Micron Technology, Inc. · NASDAQ:MU
Micron Technology — Flash Memory Summit 2008
19 slides from Micron Technology's conference presentation, August 2008
Slide 1: Endurance Requirements for Next-Generation SLC NAND on Mobile Systems
Endurance Requirements for Next-Generation SLC NAND on Mobile Systems Khurram Ismail Sr. Applications Engineer Micron Technology, Inc. ©2008 Micron Technology, Inc. All rights reserved. Products are warranted only to meet Micron’s production data sheet specifications. Information, products and/or sp
Slide 2: Agenda
Agenda Overview Program/erase cycling effects on NAND Flash How to minimize program/erase cycling effects Cycling of Flash devices in mobile systems • Usage models and their effects on NAND Flash • Usage models and how to predict the life of NAND Flash in a mobile system Santa Clara, CA USA
Slide 3: Overview
Overview In the mobile market, the SLC 100K endurance requirement has remained unchanged since the small-page NAND With the advancement in process nodes, it is becoming difficult to achieve 100,000 program/erase cycles As the densities/page sizes are increasing, should the requirement be same
Slide 4: Program/Erase Cycling (Endurance)
Program/Erase Cycling (Endurance) Program/erase (high I/O I/O I/O I/O voltage stress) cycles cause charge to be trapped in the dielectric Causes a permanent shift Damaged in cell characteristics—not Cells recovered by erase Observed as failed program or erase status Blocks that fail should b
Slide 5: Data Retention
Data Retention Charge loss/gain I/O I/O I/O I/O occurs on the floating gate over time; device threshold voltage trends to a quiescent Charge Loss level Cell is undamaged; block can be reliably erased and reprogrammed Note: Circuit structures and voltages are representative only. Details vary by
Slide 6: Endurance and Data Retention
Endurance and Data Retention There is a direct correlation in the number of cycles experienced by the NAND Flash and data retention Low cycled blocks have longer retention 5 yr High cycled blocks have Retention 2 yr shorter retention required (arbitrary time) 0.5 yr Block cycles 10 cyc 1K cyc 10K
Slide 7: Maximizing Endurance and Data Retention
Maximizing Endurance and Data Retention Robust wear-leveling mechanism Bad block management ECC Santa Clara, CA USA August 2008 7
Slide 8: Good Wear-Leveling
Good Wear-Leveling Makes sure that each block gets approximately equal amounts of Number of erases are even program/erase cycles across all blocks Elongates the life expectancy of the NAND Flash device Ensures more data can be stored before the bad block limits is reached Santa Clara, CA USA A
Slide 9: Bad Wear-Leveling
Bad Wear-Leveling Does not move around static data and only cycles the portion that is dynamic Shortens the life expectancy of NAND Most of the blocks don’t Flash device see erases Stores less data before the bad block limit is reached Santa Clara, CA USA August 2008 9
Slide 10: Theoretical Expected Life of SLC NAND Flash
Theoretical Expected Life of SLC NAND Flash Assumes cycling of 1GB daily until 100K Flash expected life through 100K cycles with daily 1GB storage cycles are reached 116.4 Nand Flash life(years) 140.0 120.0 Assumes that the 100.0 80.0 wear-leveling is perfect 60.0 58.2 40.0 20.0 0.9 1.9 3.7 7.5
Slide 11: Theoretical Expected Amount of Cycles
Theoretical Expected Amount of Cycles Taking the smallest density, Amount of Cycles requiredper density tostore 800000MB such as 64Mb, and multiplying this by 100K 120000 cycles, we get 800,000MB Program/Erase cycles 100000 100000 worth of data that can be 80000 cycled through 64Mb device 60000 50
Slide 12: System Information
System Information Lets examine some use cases on a mobile platform running with operating system 1 (OS1) with no wear-leveling and operating system 2 (OS2) with wear-leveling Both OSs use NAND Flash as a boot solution The file system resides inside the NAND Flash The wear-leveling also resi
Slide 13: Use Cases
Use Cases All use cases assume 1GB of data is cycled daily through NAND Flash device 256MB device is used; the endurance spec on this device is 100,000 program/erase cycles The use cases keep certain portions of the NAND Flash with static data The system is booted, and a test application is
Slide 14: Use Case (OS1)
Use Case 1 (OS1) Number of Erase per block 97% of NAND Flash filled 120000 with static data 100000 Test application creates random files with sizes 80000 No wear-leveling to ranging from 64KB to Number of erase the blocks with static data 1MB Bad; not all the good 60000 Number of erases 40000
Slide 15: Use Case (OS1 To Be Collected)
Use Case 2 (OS1 To Be Collected) Number of Erase per block 70% of NAND Flash filled 120000 with static data Test application creates 100000 random files with sizes No wear-leveling to the ranging from 64KB to Noumber of Erase 80000 blocks with static data 1MB 60000 Number of Erase Bad; not all
Slide 16: Use Case (OS2)
Use Case 1 (OS2) 97% of NAND Flash filled with static data Test application creates 1 file with fixed size Better; a lot of blocks see even amount of erases With this use case, the 256MB device will last 7 years, cycling 1GB daily Erase-to-write ratio is 1.79 Santa Clara, CA USA August 200
Slide 17: Revised Expected Life of SLC NAND Flash (No Wear-Leveling)
Revised Expected Life of SLC NAND Flash (No Wear-Leveling) Flash expected life through 100K cycles with daily 1GB storage This model assumes cycling of 1GB daily N a n d F la s h lif e ( y e a r s ) 7.0 6.3 until 100K cycles are 6.0 5.0 reached 4.0 3.2 97% of NAND Flash 3.0 2.0 0.8 has static da
Slide 18: Revised Expected Life of SLC NAND Flash (With Wear-Leveling)
Revised Expected Life of SLC NAND Flash (With Wear-Leveling) Flash expected life through 100K cycles with daily 1GB storage This model assumes cycling of 1GB daily 27.7 N a n d F la s h l if e ( y e a r s ) 30.0 until 100K cycles are 25.0 20.0 reached 15.0 13.8 97% of NAND Flash 10.0 3.5 has sta
Slide 19: Conclusion
Conclusion With wear-leveling and 1GB write daily, we get about 7 years on a 256MB device; is the 100K cycles really required On large page 1G, 2G, 4G SLC, the requirement should be reviewed From the initial data, 100K requirement looks like overkill Santa Clara, CA USA August 2008 19