verres de l'ordre de 200 µm avec une précision de positionnement du nm.L'objectif visé de ce manuscrit, composé de 3 parties, est le pilotage de cet interféromètre. La première partie, composée des chapitres 1 et 2, présente les généralités sur l'interférométrie de Fabry-Perot puis décrit les instruments 3DNTT et BTFI où seront installés l'interféromètre de nouvelle génération et son contrôleur associé. La conception et la réalisation d'un amplificateur hybride en vue de la réduction des non-linéarités des actionneurs piézoélectriques mettent fin à cette partie.La deuxième partie, chapitres 3 et 4, décrit le développement et la réalisation du contrôleur. Pour cela, une démarche basée sur le concept du Co-design a été adoptée.Le contrôleur ainsi réalisé est composé d'une carte de développement Microzed dont le cœur est un système sur puce de la série Zynq 7000 EPP et d'une carte d'interfaçage comportant des convertisseurs 3 ADC et 3 DAC et des circuits d'alimentation. La troisième et dernière partie, chapitres 5 et 6, traite de la modélisation d'état de l’interféromètre de Fabry-Perot et de son contrôle : un contrôle classique basé sur la régulation PID et un contrôle robuste et optimal basé sur le filtrage de KALMAN. Cette dernière partie conclut sur les perspectives pouvant découler des contributions de ce travail sur le contrôle et la commande Fabry-Perot. / The Physics of Galaxies Team of Laboratoire d'Astrophysique de Marseille (LAM) has developed a new type of Fabry-Perot, with three amplified piezoelectric actuators and three capacitive sensors to control the spacing and parallelism of mirror plates of approximately 200 µm with a positioning accuracy of 3 nm.The purpose of this manuscript, consisting of 3 parts is the control of this interferometer.The first part, consisting of Chapters 1 and 2 presents the general interferometry Fabry-Perot and then describes 3DNTT and BTFI instruments which will be installed the next generation interferometer and its associated controller. The design and implementation of a hybrid amplifier to reduce non-linearities of the piezoelectric actuators (hysteresis and creep) end this first part.The second part, Chapters 3 and 4, describes the development and implementation of the controller.For this, an approach based on codesign concept was adopted. The thus achieved controller consists of a Microzed development board whose heart is a system on chip of the 7000 series Zynq EPP (FPGA + Dual-Core ARM Cortex A9) and an interface card with converters (3 ADC and 3 DAC) and power supply circuits.For the finalization of the controller, two steps are necessary: hardware design in Xilinx Vivado and software design in Xilinx SDK.The third and final section, chapters 5 and 6 deals with the Fabry-Perot space-state modeling and its control: a classic control based on PID control and a robust and optimal control based on KALMAN filtering. This last part concludes the outlook may result from contributions of this work on the monitoring and control of the Fabry-Perot.

De nos jours, les systèmes de contrôle-commande temps-réel distribués à travers un réseau de communication sont de plus en plus utilisés dans les secteurs de l’automobile, de l’avionique, de la robotique mobile, de la télécommunication et plus généralement de la conduite de procédés industriels. En comparaison avec les systèmes de contrôle point-à-point conventionnel, un système contrôlé en réseau (SCR) permet non seulement de réduire le câblage et les coûts d’installation, mais offre aussi plus de flexibilité pour faire évoluer une installation existante et favorise les actions de diagnostic et de maintenance. Comme le réseau est partagé par plusieurs boucles de contrôle et par d’autres applications, la conséquence est que le trafic de communication est difficilement maîtrisable, ce qui peut conduire à des pertes de messages et engendre des délais aléatoires. Par ailleurs, le diagnostic et la tolérance aux défauts sont des enjeux importants pour les SCR, particulièrement dans le cas ou le domaine d’utilisation exige une grande sécurité. Il est évident que la théorie et l’application des approches classiques de diagnostic et de tolérance aux défauts doivent être révisées lorsqu’il s’agit de SCR.L’étude des SCR, reposant sur des compétences en automatique, en informatique et en réseau propose naturellement des solutions propres à chaque domaine. La problématique du diagnostic des SCR consiste non seulement à détecter et localiser des défauts affectant l’ensemble du système mais aussi à distinguer, les perturbations et défaillances affectant le réseau de communication de celles du système contrôle. L’objectif de nos travaux de thèse est de proposer des modèles intégrés permettant de représenter le comportement des SCR et de contribuer à leurs diagnostics. / Real-time control systems distributed across communication networks are increasingly used in automotive, avionics, mobile robotics, and telecommunications and more generally in the conduct of industrial processes. Compared with point- to-point conventional control systems, a networked control system (NCS) can not only reduce wiring and installation costs, but also offer more flexibility to expand an existing facility and promote actions of diagnosis and maintenance. As the network is shared by multiple control loops and other applications, the result is that the communication traffic is difficult to control, which can lead to loss of messages and generate random delays.Diagnosis and fault tolerance are important issues for NCS, especially in the case where the area of application requires security. It’s obvious that the theory and application of conventional approaches to diagnosis and fault tolerance need to be revised when it comes to NCS.The study of the NCS, based on automatic skills, computer science and network competences naturally provides adequate solutions to each area. The problem of NCS diagnosis is the one hand to detect and locate faults affecting the whole system and on the other hand to distinguish, disturbances and malfunctions affecting the communication network of the control system.Our work aims is to propose integrated models to represent the behavior of NCS and contribute to its diagnosis.

