REAL-TIME AND LOW-POWER EMBEDDED MULTI-TASKING SYSTEMS

This study proposes a technique which leverages data cache reconfigura-bility to address the problem of cache interference in multitasking embedded systems. Modern embedded systems often implement complex applications, comprising of multiple execution tasks with heavy memory requirements. Data caches are necessary to provide the required memory bandwidth. However, caches introduce two important problems for embedded systems. Cache out-comes in multi-tasking environments are difficult to predict, thus resulting in very poor real-time performance guarantees. Additionally, caches contribute to a significant amount of power. We study the effect that multiple concurrent tasks have on the cache and, subsequently, propose a technique which leverages work on reconfigurable cache architectures to eliminate cache interference and reduce power consumption using application specific information. By mapping parallel tasks to different cache partitions, inter-task interference is completely eliminated with minimal performance impact. Furthermore, both leakage and dynamic power is significantly improved