Download From Spin Torque Random Access Memory to Spintronic Memristor

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Opto-isolator wikipedia , lookup

Switched-mode power supply wikipedia , lookup

Electrical ballast wikipedia , lookup

Stepper motor wikipedia , lookup

Current source wikipedia , lookup

Buck converter wikipedia , lookup

Resistive opto-isolator wikipedia , lookup

Thermal runaway wikipedia , lookup

Alternating current wikipedia , lookup

Ohm's law wikipedia , lookup

Memristor wikipedia , lookup

Transcript
From Spin Torque Random Access Memory
to Spintronic Memristor
Xiaobin Wang
Seagate Technology
Contents
Spin Torque Random Access Memory:
dynamics characterization, device scale down challenges and opportunities
Spin Torque Memristor:
concept, device and application examples
X. Wang − 2
Spin Torque Random Access Memory
(SPRAM) Working Principle
Top electrode
Magnetic
tunneling junction
(MTJ)
MTJ
Reference layer
Bottom electrode
.
Select
Transistor
(1)
Tunneling insulating
barrier
Free layer
Integrate magnetic tunneling junction (MTJ) with CMOS
Reading through magneto-resistance principle
Writing through spin torque excitation
X. Wang − 3
SPRAM Switching Behavior
1
I
RH
6
MTJ coupled with CMOS
2
RAP
4
5
RL
-
0
3
+
V
Resistance (Ohm)
MTJ static picture
Ref: Y. Chen, X. Wang, H. Li, D. Dimitar, H. Liu,
International Symposium on Quality Electronic Design, 2008.
MTJ switching asymmetry
Spin torque switching
asymmetry for a field
balanced MTJ stack
Reference: X. Wang, W. Zhu, D. Dimitrov,
Phys. Rev. B, 79, 104408, 2009.
X. Wang − 4
Magnetic field (Oe)
SPRAM System Dynamical Approach:
Quantum-Micromagnetic-SPICE
Macroscopic magnetic
(size, Ms, etc)
Dynamical
approach
Dynamic Magnetization
25
20
MTJ electronic spin transport
(band structure, tunneling)
15
15
10
10
5
CMOS I-V curve,
equivalent capacitors,etc
5
Ref: X. Wang, Y. Chen, H. Li, D. Dimitar, H. Liu,
IEEE Tran. On Magn., 44(11), 2470, 2008
Dynamic CMOS circuit
VDD
BL
I M + I GD = I DB + I T
CGD
CDB
WL
CSB
CGS
CGB
SL
 dV 
I GD = C GD ⋅  −

