Recently, deep neural networks (DNNs) have been utilized in real-time systems such as autonomous vehicles, where meeting temporal constraints is essential. However, executing such systems on CPU-GPU architectures can make scheduling analysis challenging due to the added delays caused by computing and memory contention. In addition, classic task models are not directly able to model accurately such systems. In this article, we propose a new task model called DNN Task Model (DTM). This model considers both DNN properties and GPU architecture at the same time. It allows us to distinguish between CPU and GPU tasks, provides information about the DNN application and give more accurate execution time analysis through consideration of data quality. We compute DTM from a source CUDA file and a set of real-time specifications of the system. The proposed model is extensible enough to be adopted to various DNN type applications allowing designer to compare candidate software and GPU architectures. Furthermore, we propose a graph optimization inspired by Tensor-RT.

Network-on-Chips (NoC) are widely used in industrial applications since they provide communication parallelism and reduce energy consumption. The use of NoC has been recently extended to real-time systems, whose execution has to meet temporal constraints. Communication delays introduced by the network make the scheduling analysis challenging. In this article, we propose a new NoC communication model called ECTM. The main goal of this model is to assess the schedulability of dependent periodic tasks exchanging messages on a NoC. ECTM is a model allowing schedulability analysis of messages and tasks of the NoC. To achieve schedulability, ECTM produces an analysis model by transforming NoC messages to tasks in order to take into account communication delays during the scheduling analysis. Schedulability of the system is assessed using simulation over the feasibility interval with a list scheduling, ECTM supports Store-And-Forward and Wormhole NoC. In this article, we have demonstrated the correctness of the transformations of ECTM. ECTM has been implemented in a real-time scheduling analysis tool called Cheddar and we performed experiments to assess its efficiency. ECTM is more efficient than existing solutions with an improvement of 30% for Store-And-Forward NoCs and up to 100% for Wormhole NoCs, while the proposed model requires a larger computation time about 17% for Store-And-Forward NoCs.

A Mixed Criticality System (MCS) combines real-time software tasks with different criticality levels. In a MCS, the criticality level specifies the level of assurance against system failure. For high-critical flows of messages, it is imperative to meet deadlines; otherwise, the whole system might fail, leading to catastrophic results, like loss of life or serious damage to the environment. In contrast, low-critical flows may tolerate some delays. Furthermore, in MCS, flow performances such as the Worst Case Communication Time (WCCT) may vary depending on the criticality level of the applications. Then execution platforms must provide different operating modes for applications with different levels of criticality. To conclude, in Network-On-Chip (NoC), sharing resources between communication flows can lead to unpredictable latencies and subsequently turns the implementation of MCS in many-core architectures challenging. In this article, we propose and evaluate a new NoC router to support MCS based on an accurate WCCT analysis for high-critical flows. The proposed router, called Double Arbiter and Switching router (DAS), jointly uses Wormhole and Store And Forward communication techniques for low- and high-critical flows, respectively. It ensures that high-critical flows meet their deadlines while maximizing the bandwidth remaining for the low-critical flows. We also propose a new method for high-critical communication time analysis, applied to Store And Forward switching mode with virtual channels. For low-critical flows communication time analysis, we adapt an existing wormhole communication time analysis with share policy to our context. The second contribution of this article is a multi-abstraction-level evaluation of DAS. We evaluate the communication time of flows, the system mode change, the cost, and four properties of DAS. Simulations with a cycle-accurate SystemC NoC simulator show that, with a 15% network use rate, the communication delay of high-critical flows is reduced by 80% while communication delay of low-critical flow is increased by 18% compared to solutions based on routers with multiple virtual channels. For 10% of network interferences, using system mode change, DAS reduces the high-critical communication delays about 66%. We synthesize our router with a 28nm SOI technology and show that the size overhead is limited of 2.5% compared to the solution based on virtual channel router. Finally, we applied model checking verification techniques to automatically prove several DAS properties required by critical systems designers.

Mixed-Criticality Systems (MCS) are real-time systems characterized by two or more distinct levels of criticality. In MCS, it is imperative that high-critical flows meet their deadlines while low critical flows can tolerate some delays. Sharing resources between flows in Network-On-Chip (NoC) can lead to different unpredictable latencies and subsequently complicate the implementation of MCS in many-core architectures. This paper proposes a new virtual channel router designed for MCS deployed over NoCs. The first objective of this router is to reduce the worst-case communication latency of high-critical flows. The second aim is to improve the network use rate and reduce the communication latency for low-critical flows. The proposed router, called DAS (Double Arbiter and Switching router), jointly uses Wormhole and Store And Forward techniques for low and high-critical flows respectively. Simulations with a cycle-accurate SystemC NoC simulator show that, with a 15% network use rate, the communication delay of high-critical flows is reduced by 80% while communication delay of low-critical flow is increased by 18% compared to usual solutions based on routers with multiple virtual channels.