An Evolutionary Approach to the Soft Error Mitigation Technique for Cell-Based Design

Abstract

In this paper, we present a soft error mitigation algorithm that searches for the proper gate sizes within constrained gate-level designs. The individual gate sizing has an impact on the former optimization results and degrades the quality of the solution. In order to address this inefficiency, we utilize a modified topological sort that preserves the preceding local optima. Using a new local searcher, a hybrid genetic optimization technique for soft error mitigation is proposed. This evolutionary search algorithm has general genetic operators: the initialization of the population, crossover, mutation and selection operators. The local searcher consists of two subsequent heuristics. These search algorithms make the individual chromosome move to better search regions in a short time and then, the population acquires various candidates for the global optimum with the help of other genetic operators. The experiments show that the proposed genetic algorithm achieves an approximately 90% reduction in the number of soft errors when compared to the conventional greedy approach with at most 30% overhead for the area and critical path delay.

Keywords

• soft error mitigation
• single event transient
• cell-based design
• cell sizing
• hybrid genetic algorithm