1
Wei Lu
University of Michigan Electrical Engineering and Computer
Science
Resistive Memories Based on Amorphous
Films
Crossbar Inc
Lu Group EECS, UM 2
Introduction
Hysteretic resistive switches and crossbar structures
– Simple structure
• Formed by two-terminal devices
• Not limited by transistor scaling
– Ultra-high density
• NAND-like layout, cell size 4F2
• Terabit potential
– Large connectivity
– Memory, logic/neuromorphic applications crossbar Structure
single-cell structure
Lu Group EECS, UM 3
ITRS_ERD workshop, April 2010
Resistive Switching Memory
Lu Group EECS, UM 4
• Switching type: bipolar
• Electric field-driven redox
chemical effect
• Metal filament formation
• Electrode plays active role
(Ag or Cu)
• Materials: chalcogenides (e.g. GeS, GeSe, …), other amorphous films (e.g. oxides, a-Si, a-C, …)
RRAM – ECM Cell
Schindler et al., Proc. IEEE Non-volatile Memory Technology Symp. 82, 2007.
ElectroChemical Metallization Cell Ag/Ag-Ge-Se/Pt
Lu Group EECS, UM 5
•Ag/SiO2/Pt structure, sputtered SiO2 film •The filament grows from the IE backwards toward the AE •Branched structures were observed with wider branches pointing to the AE •Single filament dominates
ECM: Visualization of Filament
Partially formed filaments Completed filament
+ -
-
Ag
Pt
200nm
Yang, Gao, Chang, Gaba, Pan, and W. Lu, Nature Communications, 3, 732, 2012.
Lu Group EECS, UM 6
Before erasing After erasing
Pt
Ag
Pt
Ag
Ag/SiO2/Pt structure
Visualization of Filament, TEM
•A single filament dominates the switching process
Yang, Gao, Chang, Gaba, Pan, and W. Lu, Nature Communications, 3, 732, 2012.
Lu Group EECS, UM 7
Compositional Analysis of the filament
The filament was verified to be composed of elemental fcc Ag particles (i.e. not Ag ions or oxides) Thin Ag filaments are not stable and naturally break into discrete Ag particles High conductance can be maintained when the particles are closely spaced
Yang, Gao, Chang, Gaba, Pan, and W. Lu, Nature Communications, 3, 732, 2012.
Ag filament in SiO2
Lu Group EECS, UM 8
HRTEM, showing the particles are (111) Ag with fcc structure
Ag particles forming the filament
2 nm
• The filament was verified to be composed of elemental fcc Ag particles
• Thin Ag filaments are not stable and naturally break into discrete Ag particles
Compositional Analysis of the filament
Yang, Gao, Chang, Gaba, Pan, and W. Lu, Nature Communications, 3, 732, 2012.
Ag filament in a-Si
Lu Group EECS, UM 9
Visualization of Ag Filament, in-situ TEM
c d e f gAg Ag Ag Ag Ag
W W W W W
+ + + + +
- - - - -
W
a-Si
Ag
a b
Cu
rre
nt
(μA
)
c d e
g
f
0 100 200 300 400 500
Time (s)
0.0
0.5
1.0
1.5
2.0
2.5
20nm 20nm 20nm 20nm
100nm
Yang, et al. Nature Communications, 3, 732, 2012.
