Date post: | 29-Nov-2014 |
Category: |
Documents |
Upload: | flashdomain |
View: | 1,594 times |
Download: | 4 times |
Flash-Memory Storage Systems: Design Issues and Challenges
Department of Computer Science & Information Engineering
National Taiwan University
Tei-Wei Kuo
2008/7/14 Embedded Systems and Wireless Networking Lab. 2
Agenda
IntroductionManagement IssuesBehavior AnalysisReliability – Wear LevelingOther Challenging IssuesConclusion
2008/7/14 Embedded Systems and Wireless Networking Lab. 3
Introduction – Why Flash MemoryDiversified Application Domains
Portable Storage DevicesConsumer ElectronicsIndustrial ApplicationsCritical System Components
2008/7/14 Embedded Systems and Wireless Networking Lab. 4
Trends in VLSI Technology
Source: www.icknowledge.com
* This slide was from the ASP-DAC’06 talk delivered by Prof. SangLyul. Min from the Seoul National University.
Flash Makers – NAND Flash Memory
2008/7/14 Embedded Systems and Wireless Networking Lab. 5
http://hugoleijtens.spaces.live.com/blog/cns!4B94B7453D4BFD9E!988.entry
Source: iSuppli Corp (Unit: Million Dollars)
[EE Times,11/30/2007]For years, NAND priceshave dropped by anaverage of 40 percentor more per year.
2008/7/14 Embedded Systems and Wireless Networking Lab. 6
Trends – Storage Media
Source: Using multilevel cell NAND flash technology in consumer applications, Electronic Engineering Times, July ,2005
Samsung 2GB USB 2.0 Flash DrivePrice: $49.99Less Rebate: - $25.00 Final Price: $24.99*
T-One 2GB Microdrive/3600RPM$144.99
September 2006
2008/7/14 Embedded Systems and Wireless Networking Lab. 7
Trends – Storage Media
Source: Using multilevel cell NAND flash technology in consumer applications, Electronic Engineering Times, July ,2005. Amazon.com
Transcend 8GBCompactFlash CardPrice: $84.85
Microdrive 4GBCompact Flash Type II
Price: $116
SanDisk 4GBCompactFlash CardPrice: $55.99
March 2007
Trends – Storage Media
2008/7/14 Embedded Systems and Wireless Networking Lab. 8
Source: Using multilevel cell NAND flash technology in consumer applications, Electronic Engineering Times, July ,2005Component Times, Nov 2007.; March/July 2008, www. amazon.com
20Gb IBM HDD
Transcend 8GB SDHC SD CARD (USD23.77)
HITACHI 6GB Microdrive MD6GBBP (USD169.95, Feb 2008No Longer Carried in July)
July 2008
Seagate FreeAgent Desktop 500 GB 3.5" USB 2.0 External Hard Drive (USD104.99)
Transcend 32GB SSD, 2.5- Inch, SATA, MLC (USD343.30 (2008/03) USD202.79 (2008/07))
Flash Stories – Solid-State Disks< 50% Heat, Ultra Silence, Light Weight, MTTF – 200M hours (HDD – 30M hours), Energy Efficiency
2008/7/14 Embedded Systems and Wireless Networking Lab. 9IDC, SanDisk - Component Times, Nov 2007
Flash Stories – Flash WarsFab
300mm Wafer Fab by Toshiba and SanDisk at Yokkaichi, Japan
80,000 Wafers per Month (2008)210,000 Wafers per Month
300mm Wafer Fab by IM Flash
Technologies (Intel & Micron) at UtahJoint Venture: Sony and Qimonda, Hynix and Sandisk
Technology32nm (Samsung, Intel), 43nm (Toshiba), 50nm (IM)
2008/7/14 Embedded Systems and Wireless Networking Lab. 10
Introduction – The Characteristics of Storage Media
*Flash Erase Time: NOR(10ms), NAND (1.5ms)[Reference] DRAM: DDR-400. NOR FLASH: Intel 28F128J3A-150. NAND FLASH: Samsung K9K8G08U0M. Disk: Segate Barracuda ATA II.11. Jian-Hong Lin, Yuan-Hao Chang, Jen-Wei Hsieh, Tei-Wei Kuo, and Cheng-Chih Yang, "A NOR Emulation Strategy over NAND Flash Memory," the 13th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA), Daegu, Korea , August 21-24, 2007.
