1

L12 – Magnetization Dynamics and Applications

With many thanks to Prof. G. Csaba, Pazmany University, Budapest

Many slides are adopted from a lecture he gave at TUM.

Prof. Markus Becherer

Technische Universität München

Department of Electrical and Computer Engineering
Chip-Based Magnetic Sensor Technology

Discover the world of

Nanoelectronics magnetism
 2

L01 Kick-off + ERD roadmap

Nanofabrication Nanometrology/Patterning Applications Magnetic devices

L02 Conversion + PVD
L03 CVD + Others
L04 Nanometrology I - Electrical
L05 Nanometrology II - Optical and SPM
L06 Nanometrology III - Beams of Electrons/Ions
L07 Patterning - Lithography
L08 Metallic Nanowires I
L09 Metallic Nanowires II
L10 Quantum LEDs with Si NP
o Landau-Lifshitz-Gilbert (LLG) equation
o Switching of magnetic domains L11 Introduction (nano)magnetism
L12 Magnetization dynamics
o Micromagnetic simulations and modeling
L13 Spin-wave devices
o Magnetic switching and electric/magnetic interactions
L14 Nanomagnetic Logic
o Magnetic random access memory (MRAM) and L15 Nanomagnetic Devices 2.0
magnetoelectric effects

We have been discussing magnetostatic problems before. Now, if we think of applications,

we have to consider dynamic behavior and switching phenomena in magnetism as well.
 3

Magnetic
Domain switching in Reality North-pole

Multi domain states Single domain states Magnetic

South-pole

More movies see images at :

[Link]
omez/[Link]
As the magnetic field changes, different domain structures are appearing 3

On the left graph, you see snapshots of a magnetic multidomain state, that is put together
to a small video clip. By applying (external) positive and negative magnetic fields, the
magnetic domains are shifted. As this is a PDF only, you should visit the website to
see it. There is a micrometer-sized Permalloy (soft-magnetic FeNi) thin film island. On
the right, you see a (1) topography image of FIB milled Permalloy needles with two
extra magnets on the side. The rightmost image is showing the magnetic contrast of
the needle-like structures: north-poles appear dark, and south-poles appear bright. No
dynamic behavior so far, but it was made to test domain wall movement and its capturing
between the two side magnets, depending on their stray fields.
 4

Modeling
hierarchy
in
magnetism

If we look at dynamic behavior and its modeling, micromagnetic analysis is the most
powerful tool — at least if we consider nanomagnetic device behavior and dynamics. The
vector fields and quasi-classical description of quantum behavior by means of effective
fields (i.e. exchange interactions are described by exchange fields) is a smart way to
handle the length-scales of up to several µm.
 5

The Landau-Lifshitz-Gilbert equation

5
 6

Torque on the angular momentum

We use the dynamic equation for two purposes:
• To really understand dynamics
• To find the minima of the energy equation without doing too much mathematics ☺
A quick summary on Lorentz forces (FL = q * (v x B)), torques and angular momentum
Important: there is no net force on the current loop, only torque

B B
m m
τ =mxB
vxB vxB
v v
If the magnetic induction field B is parallel to the In this case there will be a torque on the current loop
magnetic moment, then there will be no torque τ and m will make a precessional motion around B
on the current loop.

The torque τ (which is a vector) is perpendicular both to m and B : τ = m x B

6

In order to model magnetization dynamics, we have to describe the dynamic behavior of

spins. Consider a current loop, which is the orbital movement of an electron: basically,
when changing the magnetic inductions field B in a manner, that is not co-linear with
the magnetization m a torque τ is generated, due to the Lorentz force (right sketch). As
a result, m (which is the local magnetization) will make a precessional motion around
B.
 7

Recap: Origin of Magnetism: Orbital magnetic moment

orbital magnetic moment and mechanical
angular momentum are linked

Since angular momentum is quantized in an atom, the magnetic

moment also must be quantized and must be an integer multiple
of the Bohr magneton:

J. M. D. Coey, Magnetism and Magnetic Materials, Cambridge University Press, 2009 7

The recap should be more or less clear: angular momentum is proportionally linked to
magnetic orbital momentum.
 8

orbital magnetic moment and

mechanical angular momentum
Equation of motion for a magnetic moment are linked

g=1 for orbital motion

g=2 for spin
g=2.002319 (considering a lot of
quantum electrodynamics)

