Scrubbing optimization via availability prediction (SOAP) for reconfigurable space computing

Abstract

Reconfigurable computing with FPGAs can be highly effective in terms of performance, adaptability, and power for accelerating space applications, but their configuration memory must be scrubbed to prevent the accumulation of single-event upsets. Many scrubbing techniques currently exist, each with different advantages, making it difficult for the system designer to choose the optimal scrubbing strategy for a given mission. This paper surveys the currently available scrubbing techniques and introduces the SOAP method for predicting system availability for various scrubbing strategies using Markov models. We then apply the method to compare hypothetical Virtex-5 and Virtex-6 systems for blind, CRC-32, and Frame ECC scrubbing strategies in LEO and HEO. We show that availability in excess of 5 nines can be obtained with modern, FPGA-based systems using scrubbing. Furthermore, we show the value of the SOAP method by observing that different scrubbing strategies are optimal for different types of missions.