Redes neurais estão se tornando soluções atrativas para a automação de veículos nos mercados automotivo, militar e aeroespacial. Todas essas aplicações são de segurança crítica e, portanto, precisam ter a confiabilidade como um dos principais requisitos. Graças ao baixo custo, baixo consumo de energia, e flexibilidade, FPGAs estão entre os dispositivos mais promissores para implementar redes neurais. Entretanto, FPGAs também são conhecidas por sua susceptibilidade à falhas induzidas por partículas ionizadas. Neste trabalho, nós avaliamos os efeitos de erros induzios por radiação nas saídas de duas redes neurais (Iris Flower e MNIST), implementadas em FPGAs baseadas em SRAM. Em particular, via experimentos com feixe acelerado de nêutrons, nós percebemos que a radiação pode induzir erros que modificam a saída da rede afetando ou não a corretude funcional da mesma. Chamamos o primeiro caso de erro crítico e o segundo de error tolerável. Nós exploramos aspectos das redes neurais que podem impactar tanto seu desempenho quanto sua confiabilidade, tais como os níveis de precisão na representação dos dados e diferentes métodos de implementação de alguns tipos de camadas. Usando campanhas exaustivas de injeção de falhas, nós identificamos porções das implementações da Iris Flower e da MNIST em FPGAs que são mais prováveis de gerar erros critícos ou toleráveis, quando corrompidas. Baseado nessa análise, nós propusemos estratégias de ABFT para algumas camadas das redes, bem como uma estratégia de proteção seletiva que triplica somente as camadas mais vulneráveis das redes neurais. Nós validamos essas estratégias de proteção usando testes de radiação com nêutrons, a vemos que nossa solução de proteção seletiva conseguiu mascarar 68% das falhas na Iris Flower com um custo adicional de 45%, e 40% das falhas na MNIST com um custo adicional de 8%. / Neural networks are becoming an attractive solution for automatizing vehicles in the automotive, military, and aerospace markets. All of these applications are safety-critical and, thus, must have reliability as one of the main constraints. Thanks to their low-cost, low power-consumption, and flexibility, Field-Programmable Gate Arrays (FPGAs) are among the most promising devices to implement neural networks. Unfortunately, FPGAs are also known to be susceptible to faults induced by ionizing particles. In this work, we evaluate the effects of radiation-induced errors in the outputs of two neural networks (Iris Flower and MNIST), implemented in SRAM-based FPGAs. In particular, through accelerated neutron beam experiments, we notice that radiation can induce errors that modify the output of the network with or without affecting the neural network’s functionality. We call the former critical errors and the latter tolerable errors. We explore aspects of the neural networks that can have impacts on both performance and reliability, such as levels of data precision and different methods of implementation for some types of layers. Through exhaustive fault-injection campaigns, we identify the portions of Iris Flower and MNIST implementations on FPGAs that are more likely, once corrupted, to generate a critical or a tolerable error. Based on this analysis, we propose Algorithm-Based Fault Tolerance (ABFT) strategies for certain layers in the networks, as well as a selective hardening strategy that triplicates only the most vulnerable layers of the neural network. We validate these hardening approaches with neutron radiation testing, and see that our selective hardening solution

Redes neurais estão se tornando soluções atrativas para a automação de veículos nos mercados automotivo, militar e aeroespacial. Todas essas aplicações são de segurança crítica e, portanto, precisam ter a confiabilidade como um dos principais requisitos. Graças ao baixo custo, baixo consumo de energia, e flexibilidade, FPGAs estão entre os dispositivos mais promissores para implementar redes neurais. Entretanto, FPGAs também são conhecidas por sua susceptibilidade à falhas induzidas por partículas ionizadas. Neste trabalho, nós avaliamos os efeitos de erros induzios por radiação nas saídas de duas redes neurais (Iris Flower e MNIST), implementadas em FPGAs baseadas em SRAM. Em particular, via experimentos com feixe acelerado de nêutrons, nós percebemos que a radiação pode induzir erros que modificam a saída da rede afetando ou não a corretude funcional da mesma. Chamamos o primeiro caso de erro crítico e o