Adaptive time-triggered network-on-chip-based multi-core architecture: enhancing safety and energy efficiency

Real-time computing systems are designed to meet strict timing constraints and respond to events or inputs within specified deadlines. These systems are commonly used in safety-critical applications such as spacecraft, medical devices, industrial control, and automotive systems. Engineers rely on various scheduling techniques to ensure that timing constraints are met. One such technique is static resource allocation in time-triggered systems. Static resource allocation offers valuable advantages in terms of system dependability by minimizing message congestion and contention, enabling efficient resource usage in network-on-chip (NoC) architectures. This is achieved through the pre-allocation of resources and scheduling of tasks, resulting in improved system throughput and reduced jitter. The time-triggered concept in NoC architectures provides precise knowledge about the permitted points in time for message exchanges between cores, serving as a fundamental building block for fault containment, real-time support, and enhanced system performance.
While static resource allocation excels in minimizing congestion and contention and contributes to system dependability, it may pose challenges in accommodating dynamic workloads and evolving requirements. Additionally, it can limit the achievement of fault tolerance, a crucial aspect of ensuring safety in safety-critical systems. To address these limitations, this thesis focuses on developing fault tolerance and energy-saving techniques tailored explicitly for NoC-based multi-core architectures to enhance their safety and energy efficiency.
The main goal is to incorporate fault tolerance mechanisms, such as adaptation and redundancy, into time-triggered systems without compromising the benefits of static resource allocation. The adaptation technique within the NoC is designed to support multiple schedules, allowing the NoC to switch schedules during run-time in response to context events, such as permanent faults in NoC resources (e.g., routers, links, network interfaces, and cores). By dynamically reconfiguring the schedule upon the occurrence of a permanent fault, the faulty component is effectively isolated, and tasks or messages are redistributed to other available resources. This ensures the system’s operational continuity despite faults that could lead to message corruption, delays, or losses within NoC resources. This adaptation technique improves the system’s safety by providing flexibility in resource allocation without sacrificing the benefits of static resource allocation.
Furthermore, this thesis incorporates seamless redundancy techniques to enhance the system’s safety, especially in scenarios involving transient and permanent faults. This technique selectively applies message replication and fusion to safety-critical messages at the network interface, minimizing overhead in non-critical parts of the system. It safeguards critical data from potential failures caused by message corruption, delays, and losses in routers or links during message exchanges.
The thesis also focuses on improving energy efficiency in multi-core chips by providing low-power services. By incorporating time-triggered communication into NoC-based multi-core architectures, deterministic communication is achieved by scheduling the message’s injection time and specifying the frequency to be used by each router at different points in time. This predetermined frequency in the schedule allows routers to adjust their frequencies accordingly during their active time and to clock gate the idle routers, enhancing energy efficiency and preserving the deterministic behaviour of the NoC communication.
Moreover, the adaptation techniques in the NoC are used to reconfigure the operating frequency of the NoC based on workload or power requirement variations by switching between schedules, further optimizing energy consumption. Integrating features such as time-triggered capability, adaptation, time-triggered frequency scaling, and seamless redundancy mechanisms into NoC-based multi-core architectures represents a significant advancement over the current state of the art. The results of this work have significant implications for applications relying on high-performance, safe, and energy-efficient multi-core systems in various domains, such as healthcare and transportation.