fundamentals of oxide resistive random access memory (rram)
TRANSCRIPT
Copyright ©2010
SEMATECH, Inc. SEMATECH, and the SEMATECH logo are registered servicemarks of SEMATECH, Inc. International SEMATECH Manufacturing Initiative, ISMI, Advanced Materials Research Center
and AMRC are servicemarks of SEMATECH, Inc. All other servicemarks and trademarks are the property of their respective owners.
Fundamentals of
Oxide Resistive Random Access Memory
(RRAM)
David Gilmer
*Nano & Giga Challenges Forum 2017
Memory Technology Trends: New Drivers
26% CAGR
37% CAGR
Future may have different needs:
1. Mobile: smaller, lower power, working memory
than SRAM & lower power storage than NAND
2. Data Center: faster storage than HDD
3. SOC: smaller, lower power working memory
than SRAM & embedded memory
2
3
[1] W. Y. Choi, and T.-J. King Liu IEDM p. 603 (2007).
[2] A. Driskill-Smith, Y. Huai Future Fab p. 28 (2007).
[3] J. E. Green Nature v. 445 p. 414 (2007).
[4] B. Yu IEEE Trans on Nanotech 7, p. 496 (2008).
[5] Kryder, et. al. IEEE TRANS ON MAGNETICS, 45, NO. 10, (2009)
• MLC like NAND
• Cost structure like NAND
• Function, reliability like NAND
• Density > NAND
• Speed > NAND.
Benchmarking Sub 2Xnm Memory Candidates
Goals:
BL1
BL2
BL3
WL1 WL2 WL3
1F Unit cell = 4F2
F= minimum litho CD
A=multiplier of min CD (cell size)
4
NV memory options
MRAM, STT-RAM RRAM
[high density, low power,
“immune” to scaling issues]
Flash (NAND)
Increasingly difficult scaling
Alternative
Storage
Mechanisms
…Changing e- Spins …To Moving Atoms:
Incumbents :
From: Trapping Electrons…To: Moving Atoms
Micron Technologies
http://blogs.zdnet.com/
Aug. 2009
Toshiba T. Kamigaichi,
EDM 2008.
SAMSUNG
2009
25nm
34nm Floating gate
Trap electrons…
Incumbents
*Hard to maintain charge
When aggressively scaled
WL
BL
O
*Hard to maintain polarization
When aggressively scaled
What is Resistive Random Access Memory?
(RRAM)
• For any memory system to work, it needs at least
two different detectable states
• RRAM detectable memory states are “high” or
“low” resistance…. A resistance change memory
5
V
V
V
V
V
VV
V
V
V
V
V
V
VV
V
Electric
Current
Low
High
Resistance
+
-
Electric
Current
High
Low
Resistance
+
-
Why metal-oxide filament-based RRAM?
6
V
V
V
V
V
VV
V
V
V
V
V
V
VV
V
V
V
V
V
V
V
VV
V
= =
Challenge for filament-based RRAM
7
*Filament forming is a random processControl conductive filament
Uniformity
Key Understand operating mechanisms
8
Dielectric
Breakdown
Filament formation is key
to well behaved switching
Dielectric Fresh State:
starting defect profile
important
Features of filamental Bipolar RRAM
I-compliance: limit with
internal FET for low
parasitics, overshoot
I
On
(LRS)
Off (HRS)
9
1.E-08
1.E-07
1.E-06
1.E-05
1.E-04
1.E-03
-2 -1 0 1 2 3
Resistive switching requires
Metal-Oxide defects
Vg
I (A)
HfO2 HfO2
Ar
1.E-11
1.E-10
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
-2 -1 0 1 2 3
No Ion-Bombardment
Ar bombarded of HfO2
Oxygen gettering for sub-stoichiometry
*National Synchrotron Light Source (NSLS) at (BNL)
“Oxygen exchange layer (OEL) for
Generating O-deficiency in HfO2”
Evidence of HfO2 Reduction by OEL
10
11
12
Filament Mechanism: Grain boundaries are weak point;
likely location for filaments
• Different conduction mechanisms
along GB and Grains due to
accumulated defects at GB’s
over GB
over grains
C-AFM data
SiO2
Si
HfO2
C-AFM
*YEW, K. S. et al., Proceedings of the International Conference on Solid State
Devices and Materials (SSDM), Japan, 2009.
