phase change memory - nvmw.eng.ucsd.edu
TRANSCRIPT
Phase Change Memory: Development Progress and System Opportunities
Greg AtwoodNumonyx Sr. FellowCorporate Technology Office
#2
Outline
Motivation for Phase Change Memory
Technology Status
Development Directions
System Opportunities
#3
Memory Consideration:Past, Present, and Future
“New Memory” Motivation (Recognized ~1999) :– Current NVM (and DRAM) are becoming Electrostatics Limited– Starting to Encounter Physical Scaling Limitations– Manifesting First as Degradation in Reliability (Endurance/Retention)
Realization:– Next Generation NVM will Exploit New Storage Physics– Significant Innovation Will take Time (History says ~10yr “Gestation”)
Response:– Phase Change Memory Identified as the Best Candidate– Research Efforts Validated Assumptions and Initiated Development– Phase Change Memory Moving into Production Today
PCM Features Will Enable New Usage Models Approaching the Goals of “Storage Class Memory”
#4
NAND Scaling Challenges
Reducing Cell Capacitance Results in Fewer Stored
Electrons with Scaling – Effect Magnified with MLC
Degrading Reliability Results in Increased Usage of Error Correction – More System
Management Required
SLC R/W
NAND Electron Scaling
20122011
20102009
2008
2007
2006
2004
10
100
1000
10000
20406080100
Litho Node
# E
lect
rons
SingleBit/Cell
0
5
10
15
20
25
30
0
20
40
60
80
100
120
90 70 60 50 40 30 20Lithography (nm)
# of
EC
C B
its
R/W
Cyc
les
(k)
Cycles/ECC vs NAND Litho
SLC R/W
MLC R/W
SLC ECC
MLC ECC
0
5
10
15
20
25
30
0
20
40
60
80
100
120
90 70 60 50 40 30 20Lithography (nm)
# of
EC
C B
its
R/W
Cyc
les
(k)
Cycles/ECC vs NAND Litho
SLC R/W
MLC R/W
SLC ECC
MLC ECC
#5
DRAM Scaling Challenges
Major Changes Required
New Periphery Transistor Every Gen
Frequent Capacitor Structure and Material Changes with Highest “known” HiK Material at ~32nm Generation
New Cell Transistor Every Generation
#6
Phase Chang Memory Key Attributes
105 -106
200-200 ns
13-30
KB/s
~Flash
Easy
No
Small/Byte
Yes
EEPROM
NoYesYesNoErase
20 - 80 ns15 - 50 us70-100 ns50 - 100 nsRead Latency
100+
MB/s
10+
MB/s
0.5-2
MB/s
1- 15+
MB/sWrite
Bandwidth
EasyHardModerateEasySoftware
High~Flash~Flash~FlashPower
Small/ByteLargeLargeSmall/ByteGranularity
NoYesYesYesNon-Volatile
104-5
NAND
Unlimited
DRAM
105
NOR
106+Endurance
PCMAttribute
PCM provides an new set of features combining components of NVM with DRAM
#7
Outline
Motivation for Phase Change Memory
Technology Status
Development Directions
System Opportunities
#8
Phase Change Material Properties
amorph fcc hexagonal
XRD-measurements
M.Wuttig et al., RWTH Aachen FP6 Project CAMELS
Certain alloys containing one/more group VI elements (chalcogenides) exhibit reversible transition between the disordered (amorphous) and ordered (crystalline) atomic structure
Ge2Sb2Te5
#9
Phase Change Memory Mechanisms
Storage– GST: Germanium-Antimony-Tellurium
Chalcogenide glass
– Cell states varying from amorphous (high resistance) to crystalline (low resistance) states
Read Operation– Measure resistance of the GST
Write Operation– Heat GST via current flow (Joule effect)
– Time at critical temperature determines cell state
Crystalline Amorphous
Time
Tx
Tm
Reset (amorphization)
Set (crystallization)
TimeTem
pera
ture
Tx
TmReset (amorphization)
Set (crystallization)
I
V
I
VTemperature during write
0.0 0.5 1.0 1.50.00
