improving system performance and longgyevity with a new ... · nand block size trend epm = 128...
TRANSCRIPT
Improving System Performance and Longevity with a New NAND Flash g yArchitecture
Jin-Ki KimVi P id t R h & D l tVice President, Research & Development
MOSAID Technologies Inc.
Santa Clara, CA USAAugust 2009 1
Agenda
Background
Technical InnovationsTechnical Innovations
Solutions for Emerging Applications
Summary
Santa Clara, CA USAAugust 2009 2
NAND Technology Timeline
NAND Cell Invented
2Mb NAND Prototype & 1st Specification
Entered Gb Flash Era(1Gb NAND Flash Developed in 90nm)
SSD Emerged(8Gb SLC & 16Gb MLC in 50nm)
1984 1987 1992 1995 2002 2005 2007 2009
MCL (2bits/cell) for Mainstream
Confirmed NAND Technology
Key Circuit Techniques (Self boosting ISSP)
3bits/cell & 4bits/cell(32Gb MLC in 34nm)Mainstream
(4Gb SLC & 8Gb MLC in 65nm)
NAND Technology& Manufacturability
(Self-boosting, ISSP) Reported(32Mb NAND Developed)
(32Gb MLC in 34nm)
NAND Flash ProgressionDensity
The single-minded focus on bit density
32Gb MLC
improvements has brought Flash technology to current multi-Gb NAND devices
8Gb MLC
16Gb MLC
2Gb SLC
4Gb MLC
1Gb SLC1Gb SLC
Santa Clara, CA USAAugust 2009 4
120nm 90nm 73nm 65nm 51nm 34nm
Process Technology
NAND Architecture Trend
Primary focus is density for cost and market adoptionadoptionPage-based program and block-based erase
1 2 4B: # of Bitline/Page Buffer
1 2 3A: # of Bit/Cell 4
512B 2KB 4KB 8KB 16KBD: Page Size
8 16 32 64C: # of Cell/NAND String
512B 2KB 4KB 8KB 16KBD: Page Size
Block Size = (A * B * C) * D
Santa Clara, CA USAAugust 2009 5
( )Erase Program Mismatch (EPM) = A*B*C
NAND Block Size Trend
EPM = 128 (SLC) & 256(MLC)
Erase Program Mismatch (EPM) is a key parameter to degrading write efficiency (i.e. increase write amplification factor)
512KB
8Gb/16Gb/32GbMLC
• 64 cells/NAND• 8KB Page Buf.• 2 BLs/PB
512KB
8Gb/16Gb/32GbMLC
• 64 cells/NAND• 8KB Page Buf.• 2 BLs/PB
256(MLC)factor)
256KB
2Gb/4Gb/8GbMLC
256KB
2Gb/4Gb/8GbMLC
EPM = 64 (SLC) & 128(MLC)
128KB4Gb/8Gb/16Gb
128KB4Gb/8Gb/16Gb
EPM = 16
EPM = 32
4KB8KB
16KB
8Mb/16Mb 32Mb/64Mb128Mb/256Mb/512Mb
1Gb/2Gb/4Gb
4KB8KB
16KB
8Mb/16Mb 32Mb/64Mb128Mb/256Mb/512Mb
1Gb/2Gb/4GbEPM = 8
Santa Clara, CA USAAugust 2009 6
90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 • 8 cells/NAND• 512B Page Buf.
• 16 cells/NAND• 512B Page Buf.
• 16 cells/NAND• 512B Page Buf.• 2 BLs/PB
• 32 cells/NAND• 2KB Page Buf.• 2 BLs/PB
• 32 cells/NAND• 4KB Page Buf.• 2 BLs/PB
90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 • 8 cells/NAND• 512B Page Buf.
• 16 cells/NAND• 512B Page Buf.
• 16 cells/NAND• 512B Page Buf.• 2 BLs/PB
• 32 cells/NAND• 2KB Page Buf.• 2 BLs/PB
• 32 cells/NAND• 4KB Page Buf.• 2 BLs/PB
NAND Challenges in Computing Apps
100KSLC (1bit/cell)
Cell Endurance: # of Reprogram Cycles
10years
Data Retention: Ability to Retain Charge
100K
MLC (2bi / ll)
10years
10K
5K
MLC (2bits/cell)
2.5years
5years
2002 2006 2007 2008 2009 2010
90 60 0 40 30 20
3K
1K
Year 2002 2006 2007 2008 2009 2010
90 60 0 40 30 20
1years
Year
90nm 60nm 50nm 40nm 30nm 20nmTech. 90nm 60nm 50nm 40nm 30nm 20nmTech.
