Online Profiling for cluster-specific variable rate refreshing in high-density DRAM systems
Rasool Sharifi, Zainalabedin Navabi · 2017
Multi-rate refresh techniques are among the methods that use non-uniformity in retention time of DRAM cells to reduce the DRAM refresh overheads. Unfortunately, retention time of some DRAM cells may change unpredictably over time due to variable retention time (VRT). In this paper, we propose an Online Profiler that divides DRAM cells into clusters and proactively tests and measures retention time of each cluster over time. The Online Profiler decides on increasing refresh period of a cluster based on a measured retention time, where this retention time has passed all tests of different data sets. Also, for ensuring maximum data integrity, the Online Profiler reads the entire memory periodically for correction of possible errors. We show that our proposed mechanism, that uses cluster-specific variable rate refreshing, can provide reliable operation while reducing refresh overhead of the performance by 6%, 13%, and 23%, and Energy-Delay Product (EDP) by 7%, 13%, and 27% for 32GB, 64GB, and 128GB DRAM modules, respectively.