MORSE: Memory Overwrite Time Guided Soft Writes to Improve ReRAM Energy and Endurance

ReRAM is an attractive main memory technology due to its high density and low idle power. However, ReRAM exhibits costly writes, especially in terms of energy and endurance. Prior device studies show that retention can be traded off for write energy and endurance by employing soft write operations with lower currents. But given their reduced retention times, soft writes require refresh operations to prevent data loss. Unfortunately, a large number of refreshes are needed in between writes to infrequently updated data. Hence, a non-volatile memory system with soft writes still needs traditional hard writes, and a way to choose between them.

The efficacy of soft writes hinges on the time gap between consecutive writes to the same data, which we call the overwrite time. If the combined cost of a soft write and its refreshes within the overwrite time window is less than that of a hard write, the original soft write is profitable. Otherwise, it would have been better to perform a hard write to eliminate the refreshes.

To address this, we propose MORSE, a predictor that learns the overwrite times between back-to-back writes to main memory, and associates them with static store instructions in a prediction table. As dynamic stores execute, MORSE predicts the optimal write type based on the predicted overwrite time’s magnitude. This soft write decision is placed in the cache hierarchy, and eventually informs the writeback to main memory to use either a soft or hard write. Our results show MORSE provides 2.5x - 4.1x improvement in endurance and 2.9x - 4.2x reduction in write energy over a state-of-the-art predictor. Moreover, we demonstrate that MORSE is within 19.8% of the Oracle policy. Finally, we integrate MORSE with a prior wear leveling technique, called Ouroboros, and show that MORSE improves actual memory system lifetime by 6.2x over a baseline that only employs hard writes.