Exploiting error detection latency for parity-based soft error detection

Abstract

Local triple modular redundancy (LTMR) is often the first choice to harden a flash-based FPGA application against soft errors in space. Unfortunately, LTMR leads to at least 300% area overhead. We propose a parity-based error detection approach, to use the limited resources of space-proven flash-based FPGAs more area-efficiently; this method can be the key for fitting the application onto the FPGA. A drawback of parity-based hardening is the significant impact on the critical path. To alleviate this error detection latency, pipeline structures in the design can be utilized. According to our results, this eliminates from 22% to 65% of the critical path overhead of the unpipelined error detection. Compared with LTMR, the new approach increases the critical path overhead of LTMR by a factor varying from 2 to 7.