development of ferromagnetic/insulator/ferromagnetic
TRANSCRIPT
Developmentof
Ferromagnetic/Insulator/Ferromagnetic DevicesFor
Digital Magnetic Data Storage and Magnetic Field Sensors
ThesisSubmitted to the Faculty of Physics and Astronomy
Ruhr-Universität Bochum
For the degree ofDoktor der Naturwissenschaften (Dr. rer.nat.)
By
Emad Azmy Sultan Girgis
Institute For Thin Film and Ion TechnologyForschungszentrum Juelich
Germany
2000
1
I. Introduction ....................................................................................................................................................4
II. Literature ........................................................................................................................................................6
II.1. Tunnel magnetoresistance (TMR) ....................................................................................................6
II.1.1. Electron tunneling ....................................................................................................................6
II.1.2. Spin dependent tunneling .......................................................................................................8
II.1.3. Preparation of tunnel magnetoresistance devices (TMR) .............................................10
II.1.3.1. Preparation of the electrodes ..................................................................................10
II.1.3.2. Preparation of the barrier .........................................................................................11
II.1.4. Applications of tunnel magetoresistance devices ............................................................13
II.1.4.1. Magnetic data storage ...............................................................................................13
II.1.4.2. Magnetic field sensors ..............................................................................................16
III. Experimental ................................................................................................................................................18
III. 1. Preparation of the samples ........................................................................................................18
III.1.1. Natural oxidation method ................................................................................................18
III.1.2. Thermal oxidation method ..............................................................................................19
III.1.3. Ultraviolet oxidation assisted in oxygen method .....................................................19
III.2. Processing the samples .................................................................................................................21
III.2.1. Optical lithography ............................................................................................................21
III.2.2. Electron beam lithography ..............................................................................................22
III.3. Materials characterization methods .......................................................................................23
III.3.1. Alternative gradient magnetometer (AGM) ...............................................................23
III.3.2. SQUID magnetometer ......................................................................................................25
III.3.3. Atomic force microscopy (AFM) ..................................................................................26
III.3.4. Magnetic force microscopy (MFM) .............................................................................26
III.3.5. Magnetic Optic Kerr Effect (MOKE) ..........................................................................27
III.3.6. X-ray Photoemission Spectrum (XPS) .......................................................................28
III.4. TMR devices characterization methods ................................................................................28
III.4.1. Characterization of the electrical properties of TMR devices .............................28
III.4.2. Characterization of the magnetic properties of TMR devices...............................29
III.4.3. Realibility and breakdown of TMR devices ..............................................................30
III.4.4. Micromagnetic simulation of TMR devices based
on solution of Landau-Lifshitz-Gilbert equation .........................................................30
2
III.4.5. Characterization of the TMR elements using MRAM architecture ....................32
IV. Results and Discussions .............................................................................................................................34
IV.1. Characterization of the substrate ......................................................................................................34
IV.1.1. Surface roughness for Si/SiO2 wafer using AFM .........................................................34
IV.2. Characterization of the magnetic materials ....................................................................................35
IV.2.1 Characterization of the Cobalt thin film ..............................................................................36
IV.2.1.1. Magnetic switching characteristics of thin film Cobalt
in nanometer-scale patterned arrays of elements...........................................36
IV.2.1.2. Magnetic switching characteristics of thin film Cobalt
at different temperature .........................................................................................39
IV.2.1.3. Images of Cobalt thin film patterned element arrays
by AFM and MFM microscopes.........................................................................42
IV.2.1.4. Magnetic switching simulation of Cobalt thin film
at different aspect ratios ......................................................................................45
IV.2.2. Characterization of Nickel-Iron thin film .........................................................................47
IV.2.2.1. Magnetic switching characteristics of Nickel-Iron thin
film in nanometer-scale patterned arrays of elements..................................47
IV.2.2.2. Images of Nickel-Iron thin film patterned element arrays
by AFM and MFM microscopes .......................................................................50
IV.2.2.3. Magnetic switching simulation of Nickel-Iron thin film
at different geometry .............................................................................................51
IV.3. Characterization of the barrier .........................................................................................52
IV. 3.1. Barrier thickness variations......................................................................................52
IV.3.2. Structural characterization of the barrier using (XPS) .....................................54
IV.4. Characterization of TMR devices .....................................................................................55
IV.4.1. Magnetic switching characteristics of NiFe/Al2O3/Co trilayers
in nanometer-scale patterned arrays of elements................................................55
IV.4.2. Images of NiFe/Al2O3/Co trilayers patterned element
arrays by AFM and MFM microscopes ................................................................59
IV.4.3. Magnetic Optic Kerr Effect (MOKE) for NiFe/Al2O3/Co trilayers ..............60
IV.4.4. Magnetic switching simulation of NiFe/Al2O3/Co trilayers
at different aspect ratios .............................................................................................61
3
IV.4.5. Electrical and magnetic properties of the TMR devices
at different oxidation methods ..................................................................................63
IV.4.6. Electrical and magnetic properties of the TMR devices
at different temperatures ............................................................................................65
IV.4.7. Further Parameters which effect on the tunnel magnetoresistance
devices .............................................................................................................................68
IV.4.8. Realibility and breakdown of TMR devices ........................................................73
IV.4.9. MRAM array of tunnel magnetoresistance elements characterization.........75
Summary..................................................................................................................................................................82
References.............................................................................................................................................................. .85
Zusammenfassung ...............................................................................................................................................90
Acknowledgment...................................................................................................................................................98
4
I. Introduction
In science and technology, as well as in daily life, magnetism plays an important role. One of the
applications of magnetism in the present technology can be found in the ability of magnetic
materials to store information. This includes hard disks, floppy disks, optical storage devices and
magnetic strips on credit cards. Recent advances in this area have been driven in many respects
by the improvements of the magnetic materials.