MediaAccess time
Read Write Erase
DRAM 5ns (1B)2.56us (512B)
5ns (1B)2.56us (512B)
-
NOR FLASH 150ns (1B)14.4us (512B)
211us (1B)3.52ms (512B)
1.2s (16KB)
NAND FLASH 20us (1B)32.8us (512B)
200us (1B)212us (512B)
1.5ms (128KB)
DISK 12.4ms (512B)(average)
12.4 ms(512B)(average)
-
13X 83X
400X 50X
2008/7/14 Embedded Systems and Wireless Networking Lab. 12
Introduction – Single-Level Cell (SLC)
Selected cellIDS
Control GateDrain
Source
Each Word Line is connected to control gates.Each Bit Line is connected to the drain.
Cell
2008/7/14 Embedded Systems and Wireless Networking Lab. 13
Introduction – Multi-Level Cell (MLC) vs SLC
NU
MB
ER
OF
CE
LLS
2008/7/14 Embedded Systems and Wireless Networking Lab. 14
Introduction – Consumer Applications
Electronic Engineering Times, July 2005
2008/7/14 Embedded Systems and Wireless Networking Lab. 15
Bandwidth Requirements – Video
ˇ
ˇ
Electronic Engineering Times, July 2005
2008/7/14 Embedded Systems and Wireless Networking Lab. 16
Bandwidth Requirements – Audio
ˇ
ˇ
Electronic Engineering Times, July 2005
2008/7/14 Embedded Systems and Wireless Networking Lab. 17
Introduction – Challenges in Flash-Memory Storage Designs
Requirements in Good PerformanceLimited Cost per Unit Increasing in Access FrequenciesStrong Demands in ReliabilityTight Coupling with Other ComponentsLow Compatibility among Vendors
2008/7/14 Embedded Systems and Wireless Networking Lab. 18
Agenda
IntroductionManagement IssuesBehavior AnalysisReliability – Wear LevelingOther Challenging IssuesConclusion
2008/7/14 Embedded Systems and Wireless Networking Lab. 19
Management Issues – System Architectures
AP
File-System Layer
AP AP
Block Device Layer (FTL emulation)
Flash Memory
MTD drivers
AP
2008/7/14 Embedded Systems and Wireless Networking Lab. 20
Management Issues – Flash-Memory Characteristics
……
Block 0Block 1
Block 2Block 3
Erase one block
1 Page = 512B1 Block = 32 pages(16KB)
……
Write one page
2008/7/14 Embedded Systems and Wireless Networking Lab. 21
Management Issues – Flash-Memory Characteristics
Write-OnceNo writing on the same page unless its residing block is erased!Pages are classified into valid, invalid, and free pages.
Bulk-Erasing Pages are erased in a block unit to recycle used but invalid pages.
Wear-LevelingEach block has a limited lifetime in erasing counts.
2008/7/14 Embedded Systems and Wireless Networking Lab. 22
Management Issues – Flash-Memory Characteristics
Example 1: Out-place Update
Live pages Free pages
A B C D
Suppose that we want to update data A and B…
2008/7/14 Embedded Systems and Wireless Networking Lab. 23
Dead pages
A B C D A B
Management Issues – Flash-Memory Characteristics
Example 1: Out-place Update
2008/7/14 Embedded Systems and Wireless Networking Lab. 24
A live pageA dead pageA free page
This block is to be recycled. (3 live pages and 5 dead pages)
L D D L D D L D
L L D L L L F D
L F L L L L D F
F L L F L L F D
Management Issues – Flash-Memory Characteristics
Example 2: Garbage Collection
2008/7/14 Embedded Systems and Wireless Networking Lab. 25
L L D L L L D
L F L L L L D
L L F L L F D
L
L
D D D D
A live pageA dead pageA free page
Live data are copied to somewhere else.
L
D DDD
Management Issues – Flash-Memory Characteristics
Example 2: Garbage Collection
2008/7/14 Embedded Systems and Wireless Networking Lab. 26
A live pageA dead pageA free page
The block is then erased.
Overheads: •live data copying •block erasing.
L L D L L L D
L F L L L L D
L L F L L F D
L
L
F F F F F F F F
L
Management Issues – Flash-Memory Characteristics
Example 2: Garbage Collection
2008/7/14 Embedded Systems and Wireless Networking Lab. 27
Management Issues – Flash-Memory Characteristics
Example 3: Wear-Leveling
L D D L D D L D
L L D L L L F D
L F L L L L D F
F L L F L L F D
100
10
20
15
Erase cycle counts
Wear-leveling might interfere with the decisions of the block-recycling policy.