This we call Landau-Lifshitz gyromagnetic ratio, often denoted by simply γ

→ This equation gives the dynamics of an undamped magnetic moment in a field

mr

mz
m(t)

The precession frequency of the magnetic moment will depend on the

applied external field (B). The proportionality constant is 28 GHz / T.
8

Now, to describe the dynamics, we perform the time-derivative of mechanical momentum

which equals the torque, and that has to be equal to the cross-product of m and H. By
feeding in the right value for g, which was 1 for the orbital motion and 2 for spin behavior
and something more precise by considering quantum electrodynamics, we can substitute
all the constants in the equation with the so-called Landau-Lifshitz gyromagnetic ra-
tio denoted γLLG . But more importantly, this equation is the undamped precession of
magnetic moment around the applied magnetic induction field. And, the stronger the
magnetic field, the faster the precession. 28 GHz T−1 , which is a technically interesting
range.
 9

Including dissipation in the dynamic equation

We try to construct a vector that dampens (breaks) the
precession movement. H eff
This is a phenomenological argument

Should be proportional to M x Heff so

→ the faster it precesses, the faster it is ‘braking’
Red arrow
is the damping
Should be perpendicular to M x Heff so
→ that it is purely dissipative
Should be perpendicular to M so
→ that it preserves the magnitude of M

M
Physically, dissipation is a complex multi-step process that transfers energy from the magnetic Blue arrow is the precession
degree of freedom to heat. A typical process:
1. Change in M → change in volume (Magnetostriction) → sound waves (phonons) → heat
2. Change in M → electromagnetic radiation → absorption of radiation → heat
9

You can imagine, that an undamped precessional motion is not very physical. There have
to be losses and the phenomenological arguments are listed here. The damping vector
is perpendicular to the effective magnetic field Heff , summing all field contributions
like anisotropy field, exchange field, demagnetization field, Zeeman field etc. Physically
speaking, dissipation is a complex multi-step process that transfers energy from the
magnetic degree of freedom to heat.
 10

The Landau-Lifshitz-Gilbert equation

The magnitude of the α dimensionless damping constant characterizes H eff

the relative strength of the precession and damping terms
• α ~ 1 → magnetization falls in the effective field (high damping)
• α << 1 → magnetization makes several oscillations before settling

Typical values:
α ~ 1 for small or very rough particles (surface effects)
α ~ 0.01 – 0.1 for anisotropic metallic ferromagnets (Fe, Co, CoPt)
α ~ 0.008 for Permalloy
α ~ 0.00001 for Yttrium-Iron Garnet (YIG) M
α ~ 0.5 often used for micromagnetic simulations
Fast convergence and still some precession – it is somewhat heuristic
Energy minimum is reached when magnetization becomes parallel with the local effective field 10

Now, we are ready for the famous Landau-Lifshitz-Gilbert equation. A partial differ-
ential equation describing the time-derivative of the magnetization vector field M. We
acknowledge a precessional term (red) with magnitude γ and the damping term (blue),
with damping parameter α. The smaller α the smaller the damping, the more oscillations
are expected. For example, the YIG films in your lab course should’ve had an α in the
range of 2 × 10−4
 11

Precession term: Dissipative term:

H eff
Side view: Side view:

H eff H eff

M M

M
Top view: Top view:
The precession term makes the
magnetization going around for
forever (without changing energy)

The dissipative term pulls the

magnetization toward the effective
field vector.
11

Another summary on precession and damping terms. The LL equation is the most
important equation for micromagnetic simulations, therefore I spent many slides on it.
 12

Three ways of interpreting and using the LLG equation

1: It is the equation of motion for a really small magnetic moment
Often a good approximation for single-domain nanomagnets

2: It is the equation of motion for each magnetic moment in a ferromagnetic volume

In order to find the dynamic behavior of any magnetic material, we split (discretize) it
into elementary volumes. Any micromagnetic problem can be formulated as a solution
to the time dependent LLG equation.

3: It is a way of finding stationary domain patterns

Stationary solutions of the LLG equation are global or local energy minima of the
¶M (r,t)
ferromagnetic material. The LLG equation is a mathematical tool to find this =0
minima. For faster convergence the LLG equation is often overdamped (higher ¶t
than real α is used).
12

Interpretation is important, especially when it comes to physics.