Switching at grain boundaries induced by AFM tip
amorphous HfO2 crystalline HfO2
Switching observed only on GBs oxygen deficient regions
Forming
13Lanza et al., Appl. Phys. Lett. 101, 193502 (2012)
14
• Bond breakage followed by oxygen
out-diffusion• Vacancy diffusion
Alternative description
IfIf
Metal-Oxide RRAM Material Model;
Form a Conductive Filament
15
kT
EEETG
oxact exp),(
Trap generation rate:
Heat dissipation increases deeper in the oxide
larger O-deficient region near the forming anode
Forming process and filament geometry
Cone-shaped filament
0.E+00
2.E+02
4.E+02
6.E+02
8.E+02
1.E+03
1.E+03
02468
power [W/m]
z[nm]
Higher power
dissipation
closer to anode
anode cathode
eCF
anode cathode
eCF eCF
e
ehw
hwO
defect generation
by Temp/Eox
*G. Bersuker et al, IEDM 2010; G. Bersuker et al, JAP 2011
(-)(+)
16
GB
O2-
O2-
O2-
TiN
HfO2
Ti
TiN
HfO2GB
O2-
O2-
O2-
O2-
TiN
O2-
O2-
O2-
TiN
Ti
Filament growth model • Initial: O-vacancies along grain
boundary
• Two-step process for each O atom
removal during filament growth:
1) Breakage of Hf-O bond
2) Diffusion of released O ion
Breakage of Hf-O bonds by thermo-
electrical stress:
O-ion diffusion by E-field assisted
substitutional hopping, :
G. Bersuker, ME 2001; J. McPherson, APL 2003
),,(
),,(exp),,( 0
zyxkT
zyxEEGzyxG
oxA
𝑫 = 𝝊 ∗ 𝒆−
𝑬𝑫−𝑸𝝀𝟐 𝑭𝒆𝒙𝒕
𝑲𝒃𝑻
Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
17
formingFilament reset mechanism
• Most voltage drops across the narrow filament tip
-1.0 -0.5 0.0
5.0x10-5
1.0x10-4
1.5x10-4
2.0x10-4
2.5x10-4
3.0x10-4
I [A
]
Vg [V]
ResetModelOhmic
*Need ~1nm
Tunnel barrier to fit
*Tunnel gap ~ 1nm
• Current deviates from Ohmic initially due to Resistance from Temp
• Reset occurs when temp sufficient for oxidation and oxygen is available
• Asymmetric structure and switching supports model
*G. Bersuker et al, IEDM 2010
_
+
ReSet process
*G. Bersuker et al, IEDM 2010; G. Bersuker et al, JAP 2011
Reset process
O(2-) hopping between interstitial
positions and Hf oxidation at
oxide/filament interface
18
TiN
O2-
O2-
TiN
O2-
O2-
O2-
O2-O2-
O2- O2-
O2-
O2-
O2-
O2-O2-
O2-
HfO2
O2-O2-
Eox
.
Radial T-profile
Re-oxidation of the filament tip
driven by O-ion Coulomb repulsion
and density gradient
18
Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
Positive forming/set, negative reset
Reset has Polarity Dependency!!
FormingReset
Set
Asymmetric defect/vacancy profile “metal-OEL”/HfO2-x dielectrics ;
Vg [V]
I g[A
]
Oxygen Poor near the reset anode prevents barrier No reset
Oxygen Rich near the reset anode barrier forms Reset
Bias to TE
Barrier forms
*D. Gilmer, et al, IMW 2012
X
Negative forming/set, positive reset
Barrier does
NOT form
Bias to TE
Oxidation for Asymmetric Defect Profile
20
Relative O% profile DC switching showing 100% yield for a 10-cycle
set/rest dc sweep for 35 die (350 total sweeps)
[Inset; characteristic endurance]
200ns pulsed endurance to 108
• Operating the Asymmetric O-vacancy stack in preferred polarity = 100% yield
• Operating the Asymmetric O-vacancy stack in NON-preferred polarity = 0% yield
TE
BE
HfO2
HfOx
*D. Gilmer, et al, IMW 2012
for various
asymmetric O-vacancy HfOx
Reset Anode against “Good” vs “Bad” dielectic
• Even non-preferred bias polarity has similar current deviation
near reaching the forming compliance during reset sweep….