A live pageA dead pageA free page
A
B
C
D
2008/7/14 Embedded Systems and Wireless Networking Lab. 28
Management Issues – Challenges
The write throughput drops significantly after garbage collection starts!The capacity of flash-memory storage systems increases very quickly such that memory space requirements grows quickly.Reliability becomes more and more critical when the manufacturing capacity increases!The significant increment of flash-memory access rates seriously exaggerates the Read/Program Disturb Problems!
2008/7/14 Embedded Systems and Wireless Networking Lab. 29
Agenda
IntroductionManagement IssuesBehavior Analysis
Performance vs Overheads – FTL vs NFTLTrace Study
Reliability – Wear LevelingOther Challenging IssuesConclusion
2008/7/14 Embedded Systems and Wireless Networking Lab. 30
System Architecture
GarbageCollection
AddressTranslation
FTL/NFTLLayer
File system (FAT, EXT2, NTFS......)
Device Driver
fwrite(file,data)
Block write(LBA,size)
Flash I/O Requests
Controlsignals
File Systems
process processprocess Applications
Flash-MemoryStorage System
Physical Devices(Flash Memory Banks)
process
31
31
System Architecture
*FTL: Flash Translation Layer, MTD: Memory Technology Device
SD, xD,MemoryStick,SmartMedia
CompactFlash
2008/1/30 Embedded Systems and Wireless Networking Lab. 312008/7/14
2008/7/14Embedded Systems and Wireless Networking Lab. 32
Policies – FTLFTL adopts a page-level address translation mechanism.
The main problem of FTL is on large memory space requirements for storing the address translation information.
2008/7/14 Embedded Systems and Wireless Networking Lab. 33
Policies – NFTL (Type 1)
.
.
.
(9)
Write data to LBA=1011
.
.
.
NFTLAddress Translation Table
(in main-memory)
FreeFreeFreeUsedFreeFreeFreeFree
FreeFreeFreeFreeFreeFreeFreeFree
A Chain Block
Address = 9
A Chain Block
Address = 23
VBA=126
Block Offset=3
If the page has been used
Write to the page with block
offset=3
A logical address under NFTL is divided into a virtual block address and a block offset.
e.g., LBA=1011 => virtual block address (VBA) = 1011 / 8 = 126 and block offset = 1011 % 8 = 3
FreeFreeFreeFreeFreeFreeFreeFree
A Chain Block
Address = 50
Used
Write to the page with block
offset=3
2008/7/14 Embedded Systems and Wireless Networking Lab. 34
Policies – NFTL (Type 2)A logical address under NFTL is divided into a virtual block address and a block offset.
e.g., LBA=1011 => virtual block address (VBA) = 1011 / 8 = 126 and block offset = 1011 % 8 = 3
.
.
.
(9,23)
Write data to LBA=1011
.
.
.
NFTLAddress Translation Table
(in main-memory)
FreeFreeFreeUsedFreeFreeFreeFree
UsedUsedUsedFreeFreeFreeFreeFree
A Primary Block
Address = 9
A Replacement Block
Address = 23
VBA=126
Block Offset=3
If the page has been usedWrite to the
first free page
2008/7/14 Embedded Systems and Wireless Networking Lab. 35
Policies – NFTL
NFTL is proposed for the large-scale NAND flash storage systems because NFTL adopts a block-level address translation.
However, the address translation performance of read and write requests might deteriorate, due to linear searches of address translation information in primary and replacement blocks.
2008/7/14 Embedded Systems and Wireless Networking Lab. 36
Policies – FTL or NFTLFTL NFTL
Memory Space Requirements Large SmallAddress Translation Time Short Long
Garbage Collection Overhead Less MoreSpace Utilization High Low
The Memory Space Requirements for one 1GB NAND (512B/Page, 4B/Table Entry, 32 Pages/Block)
FTL: 8,192KB (= 4*(1024*1024*1024)/512)NFTL: 256KB (= 4*(1024*1024*1024)/(512*32))
Remark: Each page of small-block(/large-block) SLC NAND can store 512B(/2KB) data, and there are 32(/64) pages per block. Each page of MLCx2 NAND can store 2KB, and there are 128 pages per block.