 13

Micro
Two vector fields: M (r, t ) H eff (r, t )
mag
¶M(r,t) ag é
netic
¶t
= -g M(r,t) ´ H eff (r,t) -
Ms ë
(
M(r,t) ´ M(r,t) ´ H eff (r,t) ùû )
theory
summa Landau – Lifshitz equation: how the magnetization distribution
changes under the influence of an effective field H eff (r, t )
rized

The effective field: H eff (r, t ) = H(r, t ) + H exch (r, t )

2 Ax M(r, t )
B(r, t ) Hexch (r,t) =
Maxwell equations H(r, t ) = − M(r, t ) 0 M s2
“Real” magnetic field 0 Exchange field: a quasi –
classical inclusion of
div B(r, t ) = 0 quantum – mechanical
interactions
Magnetic domain theory 13

This slide highlights the clever trick of including quantum-mechanical behavior like ex-
change interaction as vector field and treating it as such in the Landau - Lifshitz equation.
 14

Micromagnetic modeling

14
 15

Flow of micromagnetic simulation

Initialize magnetization distribution

Calculate volume charges
Calculate the effective magnetic field
Calculate the change of magnetization

Now, we are all set to run micromagnetic simulations

15

Even though, the LL equations look compact, the implementation of the LL equation
in a micromagnetic simulation tool needs quite some effort. Just to give you an idea of
what is inside of the tools available.
 16

An open source micromagnetic solver OOMMF

First download the active Tcl system:
→[Link]

Go to [Link]
→ Download the latest version under ‘Development snapshots’

For windows: just extract the zip file in a folder you like
→‘[Link]’ is the script that starts the program

16

But of course, you will probably never write your own micromagnetic simulator but use
an open source or commercial tool instead. Here is a well-established software, OOMMF,
that does it for you. A bit old-fashioned but well documented.
 17

A second
open-source
micromag-
netic solver
mumax3

with GPU support

and web interface
based frontend

17

This is a kind of newer simulator with GPU support. If you do micromagnetic simula-
tions, I can really recommend this open-source simulator — the learning curve will be
really steep.
 18

Shape dependence of coercivity

B
0.25
the smaller the magnet, M
the higher the switching field
B 50nm wide

Switching field [T]

M
1 0.2
100nm wide
Applied
external 0.15

field 200nm wide

2 0.1

300nm wide
0.05

3
0
0 200 400 600 800 1000 1200 1400
Major axis [nm]

Coherent reversal Incoherent reversal 18

Now the question arises, how do we switch or inverse magnetization of such a single
domain magnet? Let’s look at the simulated switching field that is plotted versus the
length of the major magnet axis. For that, in step (1) a magnetic induction field B is
applied in the opposite direction to M of a single domain magnet. After that, the field
amplitude, where a reversal of magnetization occurs (2 −→ 3) is recorded. For small
major axis, we see a coherent reversal, whereas for larger magnets we see an incoherent
reversal. The higher the aspect ratio, the larger the coercivity = switching field of the
magnet. Even more, when the aspect ratio is too little, and if we are dealing with larger
magnets, incoherent reversal, and a kind of domain wall forms during reversal.
 19

Varga et al. IEEE International conference on Nanotechnology, Portland, 2011 →

Single-domain hysteresis curves Application →

Applied
Low aspect ratio magnet High aspect ratio magnet
field
direction Mx Mx
1.0 My 1.0 My
0.8 Mz Mz
0.8
0.6 0.6

M 0.4
0.2
0.4
0.2
B 0.0
-0.2
0.0
-0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1.0 -1.0
-8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
Field x 10
5
5
Field x 10

1.0 Mx
My Mx
Mz 1.0 My
0.8
Mz
0.8
0.6
0.6
0.4
0.4
0.2
B 0.0
0.2
0.0
-0.2
-0.2
-0.4
-0.4
-0.6
-0.6
-0.8
-0.8
a) SEM and b) MFM images of a full
-1.0
-1.0 adder built from 53 nanomagnets.
-8 -6 -4 -2 0 2 4 6 8
Field x 10
5
-8 -6 -4 -2 0
Field
2 4 6 8
x 10
5 19