A special class of the magnetic materials are the so-called magnetic multilayers or
superlattices, consisting of alternate layers of magnetic and non-magnetic metallic layers with
typical thickness of the order of nm (10-9m). In 1988 it has been discovered that a multilayer
consisting of alternating magnetic and non-magnetic layers shows a large magnetoresistance
effect when a magnetic field was applied upon the system, denoted as “giant magnetoresistance
or GMR “ [1]. These effect results from spin dependent scattering of electrons within the
magnetic layers or at their interfaces. This effect arises as the result of the mutual alignment of
the magnetization directions of the magnetic layers which varies as a function of applied
magnetic field: antiparallel in the absence of a field but parallel in a sufficiently large external
field. Since the scattering of the electrons is spin dependent, the two spin currents lead to
different resistances for both situations resulting in the observed magnetoresistance. The
discovery of this effect led to an enormous increase of effort in scientific research on layered
magnetic films. The GMR effect was found in many other combinations of
magnetic/nonmagnetic (e.g. Co/Cu/Nife) layers. The effect exists irrespective of the direction of
the current with respect to the layers, for which two configurations are identified: current in the
layer plane (CIP) and current perpendicular to the layer plane (CPP). The magnetoresistance
effect is larger in the CPP than in the CIP while the resistance is extremely small.
More recently, the development of magnetoresistive devices is focused on tunnel devices,
which consist of two ferromagnetic metal layers, separated by a very thin insulation layer (type
CPP). The electron transport is perpendicular to the plane of layers and is determined by
tunneling of electrons through the insulator barrier [2]. The magnetoresistivity is based on a spin-
dependent tunneling probability caused by an energetic splitting of the energy bands with spin-up
and spin-down electrons. The magnetoresistivity is directly proportional to the polarization
values of the electrons at the two-insulator/metal interfaces, which might be different from the
bulk polarization of the two magnetic films. Nowadays, the tunnel magnetoresistance is
becoming more important when patterned into micron or submicron structure which might serve
5
the high technology industry in two major ways, as a means of storing information and as sensors
to read out this stored information [3]. In the present time different types of storage are used. The
difference between the types of the storage based on the relation between the capacity vs. average
access time. One of the different types of memories which is widely used is dynamic random
access memory (DRAM), which is based on Si technology. One of disadvantage of this memory
is the backup battery, which is important to refresh the memory. It means that when users turn on
their computers, they must wait for transfer the information form the hard disc to the memory
during the “boot up “ process (and vice versa during shutdown). Nowadays there is great interest
in the possibility of fabricating DRAM, which retains its memory even after removing power
from the device. Non-volatile memories do not have this problem; therefore these memories have
important applications, e.g. for missiles as satellites. Magnetoresistive Random Access Memory
(MRAM) is one of the non-volatile memories, which is the newest technique based on the
integration of Si CMOS with magnetic memory element. One of the most important
characteristics of MRAM is the fact that it uses the spin of an electron, rather than the charge.
MRAM has unlimited read and write endurance, and could enable truly non-volatile RAM with
both the high speed of today’ s Static RAM and the high density of DRAM. Recent advances in
giant magnetoresistance (GMR) give MRAM the potential for high speed, low operating
voltages, and high density. Nowadays there is a great interest to fabricate the MRAM based on
magnetic tunnel junctions
In this thesis, the results of experiments on the spin dependent transport in structured
multilayered systems are presented. The first part deals with the characterization of the magnetic
materials using different techniques. In the second part results on the fabrication of tunnel
magnetoresistance devices with different oxidation methods are presented. These different
oxidation methods are natural oxidation and ultraviolet radiation assisted oxidation in oxygen,
which is used in the preparation of tunnel magnetoresistance devices for the first time worldwide.
Also the electrical and magnetic properties of the magnetic tunnel junctions compared with
micromagnetic modeling are presented.
The third part is the focused on the applicability of magnetoresistive devices as
magnetoresistive random access memory (MRAM) and magnetic field sensors. However, before
these structures will be actually applied in commercial devices, a good physical understanding of
their properties and their limitations is needed, which is the aim of the research described in this
thesis.
6
II. Literature
II.1. Tunnel magnetoresistance (TMR)
The Magnetoresistance effect was discovered in 1857, which is defined as the changes in the
electrical resistivity of a conductor carrying a current as a function of external magnetic field.
The magnetoresistance is expressed as ∆ρ/ρo, where ∆ρ=ρ (B) - ρo is the change in resistivity
when an external magnetic field is applied, where ρo is the resistivtity at zero external magnetic
field [4]. While the research work was carried out to understand the physical importance of this
effect, different types of the magnetoresistance effect were discovered like anisotropy
magnetoresistance (AMR) [5], Giant magnetoresistance (GMR) [6] and tunnel magnetoresistance
(TMR) [7].
