A High Performance and Energy-Efficient Cold Data Eviction Algorithm for 3D-TSV Hybrid ReRAM/MLC NAND SSD

• Published in: IEEE transactions on circuits and systems. I, Regular papers Regular papers, 2014-02, Vol.61 (2), p.382-392
• Main Authors: Chao Sun, Miyaji, Kousuke, Johguchi, Koh, Takeuchi, Ken
• Format: Article
• Language: English
• Subjects: 3D-TSV; Adaptive algorithms; Algorithms; Circuits; Endurance; Energy consumption; Flash memory (computers); Information storage; Migration; MLC NAND flash; Power consumption; Random access memory; ReRAM; SSD
• ISSN: 1549-8328; 1558-0806
• DOI: 10.1109/TCSI.2013.2268111

A hybrid 3D-TSV ReRAM/MLC NAND SSD with cold data eviction (CDE) algorithm is proposed. In the proposed hybrid SSD, the lifetime and energy consumption are dominated by MLC NAND flash memory due to ReRAM's high endurance and low power consumption. In addition, partial page overwrites are possible in ReRAM. Thus, the write accesses to MLC NAND flash memory are largely reduced by storing hot data in ReRAM. As a result, the SSD energy consumption decreases and the lifetime is prolonged. With the CDE algorithm, a page-level adaptive data migration is achieved, which is transparent to the file system. Compared to the previous work, 8-times write throughput increase, 83% energy reduction and 6.5-times longer longevity are achieved with 3D-TSV technology. Moreover, from the experimental results, the data eviction should be triggered when ReRAM free space ratio decreases to a range of 8%-20%. Hence, the eviction frequency is adaptive to the data pattern in the hybrid SSD. The experimental results also suggest the requirements for ReRAM. To obtain the best effect, both the read and write latency of ReRAM should be below 3 μs for 512 Bytes.