The same result as seen before, now with bar magnets and an externally applied field
(red arrow) parallel to the major axis of the magnet (top) and with some angle to the
major axis (bottom). The hysteresis loops recording Mx, My, and Mz are narrower for
tilted fields and narrower for low aspect ratio magnets. In other words, by playing with
the aspect ratio and tilted externally applied magnetic fields, we can significantly change
the switching field (coercivity) of magnets. One application was the use of magnets with
different aspect ratios in nanomagnetic logic devices from the group of the University of
Notre Dame, Indiana, USA as shown on the right.
 20

Domain wall nucleation and motion

Nucleation region
Switching by domain wall motion is always a two-step process:
• Nucleation
• Propagation
Hard end

Creation of a head-to-head wall Creation of a tail-to-tail wall

Initial state:

M M

After wall starts turning in ‘head’ After wall starts turning in ‘head’
B
B
DW moves right DW moves right

M M
20

There were device proposals (and some current sensors are really using this effect), where
one plays with the movement of domain walls during reversal. Let’s look at the pin-like
shape of magnetic thin film. There is a large area acting as a nucleation region, a
magnetic strip, and a tapered hard end. This structure is injections domain walls, when
applying a magnetic field opposite to the dipole that is formed. On the left, a so-called
head-to-head domain wall is formed, on the right a tail-to-tail wall is formed, depending
on the initial magnetization and applied external field. Important here, a domain wall is
moving along the thin strip and reversing the initial dipole magnetization.
 21

Applications: In plane magnetic logic

2) Allwood et al., Science, 2002. — 3) Varga et al., IEEE Int. Conf. Nanotechnology, 2011. 21

In this science paper on the left, a famous group in GB proposed logic devices out of
such domain wall conduits (strips). By applying an in-plane rotating field, head-to-head
or tail-to-tail domain walls (DWs) were circulating, and by constrictions and correctly
chosen tapering, the type of DW was changed. The shown device is a NOT-gate, inverting
the type of wall which is the binary state (digital 0 and 1). By contrast, the devices
from Notre Dame on the right, used magnetostatic interaction between single domain
magnets to form complex gates like a 1-bit full-adder and the binary state is encoded in
the direction of the magnetization at the outputs C0 and S.
 22

Magnetic multi- MS

Magnetization
domain hysteresis Hc

M (r, t )

Hc

-20 -15 -10 -5 0 5 10 15 20

Applied induction field (mT)
2μm  2μm  30nm
Barkhausen noise is the
consequence of discrete jumps
on the hysteresis curve
22

Of the many hysteresis loops I have shown you so far, this is one on the micromagnetic
scale. You can see many jumps in M when sweeping the field. But the processes are
not reversible. The two cycles are different, because, the domains form in a different
way. And the jumps you see, are the world-famous Barkhausen noise one can even see in
macroscopic hysteresis loops —- you just have to look closely (and be sensitive) enough.
 23

A note about permeabilities Sometimes it is more relevant to use

differential permeability, i.e. taking the slope of the
The flux density (B) that the sample is generating M(H) curve around a certain field value.
Binduced =  0 ( H + M )   0 M
A
M = M s = 8.6 105
m
Binduced = 1.08 T
Above value for saturated Permalloy
(Ni80%, Fe20%), a very common soft-
magnetic material.

Permeability results as a rough linear fit:

23

Let’s look at the previous hysteresis loop again. When talking about magnetic materials
on a macroscopic level, the magnetic permeability is a constant or proportionality factor
connecting the induced and applied induction field. In the graph on the right it is the
approximated linear fit to the hysteresis loop, but it makes only sense for small magnetic
fields until the material saturates. Hence, a differential permeability for each operating
point is often the better description. Hysteresis very generally is a phenomenon not easy
to handle, as one has to consider the history of the sample/material: complete textbooks
are written on hysteresis and how to treat it.
 24

Magnetic switching and electric / magnetic interactions

24
 25

Energy barriers and thermally activated switching

æ -DE ö
E fswitching = f0 exp ç
è kT ÷ø
g Hc
kT f0 » » 0.1 GHz - 10 GHz
DE 2p
1
Tmean =
fswitching
u

This is a widely-used design criterion in data storage: choosing potential

barriers such as ΔE > 60 kT, information is safely retained for t > 10 years. 25

First, let’s look at the limit, where ferromagnetic behavior becomes random due to el-
evated temperatures. This, we call superparamagnetic limit. Imagine, that we have a
single domain state with two possible states. It turns out, that the switching probability
follows the Arrhenius law and in order to have a stable state for more than 10 years
a barrier height of 60 kT separating the states is needed. A finding coming from HDD
research. You do not want to lose your bits earlier.
 26