The tunnel magnetoresistance, which is the newest type of the magnetoresistance effect, has
attracted more interest than AMR and GMR because of its high magnetoresistance ratio at room
temperature. The multilayer device of the tunnel magnetoresistanec structure consists of two
ferromagnetic electrodes separated by a very thin nonmagnetic insulator layer. The tunnel current
through the insulator layer depends on the magnetization direction of the two ferromagnetic
electrodes relative to each other in the presence of an external magnetic field. The tunnel current
for a given voltage is higher if the magnetization directions of both electrodes are aligned in
parallel while the tunnel current for the same given voltage is lower if the magnetization of both
electrodes are aligned in antiparallel. The change of the tunnel current when the magnetization
directions of the two electrodes are parallel and antiparallel as a function of external magnetic
field is called tunnel magnetoresistance [8]. The essential effect of the TMR is based on the spin
state of the electron, therefore the tunnel magnetoresistance effect is called spin dependent
tunneling. More details about the tunneling process of the electron in metal/insulator/metal and in
ferromagnetic/insulator/ ferromagnetic film structures will be described in the next part.
II.1.1. Electron tunneling
The phenomenon of electron tunneling in metal/insulator/metal (MIM) structures describe the
electron transport from one metal to the other through the barrier. This phenomenon was studied
long time ago [9], in which the insulator (barrier) of the MIM junction was fabricated with a
thermally grown oxide on the bottom electrode metal. In the literature different models describe
the electron tunneling phenomenon; the simplest model for the electronic potential across a
7
tunnel junction can be expressed as the electronic potential of the two regions of the electrode,
which is equal to the bulk materials potential. The two electrodes are separated by a barrier
region in which the potential is flat and higher than the Fermi levels in the electrodes. The barrier
height Φ is defined as the height of the barrier above the Fermi level, and d is the barrier
thickness. Different pairs metal/oxide have different Φ, it is approximately linear with the
metal/vacuum work function for a given oxide [10]. It is noted that Φ is not equal to the work
function of the material in vacuum, it depends on the electronic structure of the insulator (e.g.
size of the band gap) and the detailed electronic structure at the interface. Fig. 1-a shows the
energy diagram for such a metallic junction when no external voltage is applied. In this case the
Fermi level of both electrodes is at the same energy level, so no tunneling occurs.
Figure 1: Schematic representation of the energy diagrams of a metallic tunnel junction. (a) A square barrier withzero applied voltage, (b) the same barrier as in (a) with an applied voltage V. (c) An asymmetric barrier, with twodifferent barrier heights Φ1 and Φ2 (at zero V). The energy range within the electrodes for which electron states areoccupied (E<EF) is shaded.
The sign and the direction of the applied voltage on the MIM play an important role in the
tunneling process. If the right electrode is positively biased with voltage V, an energy difference
of eV appears across the barrier. The probability of the electrons to tunnel from the top of the
conduction band of the grounded metal to the empty levels of the positively biased metal
increased. This situation is depicted in Fig. 1-b. In this case a net electron current will flow from
the left electrode to the right electrode. The number of empty levels that can receive electrons is
proportional to the bias, so the tunnel current flow is also proportional to the bias. When two
different metals, which give rise to different barrier potentials, are used as electrodes, the barrier
will be asymmetric. For asymmetric barriers, the I-V characteristic is different to (a and b). The
minimum of the differential conductance is shifted from zero to a finite voltage, depending on the
side at which the barrier height is the lowest. Fig. 1-c depicts the energy diagram of an
EF,1EF,2
d
electrode1 electrode2
EF,1
EF,2
EF,1 EF,2
Φ0 Φ0Φ1
Φ2
eV
(a) (b) (c)
8
asymmetric barrier with two different barrier heights Φ1, Φ2 for the left and right electrodes,
respectively, at zero applied voltage. The voltage dependent tunnel current can be evaluated by
Simmons’s formula, where the tunnel barrier is treated as a square barrier [11]. The occupation
probability of electron states is defined by the Fermi-Dirac distribution, which is taken into
account for calculating the rate of tunnel electrons at a certain applied voltage and temperature.
II.1.2. Spin dependent tunneling
As the electron has charge, it has spin also. The spin of the electron has two states, which are
called spin up and spin down. The nonmagnetic metal such as copper has equal numbers of
electrons with spin up and down. Therefore it has no net moment and the current carrying
electrons at the top of the Fermi level, is unpolarised [12]. The ferromagnetic metals such as
cobalt has unequal numbers of electrons with spin up and spin down see Fig. 2. The magnetic
moment of cobalt is simply proportional to the difference between the occupations of the two
spin bands. In the spin dependent tunneling of the two ferromagnetic films separated by a very
thin insulator, the electron of one spin state at the Fermi level of the first film can tunnel through
the barrier to the unoccupied states of the same spin state at the Fermi level of the second film.
Figure 2: Scheme showing the density of spin up and spin down states for nonmagnetic and magnetic metal.
If the two ferromagnetic films were magnetized parallel to each other, then the minority (spin
state up or down) of the electrons would go into minority states, and majority electrons would
pass into the majority states which leads to high tunnel current. If the two films were magnetized
in opposite directions, the identity of majority and minority would be reversed, minority electrons
Energy Energy Fermi level EF
Density of States N (E) Density of States N(E)
Copper Cobalt
9
from the first film would seek empty majority states in the second, and the majority electrons
from the first film would seek minority empty states in the second which leads to low tunnel
current, see Fig. 3.
