Upload
aimee-cryer
View
234
Download
1
Tags:
Embed Size (px)
Citation preview
Introduction to Flash MemoriesAnd Flash Translation Layer
장보길한국 외국어대학교 디지털정보공학과System Software [email protected]
Storage Technologies
Magnetic recording
Optical recording
Electronic memories
Applications of Flash Memories
Flash memory is the major type of NVM (more than 90% of NVM market)
mobile devices
embedded systems flash disks
What is a flash memory cell?
Physics of Flash Memory
Cell array in a flash memory A flash memory cell(Floating gate)
A flash memory cell can store charge. And the charge level represents data.
How does a cell store a bit?
0
How does a cell store a bit?
1 Inject electrons: Hot electron injection mechanism, or
Fowler-Nordheim tunneling mechanism
Remove electrons: Fowler-Nordheim tunneling mechanism
Single-level cell and Multi-level cell
Single-level cell: Two levels One bit
0cell
Multi-level cell: q levels bits q2log
1cell
0cell
1cell
2cell
q-1cell
Typical number of cell levels: 2, 4, 8, 16
How is a cell programmed?
Through multiple rounds of charge injection
Target level
A flash cell Source: [Bandyopadhyay, Serrano, Hasler 2005]
Speed and physical limits
Read: Fast
Write: Slower (due to multiple rounds of programming)
Erase: Very slow
Speed of operations:
Physical limits:
Endurance. (In NOR, a block can stand about 100,000 to 1,000,000 erasures. In NAND, it can stand 10,000 to 100,000 erasures.)
Physical size (e.g., 34nm).
Voltage.
Number of electrons.
NOR and NAND Flash Memories
NOR: Older, still used.NAND: Newer, much more popular now.
What is a NOR flash memory1. Cells form blocks. A block has about 100,000 cells.
2. NOR is a random-access device. Every cell is directly addressable by the processor. That is, a cell can be individually read and programmed.
Word line
Bit line
Control gate
Floating gate(-) (-) (-) (-)
Oxidelayer
Drain Sourceflash memory cell
What is a NOR flash memory
Block erasure!!!!!!
Block erasureIn NOR, the level of a cell can be increased individually and multiple times. But to lower any cell level, the whole blockmust be erased at the same time.
block of cells
Block
What is a NAND flash memory
Cells form blocks. Every block is an array. Every row is a page.
Block
page
page
page
page
Typically: 512 to 2048 cells in a row (page)
Typically: 32 to 128 rows (pages)
Block erasure!!!!!!
Read and write: A page as a unit.
Writing a page in NAND
Why?: Programming is not very accurate, especially with multiple times of writing (for the same page, and for the interference between pages).
A page can be written only once before the block is erased.It is even recommended that the pages are written sequentially.
Page 1
Page 2
…………
Page 64
Partial writing: A page is partitionedinto 4 parts, and we can write a partat a time.
Part 1 Part 2 Part 3 Part 4
A page Note: This is logic partition.
A typical NAND page with spare bytes
2KB of dataA page:
64 Bytes of spare area
MetadataECCUndefined bits
Comparison of NOR & NAND flash
Basic difference: Different ways to connect cells in a block.
Additional difference: Ways to inject charge, used voltages.
NOR: cellsare independent
NAND: Cells in thesame column are connected (and disturbeach other).
Comparison of NOR & NAND flash
NOR: NAND:
6. Mainly to store code. 6. Mainly to store data.
1. Lower density. 1. Higher density.
2. Random access. 2. Page access.
3. More reliable. 3. Less reliable, error-prone. (Requires ECCs.)
4. Slower erase. 4. Faster erase.
5. Faster random read. 5. Faster streaming read.
Flash File System
Wear leveling
Garbage collection
Mapping
Wear leveling
How to know the block’s level of wearing out:
Count the number of erasures, or
Measure the performance of its cells (e.g., erase latencies), or
Other methods?
Alternative approach: Just use randomization (i.e., randomly use the blocks, and hopefully, things will even out).
Wear leveling: Let the blocks be erased about the same number of times.
Method: Write data in different places (instead of the same block).
Wear leveling techniques
If all the data in a block are obsolete, just erase it.Write in blocks that are less worn out.
What if the blocks contain both obsolete and valid data?
Page 3: obsolete
Page 4: obsolete
…………
Page 64: valid
Page 1: valid
Page 2: obsolete
Simple case:
Combining wear leveling with garbage collection
This happens when we want to re-use those blocks that containboth valid and invalid (obsolete) data.