At which size do they start to

Superparamagnetic behavior at T > 0K randomly switch
→ Superparamagnetic limit

tflip = 1 / f0

t = 1 ms
Magnetization jiggles and does not reach Particle can spontaneously flip between states
steady state
26

Micromagnetic simulations can again help in understanding this phenomenon. A mall

volume magnet is simulated at 300 K (left) and the magnetization is jiggling and not
coming to a stable settled state. Even worse when having a smaller magnet (right) at
150 K, the magnet starts to flip as seen in the telegraph noise of Mz over time.
 27

Switching field distribution induced by thermal agitation

Switching field distribution of a 60 nm x 90 nm x 10 nm size particle at T = 300 K

Sweep time = 10 μs
# switching events

# switching events
Sweep time = 100 ns
Bmean = 53 mT
Bmean = 61 mT

DB » 10 mT Switching field [mT] DB » 5 mT

Slow sweep rate → longer reversal time → more

‘chance’ for the thermal fluctuations to help the
switching of the magnetization
27

Temperature has also an influence on switching field distributions. The slower the sweep
rate of the magnetic field in a switching experiment, the smaller the distribution (right
plot). If we sweep the magnetic field fast, the width of the distribution becomes twice
as large (left graph). Temperature helps in increasing the probability of overcoming an
energy barrier.
 28

Application: Integrated memory devices with ferromagnets

→ Magnetic Random Access Memory MRAM

28
 29

Interaction between electricity and magnetism

Magnetoresistive Spin-torque effect
Field-interaction effects Current injection switches
Two-way interaction Magnetization state magnetization state
influences current
Oersted field from wire Current driven domain
switches magnet: wall propagation
GMR and MTJ

[Link]
29

Of course, it would be nice to switch magnetization electrically — or measure it elec-

trically with high precision. For that, three concepts shall be briefly discussed here.
1) A current-carrying wire generates a magnetic field — this is called the Oersted field
(Oersted is a non-SI unit sometimes used for magnetic field). 2) If we form stacks of
ferromagnets separated by heavy metal or dielectrics, we get magnetoresistive effects.
3) By applying current that is spin-polarized to domain walls, they will move forward
(backward) depending on the sign of the current — spin-torque effect.
 30

Field interaction

Oersted field from wire

switches magnet:

PhD Thesis, Kiermaier 2013 30

An example of using the generated field of a current-carrying wire is taken from our
group. An Au strip is put on a Si-chip and the current generates a close-by field that
can be calculated. High currents, do even heat up the vicinity, which was modeled in
this work. Classical physics.
 31

Basic Structures for Spin Manipulation: Spin Valve and

Magnetic Tunnel Junction

These are the basic spin filter structures, where the orientation of the magnetic layers / external
magnetic field determines the current flowing through them. They can be used as magnetic
field sensors or for many other magneto-electronic purposes.
31

A bit more complicated is this structure: a spin valve. A) A pinned ferromagnet (the
upper magnet is pinned to an anti-ferromagnet and will not switch when applying a
magnetic field), a conductor followed by a ferromagnet that can be easily switched with
magnetic fields. When now applying currents, the resistance of the structure is highly
dependent on the magnetization orientation of the two ferromagnets. B) More com-
plicated is the magnetic tunnel junction, where again a pinned and free ferromagnet
are integrated, with the difference, that a dielectric tunneling barrier is added instead
of the conductor. Tunneling in turn is dependent on the magnetic polarization of the
ferromagnetic layers.
 32

The GMR effect:

[Link]
spin dependent scattering

mmons/9/93/Spin-valve_GMR.svg
Low resistance High resistance

Magnetoresistive effects were well-known phenomena, but the

discovery of giant magnetoresistance (GMR) revitalized the field,
resulted in numerous applications and proposals for new devices.
GMR was used in commercial hard disk drives six years after its
discovery(!). Grünberg and Fert were awarded the 2007 Physics
Nobel prize for it.
32

The giant magnetoresistance (GMR) was discovered in 1988 and later on, the Nobel price
was awarded for those experiments. Repetitive layers of Fe and Cr showed a big change in
magnetization depending on the orientation of the magnetization. The equivalent circuit
is shown on the right. The parallel orientation of the magnetization gives an up to 80%
reduced resistance.
 33