The effective density of states for the tunneling electrons with spin up and down for both
electrodes are (D1 up & D1 down) and (D2 up & D2 down) respectively. The tunnel current for
the parallel (Ip) and antiparallel (Ia) case can be expressed as:
Ip α D1↑ D2 ↑ +D1 ↓ D2 ↓ (1.a)
Ia α D1↑ D2 ↓ + D1 ↓ D2 ↑ (1.b)
The effective D↑ and D↓ are not the real densities of states of the ferromagnetic, because the
tunneling electrons are influenced by the interface between electrode and insulator [7]. The
asymmetry in the effective density of states of the up and down electrons is described by the
polarization P of the ferromagnetic:
P= (D↑ – D↓)/ (D↑ + D↓) (2)
The magnetoresistance ratio, can be written as
∆R/R= 2P1P2/ (1-P1P2) (3)
where P1 and P2 represent the polarization of the tunneling electrons at the first and the second
electrode. The TMR ratio becomes infinite if a material has a polarization of 100 % [13,14].
Figure 3: Scheme of spin-dependent tunneling: the similar density of states for a parallel alignment of themagnetization direction of the ferromagnetic layers (top) leads to a greater conductance as compared to theantiparallel alignment (bottom).
e-EF
M1 M2
EF
M1 M2
10
II.1.3. Preparation of tunnel magnetoresistance devices (TMR)
Evaporation deposition and sputter depositions are two methods, which were used to prepare the
electrodes and the barrier of the TMR devices. In the case of evaporation, a crucible with source
material is heated, leading to evaporation of the source atoms. Due to the low vacuum pressure,
the evaporated atoms move collision-free through the vacuum chamber and condense on the
substrate. In the case of sputter deposition, the source atoms are sputtered from a target due to the
impact of highly energetic Ar ions. The chamber is filled with Ar gas at a low pressure, and
because the substrate is close enough to the source (i.e. 5cm), the atoms will be deposited on the
substrate with few collisions in the sputter gas. In this case the atoms have a much larger kinetic
energy when deposited compared to evaporation. This results in a high mobility of the incoming
atoms on the surface and a chance of the impinging atoms being back scattered or displacing
substrate atoms. The higher mobility results in fewer nucleation sites as compared with
evaporation since the atoms are able to move around to find an island, instead of nucleating
almost at the site of incidence. Generally, sputtered layers are polycrystalline, have a low surface
roughness, and relatively large grains. The preparation process for TMR devices with the
electrodes and tunnel barrier are deposited as follows; the first electrode is sputter deposited on
the top of Si/SiO2 followed by a very thin aluminum film, which is oxidized using different
oxidation methods [15,16]. Finally the top electrode is sputter deposited on the top of the
aluminum oxide. The junction is defined using optical and electron beam lithography with
different size areas. The junction is surrounded by SiOx in order to insulate the top and the
bottom electrodes, followed by depositing a gold film on the top of the junction as contact pads to
measure the tunnel current.
II.1.3.1. Preparation of the electrodes
A flat and smooth bottom electrode will improve growth of a flat and pinhole free barrier,
therefore a special care is taken when growing the bottom electrode. The magnetron sputtering
method is used to deposit the two electrodes in which the effect of the magnetron can be
described as a closed drift path of crossed electric and magnetic fields for electrons in a plasma
discharge. For a simple planar magnetron cathode, the arrangement consists of the planar cathode
(target) backed by permanent magnets that provide a torodial field, with field lines forming a
closed path on the cathode surface as shown in Fig. 4. The difference in the mobilities of the ions
and the electrons causes a positive ion sheath to be developed close to the target cathode, floating
11
at a negative potential relative to the plasma. Because of the field due to the ion sheath at the
cathode, ions are extracted from the plasma and accelerated to strike the target, resulting in the
sputtering of the target material. The produced secondary electrons, upon entering the region of
crossed electric (E) and magnetic (B) fields, are trapped in orbits that permit long travel distances
close to the cathode. In the zones of the efficient electron trapping, the electron density reaches a
critical value, at which the ionization probability due to the trapped electrons is at its maximum.
This means that a higher rate of secondary electron production by high-energy positive ions is not
necessary for effective sputtering. Most magnetron sources operate in the pressure range from 1-
20 mbar and a cathode potential of 300-700V. The sputtering rates are primarily determined by
the ion current density at the target, and the deposition rates are affected by factors such as
applied power, source-substrate distance, target material, pressure, and sputtering gas
Composition [17, 18].
Figure 4: Scheme of a magnetron sputtering apparatus.
II.1.3.2. Preparation of the barrier
After depositing the bottom electrode the aluminum target rotate to the central position above the
sample and a very thin aluminum film is deposited on top of the bottom electrode using the radio
frequency (RF) sputtering method. The usefulness of RF methods for sputtering nonconducting
materials is based upon the fact that a self-bias voltage, negative with respect to the plasma
floating potential, develops on any surface that is capacitively coupled to a glow discharge. When
an alternating voltage is applied to such an electrode, more electron current flows when the
+
-
+
-
V c
V c
V acuumPu m ps
U nifo rmM agn eticF ie ld
Ca rbo nBo ttom Plate
M a gnetic F ie ldL ines
Carbon SteelO uter C y lind er
C y linde rical M a gn etro nSp u tte r ing So u rce
ST op P la teM ag netic
F ield C oils
Carbon teel
Stanless SteelVacuum ChamberWall
12
electrode is positive relative to the floating potential than ion current flows when the electrode is
negative relative to the floating potential. Fig. 5 shows a schematic drawing of a typical RF
planar-diode-sputtering configuration in which a nonconducting target is placed over one
electrode and substrates are placed on the other one [19].
Figure 5: Schematic drawing of a RF sputter apparatus.