 dt 
 dV 
I DB = C DB ⋅ 

 dt 
I M ⋅ R(t ) = VDD − V (BL → SL )or
I M ⋅ R(t ) = −V
(SL → BL )
Reference: X. Wang, W. Zhu, D. Dimitrov,
Phys. Rev. B, 79, 104408, 2009.
Quantum
electronic spin transport
X. Wang − 5
Challenge for SPRAM to Scale Down
150
CMOS driving current (uA)
Field driving: switching field
scales up with 1/volume or
1/surface dimension
140
130
120
110
100
22nm x 75nm
CMOS
22nm x 45nm
MTJ, 13.2F2
90
80
70
1500
Spin torque driving:
switching current density
scales up with 1/volume or
1/surface dimension
22nm
22 nm x 100nm
C
x 45 n
m MT MOS
J, 17 . 2
4F
22nm x 45nm
CMOS
22nm x 22nm
MTJ, 9F2
2000
2500
3000
MTJ Resistance (Ohm)
As magnetic device scales down, with CMOS driving stength
decreasing, achieving fast nanosecond time scale
magnetization switching and maintaining thermal stability at
years time scale become a challenge.
X. Wang − 6
Challenge for SPRAM to Scale Down
Variation of single device different switching
Spin torque MRAM device to device variations
reduce sensing and writing margins
As device scales down, increased
variability is another challenge.
X. Wang − 7
Current Reduction Through Magnetization Dynamics
critical switching current ∝
M s2Vα
η
( Dz − D y )( Dz − Dx )
Dx , D y : demag factors in - plane, Dz : demag factor out - plane
Stable energy in equilibrium : M s2V ( D y − Dx )
Dz >> Dx, Dy, different scaling behavior of MTJ writing current magnitude and thermal stability energy
Geometric wise: thin film for TMR etc.
Magnetization dynamics wise: Symmetric
magnetization motion: Dz =Dy
Road leads to same scaling of
switching current and thermal
stability energy
X. Wang − 8
Current Reduction Through Magnetization Dynamics
RF frequency time varying polarization angle
s
ion
t
u
ol
e
dir
n
tio
a
z
i
lar
o
p
nt
a
t
ns
on solution
ti
c
e
ir
d
n
o
ti
Co
a
riz
varying pola
e
m
ti
l
a
m
ti
p
O
Hard magnetic layer to
keep thermal stability
Normalized magnetization Mz/Ms
on
cti
Composite material stack
Soft magnetic layer to
decrease switching
field/current magnitude
1
0.8
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
0.2
0.4
0.6
0.8
1
1.2
1.4 x 10-8
Time (sec)
X. Wang − 9
Current Reduction Through Electronic and Spin
Transport Across Interface
More opaque
barrier, less
magnetization
precessing induced
spin pumping,
effective damping
reduction
MTJ with spin moment conservation
layer between free layer and top cap
layer
Increase insulating layer thickness only, RA increases 5 times, critical
switching voltage only increases 2.5 times
150
130
1.0
110
90
0.5
Jc(MA/cm^2)
MR(%)
Ref: X. Wang, W. Zhu, Y. Zheng, Z. Gao, H. Xi, INTERMAG 2009
70
50
30
10
-10
-30
-50
Double AP->P
Single0.0
Double P->AP
Double
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
-0.5
-1.0
0.000001
0.0001
0.01
1
tp(s)
J(MA/cm^2)
Dual tunneling barrier, critical switching current down to 0.6MA/cm2, quantum confinement effects ?
Ref: Y. Zheng, W. Zhu, X. Wang, Z. Gao, D. Wang, D. Dimitrov et. al. submitted to Appli. Phys. Letters
X. Wang − 10
Ω)
MTJ Resistance (Ω
Variation Controlling at Device Level
2800
2600
2400
2200
2000
1800
1600
1400
1200
1000
-1000
40ns pulse
DC extrapolation
RH
destructive
RL
-500
0
IR
Sensing Current (µ
µ A)
500
1000
Unique Device characters leads to new
sensing scheme
Self-reference
sensing
non-destructive
Ref. Y. Chen, H. Li, X. Wang, W. Zhu, W. Xu and T. Zhang,
Design, Automation & Test in Europe Conference and Exhibition, 2010.
X. Wang − 11
Variation Controlling at System Level
Reduce switching current requirement by intentionally move away from worst case scenario design
a smaller-than-worst-case transistor sizing approach. For 256Mb SPRAM design at 45nm node, under
a normalized write current threshold deviation of 20%, the overall memory die size can be reduced by
more than 20% compared with the conventional worst-case transistor sizing design.
Ref: W. Xu, Y. Chen, X. Wang, and T. Zhang, 46th Design Automation Conference, 2009.
X. Wang − 12
Heat Asssited Solutions
Surface anti-ferromagnetic coupled (AFC) magnetic layer with relatively low Curier temperature:
At room temperature, the AFC coupling provides surface anisotropy to maintain thermal stability of the square magnetic
element.
During the writing process, the joule heating of spin torque current raises temperature of AFC surface magnetic layer above
Curier temperature and the AFC induced surface anisotropy disappears. The element can be switched at low threshold current.
500
1.00
6 nsec
10 nsec
300
Cell size (um2)
I (µA)
400
64Mb
Required heating current
200
256Mb
0.10
65nm, 6nsec
65nm, 10nsec
1Gb
100
0.01
0
0
20
40
60
80
D (nm)
100
120
140
40nm, 6nsec
40nm, 10nsec
RA = 12.5 Ωµm2
0
20
40
60
80
100
120
140
D (nm)
Write current decreases with decreasing memory element dimension D. Meanwhile, the MTJ resistance increases
with decreasing memory element dimension. For a given technology node, an optimal memory cell size can be
found due to the tradeoff between critical switching current and MTJ resistance
Ref: H. Xi, J. Sreicklin, H. Li, Y. Chen, X. Wang, and Y. Zheng, Z. Gao, M. Tang, IEEE Trans. Magn, 46,no. 3, 860, 2010
X. Wang − 13
MTJ Memristor and Mutibit Data Storage and Logic
00
demonstrated multi-bit MTJ devices
01
10
Ref: X. Lou, Z. Gao, D. Dimitrov, and
M. Tang, Appli. Phys. Letters, 93,
242502, 2008
11