0.25
0.50
0.75
Vth
Crystal Amorphous
Cur
rent
[mA
]
Voltage [V]
Read
Write
#10
BJT-PCM Cell and Array Construction
Chalcogenide
n-array
p+
n-array
p+
n-array
p+
Bitline (Metal 0)
Wordline(Metal 1)
B-B’ cross section
A A’
B
B’
Basic layout
BaseContact
EmitterContact
A-A’ cross section
Wordline
Chal
Bitline
Chal
Bitline
n+n+
Chal
Bitline
Chal
Bitline
N - Base
P – SubstrateCollector
PEmitter
PEmitter
PEmitter
PEmitter
4F2 Basic Cell, 5.5F2 Including Base Contacts
#11
PCM Technology Roadmap
20072005 2009
Numonyx
PCM Roadmap
#12
Summary of Key PCM Reliability Metrics
Retention decoupled from endurance
P/E Cycles
Ret
entio
nT
ime
PCM
NAND
No issue with Read disturb
P/E Cycles
Rea
d
End
uran
ce PCM
NAND
Result:
Greatly simplified management
No software overhead for some apps
Post-stress shelf-life problem solved
Endurance can be increased significantly through management of bit errors
Data is stable once successfully writtenP/E Cycles
Writ
e E
rror
R
ate
PCM
NAND
Wearout is Well Behaved
#13
PCM Reliability Data – 90nm 128Mb
0.01
0.1
1
10 100 1000 10000 100000
Prior Write Cycles
Nor
mal
ized
Cel
l Fai
ls A
fter B
ake Retention/Endurance
Independence
0.01
0.1
1
10 100 1000 10000 100000
Prior Write Cycles
Nor
mal
ized
Cel
l Fai
ls A
fter B
ake Retention/Endurance
Independence
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Each Point is the average of ~200die
(log)
10x – 100x Margin to Goal
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Each Point is the average of ~200die
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Lot #
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1 2 3 4 5 6 7 8 9 1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
Goal
Cel
l Fai
ls
Data Retention
Each Point is the average of ~200die
(log)
10x – 100x Margin to Goal
1Mc Endurance
0
20
40
60
80
100
1 2 3 4 5Lot #
% U
nits
pas
sing
Process + Programmingoptimization
Each Point represents ~30die
Improvements due to Process and Programming
Optimization
1Mc Endurance
0
20
40
60
80
100
1 2 3 4 5Lot #
% U
nits
pas
sing
Process + Programmingoptimization
Each Point represents ~30die
1Mc Endurance
0
20
40
60
80
100
1 2 3 4 5Lot #
% U
nits
pas
sing
Process + Programmingoptimization
Each Point represents ~30die
Improvements due to Process and Programming
Optimization
Cell Current
Dis
trib
utio
n #
Sig
ma
Array DistributionReset Set
Cell Current
Dis
trib
utio
n #
Sig
ma
Array DistributionReset Set
128M
b V
ehic
le12
8Mb
Veh
icle
DiePhoto
128M
b V
ehic
le12
8Mb
Veh
icle
DiePhoto
0
20
40
60
80
100
1.E+05 1.E+06 1.E+07 1.E+08 1.E+09
Writes
% P
assi
ng U
nits
100Mc FeasibilitySmall Sample
0
20
40
60
80
100
1.E+05 1.E+06 1.E+07 1.E+08 1.E+09
Writes
% P
assi
ng U
nits
100Mc FeasibilitySmall Sample
#14
Outline
Motivation for Phase Change Memory
Technology Status
Development Directions
System Opportunities
#15
45nm Technology – 1Gb “Bonelli” Product
Process Architecture
PCM cell
– Salicided minimum-size BJT selector
– Self-aligned “Wall” Structure
– 1 Base Contact / 4 Emitter
– 0.015 µm2 cell size
CMOS periphery– Dual gate oxide (3nm for 1.8V and
8nm for 3V operation)
3 Cu levels for tight interconnects
#16
Scalability of PCM
Thermal Disturb Scales if Cell Geometry is Properly Scaled
Scaling of the “Wall” structure for the storage element and of the BJT for the selector
#17
CMOS SupportCircuitry
PCMS Array