segundo de error tolerável. Nós exploramos aspectos das redes neurais que podem impactar tanto seu desempenho quanto sua confiabilidade, tais como os níveis de precisão na representação dos dados e diferentes métodos de implementação de alguns tipos de camadas. Usando campanhas exaustivas de injeção de falhas, nós identificamos porções das implementações da Iris Flower e da MNIST em FPGAs que são mais prováveis de gerar erros critícos ou toleráveis, quando corrompidas. Baseado nessa análise, nós propusemos estratégias de ABFT para algumas camadas das redes, bem como uma estratégia de proteção seletiva que triplica somente as camadas mais vulneráveis das redes neurais. Nós validamos essas estratégias de proteção usando testes de radiação com nêutrons, a vemos que nossa solução de proteção seletiva conseguiu mascarar 68% das falhas na Iris Flower com um custo adicional de 45%, e 40% das falhas na MNIST com um custo adicional de 8%. / Neural networks are becoming an attractive solution for automatizing vehicles in the automotive, military, and aerospace markets. All of these applications are safety-critical and, thus, must have reliability as one of the main constraints. Thanks to their low-cost, low power-consumption, and flexibility, Field-Programmable Gate Arrays (FPGAs) are among the most promising devices to implement neural networks. Unfortunately, FPGAs are also known to be susceptible to faults induced by ionizing particles. In this work, we evaluate the effects of radiation-induced errors in the outputs of two neural networks (Iris Flower and MNIST), implemented in SRAM-based FPGAs. In particular, through accelerated neutron beam experiments, we notice that radiation can induce errors that modify the output of the network with or without affecting the neural network’s functionality. We call the former critical errors and the latter tolerable errors. We explore aspects of the neural networks that can have impacts on both performance and reliability, such as levels of data precision and different methods of implementation for some types of layers. Through exhaustive fault-injection campaigns, we identify the portions of Iris Flower and MNIST implementations on FPGAs that are more likely, once corrupted, to generate a critical or a tolerable error. Based on this analysis, we propose Algorithm-Based Fault Tolerance (ABFT) strategies for certain layers in the networks, as well as a selective hardening strategy that triplicates only the most vulnerable layers of the neural network. We validate these hardening approaches with neutron radiation testing, and see that our selective hardening solution

<p>Day by day, we are witnessing a considerable increase in number and range of applications which entail the use of embedded computer systems. This increase is closely followed by the growth in complexity of applications controlled by embedded systems, often involving strict timing requirements, like in the case of safety-critical applications. Efficient design of such complex systems requires powerful and accurate tools that support the designer from the early phases of the design process.</p><p>This thesis focuses on the study of real-time distributed embedded systems and, in particular, we concentrate on a certain aspect of their real-time behavior and implementation: the time-triggered (TT) and event-triggered (ET) nature of the applications and of the communication protocols. Over the years, TT and ET systems have been usually considered independently, assuming that an application was entirely ET or TT. However, nowadays, the growing complexity of current applications has generated the need for intermixing TT and ET functionality. Such a development has led us to the identification of several interesting problems that are approached in this thesis. First, we focus on the elaboration of a holistic schedulability analysis for heterogeneous TT/ET task sets which interact according to a communication protocol based on both static and dynamic messages. Second, we use the holistic schedulability analysis in order to guide decisions during the design process. We propose a design optimisation heuristic that partitions the task-set and the messages into the TT and ET domains, maps and schedules the partitioned functionality, and optimises the communication protocol parameters. Experiments have been carried out in order to measure the efficiency of the proposed techniques.</p> / Report code: LiU-Tek-Lic-2003:21.

