A constante redução nas dimensões dos transistores foi o principal combustível capaz de manter o crescente desempenho exigido por aplicações. Ao mesmo tempo, as tensões de alimentação dos circuitos também são reduzidas a cada novo nó tecnológico, fazendo com que partículas como nêutrons e partículas alpha, portando quantidades de energia cada vez menores sejam capazes de gerar os chamados soft errors, que impactam diretamente na redução da confiabilidade dos sistemas embarcados atuais. Isto faz com que a implementação de técnicas de tolerância a falhas se tornem praticamente obrigatórias para tecnologias atuais e futuras. Estes mesmos sistemas embarcados, como smartphones, devem apresentar alto poder de processamento, visando atender um crescente conjunto de aplicações de natureza heterogênea, consumindo a mínima potência possível. Nestes sistemas, algumas dessas principais aplicações como codec GSM, cancelamento de eco acústico, processamento de áudio e vídeo apresentam em comum a necessidade de multiplicar matrizes de diferentes dimensões em determinados intervalos de tempo. Pensando nestas demandas, será proposta a arquitetura RA3, cujo objetivo é executar o algoritmo de multiplicação de matrizes em paralelo com a técnica de tolerância a falhas conhecida na literatura como ABFT, visando a aumentar a confiabilidade da mesma. Além disso, a RA3 possui uma estrutura adaptativa que permite que unidades internas como memórias, multiplicadores e somadores sejam ligadas ou desligadas através da aplicação da técnica de power gating em tempo de execução, conforme restrições impostas pela largura da banda de memória, power budgets e deadlines impostos por aplicações de tempo real, visando executar tarefas consumindo a mínima potência possível. Para avaliar as funcionalidades propostas, dois estudos de caso reais são apresentados e o comportamento da arquitetura é avaliado sobre diversos aspectos como desempenho, área, consumo de potência e cobertura de falhas. Finalmente é possível comprovar que a adaptabilidade proposta pela arquitetura RA3 permite que seja encontrada, em diversos cenários, a quantidade exata de recursos necessários para executar determinadas aplicações sem comprometer as restrições impostas principalmente no consumo de potência e por aplicações com deadlines críticos, mantendo ainda altas taxas de cobertura de falhas. / The continuous reduction of transistors’ dimensions was the main drive capable of maintaining the performance increase required by applications. At the same time, supply voltages of the circuits are also reduced with each new technology node, causing particles such as neutrons or alpha particles, even with reduced amounts of energy, to generate so-called soft errors that directly impact on the reliability of embedded systems. This scenario makes the implementation of techniques for fault tolerance mandatory for current and future technologies. Still, embedded systems, such as smartphones, must provide high processing power to execute a growing set of applications of heterogeneous nature, consuming the least possible power. In these systems, applications like GSM codec, acoustic echo cancellation, audio and video processing have in common the need for matrix multiplication operations of different dimensions at certain time intervals. To efficiently support the aforementioned scenario, this dissertation proposes the RA3 architecture whose goal is run the matrix multiplication algorithm in parallel with the fault tolerance technique know in the literature as ABFT, aiming to support software execution with high reliability. Furthermore, the RA3 architecture provides adaptive internal units such as memories, multipliers and adders with adaptive powering on or off by applying power gating at runtime. Runtime power gating enables to meet restrictions imposed by real-time applications or memory bandwidth with minimum power. To evaluate the proposed architecture, two case studies are presented and the behavior of the architecture is evaluated in terms of performance, area, power consumption and fault coverage. Finally, a comprehensive design space exploration shows that the adaptability provided by the RA3 architecture allows the system designer to find, in many scenarios, the exact amount of resources needed to run a set of applications without compromising the restrictions imposed mainly in power consumption and real-time deadlines, while still maintaining a high fault coverage rate.

