Exploration of Magnetic RAM Based Memory Hierarchy for Multicore Architecture
read more
Citations
Emerging NVM: A Survey on Architectural Integration and Research Challenges
Exploring MRAM Technologies for Energy Efficient Systems-On-Chip
A Light-Weighted Software-Controlled Cache for PCM-based Main Memory Systems
Comparative Analysis of Spintronic Memories for Low Power on-chip Caches
A Survey of Low Power Design Techniques for Last Level Caches
References
The gem5 simulator
NVSim: A Circuit-Level Performance, Energy, and Area Model for Emerging Nonvolatile Memory
CACTI 6.0: A Tool to Model Large Caches
A novel architecture of the 3D stacked MRAM L2 cache for CMPs
Hybrid cache architecture with disparate memory technologies
Related Papers (5)
Frequently Asked Questions (12)
Q2. What are the main benefits of using STT-MRAM?
Optimizations techniques such as early write termination which prevent unnecessary writes, or write buffers, to deal with high write latency and high write dynamic energy of MRAM were proposed in [9] and [10].
Q3. What is the name of the simulator?
NVSIM [2], a modified environment of CACTI [3], is a circuit-level model for NVM performance, energy, and area estimation, which supports various NVM technologies, including STT-MRAM, PCRAM, RRAM, and legacy NAND Flash.
Q4. How much static energy is saved for the L1 cache?
For total L1 cache, i.e. including all the L1 caches of each core, the authors save more than 80%, 40%, and 25% of static energy for the STT_SRAM, iSTT/dSRAM_SRAM and dSTT/iSRAM_SRAM scenarios respectively.
Q5. What is the critical part of the memory hierarchy?
Since these benchmarks compute a very large amount of data, the most critical part in the memory hierarchy is the L1 D-cache memory.
Q6. What is the main purpose of this article?
GEM5 can simulate a complete processor-based system with devices and operating system in full system mode and it supports also simulation of multi-core systems.
Q7. Why is the write access energy higher in STT-MRAM?
Because intrinsically non-volatile, STT-MRAM cell has zero standby power, and the CMOS access transistor does not need to be power supplied.
Q8. How much static energy is saved by using STT-MRAM?
their objective is to explore all cache memory hierarchy strategies directly replacing SRAM with MRAM, taking into account that, for instance, MRAM can be up to seven times larger than SRAM for a same die footprint.
Q9. What is the purpose of this article?
EXPERIMENTAL SETUPFor their study, the authors propose to use some applications of SPLASH-2 benchmark suite [6], which are mostly in the area of High Performance Computing (HPC), to evaluate the impact of STT-MRAM for shared L2 cache on four-core processor architecture and its impact for L1 cache on two-core architecture.
Q10. What is the purpose of this study?
For future work, the authors plan to extend this study with the Thermally Assisted Switching MRAM technology whose implementation can lead to Magnetic Logic Unit (MLU) [12] presenting new logic functionalities compared with a standard MRAM.
Q11. Why is the L2 cache more accessed by read operations?
because L2 cache is much more accessed by read operations, the total L2 dynamic energy is not so high using STT-MRAM instead of SRAM.
Q12. What is the difference between STT-MRAM and SRAM?
But the considerable gain of STT-MRAM over SRAM is on the leakage power: STT-MRAM is more than 10x less power consuming than SRAM.