Today, learning spaces are stuck in the industrial age with rows of desks and chairs. Differentiated teaching and personalised learning are not effective in traditional learning spaces and should focus on creating future classrooms (Kariippanon, 2017). This thesis is covering how furnishing for learning environments can be improved and designed through a participatory design process with Katedralskolan in Växjö by using recycled materials and additive manufacturing technology. The thesis is also exploring the opportunities of involving stakeholders to create new learning through the additive manufacturing process in interior and furniture design. Together with Katedralskolan and Sculptur, this project is exploring a concept for schools to have integrated education in interior design through semiotics that will contribute to students’ learning and explore large-scale additive manufacturing. The objective is to design a collection of interior products that will inform and communicate at an educational level and how a circular manufacturing technique is possible through 3D-printing using communicating design and semiotics. The aim is also to understand Sculptur’s product development- and manufacturing process through large-scale 3D-printing. The mission statement whereas follows: Develop a furniture concept based on an understanding of the needs of, and participation with, the user group in a co-design process as a case to study the large-scale additive manufacturing techniques together with the given conditions provided by Sculptur. The thesis process has been following an iterative design process called the design thinking process (The Interaction Design Foundation, 2021) and a co-designing process (Sanders, 2018). The design thinking process is a design methodology that provides a solution-based approach to solving problems. The five stages of Design Thinking are as follows: Empathise, Define, Ideate, Prototype, Test. Through studies, surveys, and observations a list of stakeholder needs was created and was used when developing ideas through workshops, drawings, and prototyping. The ideas were then developed into concepts that were tested through both desktop 3D-printing and large-scale additive manufacturing. The concepts were also evaluated by stakeholders as well as through a concept evaluating matrix (Wikberg N., et.al., 2015). The result of this master thesis is the conclusion of the furniture concept as well as the study of large-scale additive manufacturing as an industrial designer. The furniture concept “Unfold lounge chair” is based on stakeholder needs and manufacturing restrictions. It is also an attempt to use theory to make the next generation of pedagogical furnishings using sustainable and circular manufacturing techniques. Through design thinking, the master thesis result was created with a human-centred approach to integrate the needs of people, the possibilities of technology and the requirements for business success (IDEO, n.d.). / Idag sitter läromiljöer fast i den industriella epoken med rader av skrivbord och stolar. Differentierad undervisning och personlig inlärning är inte effektiv i traditionella läromiljöer och bör vara i fokus för att skapa framtidens klassrum (Kariippanon, 2017). Detta examensarbete tar upp hur inredning för inlärningsmiljöer kan förbättras och utformas genom en deltagande designprocess med Katedralskolan i Växjö med hjälp av återvunna material och additiv tillverkningsteknik. Arbetet har också undersökt möjligheterna att involvera intressenter för att skapa nytt lärande genom tillämpning av additiv tillverkning inom inredning och möbeldesign. Tillsammans med Katedralskolan och Sculptur har detta projekt undersökt ett koncept för skolor att ha integrerad utbildning i möbler genom semiotik och pedagogisk design som kommer att bidra till elevernas lärande samt utforska storskalig additiv tillverkning. Målet är att utforma en samling möbelkoncept som informerar och kommunicerar på utbildningsnivå och hur en cirkulär tillverkningsteknik är möjlig genom 3D-printnig med hjälp av kommuniationsdesign och semiotik. Målet är också att förstå Sculpturs produktutvecklings- och tillverkningsprocess genom storskalig additiv tillverkning. Projektets Mission statement var följande: Utveckla ett möbelkoncept baserat på en förståelse av behoven hos användargruppen i en samdesignprocess som ett fall för att studera storskalig additiv tillverkning tillsammans med de givna förutsättningarna från Sculptur. Examensarbetet har följt en iterativ designprocess som kallas design thinking process (The Interaction Design Foundation, 2021) tillsammans med en co-designprocess (Sanders, 2018). Design thinking är en designmetodik som ger en lösningsbaserad metod för att lösa problem. De fem faserna i design thinking är följande: Empathise, Define, Ideate, Prototype, Test. Genom studier, undersökningar och observationer skapades en lista över intressenters behov och användes när idéer utvecklades genom workshops, skisser och prototyper. Idéerna utvecklades sedan till koncept som sedan testades genom både mindre 3D-utskrift och storskalig additiv tillverkning. Koncepten utvärderades också av intressenter samt genom en konceptviktningsmatris (Wikberg N., et.al., 2015). Resultatet av detta examensarbete är sammanfattningen av möbelkonceptet samt studien av storskalig additiv tillverkning som industridesigner. Möbelkonceptet ”Unfold lounge chair” bygger på intressenternas behov samt tillverkningsrestriktioner. Det är också ett försök att använda teori för att skapa nästa generation av pedagogiska möbler med hållbara och cirkulära tillverkningstekniker. Genom design thinking skapades resultatet med ett mänskligt centrerat tillvägagångssätt för att integrera människors behov, teknikens möjligheter och kraven för produktens framgång (IDEO, n.d.).