From nonlinear resistor point of view:
reversible branch: 11 to 10, 00 to 01
Irreversible branch: 00 to 10
MTJ Memristor and Mutibit Data Storage and Logic
Switching current (uA)
1400
1200
From memristor point of view:
1000
800
600
400
e
re
o
m
200
0
0
0.5
te
a
g
lo n
1
nd
a
pe
a
h
ds
fac
r
u
ss
s
e
l
switch
00->01
1.5
2
Switching from 00 to 10 and from
00 to 01 are all possible by
adjusting pulse width
rea
a
e
Different slope, switching depends
upon integration of current, not
current magnitude !
switch
00->10
2.5
3 x 10
8
Inverse of pulse width (sec-1)
Ref. X. Wang, Y. Chen,
Design, Automation & Test in Europe Conference and Exhibition, 2010.
X. Wang − 15
Memristor as Fourth Circuit Element
Resistor
Memristor
What makes memristor different from an ordinary constant resistor or even a current
or voltage dependent nonlinear resistor: memristance is a function of charge, which
depends upon the hysteretic behavior of the current (or voltage) profile.
V
ϕ = ∫ Vdt
capacitor
Leon O. Chua, Memristor-the missing circuit element,
IEEE Trans. Circuit Theory, vol. 18, no. 5, 1971
Chua (1971) proposed memristor for
logic completeness of circuit element
q
q = ∫ Idt
What makes memristor different from a capacitor or inductor is
the ability to accumulate current or voltage information at X. Wang − 16
constant voltage or current.
Memristance in Resistive Memory Stack
low resistance state
high resistance state
measured
simulated
1) Nano-scale two terminal device as two connecting resistors
(doped region and undoped region)
2) Bias electric current drives the front of the doped region
1) J. M. Tour and T. He, Nature 453, 42, (2008).
2) D. B. Strukov; G. S. Snider; D. R. Stewart and R. S. Williams, Nature, 453, 80, (2008)
X. Wang − 17
Spintronic Memristor Through Spin
Torque Induced Magnetization Motion
GMR/TMR device:
resistance depends upon
magnetization state
Spintronic memristor:
resistance depends upon the integral
effects of its current profile.
Spin Torque device: current
electronic spin changes the
magnetization state of the device.
The magnetization state of the
device depends upon the
cumulative effects of electron spin
excitations.
X. Wang, Y. Chen. H. Xi, H. Li, D. Dimitrov:
IEEE electronic device letters vol. 30, no. 3,
pp.294, 2009.
X. Wang − 18
Domain Wall Memristor and Temperature Sensor
interactions between
macroscopic coherent
magnetic structures and
random fluctuations
Nano-scale feature size, low cost, and
mature integration technology with the
CMOS process.
Targeting the highly integrated on-chip
thermal detection applications
X. Wang, Y. Chen, Y. Gu, H. Li: IEEE electronic device letters vol. 31, no. 1, pp.20, 2010.
X. Wang − 19
Memristor for Power Management
The ability to accumulate current/voltage through
constant current and/or voltage driving strength:
power monitor device
∫
∫
E = VIdt = V Idt
Negative feedback: power control device
Ref. X. Wang, Y. Chen,
Design, Automation & Test in Europe Conference and Exhibition, 2010.
X. Wang − 20
Memristor for Data Security
Question: Administrator and user. Design a scheme to fight against following action: after reading the stored data information, the
user tries to restore the device state to the same state as before reading, so that administrator could not find whether the data has
been accessed.
Solution depends upon whether the user is given the limited or the same authority on device writing and reading compared to that
of the administrator. For the case of user with limited reading and/or writing authority, data information security task can be
achieved more easily. For limited writing authority case, user may not be able to restore the device to the state set by the
administrator before. For reading limited case, the user may not know the state of the device set by the administrator.
A solution giving user and administrator the same authority on
reading and writing:
1) Writing process: administrator sets high resistance state (0)
fully saturated and low resistance state (1) partially saturated.
2) Reading process: two constant voltage pulses excite device a
few times (order of 10). One pulse pushes domain wall toward
high resistance end and the other pulse tries to push domain
wall toward low resistance end.
Final domain wall settled position
0.8
0.7
0.6
0.5
0.4
0.3
0.2
Administrator reads the data
using pulses with a particular
second pulse driving strength.
0.1
3) The device reports two values for reading: 1) final state of
00.8
1
1.2 1.4 1.6 1.8
2
2.2
the device close to high or low resistance state (High or Low)
and 2) whether the device resistance has been significantly
Second pulse magnitude
changed during reading (Yes or No).
Ref. X. Wang, Y. Chen,
X. Wang − 21
Design, Automation & Test in Europe Conference and Exhibition, 2010.
Path to Future: Potentials
GMR with spin valve structure
Physics process is clear
and scaling behavior is
well understood:
controllable and
tunable device
TMR with tunneling structure
Easily integration on top of CMOS
New geometry structures
New material
Y. V. Pershin and M. Di Ventra, Phys. Rev. B, 78, 113309, 2008
Spin blockade, half-metallic material
X. Wang − 22
Path to Future…….
nano-scale, ultra-fast…….
variability,
memory effects…….
analogue/digit mixture
fuzzy,
neural system
……
deterministic, precise
digital
control
X. Wang − 23
……..