2008/7/14 Embedded Systems and Wireless Networking Lab. 37
Address Translation Time - NFTLThe address translation performance of read and write requests can be deteriorated, due to linear searches of physical addresses.
1. Assume that each block contains 8 pages.
2. Let LBA A, B, C, D, and E be written for 5, 5, 1, 1, and 1 times, respectively. Their data distribution could be like to what in the left figure.
3. For example, it might need to scan 9 spare areas for LBA B.
2008/7/14 Embedded Systems and Wireless Networking Lab. 38
Garbage Collection Overhead - NFTL
3. Overhead is 2 block erases and 5 page writes.
1. Copy the most-recent content to the new primary block.2. Erase the old primary block and the replacement block.
2008/7/14 Embedded Systems and Wireless Networking Lab. 39
Space Utilization - NFTL
3 free pages are wasted.
2008/7/14 Embedded Systems and Wireless Networking Lab. 40
Agenda
IntroductionManagement IssuesBehavior Analysis
Performance vs Overheads – FTL vs NFTLTrace Study
Reliability – Wear LevelingOther Challenging IssuesConclusionP.-C. Huang, Y.-H. Chang, J.-W. Hsieh, and M. Lin, “The Behavior Analysis of Flash-Memory Storage Systems, IEEE ISORC’08.
Flash Storage System Architecture
2008/7/14 Embedded Systems and Wireless Networking Lab. 41
Peak Read/Write Throughput
The sequential access pattern usually yields the best-case throughput of a flash-memory storage system
It is due to the address mapping and garbage collection overheads
ith access
LBA
Largest request packet length
2008/7/14 Embedded Systems and Wireless Networking Lab. 42
Worst-Case Write Response TimeAccess patterns with small accesses to trigger garbage collection activities
The first pass fills the whole flash memory to force out-place updatesThe second pass actually triggers the updates and garbage collection
LBA
ith access
Pass1 Pass2
2008/7/14 Embedded Systems and Wireless Networking Lab. 43
Reliability - First-Failure Time
Access patterns that repetitively write to some specific LBA’s to see whether the management facilities can evenly distribute the writes to PBA’s.
Random accesses are inserted to avoid the caching effects
LBA
ith access
2008/7/14 Embedded Systems and Wireless Networking Lab. 44
2008/7/14 Embedded Systems and Wireless Networking Lab. 45
Agenda
IntroductionManagement IssuesBehavior AnalysisReliability – Wear LevelingOther Challenging IssuesConclusion
Y.-H. Chang, J.-W. Hseuh, and T.-W. Kuo,, “Endurance Enhancement of Flash-Memory Storage Systems: An Efficient Static Wear Leveling Design,” ACM/IEEE 44-th Design Automation Conference (DAC), San Diego, USA, June 2007. [Best Paper Nomination]
2008/7/14 Embedded Systems and Wireless Networking Lab. 46
Wear Leveling versus Product LifetimeSettings
File system: FAT16 file system with 8KB cluster sizeFlash memory: 256MB small-block flash memory with 100K erase cycles
Updating of a 16MB file repeatedly with the throughput: 0.1MBs
The file requires 2K clusters = 16MB ÷ 8KB(cluster size)The FAT size of this file is 4KB (2K(clusters) x 2 bytes)The 20% of blocks in flash memory joins the dynamic wear leveling Data of a 16MB file is stored in 1K blocks (16MB ÷ 16KB(block size))Suppose flash memory is managed in the block level
File systems update the FAT in each cluster writing so that FAT is updated 2Ktimes for a 16MB fileWriting of a 16MB incurs 1K block erases because of the reclaiming of invalid space.
2008/7/14 Embedded Systems and Wireless Networking Lab. 47
Wear Leveling versus Product Lifetime
Ways in Data UpdatesIn-Place-Updates: Rewriting on the Same PageDynamic Wear Leveling: Rewriting over Another Free Page with Erasing over Blocks with Dead PagesStatic Wear Leveling: Rewriting over Another Free Page with Erasing over Any Blocks
Expected Lifetime of
)days(5.987606024)12(
1000%01(blocks)16ond)0.1(MB/sec
16(MB) Leveling Wear Static
)days(5.197606024)12(
10020%(blocks)16ond)0.1(MB/sec
16(MB) Leveling Wear Dynamic
)days(09.06060242
100ond)0.1(MB/sec
16(MB) update) place-(in Leveling Wear NO
≈×××+××
×=
≈×××+
×××=
≈×××
×=
KKKK
KKKK
KK
2008/7/14 Embedded Systems and Wireless Networking Lab. 48
Use a counter for each blockThe garbage collector always finds the block with the least erase count.