In recent experimental research the oxidation of the metals was discussed; as many metals or
perhaps all of them show very similar behavior when exposed to oxygen at a sufficiently low
temperature. Oxidation is initially extremely rapid, but after a few minutes or hours drops to very
low or negligible rates and a stable film is being formed with a thickness of 2-10 nm. This is the
behavior of the aluminum at room temperature while Copper, Iron, Barium and a few other
metals have the same behavior at the temperature of liquid air. The first explanation of this
behavior was given by Mott [20], who explains the formation of the oxide film is due to a contact
potential difference between the metal and the adsorbed oxygen, which enables the metal ions to
move through it.
In this work the aluminum oxide was used as a barrier, which is made from aluminum thin layer
oxidized by different oxidation methods (Natural oxidation, Thermal oxidation, Ultraviolet
radiation assisted in oxygen). In the literature different oxidation methods were used to prepare
the aluminum oxide, which are discussed in Table 1.
13
Barrier First author, year Preparation method RA
(Ωµm2)
MR
(%)
T
(K)
Al2O3
Al2O3
Al2O3
Al2O3
Al2O3
Al2O3
Miyazaki, 95 [21]
Moodera, 95 [22]
Lu, 98 [23]
Sousa, 98 [24]
W.Opets, 99[25]
This work
Oxidation of Al in air for 24 h
Plasma oxidation of 1.5 nm Al
Plasma oxidation of Al
Plasma oxidation of Al, anneal at 500K
Plasma oxidation of 1.5 nm Al
UV oxidation of 1-2 nm Al
6.4x103
6x107
5x105
3x104
6x106
200-10k
18
13
28
36
23
20
300
300
300
300
300
300
Table 1: Different oxidation methods of the aluminum thin film.
In 1995 Miyazaki et al presented results with Al2O3, which is formed by subjecting a thin Al
layer for a certain time (of hours) in air, Al oxidized slowly by natural oxidation. The oxidation
rate decreases logarithmically with time and the process stops when the oxidation rate becomes
low. This method shows MR ratio of 18% at room temperature.
Then Moodera presented an in-situ fabrication with Al2O3 as a barrier in which the Al is oxidized
by an oxygen glow discharge (referred to as plasma oxidation). This method shows MR ratio of
13% at room temperature.
Sousa observed that magnetoresistance values increased by annealing of the complete junction
for 30min. up to 500 K and the obtained MR ratio was 36%.
In 1999 W.Oepts et al presented results with Al2O3 is formed by plasma oxidation for 40 sec. and
the structure was 3.5nm Ta/3.0nm NiFe/20nm FeMn/2.5nm NiFe/1.5nm Co/10nm NiFe/3.5nm
Ta, the obtained MR was 23% at room temperature.
II.1.4. Applications of tunnel magnetoresistance devices
II.1.4.1. Magnetic data storage
Storing the information using Magnetic Random Access Memories (MRAM) based on the TMR
becoming more attractive in the last two years [26,27,28]. The binary information “1” or “0” is
stored in the magnetic tunnel junction by aligning the magnetization of the two magnetic
electrodes either parallel or antiparallel.
The structure of the tunnel magnetic junction consists of two magnetic electrodes with different
coercive magnetic fields, which are separated by a thin insulating barrier (e.g. A12O3) as shown
in Fig. 6. By applying magnetic field on the TMR element, the magnetic layer which has low
coercivity will rotate and the magnetization in the two electrode are antiparallel which leads to
14
large resistance and then the binary information “0” can be stored, after that the magnetization of
the second electrode will rotate and the magnetization of the two electrodes will be parallel,
which leads to small resistance and in this case the binary information “1” can be stored.
Figure 6: Two tunnel magnetic junctions have two states of the magnetization either parallel or antiparallel.
Figure 7: Shows the MRAM arrays based on TMR junctions.
The MRAM array based on tunnel magnetoresistance elements consists of an array of tunnel
magnetic junctions, which are connected, in series with silicon diodes, in this array each junction
is connected with one diode. In this case, since current only passes through a single magnetic
tunnel junction. All junctions are connected at the top with a conductor and a write line and at the
Bit line
Word line
TMR element
M1
M2
M1
M2
“1”“0”
15
bottom with a bit line as shown in Fig. 7. In the writing operation a high current is applied
through the write line and the top conductor which generates a magnetic field high enough to
switch the soft layer (first electrode) of the magnetic tunnel junction in two states either parallel
to the hard layer (second electrode), then the binary information “1” can be stored or antiparallel,
then the binary information “0” can be stored. By turning the diode on or off, the single tunnel
magnetic junction is selected for reading. In the reading operation it needs a small current than
writing, which is applied only at the top conductor, which generates a small magnetic filed
sufficient to rotate the soft layer and the resistance of the element is measured. If the element
resistance is large then the stored binary information will be “1” while if the resistance is small
then the binary stored information will be “0”. With this simple method the stored information
can be read and write [29,30,31].
Properties SRAM DRAM FLASH MRAMRead Time fast Mod. Mod. Mod-fast
Write Time fast Mod. slow Mod.-fast
Non-Volatile no no yes yes
Minimum Cell Size large small small small
Low Voltage yes limited No yes
Table 2: Shows the major parameters for the different types of RAMs.
In Table 2 the differences between nonvolatile memories and other semiconductor memories are
described. Each memory in this table has important advantages. For example, SRAM has very
fast read and write speeds, but it is volatile and requires a minimum of four transistors per cell.