The TMR effect: Two current model for parallel and

[Link]
spin dependent tunneling anti-parallel mangetisation

mmons/b/bc/TunnelSchema_TMR.svg
Low resistance High resistance

P1, P2, spin polarizations of the ferromagnets 1 and 2 33

Tunneling magnetoresistance (TMR) structures use tunneling barriers and it can be

phenomenologically described by a similar model as the GMR. Here it is spin-dependent
tunneling, which makes the difference in resistivity. In principle, the effects are larger
than in GMR structures and the absolute resistance due to the insulating barrier is higher
than in a GMR.
 34

Magnetic Random Access Memory (MRAM)

the smaller the magnet,

the higher the switching field,
the harder to switch the magnet.

In reality such structures are more difficult to build, as

currents have to stay spin polarized across the junction. Nonvolatile MRAMs 34

A beautiful example of nanomagnetic engineering is the magnetic random access memory

(MRAM), where the bit is stored in the resistance of a tunnel barrier. The first imple-
mentations used crossed current wires above and below the MRAM cell to switch the
magnetization. But it turned out, that the smaller the magnet became, the higher the
switching field was. Current-induced switching did not work anymore and writing cur-
rents increased to high values: not acceptable when scaling devices following MOORE’s
law.
 35

STT = Spint-transfer-torque.
Conventional vs STT-RAM Current trough MTJ is able to switch free layer.

Excellent write selectivity

Localized spin-injection within cell

Simpler architecture
No write line, no by-pass line and no cladding

Faster operation
Source: Driskill-Smith, Non-Volatile Memories Multibit (parallel) writing compatible
Workshop, April 2010 35

The reason for a not-scalable technology is given here for the conventional MRAM. The
write current is inversely proportional to the bit volume. The smaller the bit the larger
the current needed. But a new effect did the trick. Running a current directly through
the MTJ, a so-called spin-transfer-torque is acting on the magnetic moment, and the
switching current is proportional to the volume. That means, the smaller the cell, the
smaller the write current. STT MRAM did the trick!
 36

Spin Tranfer Torque - MRAM

Toshiba (2011)
Grandis (2007)
Sony (2005)

36

Here are the hard facts about why STT-MRAM won the race. But recently, it is the
SOT MRAM (spin-orbit-torque MRAM) that is currently overtaking.
 37

More recent progress on STT MRAM

The MTJ is fabricated in
the back-end of line.
Materials are sputter
deposited, annealed and
oxidized where needed.

Typical resistance change

when switching between
AP and P state.

S Yuasa and D D Djayaprawira 2007 J. Phys. D: Appl. Phys. 40 R337 doi:10.1088/0022-3727/40/21/R01 37

Even though the sketches look simple, real MTJ integration is complex. First, the MTJ
cells are fabricated AFTER all the transistors are fabricated. It is processed in the back-
end-of-line (BEOL). Magnetic materials are deposited after all diffusion barriers are put
on the chip. It puts a lot of extra processing costs on the technology. More interesting
here: The sputter-deposited magnetic layer is of crystalline nature and a lot of material
science is needed to optimize the stacks.
 38

Historic milestones in TMR research & development

• Theoretical prediction of over 1000% TMR using MgO barrier by W Butler (2001)

• Experimental demonstration of spin-torque switching by R Buhrman (2002)

• Over 200% TMR using MgO as a tunneling barrier by S Yuasa and S Parkin (2004)

• Commercialization of magnetic random access memory (MRAM) devices by Everspin (2006)

• Over 600% TMR using CoFeB/MgO MTJs (2008)

• Proposed commercialization of Spin Transfer Torque – RAM (by 2014)

• Flash memory, the embedded NVM of choice up until about 40nm process nodes, has
difficulties scaling below 28nm. For this reason the foundries Globalfoundries and Samsung
have both offered embedded MRAM options in their fully-depleted silicon-on-insulator
(FDSOI) processes. Intel embeds MRAM inFinFET process (2018, eeNEWS)

38

Some milestones to read.