3-d corner view of PCMS array
PCMS Cell
CMOS SupportCircuitry
PCMS Array
3-d corner view of PCMS array
PCMS Cell
Extending PCM into the 3rd DimensionResearch Program
Stacked Phase Change Memory– Same PCM memory concept
– Uses a diode (Ovonic Threshold Switch – OTS ) as select device
– OTS is the same class of materials as PCM - Calcogenide
OTS
Electrode
PCM
Electrode
OTS
Electrode
PCM
Electrode
OTS
Electrode
PCM
Electrode
Potential benefits– True cross point array -
4F2 cell
– Stackable n-layers for effective cell size of 4F2/n
– MLC NAND cost structure
– Performance of PCM
64Mb Array Distribution
64Mb Array Distribution
>1Mc Endurance Demonstrated>1Mc Endurance Demonstrated
#18
CMOS SupportCircuitry
PCMS Array
3-d corner view of PCMS array
PCMS Cell
CMOS SupportCircuitry
PCMS Array
3-d corner view of PCMS array
PCMS Cell
Numonyx PCM Development
Scaling the existing architecture, providing the smallest cell size, following the lithography roadmap
Introduction of Multi-Level-Cell exploiting the analog storage capability of PCM
Exploration of new chalcogenide alloys which may open new application fields
Exploitation of a true cross-point array which will allow vertical stacking of more than one memory layer
Ge or M (at %)
0 100
90
80
7060
5040
3020
100
1020
30
4050
6070
80
90100
60 70 80 90 10050403020100Sb (at %)Te (at %)
GeTe
Sb2Te3
GeSbTe(GST) Doped SbTe
Sb2Te
M-Sb2Te
DVD+RW
225
147
DVD+RAM
124
Mai
nstr
eam
D
evel
opm
ent
Ear
ly
Eva
luat
ions
Ong
oing
R
esea
rch
Pro
f of
Con
cept
R
esea
rch
Our Main Thrust
Expanding theEnvelope
Breakthrough Research
#19
Outline
Motivation for Phase Change Memory
Technology Status
Development Directions
System Opportunities
#20
PCM Application Opportunities
Wireless System– Direct code execution, semi-static data and file storage�Bit alterability allows direct-write memory
Solid State Storage Subsystem– Frequently accessed pages and elements easily managed when
manipulated in place�Caching with PCM will improve performance and reliability
Computing Platforms– Taking advantage of non-volatility to reduce the power�PCM offers endurance and write latency that are compelling for a
number of novel solutions
#21
PCM in Computing Applications
– More latency improvement • PCM is to SSDs as SSDs are to HDDs
• How is it possible to increase performance?– Bandwidth improvement
– Latency improvement
#22
Memory Hierarchy in Servers & StorageWhere PCM can Fit
PCM enables improved endurance for logging applications
PCM replaces battery backed up RAM-cache at cabinet and rack levels
PCM improves reliability and performance in HDD / SSD’s
PCM for performance high IOPS, low latency
Servers
ClientsNetwork Attached Storage
Storage Area Networks
Network
Local and Remote Memory
Direct Attached Storage
Storage Controllers
CPU RAM DISK TAPE
ns / GB ms / TB s / PB
SSD
us / GB
SCM
#23
Summary
Phase Change Memory (PCM) technology is demonstrating the capability to enter the broad memory market and to become a mainstream memory
PCM provides a new set of features interesting for novel applications, combining components of NVM and DRAM and being at the same time a sustaining and a disruptive technology
The technology maturity achieved, the scaling perspective and the broad application range allow predicting a key role for the PCM in the memory market for the next decade.