Some heuristic approach erases a block to maintain 2 free blocks when the garbage collector finds the erase count of the block is over a given threshold.
Problems:High extra block erases and live-page copyingsHigh main-memory consumptionHigh computation cost
Wear Leveling versus Product Lifetime Static Wear Leveling – Block-Level Mapping
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15FlashMemory
Counter
: free block : dead block
: block contains (some) valid data
: index in the selection of a victim block
: index to the selected free block 5 5 4 55 4 45 45 5 44 55 4
Update data in block 3Write new data to block 4GC starts find a victim block
5
Update block 15GC starts
5 5 5 5
49
Comparison of Different Technologies
Physical Block Addresses (PBA)
Eras
eC
ycle
s
0 20 40 60 80 1000
1000
2000
3000
4000
5000Perfect SWL
Physical Block Addresses (PBA)0 20 40 60 80 100
0
1000
2000
3000
4000
5000Intuitive SWL
Eras
eC
ycle
s
0 20 40 60 80 1000
1000
2000
3000
4000
5000
No Wear Leveling
0 20 40 60 80 1000
1000
2000
3000
4000
5000
Dynamic Wear Leveling
2008/7/14 Embedded Systems and Wireless Networking Lab.
50
An Efficient Static Wear Leveling Mechanism
A modular design for compatibility considerations A SWL mechanism
Block Erasing Table (BET)
bit flags
SW LevelerSWL-ProcedureSWL-Update
2008/7/14 Embedded Systems and Wireless Networking Lab.
51
The Block Erasing Table (BET)A bit-array: Each bit is for 2k consecutive blocks.
Small k – in favor of hot-cold data separationLarge k – in favor of small RAM space
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Flash
BET 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 1 1 1
k=0 k=2
ecnt=0fcnt =0
ecnt=0fcnt =0
ecnt=1fcnt =1ecnt=2fcnt =2ecnt=3fcnt =2
ecnt=1fcnt =1ecnt=2fcnt =2ecnt=3fcnt =2ecnt=4fcnt =2
: a block that has been erased in the current resetting interval: an index to a block that the Cleaner wants to erase
fcnt: the number of 1’s in the BET ecnt: the total number of block erases done since the BET is reset
2008/7/14Embedded Systems and Wireless Networking Lab.
52
Endurance Improvement - The First Failure Time
When k=3 and T=100, the endurance is improved by
100.2%.
When k=0 and T=100, the endurance is improved by
87.5%.
2008/7/14 Embedded Systems and Wireless Networking Lab.
53
SummaryAn Efficient Static Wear Leveling Mechanism
Small Memory Space RequirementLess than 1KB for 8GB flash memory
Efficient ImplementationAn adjustable house-keeping data structureAn cyclic queue scanning algorithm
Performance EvaluationImprovement Ratio of the Endurance:
The ratio is more than 50%
Extra Overhead: It is less than 3% of extra block eraseThere is limited overhead in live-page copyings
2008/7/14 Embedded Systems and Wireless Networking Lab.
2008/7/14 Embedded Systems and Wireless Networking Lab. 54
Agenda
IntroductionManagement IssuesBehavior AnalysisReliability – Wear LevelingOther Challenging IssuesConclusion
2008/7/14 Embedded Systems and Wireless Networking Lab. 55
Challenging Issues – Reliability
Selected cellIDS
Control GateDrain
Source
Each Word Line is connected to control gates.Each Bit Line is connected to the drain.
Cell
2008/7/14 Embedded Systems and Wireless Networking Lab. 56
Challenging Issues – ReliabilityRead Operation
When the floating gate is not charged with electrons, there is current ID (100uA)if a reading voltage is applied. (“1” state)
5V
1V
Program OperationElectrons are moved into the floating gate, and the threshold voltage is thus raised.
2008/7/14 Embedded Systems and Wireless Networking Lab. 57
Challenging Issues – ReliabilityOver-Erasing Problems
Fast Erasing Bits All of the cells connected to the same bit line of a depleted cell would be read as “1”, regardless of their values.
Read/Program Disturb ProblemsDC erasing of a programmed cell, DC programming of a non-programmed cell, drain disturb, etc.Flash memory that has thin gate oxide makes disturb problems more serious!