Within dementia care, there is a need to facilitate appropriate sensory experiences and opportunities for meaningful engagement for people living with dementia, particularly in later stages. As such, the provision of multisensory environments designed to meet the needs of these individuals and offering suitable occupation and stimulation, have increasingly gained significance. In this context, this paper explores the meaning of experience-focused, sensory design and the importance of appropriate sensory stimulation for wellbeing. It refers to recent interdisciplinary research in this area and a new guide advising on design of sensory enhanced spaces in residential dementia care. The primary focus of the paper is a case study showcasing the implementation of sensory design, based on these guidelines. The case study reports on the co-creation of a sensory, experience-focused space for residents with dementia in a London care-home. Based on this example, the paper highlights the value and impact of appropriate sensory design on the wellbeing of people with dementia and their carers. The conclusion discusses the learning and knowledge gained from this process as well as perspectives on further research activities including the analysis of user feedback and evaluation of the design intervention.

Der Einsatz von neuronalen Netzen in Edge-Geräten und deren Einbindung in unser tägliches Leben findet immer mehr Beachtung. Ihre hohen Rechenkosten machen jedoch viele eingebettete Anwendungen zu einer Herausforderung. Das Hauptziel meiner Doktorarbeit ist es, einen Beitrag zur Lösung dieses Dilemmas zu leisten: die Optimierung neuronaler Netze und die Entwicklung entsprechender neuronaler Verarbeitungseinheiten für Endgeräte. Diese Arbeit nahm die algorithmische Forschung als Ausgangspunkt und wandte dann deren Ergebnisse an, um das Architekturdesign von Neural Processing Units (NPUs) zu verbessern. Die Optimierung neuronaler Netzwerkmodelle begann mit der Quantisierung neuronaler Netzwerke mit einfacher Präzision und entwickelte sich zu gemischter Präzision. Die Entwicklung der NPU-Architektur folgte den Erkenntnissen der Algorithmusforschung, um ein Hardware/Software Co-Design zu erreichen. Darüber hinaus wurde ein neuartiger Ansatz zur gemeinsamen Entwicklung von Hardware und Software vorgeschlagen, um das Prototyping und die Leistungsbewertung von NPUs zu beschleunigen. Dieser Ansatz zielt auf die frühe Entwicklungsphase ab. Er hilft Entwicklern, sich auf das Design und die Optimierung von NPUs zu konzentrieren und verkürzt den Entwicklungszyklus erheblich. Im Abschlussprojekt wurde ein auf maschinellem Lernen basierender Ansatz angewendet, um die Rechen- und Speicherressourcen der NPU zu erkunden and optimieren. Die gesamte Arbeit umfasst mehrere verschiedene Bereiche, von der Algorithmusforschung bis zum Hardwaredesign. Sie alle arbeiten jedoch an der Verbesserung der Inferenz-Effizienz neuronaler Netze. Die Optimierung der Algorithmen zielt insbesondere darauf ab, den Speicherbedarf und die Rechenkosten von neuronalen Netzen zu verringern. Das NPU-Design hingegen konzentriert sich auf die Verbesserung der Nutzung von Hardwareressourcen. Der vorgeschlagene Ansatz zur gemeinsamen Entwicklung von Software und Hardware verkürzt den Entwurfszyklus und beschleunigt die Entwurfsiterationen. Die oben dargestellte Reihenfolge entspricht dem Aufbau dieser Dissertation. Jedes Kapitel ist einem Thema gewidmet und umfasst relevante Forschungsarbeiten, Methodik und Versuchsergebnisse.:1 Introduction 2 Convolutional Neural Networks 2.1 Convolutional layer 2.1.1 Padding 2.1.2 Convolution 2.1.3 Batch Normalization 2.1.4 Nonlinearity 2.2 Pooling Layer 2.3 Fully Connected Layer 2.4 Characterization 2.4.1 