Day by day, we are witnessing a considerable increase in number and range of applications which entail the use of embedded computer systems. This increase is closely followed by the growth in complexity of applications controlled by embedded systems, often involving strict timing requirements, like in the case of safety-critical applications. Efficient design of such complex systems requires powerful and accurate tools that support the designer from the early phases of the design process. This thesis focuses on the study of real-time distributed embedded systems and, in particular, we concentrate on a certain aspect of their real-time behavior and implementation: the time-triggered (TT) and event-triggered (ET) nature of the applications and of the communication protocols. Over the years, TT and ET systems have been usually considered independently, assuming that an application was entirely ET or TT. However, nowadays, the growing complexity of current applications has generated the need for intermixing TT and ET functionality. Such a development has led us to the identification of several interesting problems that are approached in this thesis. First, we focus on the elaboration of a holistic schedulability analysis for heterogeneous TT/ET task sets which interact according to a communication protocol based on both static and dynamic messages. Second, we use the holistic schedulability analysis in order to guide decisions during the design process. We propose a design optimisation heuristic that partitions the task-set and the messages into the TT and ET domains, maps and schedules the partitioned functionality, and optimises the communication protocol parameters. Experiments have been carried out in order to measure the efficiency of the proposed techniques.

Les accidents routiers et leurs dommages représentent un problème croissant dans le monde entier. Dans ce contexte, les réseaux véhiculaires (VANETs) peuvent être déployés pour réduire les risques et pour améliorer le confort. Ils permettent aux véhicules d'échanger différents types de données qui vont des applications de sécurité et de gestion du trafic aux applications de confort. De nos jours, les applications de sécurité sont l’objet de beaucoup d'attention des chercheurs ainsi que des fabricants d'automobiles. Dans cette thèse, nous étudierons les applications critiques pour la sécurité routière visant à fournir une assistance dans des situations dangereuses ou difficiles. Notre objectif principal sera de proposer de nouveaux protocoles de contrôle d'accès au support de transmission (MAC) et de routage, qui peuvent s’adapter dynamiquement aux changements fréquents de topologies des VANETs. Après un aperçu des protocoles d’accès sans contention dans les VANETs, nous proposons des solutions basées sur la technique de division du temps: Time Division Multiple Access (TDMA). D’abord, nous nous concentrons sur le développement d’un nouveau protocole distribué (DTMAC), qui ne repose pas sur l’utilisation d’infrastructure. DTMAC utilise les informations de localisation et un mécanisme de réutilisation des slots pour assurer que les véhicules accèdent au canal efficacement et sans collision. Les résultats obtenus ont confirmé l’efficacité de notre protocole, DTMAC se comporte très significativement mieux que VeMAC (protocole MAC basé sur TDMA.) Ensuite nous proposons TRPM, un protocole de routage basé sur une approche cross-layer. Dans TRPM, l’ordonnancement des slots TDMA construit par DTMAC et la position de la destination sont utilisés pour choisir le meilleur relais. Les résultats montrent que TRPM offre de meilleures performances, du nombre moyen de relais et de la fiabilité de livraison des messages comparé à d’autres protocoles. Dans la deuxième partie de cette thèse, nous nous focaliserons sur les mécanismes centralisés d’allocation de slots qui utilisent des coordinateurs. D’abord, nous proposons CTMAC, un protocole basé sur TDMA centralisé utilisant les RSUs (RoadSide Units) pour créer et maintenir les ordonnancements. CTMAC met en œuvre un mécanisme qui permet d’empêcher les “Access Collisions” de se produire plus que deux fois entre les véhicules qui tentent d’acquérir un même slot disponible. Les résultats ont montré que CTMAC permet de mieux minimiser les collisions, ainsi que le surcoût généré pour créer et maintenir les ordonnancements par rapport aux protocoles MAC, basés sur TDMA distribué. Cependant, dans CTMAC, les véhicules roulant vite devront acquérir des nouveaux slots après une courte période de temps à chaque fois qu’ils quittent les zones de leurs RSUs courants. Cette situation rend les protocoles centralisés inefficaces et couteux dans les réseaux à grande vitesse. Afin de pallier à ce problème inhérent à l’utilisation des RSUs, nous adaptons un algorithme d’ordonnancement basé sur le clustering dans lequel certains véhicules sont élus pour gérer l'accès au canal. Ceci permet aux véhicules de rester attachés à leurs clusters plus longtemps. Pour ce faire, nous proposons 1- un protocole de clustering nommé AWCP afin de former des clusters stables avec une longue durée de vie. AWCP est basé sur l’algorithme de clustering pour les réseaux mobiles WCA dans lequel les têtes des clusters sont élues en se basant sur une fonction de poids. 