Therefore, the current state-of-the-art SRAM requires a relatively large cell area of approximately
40F2, where F is minimum feature geometry. DRAM cell architecture is simpler and denser than
SRAM, requiring one pass transistor and a storage capacitor per cell. DRAM has moderate
read/write speeds and the cell size is currently about 10F2. The charge in the capacitor leaks
through the pass transistor and needs to be refreshed in millisecond time intervals. Flash is a
high-density (8F2) nonvolatile memory technology in which the charge is stored in a dummy
gate. In read mode, Flash has unlimited endurance, operates at low voltages, and has moderate
16
access times. However, in write mode, Flash has limited endurance of 105-106 cycles, requires
high voltage (5-12 V), and has slow program (ms) and erase (sec) times. Disadvantage of Flash;
the ratio of memory core to peripheral support circuitry, does not compete with DRAM because
of the high voltage circuit requirements.
II.1.4.2. Magnetic field sensors
Sensors are smart electronic devices which can “see”, ”hear”, ”smell”, ”taste”, and “touch”, by
converting non-electrical physical or chemical quantities into electrical signals. Microsensors
have a wide market in telecommunication, computer technology, environmental monitoring,
health care, and agriculture. A magnetic field sensor is a device, which generates electronic
signals due to the magnetic field, whereby the generated electronic signal is correlated to the
magnitude of the magnetic field and its direction [32,33].
The tunnel magnetoresistance field sensors are not available until now. Philips developed the first
magnetic read head based on TMR [34]. Actually magnetic read heads (which are used to read
the stored information from the magnetic recording medium) which based on GMR (yoke-type)
is used since a few years. For typical head design parameters, TMR- based heads are 3-5 times
more flux-efficient than GMR-based heads. In yoke-type read heads, the MR element is not
indirect contact with the medium [35]. For GMR- based yoke-type heads the efficiency of flux
transport through the yoke to the MR element is limited by the requirement of electrical
insulation of the element. However as described in the literature by Philips, the TMR-based yoke-
type heads can be very efficient, because the yoke can be at an extreme small distance from the
free magnetic layer in the TMR element. Fig. 8 shows a yoke-type read head based on a TMR
element, in which the upper and lower flux guides are soft magnetic thin films. The free magnetic
layer of the junction (F1) bridges the gap in the yoke and its easy axis is oriented perpendicular to
the gap. The magnetization of the second magnetic electrode (F2) which is pinned by the
exchange interaction with a metallic antiferromagnet (AF) in the direction perpendicular to the
medium.
17
Figure 8: Yoke-type read head based on a magnetic tunnel junction.
Back upperflux guide(at V=0)
Test WindingBottom flux guide
Contact lead (at V≠0)
AF
F2Insulator
F1
Front upper flux guide
Read gap
18
III. Experimental
III.1. Preparation of the samples
The samples were prepared by different oxidation methods and characterized by electrical and
magnetic measurements. Silicon wafers are chosen as substrate because of its smooth surface,
which is thermally oxidized. Typical oxide thickness varies between 50-150 nm. This oxide layer
is needed for electrical insulation between the junctions on the same wafer. For these oxide layers
the surface roughness increases slightly proportional to the oxide thickness. The bottom magnetic
layer is the most critical one for different reasons:
(a) The surface of this layer needs to be smooth.
(b) For patterning reasons the thickness of this layer needs to be fairly large, at least 10 nm.
(c) The deposited aluminum should exhibit a wetting on this surface. For the deposition of the
three layers Co/Al2O3/NiFe a magnetron sputter deposition system was used for the processing
with three targets of cobalt, aluminum and permalloy the layers were sputter deposited
consecutively without breaking the vacuum, at base pressure of 2 x 10-6 mbar. During the short
sputter time the substrate remains at the temperature of the water-cooled substrate holder. After
the sputter deposition of the first magnetic film a thin Al film is sputtered. The aluminum layer is
oxidized either by an ex-situ natural oxidation method (i.e. outside the chamber) in air at room
temperature or by in-situ oxidation at pure oxygen (i.e. inside the chamber) at room temperature
or by thermal oxidation or by ultraviolet radiation assisted in an oxygen atmosphere. Finally the
second magnetic film is deposited on the top of Al2O3. With optical and electron beam
lithography the cross-section of the junctions are defined and then etched by ion beam etching. A
low Ar ion energy of 250 eV was used in order to keep the sidewall damage as low as possible,
the etching process is stopped when the bottom electrode appears. The laterally structured layer
stack is then covered with a SiOx layer, which insulates the junction, followed by gold on top of
the junction. The electrical measurements were done using the four points method between the
gold contact pads and the bottom electrode. Systematically, the junction areas were varied within
the limits of 1-600 µm.
III.1.1. Natural oxidation method
The bottom electrode (NiFe or Co) was sputter deposited on top of Si/SiO2 wafer followed by
very thin aluminum film which is oxidized in air (ex-situ oxidation method i.e. out of the
19
chamber) or in pure oxygen (in-situ oxidation method) see Fig. 9. The oxidation time varied
between 24 hours to 2 weeks for natural oxidation ex-situ [36,37,38]. The oxidation time for in-
situ varied between 12 to 48 hours in pure oxygen under pressure 130mbar.
Figure 9: Schematic representation of the natural oxidation method.
III.1.2. Thermal oxidation method
After deposition of the bottom electrode the aluminum layer was sputter deposited and oxidized
in the chamber (in situ oxidation method) in the presence of pure oxygen under pressure of 100
mbar. The oxidation temperature varied between 60 C° and 90 C° see Fig. 10. The thermal
oxidation time varied between 6 and 18 hours [39,40].
Figure 10: Schematic representation of the thermal oxidation method.