c d e f gAg Ag Ag Ag Ag
W W W W W
+ + + + +
- - - - -
W
a-Si
Ag
a b
Cu
rre
nt
(μA
)
c d e
g
f
0 100 200 300 400 500
Time (s)
0.0
0.5
1.0
1.5
2.0
2.5
20nm 20nm 20nm 20nm 20nm
100nm
Bulk RRAM on W probe Ag +
W -
Lu Group EECS, UM 10
Pillar structure
a-Si RRAM Crossbar Structure
•Two terminal resistance switching device
•Ag inside a-Si matrix
•Small cell size, < 50 nmx50 nm (density > 1010/cm2)
•CMOS compatible materials and processes
Jo et al. Nano Lett., 8, 392 (2008)
Crossbar array
Kim, Jo, W. Lu, Appl. Phys. Lett. 96, 053106 (2010)
Lu Group EECS, UM 11
Resistance Switching Characteristics
Kim, Jo, W. Lu, Appl. Phys. Lett. 96, 053106 (2010) Jo, Kim, W. Lu, Nano Lett., 8, 392 (2008) 0 200 400 600 800 1000 1200
0
50
-3
-2
-1
0
1
Volta
ge
(V)
Time (ns)
Curr
ent(n
A) on on off
read1 read2 read3-2V -3.5V
-4 -2 0 2 4 6
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Voltag e (V )
Curr
ent(
100
nA)
6-4 -2 0 2 4
10-12
Voltage(V)
10-6
10-7
10-8
10-9
10-10
10-11
Ag
Pt
IREAD
Sneak path
involves one
reverse-biased
cell
Lu Group EECS, UM 12
> 1e8 W/E endurance 1e6 on/off Can be switched within 50ns
Write/Read/Erase/Read pulse : 50nsec ,5V /50usec, 0.7V /100nsec, -3.5V /50usec, 0.7V
on off on off
Endurance and Speed
Kim et al, Appl. Phys. Lett. 96, 053106 (2010) Jo et al. Nano Lett., 8, 392 (2008)
-2 -1 0 1 2 3
10-14
10-12
10-10
10-8
Cu
rre
nt
(A)
Bias (V)
Virgin
106
107
108
100
101
102
103
104
105
106
107
108
0
10
20
30
40
50
Me
asu
red
Cu
rre
nt
(nA
)
Endurance Cycle
0 100 200 300 400
0.0
0.1
0.2
0.3
0.4
0.5
Ou
tpu
t (V
)Time (uS)
Endurance Cycles 104, 10
5
106, 10
7
Lu Group EECS, UM 13
• up to 8 levels (3 bits) per cell demonstrated
• cell resistance controlled by the current-limiting control resistor
Multi-Level Storage
Jo et al. Nano Lett. 9, 496-500 (2009).
Kim et al, Appl. Phys. Lett. 96, 053106 (2010)
Lu Group EECS, UM 14
Integrated Crossbar Array/CMOS System
Kim, Gaba, Wheeler, Cruz-Albrecht, Srivinara, W. Lu Nano Lett., 12, 389–395 (2012).
CMOS
Crossbar
array
500nm
•Low-temperature process, RRAM array fabricated on top of CMOS •CMOS provides address mux/demux •RRAM array: 100nm pitch, 50nm linewidth with density of 10Gbits/cm2
•CMOS units – larger but fewer units needed. 2n CMOS cells control n2 memory cells
“1R” array, no
external
selectors or
diodes
Lu Group EECS, UM 15
East
Data A
East Address
Decoder
East
Data B
East
Address <1>
East
Address <n>
Pad
Via between nanowire and CMOS
CMOS chip I/O terminal Pad
Switch 1
Switch 2
Switch N
N n
South Connections
North Connections
West C
on
ne
ction
s
via
Integrated Crossbar Array/CMOS System
Kim, Gaba, Wheeler, Cruz-Albrecht, Srivinara, W. Lu Nano Lett., 12, 389–395 (2012).
Lu Group EECS, UM 16
I-V of the integrated Crossbar/CMOS system
Threshold voltage (V)
•Tight distribution from 256 devices measured •Devices shown good on/off and intrinsic diode characteristics
Integrated Crossbar Array/CMOS System
Average 2.30 V Stan. dev. 0.07 V
Kim, Gaba, Wheeler, Cruz-Albrecht, Srivinara, W. Lu Nano Lett., 12, 389–395 (2012).
Lu Group EECS, UM 17
- Crossbar array operation, array written followed by read - Programming and reading through integrated CMOS address decoders - Each bit written with a single 3.5V, 100us pulse
Results from a 40x40 crossbar array integrated on CMOS
Integrated Crossbar Array/CMOS System
Stored/retrieved array 1 Stored/retrieved array 2
Kim, Gaba, Wheeler, Cruz-Albrecht, Srivinara, W. Lu Nano Lett., 12, 389–395 (2012).
Lu Group EECS, UM 18
1 and 0 states are clearly distinguishable from the 1600 cells in the crossbar
Target Ron = 500k
Integrated Crossbar Array/CMOS System
Kim et al. Nano Lett., 12, 389–395 (2012).
0
50
100
150
200
50K 100K 500K 1M 5M 10M 50M 100M 500M
50K 100K 500K 1M 5M 10M 50M 100M 500M
# in
bin
s Original
0
50
100
150
200
# in
bin
s
Complementary
Lu Group EECS, UM 19
Multi-Level Storage
•Different on-states can be obtained by changing Rs •Tight resistance distribution can still be obtained for multi-level storage
Kim et al. Nano Lett., 12, 389–395 (2012).
Lu Group EECS, UM 20
Crossbar Inc - Startup company founded in 2010 to fabricate a-Si
based RRAM in commercial fab.