Approaches:
(1) Use a cost/benefit ratio to decide which block to erase. (Before erasing it, the valid data need to be moved first.)
(2) Store frequently-changing data together, and store data that do not change much together. (Reason: After a while, in a block containing frequently-changing data, most of the data are probably already invalid.)
(3) Many heuristic approaches. (And many patents.)
Most important: Design it based on the application.
When garbage collection is done
Garbage collection (of blocks) can happen when:
(1) As background work, i.e., when CPU is idle; or
(2) On demand, i.e., when there is not enough free space.
Mapping
One approach: Treat the flash memory as a block device, much like disk sectors.
Advantage: Allow standard file systems to use flash.
Problems with a simple linear mapping from virtual blocks to flash-memory pages:
Some blocks can be erased too often.
Unable (or inefficient) to write data smaller than a flash block.
Solution: Wear leveling (that is, to move data around).
Mapping between virtual blocks and physical pages is needed.
The spare part in a page may have bits indicating if the page is free/used or valid/obsolete.
How to use flash memories to store data?
Mapping
Virtual blocks Physical pages
Direct map
Stored in RAM, or partially in RAM and partially in flash.
Physical pages Virtual blocks
Inverse map
Stored in flash.
Flash Translation Layer (FTL): A technique to (1)store some of the direct map in flash, and (2) reduce the cost of updating the maps stored in flash.
Flash file systems
Tens of flash file systems (FFS) have been designed.Clearly, more of them will be designed……
Is a data structure that represents a collection of mutable random-access files in a hierarchical name space.
A flash file system:
Provides the block mapping technique.
Does wear leveling and garbage collection.
Maybe design a different, more flash-specific file system?
Do the same for data structures, such as B-trees and R-trees.
More flash-specific file systems
Most of the flash-specific file systems use the same overall principle, that of a log-structured file system.
Why? It is easier to record the small changes (and write them down sequentially), than to rewrite the whole file.
What happens when the blocksare read/erased/written again and again…
Disturb mechanismsWrite/read disturb: When a cell is programmed (or read), the cells in the same column/row are softly programmed.
For some MLC, it is even recommended that after reading the same page 1000 times, write the clean data back again.
Errors, Signal processing, and ECCs
When a block is erased, its quality goes down.
The rate of errors increases…
Types of errors:
Random errors.Fixed-position errors (because the cells really become defected).
Cells in the same column can become bad together.
Ways to correct errors:
Signal processing
ECCs (Hamming, BCH codes.) (Reed-Solomn codes? LDPC codes? Under study.)
New Area in Information Theory:
Coding for Flash Memories
Summary of recent resultsRewriting codes:
Rank modulation:
Error correcting/scrubbing codes: [Cassuto,Schwartz,Bohossian,Bruck,ISIT’07], [Jiang,ISIT’07], [Jiang,Li,Wang,CWIT’09]
Data movement: [Jiang,Mateescu,Yaakobi,Bruck,Siegel,Vardy,Wolf,ISIT’09], [Jiang,Langberg,Mateescu,Bruck,Allerton’09]
Capacity: [Jiang,Bruck,ISITA’08], [Lastras-Montano,Franceschini,Mittelholzer,Sharma,ISITA'08], [Lastras-Montano,Franceschini,Mittelholzer,Karidis,Wegman,ISIT'09], [Jiang,Li,PACRIM’09]
Worst-case performance: [Jiang,Bohossian,Bruck,ISIT’07], [Bohossian,Jiang,Bruck,ISIT’07], [Jiang,Bruck,ISIT’08], [Yaakobi,Siegel,Vardy,Wolf,Allerton’08], [Jiang,Langberg,Schwartz,Bruck,ISIT’09], [Mahdavifar,Siegel,Vardy,Wolf,Yaakobi,ISIT'09]
Expected performance: [Finucane,Liu,Mitzenmacher,Allerton’08], [Jiang,Langberg,Schwartz,Bruck,ISIT’09]
Rewriting: [Jiang,Mateescu,Schwartz,Bruck,ISIT’08]
Error-correction codes: [Jiang,Schwartz,Bruck,ISIT’08]
Sequences: [Jiang,Mateescu,Schwartz,Bruck,ISIT’08], [Wang,Jiang,Burck,ISIT’09]
Rewriting codesWOM (write-once memory) code
Floating code: Joint coding of multiple variables
Example: 2 bits are stored in 3 cells with 4 levels. Every time one bit is changed. How many rewrites can be supported?