Composition of Operations and Parameters 2.4.2 Arithmetic Intensity 2.5 Optimization 3 Quantization with Double-Stage Squeeze-and-Threshold 19 3.1 Overview 3.1.1 Binarization 3.1.2 Multi-bit Quantization 3.2 Quantization of Convolutional Neural Networks 3.2.1 Quantization Scheme 3.2.2 Operator fusion of Conv2D 3.3 Activation Quantization with Squeeze-and-Threshold 3.3.1 Double-Stage Squeeze-and-Threshold 3.3.2 Inference Optimization 3.4 Experiment 3.4.1 Ablation Study of Squeeze-and-Threshold 3.4.2 Comparison with State-of-the-art Methods 3.5 Summary 4 Low-Precision Neural Architecture Search 39 4.1 Overview 4.2 Differentiable Architecture Search 4.2.1 Gumbel Softmax 4.2.2 Disadvantage and Solution 4.3 Low-Precision Differentiable Architecture Search 4.3.1 Convolution Sharing 4.3.2 Forward-and-Backward Scaling 4.3.3 Power Estimation 4.3.4 Architecture of Supernet 4.4 Experiment 4.4.1 Effectiveness of solutions to the dominance problem 4.4.2 Softmax and Gumbel Softmax 4.4.3 Optimizer and Inverted Learning Rate Scheduler 4.4.4 NAS Method Evaluation 4.4.5 Searched Model Analysis 4.4.6 NAS Cost Analysis 4.4.7 NAS Training Analysis 4.5 Summary 5 Configurable Sparse Neural Processing Unit 65 5.1 Overview 5.2 NPU Architecture 5.2.1 Buffer 5.2.2 Reshapeable Mixed-Precision MAC Array 5.2.3 Sparsity 5.2.4 Post Process Unit 5.3 Mapping 5.3.1 Mixed-Precision MAC 5.3.2 MAC Array 5.3.3 Support of Other Operation 5.3.4 Configurability 5.4 Experiment 5.4.1 Performance Analysis of Runtime Configuration 5.4.2 Roofline Performance Analysis 5.4.3 Mixed-Precision 5.4.4 Comparison with Cortex-M7 5.5 Summary 6 Agile Development and Rapid Design Space Exploration 91 6.1 Overview 6.1.1 Agile Development 6.1.2 Design Space Exploration 6.2 Agile Development Infrastructure 6.2.1 Chisel Backend 6.2.2 NPU Software Stack 6.3 Modeling and Exploration 6.3.1 Area Modeling 6.3.2 Performance Modeling 6.3.3 Layered Exploration Framework 6.4 Experiment 6.4.1 Efficiency of Agile Development Infrastructure 6.4.2 Effectiveness of Agile Development Infrastructure 6.4.3 Area Modeling 6.4.4 Performance Modeling 6.4.5 Rapid Exploration and Pareto Front 6.5 Summary 7 Summary and Outlook 123 7.1 Summary 7.2 Outlook A Appendix of Double-Stage ST Quantization 127 A.1 Training setting of ResNet-18 in Table 3.3 A.2 Training setting of ReActNet in Table 3.4 A.3 Training setting of ResNet-18 in Table 3.4 A.4 Pseudocode Implementation of Double-Stage ST B Appendix of Low-Precision Neural Architecture Search 131 B.1 Low-Precision NAS on CIFAR-10 B.2 Low-Precision NAS on Tiny-ImageNet B.3 Low-Precision NAS on ImageNet Bibliography 137 / Deploying neural networks on edge devices and bringing them into our daily lives is attracting more and more attention. However, its expensive computational cost makes many embedded applications daunting. The primary objective of my doctoral studies is to make contributions towards resolving this predicament: optimizing neural networks and designing corresponding efficient neural processing units for edge devices. This work took algorithmic research, specifically the optimization of deep neural networks, as a starting point and then applied its findings to steer the architecture design of Neural Processing Units (NPUs). The optimization of neural network models started with single precision neural network quantization and progressed to mixed precision. The NPU architecture development followed the algorithmic research findings to achieve hardware/software co-design. Furthermore, a new approach to hardware and software co-development was introduced, aimed at expediting the prototyping and performance assessment of NPUs. This approach targets early-stage development. It helps developers to focus on the