Fully VC-funded, Silicon Valley HQ
CMOS compatible process with superior proven performance
Currently has ~ 20 full-time staffs
From Lab to Fab
• CMOS Compatible
• 3D Stackable, Scalable Architecture – Low thermal budget process
• Architectures proven include multiple Via schemes and Subtractive etching
Crossbar Inc’s non-volatile memory
technology RRAM array fabricated on 8”
wafers in a commercial fab
Lu Group EECS, UM 21
2
24
22
4
2
2/12/1
22)(2
eVm
h
lweVm
h
lw
eeV
eeV
lwh
eAI
I-V equation, tunneling through a barrier with thickness of w-l with barrier height
l is the length of the filament, determined by Eq. 2
lw
EV
lw
EV
eed
dt
dl00 //
0
de
kTE
20 d is the step length of the filament
(1)
(2)
Model Development
aS VRIV
The voltage across the device V is related to I and applied voltage Va
(3)
vvwGi ),(
),( vwfw
First order model: Two equations describing switching behavior w: state variable
P. Sheridan, K. Kim, W. Lu, Nanoscale 3, 3833 (2011).
RS
RRAM
Series resistor
w-l
Va
E=aV
Lu Group EECS, UM 22
Rs= 500kOhm
Filament length vs. applied voltage
Device current vs. applied voltage
Device voltage vs. applied voltage
SPICE Model – DC Sweep
•Direct SPICE simulation to predict RRAM switching dynamics.
•Threshold effect, multi-level, exponential dependence of switching time on voltage
captured
P. Sheridan, K. Kim, W. Lu, Nanoscale 3, 3833 (2011).
Lu Group EECS, UM 23
SPICE Model, Transient Effects
write
read
SET
SET
•RRAM switching transient effects captured in SPICE
P. Sheridan, K. Kim, W. Lu, Nanoscale 3, 3833 (2011).
Lu Group EECS, UM 24
Redox Memory – Valency Change Cell
• Materials: TiO2, HfO2, TaOx ...
• Switching type: bipolar
• Electric field-driven redox chemical effect
• Oxygen exchange between two pre-defined oxide layers
• “bulk” effect or filament effect
• On/off, uniformity
Oxide layer 1 “switching” layer
Oxide layer 2 “base” layer
“0” “1”
Lu Group EECS, UM 25
Valency-Change Devices Based on TaOx
Y. Yang, P. Sheridan, W. Lu, App. Phys. Lett. 100, 203112 (2012)
Bipolar switching
Lu Group EECS, UM 26
IREAD
Sneak path problem in passive Crossbar arrays
asymmetric
Possible device type 1
Vread-½ Vread
symmetric
½ Vread
Possible device type 2
-½ VreadVread
Valency-Change Devices Based on TaOx
Lu Group EECS, UM 27
Complementary Resistive Cell based on TaOx
•Internal distribution of VO can be used to represent the cell state
•Both 0 and 1 can have a high-resistive layer and are of high resistance at low bias
•All 4 states need to be accessed during CRS operation
Y. Yang, P. Sheridan, and W. Lu, Appl. Phys. Lett. 100, 203112 (2012)
CRS switching
Read “0”
Read “1”
Lu Group EECS, UM 28
Acknowledgements
*Sung-Hyun Jo, *Kuk-Hwan Kim
Siddharth Gaba, *Ting Chang
Patrick Sheridan, ShinHyun Choi
Jiantao Zhou, Chao Du, Jihang
Lee, Wen Ma, *Eric Dattoli
Wayne Fung, Lin Chen
*Seok-Youl Choi, *Woo Hyung Lee
Grad students:
•National Science Foundation (ECS-0601478, CCF-0621823, ECCS-0804863, CNS-0949667, CAREER ECCS-0954621). •DARPA SyNAPSE program •DARPA UPSIDE program •Air Force MURI program, Air Force q-2DEG program, Engineering Translational Research (ETR) Grant Dr. L. Liu
*Xiaojie Hao
PostDocs:
* Dr. Qing Wan
Visiting scholars:
*Dr. Zhongqing Ji
* alumni
Dr. Yuchao Yang Funding:
Collaborators:
•Dr. N. Srinivasa, Dr. D. Wheeler, Dr. T. Hussain. Dr. J. Cruz-Albrecht, HRL Labs, •Prof. Z. Zhang, Prof. P. Mazumder, Prof. M. Zochowski, UM, •Prof. D. Strukov, UCSB, •Prof. J. Hasler, GeorgiaTech, •Prof. G. Guo, Prof. T. Tu, USTC, China •Prof. R. Li, CAS, China