 39

TMR with
TMR is also in your HDD MgO barrier

Brendan A Lafferty
IMST, 2008
39

TMR is not only built into recently integrated MRAM, but it was also the needed sensor
for HDD heads from 2004 on. First in a conventional, later in a MgO-TMR version. By
the way, for some months we also deposit MgO material in our sputtering tool. Difficult
material, but fun to see the results.
 40

TMR is also in your HDD

Brendan A Lafferty
IMST, 2008
40

The TMR layer for the HDD stack. Something for magnetic experts.
 41

Brendan A Lafferty
TMR is also in your HDD IMST, 2008

41

And here is the proof, that it is really built into the reading head of the HDD. An SEM
image showing the tiny little tunneling barrier, probably around 1 nm thick, flying with
7200 rpm over the platter and reading the pictures you stored on your HDD. Amazing,
isn’t it?
 42

Brendan A Lafferty
Why we do not do TMR heads @ TUM ☺ IMST, 2008

42

Take it as a joke: Such tools are used to deposit MRAM stacks or TMR heads. Huge,
expensive, and complex machines.
 43

Some alternative approaches for memory and logic

43
 44

Memory concept with current-driven domain wall propagation

This is the race track memory concept, proposed by Stuart Parkin, IBM 44

The famous Racetrack of Stuart Parkin, at that time working at IBM in California.
 45

Vision of a 3D pNML Co-processing unit @ TUM

pNML researched @ TUM in our group 45

And here is our concept of a 3D family of non-volatile computing elements called per-
pendicular magnetic logic. If there is time in an upcoming lecture, I am happy to talk
about it.
 46

L01 Kick-off + ERD roadmap

Nanofabrication Nanometrology/Patterning Applications Magnetic devices

L02 Conversion + PVD
L03 CVD + Others
L04 Nanometrology I - Electrical
L05 Nanometrology II - Optical and SPM
L06 Nanometrology III - Beams of Electrons/Ions
L07 Patterning - Lithography
L08 Metallic Nanowires I
L09 Metallic Nanowires II
L10 Quantum LEDs with Si NP
o Landau-Lifshitz equation
o Switching of magnetic domains L11 Introduction (nano)magnetism
L12 Magnetization dynamics
o Micromagnetic simulations and modeling
L13 Spin-wave devices
o Magnetic switching and electric/magnetic interactions
L14 Nanomagnetic Logic
o Magnetic random access memory (MRAM) and L15 Nanomagnetic Devices 2.0
magnetoelectric effects

46
 47

Questions

Q 1:
The Landau-Lifshitz-Gilbert equation is a very famous differential equation describing
dynamics in micromagnetics. Explain the basic terms in your own words.
Q 2:
Sketch the precession of the magnetic moment M around the effective magnetic field
Heff .
Q 3:
Include the damping term in the precessional motion and compare the range of typically
used damping constants α. What is the damping constant good for?
Q 4:
In magnetism, we often deal with quantum-mechanical forces. Why is it convenient to
describe the Exchange-energy term as an Exchange-field when we are solving the LL
equation?
Q 5:
What are head-to-head and tail-to-tail domain walls? Describe an easy method/geometry
to generate those domain walls.
Q 6:
There is a family of domain-wall devices to perform logic operations with moving do-
main walls. Check the web for the original paper from Allwood at al., Submicrometer
Ferromagnetic NOT Gate and Shift Register, published in Science, 2002. What could be
the drawback of such a computation scheme with domain walls?
Q 7:
What is Barkhausen noise? What is the physical origin?
Q 8:
Having gathered experience on ferromagnetic materials and hysteretic behavior, why
should one be skeptical when discussing permeability µr as a constant number?
Q 9:
What is the superparamagnetic limit, and why is it finally limiting data retention when
scaling HDDs?
Q 10:
Coercive fields of single-domain magnetic particles are connected to the time scale of the
applied switching field (length of the applied field-pulse) by an Arrhenius law. Explain
in your own words the physics behind it.
Q 11:
Compare magnetic spin-valve and magnetic tunnel junction.
 48

Q 12:
Why was it impossible to scale the MRAM technology, where crossed current wires were
used to switch the magnetic bit? What effect did the trick have on novel technologies?
Explain it.
Q 13:
Explain why, in MRAM technologies, the magnetic layers are deposited in the back-
end-of-line (after all the Cu-metallization is finished) and not directly after transistor
fabrication.
Q 14:
Check the web for the latest research on racetrack memory by Stuart Parkin.