当前位置:网站首页>非易失性MRAM存储器应用于各级高速缓存
非易失性MRAM存储器应用于各级高速缓存
2020-11-06 22:00:00 【IPSILOG】
磁阻式随机存储器MRAM是一种新型存储器,其优点有读取速度快和集成度高及非挥发性等。目前许多研究主要是致力于将MRAM运用于计算机存储系统中。同时非易失性MRAM存储器也应用于各级高速缓存。
MRAM替代SRAM做L2高速缓存
首先比较具有同样面积的MRAM和SRAM。直接用相同面积的MRAM替换SRAM作L2高速缓存能降低错误率。但是写入延时较长。当写入操作强度高时,错误率降低的优势会被长延时所抵消导致性能下降。虽然这种直接替代能大大降低漏功耗,但当写入密集时,动态功耗显著增加,使减少能耗的效果变差。若直接用相同面积的MRAM替代sram ,在写入操作较密集时,其写入长延时和高能耗等缺点会抵消其优势。
MRAM作为L3高速缓存
L2容量过大会增加存取延时所以不适用。在存储体系中增加一级L3高速缓存的可行性。研究者计算出一个128MB,4-banks,16-way,256-byte block的高速缓存面积只有161mm2,适合堆叠在目前的处理器上。时间模型表明其延时只有15.82ns,远少于存储器平均存取时间。在不同的情况下的IPC速度增加了0.03%到108%。对L2高速缓存错误率较高的情况有很大改进。并且这种改进只需要增加0.4W的功耗。
MRAM用作主存
有关研究已证明片上堆叠DRAM存储器的可行性。与DRAM相比的MRAM不需要周期性刷新。但是目前还是DRAM的集成度最高。目前堆叠DRAM技术的性能提高为19%(对于整数)和40%(对于浮点数)。我们有理由相信堆叠MRAM技术因其具有更短的延时而会有更好的表现。
虽然MRAM低功耗的特点使其能够实现多层堆叠而不用担心温度方面的问题,但是延时会增加,而且堆叠层数过多会导致成品率下降。因此目前的MRAM技术要应用于主存储器还不够成熟(因为其容量不够大) ,但是可以用于对低功耗有特殊要求的嵌入式设备。
版权声明
本文为[IPSILOG]所创,转载请带上原文链接,感谢
https://my.oschina.net/u/4542041/blog/4704417
边栏推荐
- Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
- Ronglian completed US $125 million f round financing
- An article will take you to understand CSS3 fillet knowledge
- Zero basis to build a web search engine of its own
- From overseas to China, rancher wants to do research on container cloud market
- StickEngine-架构12-通信协议
- ES6 learning notes (3): teach you to use js object-oriented thinking to realize the function of adding, deleting, modifying and checking tab column
- Junit测试出现 empty test suite
- 事务的本质和死锁的原理
- 2020-08-15:什么情况下数据任务需要优化?
猜你喜欢

An article taught you to use HTML5 SVG tags

JVM内存分配 -Xms128m -Xmx512m -XX:PermSize=128m -XX:MaxPermSize=512m

The AI method put forward by China has more and more influence. Tianda et al. Mined the development law of AI from a large number of literatures

How much disk space does a new empty file take?

Digital city responds to relevant national policies and vigorously develops the construction of digital twin platform

华为云微认证考试简介

2020-08-17:详细说下数据倾斜怎么解决?

ORA-02292: 违反完整约束条件 (MIDBJDEV2.SYS_C0020757) - 已找到子记录

An article will take you to understand SVG gradient knowledge

代码生成器插件与Creator预制体文件解析
随机推荐
Flink's datasource Trilogy 2: built in connector
ES6 learning notes (4): easy to understand the new grammar of ES6
StickEngine-架构11-消息队列(MessageQueue)
The legality of IPFs / filecoin: protecting personal privacy from disclosure
Building a new generation cloud native data lake with iceberg on kubernetes
An article will introduce you to HTML tables and their main attributes
2020-08-20:GO语言中的协程与Python中的协程的区别?
IPFs rudder filecoin landing at the same time, fil currency price broke a thousand
[self taught unity2d legendary game development] map editor
Zero basis to build a web search engine of its own
Can you do it with only six characters?
The AI method put forward by China has more and more influence. Tianda et al. Mined the development law of AI from a large number of literatures
解决 WPF 绑定集合后数据变动界面却不更新的问题
Basic usage of Vue codemirror: search function, code folding function, get editor value and verify in time
Swagger 3.0 brushes the screen every day. Does it really smell good?
Call analysis of start method in JNI thread and callback analysis of run method
2020年第四届中国 BIM (数字建造)经理高峰论坛即将在杭举办
To Lianyun analysis: why is IPFs / filecoin mining so difficult?
Introduction to the development of small game cloud
谷歌浏览器实现视频播放加速功能