Criticality-Aware Negotiation-Driven Scrubbing Scheduling for Reliability Maximization in SRAM-Based FPGAs

Abstract

Memory scrubbing is a resource-efficient technique to ensure the high reliability of SRAM-based FPGAs by refreshing the configuration memory just before its execution. To maximize reliability, a scrubbing scheduling algorithm is expected to scrub as many tasks as possible. Unfortunately, contemporary scheduling algorithms either sub-optimally handle scrubbing conflicts under bursty requests from multiple user tasks or discriminate against low-criticality tasks by giving them very low scrubbing opportunities. Besides, exploring the architectural support for scrubbing problems may bring considerable potential for reliability improvements. However, this direction of scheduling-architecture co-optimization has not been well studied so far. In this paper, we propose a negotiation-based dynamic scrubbing framework, which addresses the abovementioned issues in three phases: First, we propose a negotiation-driven scrubbing scheduling algorithm, which temporarily allows and iteratively reduces the conflicts of scrubbing tasks in order to accommodate more scrubbing tasks to be scheduled. Second, we develop a logistic probability model to prevent scheduling starvation of a set of mix-criticality tasks by dynamically legalizing conflicting ones, considering both the criticality and schedulability of each task. Third, we develop a DVFS-based multi-ICAPs allocation algorithm to co-optimize with FPGA architectural features for reliability maximization. Compared to the state-of-the-art, experimental results show that our work achieves up to 31.46% improvement in terms of reliability for contemporary SRAM-based FPGAs.