However, the non-preferred biasing does not reset, it breaks
21*D. Gilmer, et al, IMW 2012
Asymmetric Defect Profile observed for
Best Behaved Bipolar RRAM
Increasing O-VACANCY
Gradient
Various RRAM stacks with O-vacancy asymmetry
• O-vacancy asymmetric stacks Preferred Reset polarity
with the “Good” dielectric against the Reset Anode (+)
• Opposites attract….O- (+); V00
(-); M+ (-)
Preferred
Reset Polarity!
+
-TE
TaOx
Ta2O5
Vo
VoVo
TiN
Vo
Vo
BE
TE
TiOx
TiO2
Vo
VoVo
TiNVo
BE
Ti
HfO2
Vo
Vo
Vo
BE
TE
HfOx
HfO2
Vo
Vo
Vo
Vo
Vo
Vo
TiN
BE
TE
Zr
HfO2
Vo
Vo
Vo
BE
TE
HfO2
TaOx
TaVo Vo
VoVo
Vo
VoVoVo
VoVo
BE
TE
22*D. Gilmer, et al, IMW 2012
Symmetrical Bias does not similarly break the
tunnel barrier Due to Asymmetric defect profile
Oscilliscope
monitored
80us pulsed
SetSet
ReSet
• Asymmetry and
Preferred
ReSet Bias *Slower to change,
‘fluctuating’
Competition between
breakdown and oxidation
23
and
pulsed Reset
• Asymmetry and
Preferred Set Bias
*Immediate and singular
resistance change
Barrier breakdown
*D. Gilmer, et al, IMW 2012
RRAM operations: From grain boundaries to
conductive filament
Forming event
CF
TE w/ strong O-1
affinity
MeOx
Reset
e-
HRS
barrier
e-e-e-
LRS
e-
induced heating/field bond breakage
Initial leakage path along GB
Set
V
V
V
V
V
V
Model-based simulations reproduce the entire forming-switching phenomenon *G.Bersuker et al, JAP 2011
switching
Vacancy
Oxygen Ion
24
Mechanism of Forming transient: Run-away
Hf-O bond breakage process
xyz
xyzoxact
oxxyzkT
EEGETG
,
0 exp),(
Bond breakage rate:
*L. Vandelli et al, IRPS’12
Time-to-Forming: Weibull statistics
breakdown event
Breakage of next-neighbor
bonds
hw
OV+
25
Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
5nm
Program Description (3D Map creation)
HfOx-Based RRAM fully ModeledSimulated unit cell
Electron transport via vacancies power dissipation T+E field
new vacancies generation + oxygen diffusion electron transport
increases power dissipation increases … new vacancy generation
+ oxygen diffusion ……filament formed
Main loop: numerically
solving Fourier heat-transfer,
charge continuity equations in
each unit volume
• Resistivity– local
values depending on
# of vacancies
• Potential
• Power dissipation
• Temp. / E-Field
3D Map Update
26Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
Why oxygen ions stay near filament?
Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
28
There is an optimum density of vacancies
out-diffused
only few generated
No vacancies
Low Density
Medium Density
High Density
Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
many initial
vacancies not
enough oxygen
ions generated
Too few
vacancies
lack of oxygen ions
in the filament
vicinity
Insufficient reset in:
Stoichiometric hafnia: lack of oxygen ions due to recombination w/ vacancies
Highly substoichiometric hafnia: low number of generated oxygen ions
Both Stoichiometric and High vacancy density
films demonstrate poor reset
29Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
A
A’
A”
‘Moderate’
vacancy
density
results in best
switching
30Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
Simulated set/reset cycle
31Bersuker & Gilmer, CH.9; Metal oxide “RRAM” technology in; “Advances in Non-
Volatile Memory and Storage Technology” 2014, Pages 288–340
Summary
32
The goal; creating vacancy profiles advantageous for the filament oxidation during reset and barrier break in set
• Model identifies material properties and operation conditions controlling
RRAM performance (supported by O-vacancy Asymmetry Device data)
• Asymmetric stacks have preferred bipolar operation polarity
• O-vacancy Asymmetry Assists the Set-ReSet Mechanism
• Many techniques can introduce vacancies; ‘OEL’, Plasma, Reactive PVD, etc…
Asymmetric TiN\HfOx\Ti\TiN
40nmx40nm x-bar80ns ±1.5V pulse,
25uA compliance
HRS
LRS
Vo Vo Vo
Vo VoHfHf
Vo Vo Vo
Vo VoHfHf
+ Set bias
-
Vo Vo Vo
Vo VoHfHf
ReSet Set