A constante redução nas dimensões dos transistores foi o principal combustível capaz de manter o crescente desempenho exigido por aplicações. Ao mesmo tempo, as tensões de alimentação dos circuitos também são reduzidas a cada novo nó tecnológico, fazendo com que partículas como nêutrons e partículas alpha, portando quantidades de energia cada vez menores sejam capazes de gerar os chamados soft errors, que impactam diretamente na redução da confiabilidade dos sistemas embarcados atuais. Isto faz com que a implementação de técnicas de tolerância a falhas se tornem praticamente obrigatórias para tecnologias atuais e futuras. Estes mesmos sistemas embarcados, como smartphones, devem apresentar alto poder de processamento, visando atender um crescente conjunto de aplicações de natureza heterogênea, consumindo a mínima potência possível. Nestes sistemas, algumas dessas principais aplicações como codec GSM, cancelamento de eco acústico, processamento de áudio e vídeo apresentam em comum a necessidade de multiplicar matrizes de diferentes dimensões em determinados intervalos de tempo. Pensando nestas demandas, será proposta a arquitetura RA3, cujo objetivo é executar o algoritmo de multiplicação de matrizes em paralelo com a técnica de tolerância a falhas conhecida na literatura como ABFT, visando a aumentar a confiabilidade da mesma. Além disso, a RA3 possui uma estrutura adaptativa que permite que unidades internas como memórias, multiplicadores e somadores sejam ligadas ou desligadas através da aplicação da técnica de power gating em tempo de execução, conforme restrições impostas pela largura da banda de memória, power budgets e deadlines impostos por aplicações de tempo real, visando executar tarefas consumindo a mínima potência possível. Para avaliar as funcionalidades propostas, dois estudos de caso reais são apresentados e o comportamento da arquitetura é avaliado sobre diversos aspectos como desempenho, área, consumo de potência e cobertura de falhas. Finalmente é possível comprovar que a adaptabilidade proposta pela arquitetura RA3 permite que seja encontrada, em diversos cenários, a quantidade exata de recursos necessários para executar determinadas aplicações sem comprometer as restrições impostas principalmente no consumo de potência e por aplicações com deadlines críticos, mantendo ainda altas taxas de cobertura de falhas. / The continuous reduction of transistors’ dimensions was the main drive capable of maintaining the performance increase required by applications. At the same time, supply voltages of the circuits are also reduced with each new technology node, causing particles such as neutrons or alpha particles, even with reduced amounts of energy, to generate so-called soft errors that directly impact on the reliability of embedded systems. This scenario makes the implementation of techniques for fault tolerance mandatory for current and future technologies. Still, embedded systems, such as smartphones, must provide high processing power to execute a growing set of applications of heterogeneous nature, consuming the least possible power. In these systems, applications like GSM codec, acoustic echo cancellation, audio and video processing have in common the need for matrix multiplication operations of different dimensions at certain time intervals. To efficiently support the aforementioned scenario, this dissertation proposes the RA3 architecture whose goal is run the matrix multiplication algorithm in parallel with the fault tolerance technique know in the literature as ABFT, aiming to support software execution with high reliability. Furthermore, the RA3 architecture provides adaptive internal units such as memories, multipliers and adders with adaptive powering on or off by applying power gating at runtime. Runtime power gating enables to meet restrictions imposed by real-time applications or memory bandwidth with minimum power. To evaluate the proposed architecture, two case studies are presented and the behavior of the architecture is evaluated in terms of performance, area, power consumption and fault coverage. Finally, a comprehensive design space exploration shows that the adaptability provided by the RA3 architecture allows the system designer to find, in many scenarios, the exact amount of resources needed to run a set of applications without compromising the restrictions imposed mainly in power consumption and real-time deadlines, while still maintaining a high fault coverage rate.

A constante redução nas dimensões dos transistores foi o principal combustível capaz de manter o crescente desempenho exigido por aplicações. Ao mesmo tempo, as tensões de alimentação dos circuitos também são reduzidas a cada novo nó tecnológico, fazendo com que partículas como nêutrons e partículas alpha, portando quantidades de energia cada vez menores sejam capazes de gerar os chamados soft errors, que impactam diretamente na redução da confiabilidade dos sistemas embarcados atuais. Isto faz com que a implementação de técnicas de tolerância a falhas se tornem praticamente obrigatórias para tecnologias atuais e futuras. Estes mesmos sistemas embarcados, como smartphones, devem apresentar alto poder de processamento, visando atender um crescente conjunto de aplicações de natureza heterogênea, consumindo a mínima potência possível. Nestes sistemas, algumas dessas principais aplicações como codec GSM, cancelamento de eco acústico, processamento de áudio e vídeo apresentam em comum a necessidade de multiplicar matrizes de diferentes dimensões em determinados intervalos de tempo. Pensando nestas demandas, será proposta a arquitetura RA3, cujo objetivo é executar o algoritmo de multiplicação de matrizes em paralelo com a técnica de tolerância a falhas conhecida na literatura como ABFT, visando a aumentar a confiabilidade da mesma. Além disso, a RA3 possui uma estrutura adaptativa que permite que unidades internas como memórias, multiplicadores e somadores sejam ligadas ou desligadas através da aplicação da técnica de power gating em tempo de execução, conforme restrições impostas pela largura da banda de memória, power budgets e deadlines impostos por aplicações de tempo