2- Nous formulons le réglage des paramètres de protocole AWCP comme un problème d’optimisation multi-objective et nous proposons un outil d’optimisation qui combine la version multi-objective de l’algorithme génétique appelé NSGA-II avec le simulateur de réseau ns-2 pour trouver les meilleurs paramètres du protocole AWCP. 3- Nous proposons ASAS, une stratégie adaptative pour l’attribution des slots temporels basée sur une approche cross-layer entre TDMA et AWCP / Road crashes and their damages represent a serious issue and are one of the main causes of people death. In this context, Vehicular Ad hoc NETworks (VANETs) are deployed to reduce the risk of road accident as well as to improve passengers’ comfort by allowing vehicles to exchange different kinds of data which ranges widely from road safety and traffic management to infotainment. Nowadays, safety applications are receiving a great deal of attention from researchers as well as from automobile manufacturers. In this thesis, we particularly focus on safety-critical applications, designed to provide drivers assistance in dangerous situations and to avoid accidents in highway environments. Such applications must guarantee to the vehicles access to the medium and have strict requirements regarding end-to-end delay and packet loss ratio. Therefore, our main goal is to propose new medium access control and routing protocols, which can efficiently adapt to frequent changing VANET network topologies. After a comprehensive overview of free-contention MAC protocols, we propose several solutions, based on Time Division Multiple Access Technique (TDMA). We have designed DTMAC, a fully distributed TDMA-based MAC protocol, which does not rely on an expensive infrastructure. DTMAC uses vehicles’ locations and a slot reuse concept to ensure that vehicles in adjacent areas have collision-free schedule. Using simulations, we prove that DTMAC provides a lower rate of access and merging collisions than VeMAC, a well-known TDMA based MAC protocol in VANET. Then, in order to ensure that event-driven safety messages can be sent over a long distance, we propose TRPM, a TDMA aware Routing Protocol for Multi-hop communication. Our routing scheme is based on a cross layer approach between the MAC and the routing layers, in which the intermediate vehicles are selected using TDMA scheduling information. Simulation results show that TRPM provides better performances in terms of average end-to-end delay, average number of hops and average delivery ratio. In the second part, we focus on coordinator-based TDMA scheduling mechanisms. First, we propose the Centralized TDMA based MAC protocol (CTMAC) which uses Road Side Units (RSUs) as a central coordinator to create and maintain the TDMA schedules. CTMAC implements an Access Collision Avoidance mechanism that can prevent the access collision problem occurring more than twice between the same vehicles that are trying to access the channel at the same time. Using simulation we show an improvement in terms of access and merging collisions as well as the overhead required to create and maintain the TDMA schedules compared to distributed scheduling mechanisms. However, in the CTMAC protocol, fast moving vehicles will need to compete for new slots after a short period of time when they leave their current RSU area, which makes a centralized scheduling approach very expensive. In order to further improve the performance of coordinator-based TDMA scheduling mechanisms, we focus on cluster-based TDMA MAC protocols in which some vehicles in the network are elected to coordinate the channel access, allowing the vehicles to remain connected with their channel coordinator for a longer period of time. To this end, first we propose an adaptive weighted clustering protocol, named AWCP, which is road map dependent and uses road IDs and vehicle directions to make the clusters’ structure as stable as possible. Then, we formulate the AWCP parameter tuning as a multi-objective problem and we propose an optimization tool to find the optimal parameters of AWCP to ensure its QoS. Next, we propose ASAS, an adaptive slot assignment strategy for a cluster-based TDMA MAC protocol. This strategy is based on a cross layer approach involving TDMA and AWCP. The objective is to overcome the inter-cluster interference issue in overlapping areas by taking into account vehicles’ locations and directions when the cluster head assign slots