design and optimization of NPUs and significantly shortens the development cycle. In the final project, a machine learning-based approach was applied to explore and optimize the computational and memory resources of the NPU. The entire work covers several different areas, from algorithmic research to hardware design. But they all work on improving the inference efficiency of neural networks. Specifically, algorithm optimization aims to reduce the memory footprint and computational cost of neural networks. The NPU design, on the other hand, focuses on improving the utilization of hardware resources. The proposed software and hardware co-development approach shortens the design cycle and speeds up the design iteration. The order presented above corresponds to the structure of this dissertation. Each chapter corresponds to a topic and covers relevant research, methodology, and experimental results.:1 Introduction 2 Convolutional Neural Networks 2.1 Convolutional layer 2.1.1 Padding 2.1.2 Convolution 2.1.3 Batch Normalization 2.1.4 Nonlinearity 2.2 Pooling Layer 2.3 Fully Connected Layer 2.4 Characterization 2.4.1 Composition of Operations and Parameters 2.4.2 Arithmetic Intensity 2.5 Optimization 3 Quantization with Double-Stage Squeeze-and-Threshold 19 3.1 Overview 3.1.1 Binarization 3.1.2 Multi-bit Quantization 3.2 Quantization of Convolutional Neural Networks 3.2.1 Quantization Scheme 3.2.2 Operator fusion of Conv2D 3.3 Activation Quantization with Squeeze-and-Threshold 3.3.1 Double-Stage Squeeze-and-Threshold 3.3.2 Inference Optimization 3.4 Experiment 3.4.1 Ablation Study of Squeeze-and-Threshold 3.4.2 Comparison with State-of-the-art Methods 3.5 Summary 4 Low-Precision Neural Architecture Search 39 4.1 Overview 4.2 Differentiable Architecture Search 4.2.1 Gumbel Softmax 4.2.2 Disadvantage and Solution 4.3 Low-Precision Differentiable Architecture Search 4.3.1 Convolution Sharing 4.3.2 Forward-and-Backward Scaling 4.3.3 Power Estimation 4.3.4 Architecture of Supernet 4.4 Experiment 4.4.1 Effectiveness of solutions to the dominance problem 4.4.2 Softmax and Gumbel Softmax 4.4.3 Optimizer and Inverted Learning Rate Scheduler 4.4.4 NAS Method Evaluation 4.4.5 Searched Model Analysis 4.4.6 NAS Cost Analysis 4.4.7 NAS Training Analysis 4.5 Summary 5 Configurable Sparse Neural Processing Unit 65 5.1 Overview 5.2 NPU Architecture 5.2.1 Buffer 5.2.2 Reshapeable Mixed-Precision MAC Array 5.2.3 Sparsity 5.2.4 Post Process Unit 5.3 Mapping 5.3.1 Mixed-Precision MAC 5.3.2 MAC Array 5.3.3 Support of Other Operation 5.3.4 Configurability 5.4 Experiment 5.4.1 Performance Analysis of Runtime Configuration 5.4.2 Roofline Performance Analysis 5.4.3 Mixed-Precision 5.4.4 Comparison with Cortex-M7 5.5 Summary 6 Agile Development and Rapid Design Space Exploration 91 6.1 Overview 6.1.1 Agile Development 6.1.2 Design Space Exploration 6.2 Agile Development Infrastructure 6.2.1 Chisel Backend 6.2.2 NPU Software Stack 6.3 Modeling and Exploration 6.3.1 Area Modeling 6.3.2 Performance Modeling 6.3.3 Layered Exploration Framework 6.4 Experiment 6.4.1 Efficiency of Agile Development Infrastructure 6.4.2 Effectiveness of Agile Development Infrastructure 6.4.3 Area Modeling 6.4.4 Performance Modeling 6.4.5 Rapid Exploration and Pareto Front 6.5 Summary 7 Summary and Outlook 123 7.1 Summary 7.2 Outlook A Appendix of Double-Stage ST Quantization 127 A.1 Training setting of ResNet-18 in Table 3.3 A.2 Training setting of ReActNet in Table 3.4 A.3 Training setting of ResNet-18 in Table 3.4 A.4 Pseudocode Implementation of Double-Stage ST B Appendix of Low-Precision Neural Architecture Search 131 B.1 Low-Precision NAS on CIFAR-10 B.2 Low-Precision NAS on Tiny-ImageNet B.3 Low-Precision NAS on ImageNet Bibliography 137