Applying non-volatile memories in future computer systems
| Author(s) | Khouzani, Hoda Aghaei | |
| Date Accessioned | 2019-05-15T12:14:41Z | |
| Date Available | 2019-05-15T12:14:41Z | |
| Publication Date | 2018 | |
| SWORD Update | 2019-02-15T20:04:07Z | |
| Abstract | The memory subsystem has been the performance and energy bottleneck of almost all computer systems. For dozens of years, the continuous advances of system design, applications, and technology make the role of memory subsystem more critical. Moreover, SRAM and DRAM as the mainstream on-chip and off-chip technologies are experiencing many technology scaling challenges that make their capacity, energy-efficiency, and reliability questionable. Fortunately, with the emergence of new memory technologies, a new path to address the limitations of these CMOS-based memories has appeared. These technologies are non-volatile and have the advantage of higher scalability beyond CMOS-based devices with near-zero static power consumption. ☐ In this dissertation, the applicability of two of these technologies, namely Phase Change Memory (PCM) and Domain Wall Memory (DWM) throughout the memory hierarchy is explored. In particular, PCM due to its close read attributes to DRAM, is employed as main memory, while DWM which has a comparable read/write latency to SRAM is employed throughout the whole hierarchy, namely main memory, CPU caches, and the stack in the processor. Alongside their benefits, PCM suffers from limited write endurance, long write latency, and high write energy, while the main challenge to employ DWM is imposed by its sequential access structure which requires shift operations to align the data of interest to the access port. Here, five different schemes are proposed to either hide these shortcomings or use them with respect to the system's advantage. ☐ Specifically, the first two techniques of this thesis consider the use of PCM as main memory. First, to hide the write performance and energy limitations of PCM, a hybrid DRAM-PCM main memory is applied, and a miss aware page allocation algorithm is proposed to further improve it. Then, a wear-resistant page allocation algorithm is proposed which leverages the existing segment information in OS to prolong PCM lifetime to the maximum extend with almost no extra overhead. The last three techniques in this thesis respectively employ DWM as main memory, lower-level CPU caches, and processor stack. In a DWM-based main memory, the latency of page table accesses is reduced using a pre-shift scheme which, based on the state of each entry in page table, pre-aligns the access port to hide read and write latency. Then, the impact of cache layout in DWM-based cache over the access latency is explored, and the most suitable layout is selected based on the data access pattern. At last, DWM is adopted in a stack-architecture in an energy harvesting system so as to be able to seamlessly recover from a power failure, and a three-folded scheme is presented to improve the performance of the proposed DWM-based stack-architecture. ☐ In sum, these five projects provide a road-map on the applicability of PCM and DWM in future computer systems, so as to take advantage of their non-volatility, high density, and near zero static power consumption. While several physical attributes create challenges in employing them directly in computer systems, architectural and runtime solutions provided in this dissertation can hide the negative impacts of these physical attributes and in some cases creatively use them with respect to the system's advantage. The results of experimental studies show that techniques in this dissertation can efficiently improve the lifetime, performance, and energy consumption of non-volatile memory hierarchy. | en_US |
| Advisor | Yang, Chengmo | |
| Degree | Ph.D. | |
| Department | University of Delaware, Department of Electrical and Computer Engineering | |
| DOI | https://doi.org/10.58088/aqm3-db27 | |
| Unique Identifier | 1101186389 | |
| URL | http://udspace.udel.edu/handle/19716/24158 | |
| Language | en | |
| Publisher | University of Delaware | en_US |
| URI | https://search.proquest.com/docview/2194351790?accountid=10457 | |
| Title | Applying non-volatile memories in future computer systems | en_US |
| Type | Thesis | en_US |