real, visando executar tarefas consumindo a mínima potência possível. Para avaliar as funcionalidades propostas, dois estudos de caso reais são apresentados e o comportamento da arquitetura é avaliado sobre diversos aspectos como desempenho, área, consumo de potência e cobertura de falhas. Finalmente é possível comprovar que a adaptabilidade proposta pela arquitetura RA3 permite que seja encontrada, em diversos cenários, a quantidade exata de recursos necessários para executar determinadas aplicações sem comprometer as restrições impostas principalmente no consumo de potência e por aplicações com deadlines críticos, mantendo ainda altas taxas de cobertura de falhas. / The continuous reduction of transistors’ dimensions was the main drive capable of maintaining the performance increase required by applications. At the same time, supply voltages of the circuits are also reduced with each new technology node, causing particles such as neutrons or alpha particles, even with reduced amounts of energy, to generate so-called soft errors that directly impact on the reliability of embedded systems. This scenario makes the implementation of techniques for fault tolerance mandatory for current and future technologies. Still, embedded systems, such as smartphones, must provide high processing power to execute a growing set of applications of heterogeneous nature, consuming the least possible power. In these systems, applications like GSM codec, acoustic echo cancellation, audio and video processing have in common the need for matrix multiplication operations of different dimensions at certain time intervals. To efficiently support the aforementioned scenario, this dissertation proposes the RA3 architecture whose goal is run the matrix multiplication algorithm in parallel with the fault tolerance technique know in the literature as ABFT, aiming to support software execution with high reliability. Furthermore, the RA3 architecture provides adaptive internal units such as memories, multipliers and adders with adaptive powering on or off by applying power gating at runtime. Runtime power gating enables to meet restrictions imposed by real-time applications or memory bandwidth with minimum power. To evaluate the proposed architecture, two case studies are presented and the behavior of the architecture is evaluated in terms of performance, area, power consumption and fault coverage. Finally, a comprehensive design space exploration shows that the adaptability provided by the RA3 architecture allows the system designer to find, in many scenarios, the exact amount of resources needed to run a set of applications without compromising the restrictions imposed mainly in power consumption and real-time deadlines, while still maintaining a high fault coverage rate.

Approved for public release, distribution is unlimited / The ability of an organization to adapt its structure to changing dynamic requirements can provide for increased effectiveness and efficiency. A better understanding of the factors that affect adaptation capabilities within an organization can facilitate implementation of changes to better fit the organization to the mission. Experiment 9, conducted for the Office of Naval Research's (ONR) Adaptive Architecture for Command and Control (A2C2) research program, provided insight into the decision making process of a small group given changes in a scenario to prompt the need for change. The experiment also provided insight into the challenges faced by an organization in the process of adaptation, and factors that affect the willingness and need for adaptation. This thesis examines how differences of emphasis within the training environment itself may affect an organization's willingness to adapt to changing circumstances. This thesis proposes changes to future experiments, focusing efforts on making cues more apparent to the test subject. This thesis also suggests modifications to the data collection system to enhance post experiment analysis. / Lieutenant, United States Navy

Ces travaux de thèse s'inscrivent dans la thématique des objets connectés, désormais connue sous le nom de Internet of Things (IoT). Elle trouve son origine dans la démocratisation d'Internet depuis le début des années 2000 et la migration vers des appareils hautement mobiles, rendue possible grâce à la miniaturisation des systèmes embarqués. Dans ce contexte, l'efficacité énergétique est primordiale puisque les projections actuelles parlent de dizaines de milliards de composants connectés à l'horizon 2020. Or pour une question de facilité de déploiement et d'usage, une grande partie des échanges de données dans ces réseaux s'effectue via une liaison sans-fil dont l'implémentation représente une part importante de la consommation. Effectivement, la question de l'efficacité énergétique est en général considérée comme un problème de perfectionnement des architectures matérielles, souvent associé à une évolution favorable de la technologie. Toutefois, ce paradigme atteint rapidement ses limites puisqu'il implique nécessairement un dimensionnement fortement contraint pour être compatible avec les pires conditions d'utilisation, même si elles ne sont pas effectives la plupart du temps. C'est typiquement le cas avec les communications sans-fil puisque le canal radio est un milieu caractérisé par une forte variabilité en raison des phénomènes de propagation et de la présence d'interférences. Notre étude a donc porté sur la conception d'une chaîne de transmission dont le budget de liaison peut être dynamiquement modifié en fonction de l'atténuation réelle du signal, afin de réduire la consommation du système. La thèse a notamment contribué à la mise au point d'un récepteur auto-adaptatif spécifique à la norme IEEE 802.15.4, en proposant à la fois une architecture de modem numérique reconfigurable et à la fois une méthode de contrôle automatique du point de fonctionnement. Plus précisément, le travail s'est appuyé sur deux approches, l'échantillonnage compressif et l'échantillonnage partiel, pour réduire la taille des données à traiter, diminuant ainsi l'activité interne des opérateurs arithmétiques. En