Data Retention ProblemsElectrons stored in a floating gate might be lost such that the lost of electrons will sooner or later affects the charging status of the gate!
2008/7/14 Embedded Systems and Wireless Networking Lab. 58
Challenging Issues – ObservationsThe write throughput drops significantly after garbage collection starts!The capacity of flash-memory storage systems increases very quickly such that memory space requirements grows quickly.Reliability becomes more and more critical when the manufacturing capacity increases!The significant increment of flash-memory access rates seriously exaggerates the Read/Program Disturb Problems!Wear-leveling technology is even more critical when flash memory is adopted in many system components or might survive in products for a long life time!
2008/7/14 Embedded Systems and Wireless Networking Lab. 59
ConclusionWhat Is Happening?
Solid-State Storage DevicesNew Designs in the Memory HierarchyMore Applications in System Components and Products
Challenging Issues: Performance, Cost, and Reliability
Scalability TechnologyReliability TechnologyInterdisciplinary Work
2008/7/14 Embedded Systems and Wireless Networking Lab. 60
Contact Information
• Professor Tei-Wei [email protected]: http://csie.ntu.edu.tw/~ktwFlash Research: http://newslab.csie.ntu.edu.tw/~flash/Office: +886-2-23625336-257Fax: +886-2-23628167Address: Dept. of Computer Science & Information Engr.National Taiwan University, Taipei, Taiwan 106
Q & A
2008/7/14 Embedded Systems and Wireless Networking Lab. 62
1-bit/Cell SLC NAND Flash100,000 Program/Erase cycles (with ECC)[1]10 years Data Retention[1]
2-bits/Cell MLC NAND Flash10,000 Program/Erase cycles (with ECC) [2]10 years Data Retention[2]
4-bits/Cell MLC NAND FLASH Developers (2006)
M-systems, Intel, Samsung, and Toshiba[1] ST Micro-electronics NAND SLC large page datasheet (NAND08GW3B2A)[2] ST Micro-electronics NAND MLC large page datasheet (NAND04GW3C2A)•USD34.65 per GB for NOR, •USD7.10 (SLC) and USD2.48 (MLCx2) per GB for NAND in 2008 Q1
Comparison of SLC and MLC
Introduction – Price and Read/Write Performance
NOR NAND SLC NAND MLCx2
Price $34.55/GB $6.79/GB $2.48/GB
Read 23.84 MB/sec 15.33 MB/sec 13.5 MB/sec
Write 0.07 MB/sec 4.57 MB/sec 2.34 MB/sec
Erase 0.22 MB/sec 85.33 MB/sec 170.66 MB/sec
*NOR: Silicon Storage Technology (SST). NAND SLC: Samsung Electronics. K9F1G08Q0M. NAND MLCx2: ST STMicroelectronics[1,2]
1. Jian-Hong Lin, Yuan-Hao Chang, Jen-Wei Hsieh, Tei-Wei Kuo, and Cheng-Chih Yang, "A NOR Emulation Strategy over NAND Flash Memory," the 13th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA), Daegu, Korea , August 21-24, 2007.2. Yuan-Hao Chang and Tei-Wei Kuo, “A Log-based Management Scheme for Low-cost Flash-memory Storage Systems of Embedded Systems”
Non-Volatile Memory
2008/7/14 Embedded Systems and Wireless Networking Lab. 64
Macronix International Co., Component Times, Nov. 2007
Performance Comparison(Still Under Development)
RAM Type R/W Speed Lifetime / Endurance Density Cost
MRAM 25~100ns/1B3, 2ns/1B in experiment5
Highly reliable, 20 yr data retention
1~4MB1GB TBA2 4USD/1MB
PRAM 50ns/1B Highly reliable, 1013 W/E cycles 4~64MB N/A
Racetrack Memory 20~32ns/1B4 Highly reliable1 Higher than
MRAM1Lower than
MRAM1
References1. http://www.digitimes.com.tw/EDM/EDM_DTF970122END/H11.pdf2. http://www.engadget.com/2008/06/02/toshiba-says-its-1gb-mram-chips-are-almost-ready-were-ready/3. http://203.66.161.5/document/mic_digi/Topology/report_topology/IA/2003/030806MRAM.pdf4. http://en.wikipedia.org/wiki/Racetrack_memory5. http://en.wikipedia.org/wiki/Magnetoresistive_Random_Access_Memory