Summary (dt.31-Aug-2020)

Title: A Framework For Designing Context-Aware Adaptive Embedded Systems

A short summary of my Ph.D. research work.

Designing modern embedded systems is becoming complex due to the presence of a variety of choices in hardware platform components, multiple tasks with different modes varying in accuracy, and the need to consider external factors (context) while building the system. We define such systems, where tasks have modes with variable accuracy and are expected to adapt to changes in context by changing their modes of operation, as Context-aware Adaptive Embedded Systems (CAES systems). During early phases of system design, there is a need for joint consideration of platforms, tasks, and context to reach efficient solutions. However, such a joint exploration results in many-fold increase in the possible space of implementation, thereby significantly increasing the complexity of designing a CAES system. Our work focuses to address such complexities associated with design process for CAES systems.

We propose a two-step design process to alleviate the complexity of designing CAES systems. The first step explores various design choices and identifies suitable platform configuration for building the system. The second step uses the reports generated from the DSE step to automatically identify suitable mapping of various contexts to task modes, known as run-time controller specification, to adapt the system at run-time to changes in context. We use three real systems to demonstrate the applicability of our proposed flow. These case studies are diverse in their end-application, in the nature of the platforms and task modes, and in the metrics and context that are relevant for these systems. For all the three systems, many new design options become feasible due to consideration of components, task modes, and context together. We also obtain a significant pruning of infeasible design points using the proposed approach which helps the designer focus on the design points of interest.