contrepartie, le processus de démodulation nécessite un SNR supérieur, dégradant la sensibilité du récepteur et donc le budget de liaison. Cette solution, portée sur une technologie STMicroelectronics CMOS 65 nm LP, offre une faible empreinte matérielle vis-à-vis d'une architecture classique avec seulement 23,4 kcellules. Grâce au modèle physique du circuit qui a été développé, la consommation pour la démodulation d'un paquet est estimée à 278 uW lorsque le modem est intégralement utilisé. Elle peut toutefois être abaissée progressivement jusqu'à 119 uW, correspondant à une baisse de la sensibilité de 10 dB. Ainsi, le modem implémenté et sa boucle de contrôle permettent d'économiser en moyenne 30 % d'énergie dans un cas d'utilisation typique. / This thesis work takes part in the connected objects theme, also known as the Internet of Things (IoT). It emerges from the Internet democratization since the early 2000's and the shift to highly mobile devices, made possible by the miniaturization of embedded systems. In this context, the energy efficiency is mandatory since today's projections are around tens of billions of connected devices in 2020. However for ease of deployment and usage, a large part of the data transfers in these networks is wireless, which implementation represents a significant part of the power consumption. Indeed, the energy efficiency question is addressed in general as a fine tuning of hardware architectures, which is often associated with a favorable technology evolution. Nevertheless, this design paradigm quickly reached its limits since it necessary implies a highly constrained sizing to be compatible with the worst operating conditions, even if they are not effective most of the time. It's typically the case with wireless communications since the radio channel is a medium characterized by a strong variability due to propagations effects and interferences. Thus, our study focused on the design of a communication chain whose link budget can be dynamically tuned depending on the actual signal attenuation, in order to reduce the system power consumption. The thesis has contributed to the design of a self-adaptive receiver dedicated to IEEE 802.15.4 standard, by proposing both a reconfigurable digital baseband architecture and an automatic control method of the operating mode. More precisely, the work relied on two approaches, the compressive sampling and the partial sampling, to reduce the data's size to process, decreasing the internal activity of arithmetics operators. In return, the demodulation processing needs a higher SNR, degrading in the same time the receiver sensitivity and thus the link budget. This solution, implemented in an STMicroelectronics CMOS 65 nm LP process, offers a low hardware overhead compared to conventional architecture with only 23,4 kgates. Thanks to the circuit physical model that has been developed, the power consumption for a packet demodulation is estimated to 278 uW when the baseband is fully activated. It can however be gradually decreased down to 119 uW, corresponding to a sensitivity reduction of 10 dB. Thus, the proposed digital baseband and its control loop save 30 % of energy in average in a typical use case.

Field Programmable Gate Array (FPGA) devices offer a suitable platform for survivable hardware architectures in mission-critical systems. In this dissertation, active dynamic redundancy-based fault-handling techniques are proposed which exploit the dynamic partial reconfiguration capability of SRAM-based FPGAs. Self-adaptation is realized by employing reconfiguration in detection, diagnosis, and recovery phases. To extend these concepts to semiconductor aging and process variation in the deep submicron era, resilient adaptable processing systems are sought to maintain quality and throughput requirements despite the vulnerabilities of the underlying computational devices. A new approach to autonomous fault-handling which addresses these goals is developed using only a uniplex hardware arrangement. It operates by observing a health metric to achieve Fault Demotion using Recon- figurable Slack (FaDReS). Here an autonomous fault isolation scheme is employed which neither requires test vectors nor suspends the computational throughput, but instead observes the value of a health metric based on runtime input. The deterministic flow of the fault isolation scheme guarantees success in a bounded number of reconfigurations of the FPGA fabric. FaDReS is then extended to the Priority Using Resource Escalation (PURE) online redundancy scheme which considers fault-isolation latency and throughput trade-offs under a dynamic spare arrangement. While deep-submicron designs introduce new challenges, use of adaptive techniques are seen to provide several promising avenues for improving resilience. The scheme developed is demonstrated by hardware design of various signal processing circuits and their implementation on a Xilinx Virtex-4 FPGA device. These include a Discrete Cosine Transform (DCT) core, Motion Estimation (ME) engine, Finite Impulse Response (FIR) Filter, Support Vector Machine (SVM), and Advanced Encryption Standard (AES) blocks in addition to MCNC benchmark circuits. A iii significant reduction in power consumption is achieved ranging from 83% for low motion-activity scenes to 12.5% for high motion activity video scenes in a novel ME engine configuration. For a typical benchmark video sequence, PURE is shown to maintain a PSNR baseline near 32dB. The diagnosability, reconfiguration latency, and resource overhead of each approach is analyzed. Compared to previous alternatives, PURE maintains a PSNR within a difference of 4.02dB to 6.67dB from the fault-free baseline by escalating healthy resources to higher-priority signal processing functions. The results indicate the benefits of priority-aware resiliency over conventional redundancy approaches in terms of fault-recovery, power consumption, and resource-area requirements. Together, these provide a broad range of strategies to achieve autonomous recovery of reconfigurable logic devices under a variety of constraints, operating conditions, and optimization criteria.