III.1.3. Ultraviolet oxidation assisted in oxygen method
After deposition of the bottom electrode the aluminum layer was sputter deposited. Followed by
an in-situ oxidation method using an ultraviolet lamp in the chamber with 100 mbar of high
purity oxygen [41,42]. By mounting a light bulb beside one of the targets, the sputter deposition
facility was modified. The low-pressure mercury bulb emits visible and UV light from a distance
of 5 cm to the substrate and the 4 W light bulb is switched on for 4-60min. The illumination is
fairly homogeneous if the substrate is positioned properly with respect to the bulb. The light of an
AlCo
SiO2
O2 O2 O2O2
O2
Hot plate 60 - 90 C°
Si
AlCo
SiO2
O2 O2 O2O2
O2
Si
20
UV lamp can act via two possible mechanisms. The first proposed by Cabrera [43], which is due
to excitation of metal electrons, so they can pass through the oxide by photoemission or an
increased tunnel current. Thus the electric field inside the oxide is enhanced which leads to a
stronger oxidation compared to in-situ oxidation. The second effect is the generation of ozone:
light dissociates the O2 bond and liberates atomic species and leads to ozone (O3) formation,
which is a more reactive environment.
O2 + hv 2O
O 2 + O O3
where hv >5.1 eV, corresponding to a wavelength shorter than 242-nm. Both mechanisms play
the major role in the oxidation method, which is determined by the absorption of UV light by the
oxygen atmosphere. Fig. 11 shows the UV oxidation method, where the UV lamp is surrounded
by polished Al surface which act as a mirror to reflect the UV radiation and concentrate the UV
radiation on the sample surface. After oxidation of the aluminum film the top electrode was
deposited using dc magnetron sputtering.
Figure 11: Schematic representation of the ultraviolet oxidation method.
The observed oxidation enhancement indicates that the rate of oxidation of Al in oxygen at room
temperature is not limited by diffusion of Al atoms through the already formed Al2O3 but by the
kinetics of the solid /gas reaction.
AlCo
SiO2
O2O2 O3 OO
O
UV lamp
O3
(+)(-)Al Cover
Si
21
III.2. Processing the samples
The fabrication of an integrated circuit requires a technique that enables the various thin-film
materials used to build up the device on a semiconductor substrate to be patterned. This technique
is the lithographic process, which developed for semiconductor devices. The lithographic process
involves transferring the circuit pattern-as might, for example, be contained in a photomask-into a
polymer film (termed a resist) and subsequently replicating that pattern in an underlying thin
conductor or dielectric film. Photolithography, which uses ultraviolet radiation 360-410 nm to
transfer the pattern from the mask to a photosensitive resist, is the dominant technology today for
integrated circuit fabrication. Nano-lithography becomes very important technique for different
applications like magnetic read heads and magnetic data storage, because the increase in density
is achieved partly by reducing the grain size in the current granular magnetic media. Thus, the
grain size must be reduced in order to increase the density of the magnetic data storage. From
different types of lithography (Photolithography, X-ray lithography, Electron-Beam lithography,
Ion-Beam lithography, stamping), only two methods are used in this work which are
photolithography and electron-beam lithography.
III.2.1. Optical Lithography
The sample has a dimension of 10x10mm, which is annealed before the lithography process, in
order to remove the water from the surface. The used photoresist AZ5214 by Hoechst can be
processed in two different ways: a positive or negative process. In the positive process the parts
exposed to light are soluble in the developer. After an additional heat treatment and overall
exposure, the parts, which were unexposed during the first exposure, are soluble. The chrome
mask was used to define the junctions, which varied between 1–600 µm. The structures were
etched with an argon ion beam (IBE). The progress of the etching was controlled by a secondary
ion mass spectrometer (SIMS), which has sensitivity for elements with high ionization
probability. Fig. 12 shows the steps necessary for preparation the trilayers and patterning the
junction. The bottom electrode is defined by etching through the complete trilayer. The second
etching process through the top electrode and the barrier is used to define the junctions. Before
removing the photoresist after the second etching, insulating of 100-nm SiOx layer is evaporated
onto the sample. The junction and the contact pads are exposed by lift-off and finally 300nm of
gold are sputter-deposited as a contact pads.
22
Figure 12: Preparation process and patterning of the junction.
III.2.2. Electron beam lithography
The electron beam can be used to write patterns of very high resolution surface which is coated
with a radiation sensitive resist. This high-resolution capability can originally led to the
development of electron beam lithography as a tool for fabricating integrated circuits. Electron
irradiation of the polymeric resist film produces microstructure changes, such as cross-linking,
that enable a pattern corresponding to the original electron-beam exposure pattern to be
developed [44]. As the resist is frequently electron beam-sensitive organic polymers, usually
Co = 15 nmAl = 1.3 nm
SiSiO2
Co = 15 nmAl = 1.3 nm + Oxidation
SiSiO2
Co = 15 nmAl2O3
SiSiO2
NiFe = 20 nm
Co = 15 nmAl2O3
SiSiO2
NiFe = 20 nm
Resist + Etcing
Co = 15 nmAl2O3
SiSiO2
NiFe = 20 nmJunction + Contact pads
Co = 15 nmAl2O3
SiSiO2
NiFe = 20 nmAu 300 nm
SiOx
Current direction
23
PMMA (polymethyal metacrylate) is used. During the positive exposure the electron beam splits
the carbon-carbon link at one or more places along the chain molecule and thus shortens the
molecule chains of the polymer. The developer dissolves the areas with low degree of
polymerization and does not dissolve the unexposure areas. Before using electron beam the
photoresist was hardened at 120C° for two min. The structure was defined by electron beam
exposure followed by a developing process. The dimensions of the sample were 3.5x3.5mm,
which were patterned into array of elements. Each sample has 108 elements and the element
dimension varies between 100 x 150 nm – 600 x 1800 nm and the distance between the elements
varies between 500 nm – 5µ depending on the structure.