0,0
0,1
1,0
1,1
0,00,1
1,01,1
Now use floating codes.
Floating codesExample: 2 bits are stored in 3 cells with 4 levels. Every time one bit is changed.
cell levels
data
Rewriting codesWOM (write-once memory) code
Floating code: Joint coding of multiple variables
Example: 2 bits are stored in 3 cells with 4 levels. Every time one bit is changed. How many rewrites can be supported?
0,0
0,1
1,0
1,1
Now use floating codes.
0,0
0,1
1,0
1,1
3 writes
7 writes
Floating codesWhen two binary variables are stored in n cells of q levels, an optimalfloating code can support rewrites.
When k variables of alphabet size L are stored in n cells of q levels, the number of rewrites that a floating code can support is:
If n is large rewrites
If k,L are large Roughly rewrites
No coding Roughly rewrites
times better
times better
More general model for rewriting
Floating codes: Every rewrite changes one variable.
State transitions of data
Example: 3 binary variables
000 100
010 110
001 101
011 111
Hypercube
Buffer codes: Remember most recent data. [BJB’07]
011010100010010010111101 State transitions of data: De Bruijn graph
More general: [JLSB’09]
The data change in a bounded-degree graph.
Maximum degree:
Trajectory code for bounded-degree rewrite
Model: The state-transition diagram of the data has bounded degree .
[Jiang, Langberg, Schwartz, Bruck, ISIT’09]
This code is asymptotically optimal.
Rank Modulation
[J, Mateescu, Schwartz, Bruck, ISIT’08]
Cell Programming
Noisy, monotonic
Trend: more levels, smaller cells
Question: How to write data reliably when cells cannot be programmed reliably?
Challenges: overshoot, worst-case constraint.
Approach: adaptive cell-ensemble programming.
Rank modulation is such an approach.
Rank Modulation
Analog cell levels induce permutations. Example: 3 cells can induce 3!=6 permutations
Permutations represent data.
Method of programming: from low to high.
Advantage: no overshoot, adaptive coding.
3c1c 2c 3c1c 2c 3c1c 2c 3c1c 2c 3c1c 2c 3c1c 2c
123 132 213 231 312 321
Rewriting; Error correction
Rewrite: How to rewrite data in the rank modulation scheme?
Error correction: How to design error-correcting codes? What does error mean?
A few other topics: (3) Data movement
)log( nnO
Block 1 Block 2 Block n Empty block
[Jiang, Mateescu, Yaakobi, Bruck, Siegel, Vardy, Wolf, ISIT’09]
No coding: erasures are needed.
With coding: 12 n erasures are needed.
Summary of recent resultsRewriting codes:
Rank modulation:
Error correcting/scrubbing codes: [Cassuto,Schwartz,Bohossian,Bruck,ISIT’07], [Jiang,ISIT’07], [Jiang,Li,Wang,CWIT’09]
Data movement: [Jiang,Mateescu,Yaakobi,Bruck,Siegel,Vardy,Wolf,ISIT’09], [Jiang,Langberg,Mateescu,Bruck,Allerton’09]
Capacity: [Jiang,Bruck,ISITA’08], [Lastras-Montano,Franceschini,Mittelholzer,Sharma,ISITA'08], [Lastras-Montano,Franceschini,Mittelholzer,Karidis,Wegman,ISIT'09], [Jiang,Li,PACRIM’09]
Worst-case performance: [Jiang,Bohossian,Bruck,ISIT’07], [Bohossian,Jiang,Bruck,ISIT’07], [Jiang,Bruck,ISIT’08], [Yaakobi,Siegel,Vardy,Wolf,Allerton’08], [Jiang,Langberg,Schwartz,Bruck,ISIT’09], [Mahdavifar,Siegel,Vardy,Wolf,Yaakobi,ISIT'09]
Expected performance: [Finucane,Liu,Mitzenmacher,Allerton’08], [Jiang,Langberg,Schwartz,Bruck,ISIT’09]
Rewriting: [Jiang,Mateescu,Schwartz,Bruck,ISIT’08]
Error-correction codes: [Jiang,Schwartz,Bruck,ISIT’08]
Sequences: [Jiang,Mateescu,Schwartz,Bruck,ISIT’08], [Wang,Jiang,Burck,ISIT’09]