Transcript Spintronic Memories -- GRC

Spintronic Memories
 Magnetic Domain Wall Motion memory
 Magnetic race-track memory
 Stuart Parkin, “Magnetic race-track – a novel storage class
spintronic memory”, Intern J. Mod. Physics B 22 (2008) 117
 Spin torque transfer MRAM
Selection criteria for new technology
entries candidates
“Minimum Requirement” Criteria
 The ‘minimum requirements’ criteria for a new
technology to be considered as a candidate for
ERD chapter is sufficient research activity, e.g.
 the technology is explored by several research groups,
or
 There is an extensive research activity of one group
 There are at least 2 publications in peer-reviewed
journals
General ‘Loose’ Criteria
 Potential for scaling
 Roadmap Driver
 On-chip integrated solutions
 Potential for embedded applications
 Outstanding research issues exist
 Guidelines for the research community and government
funding agencies
 Not in production
 Innovation phase
Spin torque transfer MRAM
Conventional MRAM
MRAM element is operated by magnetic
field generated from current lines in
the proximity of MTJ
FM free layer
W
FM pinned layer
“Wireless communication”
Proximity effects – crosstalk
Scaling issue, Heat, reliability etc.
I, mA
MRAM Scaling issue
10
9
8
7
6
5
4
3
2
1
0
10
T. Kawahara et al, “2Mb SPRAM (SpinTransfer Torque RAM)…”, IEEE J. SolidState Circ. 43 (2008) 109
100
L, nm
Hitachi group
Scaling issue: Conventional MRAM needs
a larger current for smaller dimensions
1000
Spin-torque switching
 Injected spin-polarized electrons interact with the
magnetic moment of a free layer and transfer their
angular momentum
 If sufficient current is applied, the exerted spin torque
switches the free layer either parallel or anti-parallel to
the pinned layer depending on the direction of flow of
the current
 Attractive for memory array applications,
 does not have the magnetic half-select problem
 smaller switching current
 Spin torque transfer RAM (ST-RAM)
 MTJ for spin-torque switching
MTJ for spin-torque switching
MTJ is operated by spin polarized current
passing through MTJ
Injection
efficiency
W
I c ~ N 
~Nat
~L2
Scaling promise: spin-torque MRAM needs
a smaller current for smaller dimensions
MRAM and ST-MRAM Scaling
Reciprocal Scaling Relations
MRAM and ST-MRAM Scaling
100
10
I, mA
1
electromagnetic WRITE
Spin-torque WRITE
0.1
0.01
0.001
0.0001
1
10
100
L, nm
1000
Switching time vs. current
Outstanding research issues
 Theory:
“Although the presence of spin torque has been unambiguously
observed, its quantitative behavior in MTJ, especially its bias
dependence has yet to be understood in detail”
J. C. Sankey et al., Nature Physics 4 (2008) 67
 Critical current issue
 Ic needs to be decreased
107
A/cm2
105
A/cm2
 From
to
 New material structures
 MTJ current needs to be increased
 New MTJ design
 Injection efficiency
IBM group
I c ~ N 
New concepts are needed
 Nano-current-channel (NCC) injection
 FeSiO layer with columnar NCC structure
 Current can pass through NCC only
 Provide magnetic nucleation points and induce the free layer
switching through the growth of the nucleation points
General ‘Loose’ Criteria Discussion:
Spin torque transfer MRAM
Does is belong to PIDS or to ERD?
MRAM has been transferred from ERD
to PIDS in 2003
 Potential for scaling
 Roadmap Driver
 On-chip integrated solutions
 Potential for embedded applications
 Outstanding research issues exist
 Guidelines for the research community and government funding
agencies
 Not in production
 Innovation phase
Magnetic Domain Wall Motion memory
Magnetic Domain Wall Motion memory
 Current-driven magnetic domain wall (DW) motion
 DWM
 DWM occurs in a submicron-size ferromagnetic
stripe
 Charge carriers become polarized by the
interaction between conduction electrons and local
magnetic moments
 Exert torque on the magnetic moments within DW
 Sensed by TMR or GMR device
Magnetic Race-track Memory
 A proposal for a novel storage-class memory, in which
magnetic domains are used to store information in a
“magnetic race-track”
 Shift register scheme
 A solid state memory with storage capacity same/better
than HDD
 Improved performance and reliability
 The magnetic race track is comprised of tall columns of
magnetic material arranged perpendicularly to the Si
surface
 The domains are moved up and down by current pulses
 ~ns pulses
 Sensing by magnetic tunnel junction device
DWM at ~107 has
been demonstrated
Domain wall velocity as a function of domain
wall width
Benakli et al.,
JAP 103 (2008)
Seagate Group
Planar Configuration
t=5 nm
W
L
N, bit
n,
bit/cm2
R
I
100
2.0E+05
5.0E+09
4.0E+06
5.0E-03
20000
10
2.0E+06
5.0E+11
4.0E+07
5.0E-04
20000
W, nm
V
P,
W/cm2
J,
A/cm2
1
4.8E+06
1.0E+09
1
3.3E+06
1.0E+09
L, cm
N~
W, nm
cm
2L
1
W
N, bit
n,
bit/cm2
N
N
n ~
A 4W  L  W 
R
I
V
L,L,um
um
cm
P,
P, W/cm2
J, A/cm2
1.00E-05
100
2.10E+01 4.77E+09 4.00E+02 5.00E-05
5.00E-03
5.00E-06
0.02
2
0.00
11
4.76E+06
476
1.00E-04
5
1.00E+07
1.00E+09
1.00E+06
10
5.00E-04
5.00E-06
1.00E-06
2.01E+02 4.98E+11 4.00E+03 5.00E-07
0.02
2
0.00
11
3.33E+06
333
0.0001
3
1.00E+09
1.00E+07
1.00E+06
Main Issue
 Due to the high current densities, strong heating
occurs
 DW transformations have been shown to originate not
only from spin torque effects but also from thermal
excitations
 For applications, it is a key requirement to devise ways
for efficient cooling
There is a considerable interest in DWM





IBM
Samsung
Hitachi
Seagate
Canon
Outstanding research issues
 The capacity of spin-polarized current to move a
domain wall was experimental established, but
 The mechanisms responsible for that motion
remain under debate
 Current density needs to be decreased!