Application-aware optimization of redundant resources for the reconfigurable self-healing eDNA hardware architecture

In this paper we are interested in the mapping of embedded applications on a dynamically reconfigurable self-healing hardware architecture known as the eDNA (electronic DNA) architecture. The architecture consists of an array of cells interconnected through a 2D-mesh topology. Each cell consists of a processor and an Arithmetic Logic Unit (ALU). Applications are modeled as task graphs. We propose a Tabu Search-based approach for the mapping of an application to the reconfigurable architecture, such that the performance is maximized. When faults occur, the self-healing moves the affected functionality to spare-cells. We optimize the number and placement of spare-cells such that the performance overhead is minimized in the fault-free scenario and the application degrades gracefully in case of faults. This has been done using three different spare-cell placement strategies. We use Monte Carlo simulation to determine the average performance overhead increase due to fault occurrences. The approach has been evaluated using a large number of benchmarks and have shown that the performance loss is reduced with 16% for the best spare-cell placement strategy.