Real-time Error Monitoring System Considering Endurance and Data-retention Characteristics of TaOX-based ReRAM Storage with Workloads at Data Centers

Abstract

Approximate Computing attracts attention due to reducing power consumption and improving performance by tolerating errors. However, to use Approximate Computing, systems should control amounts of errors occurred in storage. This paper proposes real-time error monitoring system using ReRAM storage in order to understand how many errors occur in storage. To evaluate this system, measured ReRAM data and one-week long workload logs at data centers are input to the proposed system. The proposed system outputs Total bit error rate (BER) calculated from Set/Reset cycles and write interval time. In addition, the proposed system reveals that Total BER of some sectors exceeds the error correction limit set to 0.9 code rate of BCH ECC for a week.