Reliability degradation combined with the current NAND architecture trend introd ces a negati e impact on comp ting applications
Santa Clara, CA USAAugust 2009 7
trend introduces a negative impact on computing applications
Flash System Lifetime Metric
System lifetime heavily relies on NAND hit t d f tarchitecture and features
Write Efficiency = Total Data Written by Host
Total Host Writes = NAND Endurance Cycle
Total Data Written to NAND
Total Host Writes NAND Endurance Cycle* System Capacity* Write Efficiency* Wear Leveling Efficiencyg ff y
System Lifetime = Total Host WritesHost Writes per Day
Santa Clara, CA USAAugust 2009 8
Host Writes per Day
NAND Architecture Innovation Current NAND Flash ArchitectureCurrent NAND Flash Architecture
Unit Block
Plane 1 Plane 2
Shared Page Buffer
A: # of Bit/CellD: Page (buffer) Size
# f / ff
Santa Clara, CA USAAugust 2009 9
C: # of Cell/NAND StringB: # of Bitline/Page Buffer
Block Size = (A * B * C) * D
NAND Architecture Innovation FlexPlane with 2-tier Row Decoder SchemeFlexPlane with 2 tier Row Decoder Scheme
Smaller & Flexible Page Size (2KB/4KB/6KB/8KB)Smaller & Flexible Erase SizeLower power consumption due to segmented pp-wellExtend system lifetime
Fl PlFlexPlane
NAND Cell Array on Sub PP-Well
t Row
Dec
oder
NAND Cell Array on Sub PP-Well
t Row
Dec
oder
NAND Cell Array on Sub PP-Well
t Row
Dec
oder
NAND Cell Array on Sub PP-Well
t Row
Dec
oder
Row
Dec
oder
Seg
men
t
Seg
men
t
Seg
men
t
Seg
men
tG
loba
l
Santa Clara, CA USAAugust 2009
10
2KB Page Buffer 2KB Page Buffer 2KB Page Buffer 2KB Page Buffer
FlexPlane Operations
NAND Architecture Innovation2-Dimensional Page Buffer Scheme2-Dimensional Page Buffer Scheme
Higher Performance
Upper PlaneLower Power ConsumptionHigher Yield
2-Dimensional Page Buffer(Shared Page Buffer)
Lower Plane0.5x Bitline Length compared to conventional NAND architecture
Santa Clara, CA USAAugust 2009 11
Micron 32Gb NAND Flash in 34nm, 2009 ISSCC
NAND Core InnovationPage-pair & Partial Block ErasePage-pair & Partial Block Erase
Minimize Erase Program Mismatch (EPM)Improve write efficiency
Block Erase Partial Block Erase Page-pair Erase
• MOSAID’s Erase Scheme*
Santa Clara, CA USAAugust 2009 12
* Jin-Ki Kim, “Low Stress Program and Single Wordline Erase Schemes for NAND Flash Memory”, IEEE NVSMW Aug. 2007.
NAND Core InnovationLower Operating VoltageLower Operating Voltage
2.7 ~ 3.6V 2.7 ~ 3.6V 3.3V
H.V. Gen
Core and Peri.
H.V. Gen
Core and Peri.
H.V. Gen
Core and Peri.
1.8V
Core and Peri.
I/O
Core and Peri.
I/O
1.8VCore and Peri.
I/O
• MOSAID’s Low Stress Program Scheme*
Santa Clara, CA USAAugust 2009 13
* Jin-Ki Kim, “Low Stress Program and Single Wordline Erase Schemes for NAND Flash Memory”, IEEE NVSMW Aug. 2007.
HyperLink (HLNAND™) Flash
Monolithic HLNAND• Device interface • Fully independent bank operations
NewDevice
independent of memory technology and density
• Packet protocol with low pin count
• Configurable data width link hit t
• Multiple, simultaneous data transactions
• Data throughput independent of core operations
• Device architecture for performance DeviceArchitecture
architecture • Flexible modular command
structure• Simultaneous Read and
Write• EDC for command and
and longevityFlexPlane ArchitectureTwo-dimensional Page BufferTwo-tier Row Decoder
New LiArch ew W
rite
eme• EDC for command and ECC for registers
• Daisy-chain cascade with point-to-point connection of up to virtually Unlimited number of devices
• NAND flash cell technology• Page/multipage erase function• Partial block erase function• Low stress program scheme
Linkrchitecture
NewSchem
MCP HLNAND
p g• Random page program in SLC• Low Vcc operations
Santa Clara, CA USAAugust 2009 14
MCP HLNAND
HyperLink Interface
Point-to-point ring topology• Synchronous DDR signaling with source termination only• Up to 255 devices in a ring without speed degradation• Dynamically configurable bus width from 1-8 bitsy y g• HL1 parallel clock distribution to 266MB/s• HL2 source synchronous clocking to 800MB/s, backward
compatible to HL1
HLNAND HLNAND
ControllerHostInterface
HLNAND HLNAND
Santa Clara, CA USAAugust 2009 15
64Gb MLC HLNAND MCP
HLNAND bridge chip MCP with 4-NANDs stackedHL1 (DDR-200/266) using NAND in MCP - fully compliant to HLNAND spec
Santa Clara, CA USAAugust 2009 16
12 x 18 100-Ball BGA
HLNAND Flash Module (HLDIMM)
Use cost effective DDR2 SDRAM 200-pin SO-DIMM f f t d k tDIMM form factor and sockets 8 x 64Gb or 8 x 128Gb HLNAND MCP (4 on each side)each side)
Santa Clara, CA USAAugust 2009 17
HLDIMM Port Configuration
Four independent HyperLink interface (266MB/s)Aggregate bandwidth of 1066MB/s regardless of # of module
Santa Clara, CA USAAugust 2009 18
HLDIMM Pin Assignment
All high speed signals fully shielded with 1:1 ratio between signal and power/groundbetween signal and power/ground6 Vss, 3 Vdd (1.8V), 3 Vdd3 (3.3V) per channel in/outVref and Reset pins shared by all channelsp y5 pin SPD interfaceForward compatible to HL2 800MB/s source synchronous clocking
r 1 r 2 r 3 r CK
O#
r KO r 4 r 5 r 6 r 7 r ncC
SO r 0r DSO
pow
eD
1/Q
1po
we
D2/
Q2
pow
eD
3/Q
3po
we
CK
#/C
pow
e
CK
/CK
pow
eD
4/Q
4po
we
D5/
Q5
pow
eD
6/Q
6po
we
D7/
Q7
pow
eC
E#/n
CSI
/Cpo
we
D0/
Q0
pow
eD
SI/D
Santa Clara, CA USAAugust 2009 19
channel input/output pin assignment
System Configurations with HLDIMM
CH0 i HL1HL1 HL1HL1
Non-Interleave - 1066MB/s Aggregate Throughput
CH0 in MCPsMCPs
CH0 out HL1MCPsHL1
MCPs
CH1 in HL1MCPsHL1
MCPs
MCPsMCPs
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
CH0 in HL1MCPsHL1
MCPs
CH0 out HL1MCPsHL1
MCPs
CH1 i HL1HL1
controller
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1 MCPsMCPs
CH1 out HL1MCPsHL1
MCPs
HLDIMM
MCPsMCPs
HL1MCPsHL1
MCPs
HLDIMMHLDIMM
Loop-around empty socket
CH1 in HL1MCPsHL1
MCPs
CH1 out HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HLDIMM
CH0 in HL1MCPsHL1
MCPsHL1
MCPsHL1
MCPsHL1
MCPsHL1
MCPsHL1
MCPsHL1
MCPs
HL1MCPs
HL1MCPs
HL1MCPs
CH0 in HL1MCPs
CH0 out HL1MCPs
HL1MCPsCH2 in
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
CH0 in HL1MCPsHL1
MCPs
CH0 out HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPsCH2 in
Interleave - 2133MB/s Aggregate Throughput
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
HL1MCPs
HL1MCPs
HL1MCPs
HL1MCPs
HL1
CH1 in HL1MCPs
CH1 out HL1MCPs
HL1MCPs
HL1MCPs
HL1
CH2 out
CH3 in
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
CH1 in HL1MCPsHL1
MCPs
CH1 out HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1MCPsHL1
MCPs
HL1HL1
CH2 out
CH3 in
Santa Clara, CA USAAugust 2009 20
CH0 out HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HLDIMM
HL1MCPs
HLDIMM
HL1MCPs
HLDIMM
CH3 outHL1
MCPsHL1
MCPs
HLDIMM
HL1MCPsHL1
MCPs
HLDIMM
CH3 out
HLDIMM PerformanceCapacity, Bandwidth and Random IOPSCapacity, Bandwidth and Random IOPS
Capacity & Bandwidth vs. # of Modules Random IOPS vs. # of Modules
Santa Clara, CA USAAugust 2009 21
HLDIMM Packing Density
Fully populated memory capacity of 512GB or 1TB i hi d i 64GB d 128GB1TB is achieved using 64GB and 128GB HLDIMM modules within only 60cm2 of motherboard areamotherboard area.
Santa Clara, CA USAAugust 2009 22
Summary
The driving forces in future Flash memory are the memory architecture & feature innovation that will support emerging system
hit t d li tiarchitectures and applications
Using HLNAND Flash Storage Class MemoryUsing HLNAND Flash, Storage Class Memory is viable today using proven NAND flash technologytechnology
Santa Clara, CA USAAugust 2009 23
Resource for HLNAND Flashwww.HLNAND.comwww.HLNAND.com
Available• 64Gb MLC MCP sample• 64GB HLDIMM sample6 G sa p e• Architectural Specification• Datasheets
White papers• White papers• Technical papers• Verilog Behavioral model
Santa Clara, CA USAAugust 2009 24