III.3. Materials characterization methods
Before preparing the tunnel magnetoresistance devices, the materials used in the two electrodes
and the barrier were investigated using different techniques. The switching characteristic of the
used materials for the two electrodes is one of the important parameter in designing the array of
MRAM and magnetic field sensors. This parameter was investigated using different techniques
like Alternative Gradient Magnetometer (AGM) at room temperature and Superconducting
Quantum Interference Devices (SQUID) at different temperatures. The samples for AGM and
SQUID were prepared by electron beam lithography. The structures were investigated by Atomic
Force Microscopy (AFM) and the magnetization behavior by Magnetic Force Microscopy
(MFM). One of the problems in the switching characteristic is the magnetic coupling between the
electrodes. Magnetic Optic Kerr Effect (MOKE) investigated this coupling after sputtering and
before patterning, the dimensions of the samples were 10x10 mm. One of the parameter, which
plays an important role in the quality of the barrier, is pinhole, which can be detected by X-ray
Photoemission Spectroscopy (XPS). Detail information about the different techniques (AGM,
SQUID, AFM, MFM, MOKE and XPS) which were used for characterization of the used
materials will be described in the next paragraphs.
III.3.1. Alternative gradient magnetometer (AGM)
Different types of magnetometers are used to measure the magnetic moment. One of these
methods is the alternating-gradient magnetometer (AGM), which has a sensitivity exceeding (10-8
emu). The magnetic sample is mounted on the end of a cantilevered rod that incorporates a
piezoelectric element. The sample is magnetized by a DC field and simultaneously subjected to a
24
small alternating field gradient. The alternating field gradient exerts an alternating force on the
sample, proportional to the magnitude of the field gradient and the magnetic moment of the
sample. The sample is mounted on the tip of a vertical extension rod, which is oriented along the
z-axis, and the gradient field is along the x or the y-axis see Fig. 13-a, b. The top end of the
sample rod is attached to the bottom end of the piezoelectric element as shown in Fig. 13-c. The
force of the field gradient on the magnetized sample produces a bending moment on the
piezoelectric element, which generates a voltage proportional to the force on the sample. The
output from the piezoelectric element is synchronously detected at the frequency of the gradient
field. The amplitude of this voltage is proportional to the magnetic moment of the sample, which
can be varied by changing the applied DC field Hx. The sensing element is composed of two
polarized sheets of a metallized piezoelectric, which are cemented back to back to both sides of a
thin brass vane [45,46]. The preamplifier is the front ends of a lock-in the amplifier, which
automatically tracks the frequency of the AC current in the gradient coils as shown in Fig. 13-d.
Figure 13: Configuration of the magnetization and gradient fields (a) and (b); the bimorph, extension, and sample(c); and the overall system (d).
A , B
Lock-IN
INREF
A
BSIG
IN
OSC
GradientCoils
OUT
(d)
Z
X
Y
-hxhx
Fx = mx dhx / dy
(a)
Clamp
Bimorph
Extension
SampleAc Field Gradient
Electrodes
Z(c)
-hx
hx
Fx = mx dhx / dy
(b)
25
III.3.2. SQUID magnetometer
Superconducting magnetometers are based on superconducting quantum interference devices
(SQUIDs). There are two types of SQUID, the DC-SQUID and the RF-SQUID. These are the
most sensitive instruments available for the measurement of magnetic fields. The DC SQUID
consists of two Josephson junctions mounted in parallel on a superconducting loop, see Fig. 14.
The device is shown in Fig. 15, which is constructed from films of Nb, with Al2O3 barriers for the
junctions. A constant current I0 larger than the maximum zero-voltage current of the two
junctions biases the junctions in the Voltage State. This voltage is periodic in the magnetic flux φ,
applied to the loop with a period of one flux quantum φ0. However, one can detect the change δØ,
which is very small compared with φ0 using the flux-locked loop. The feedback circuit generates
a current in a coil coupled to the SQUID so as to generate a flux, δφ, thereby maintaining the
total flux in the SQUID at a constant value. The output voltage V0 is proportional to δφ typically,
the dynamic range is as high 107 Hz1/2 in a 1 Hz bandwidth, and the flux resolution approaches
10-6 φ0 Hz-1/2 at frequencies above 1Hz [47,48].
Figure 14: Simplified schematic of flux-locked dc SQUID.
Figure 15: Direct-current SQUID with enclosed magnetic flux φ.
Oscillator
Lock-in detector and integratoramplifier
I0
XY
Z
V0
Vφ
I
I
Superconductor
Tunnel Barrier
26
III.3.3. Atomic force microscopy (AFM)
Atomic force microscopy is used to study the morphology and the microstructure of surfaces at
the atomic level. The atomic force microscopy scanned the surface of the sample using a
sharpened tip, which is attached to a cantilever see Fig. 16. The atomic force microscopy tip
usually consisted of a micron-sized silicon nitride pyramid on the end of a silicon nitride
microcantilever has a few hundred microns long and fabricated by semiconductor lithography
and etching techniques. The electronic force between the tip and the surface atoms cause a
deflection of the cantilever. These deflections of the cantilever can be monitored using laser
beam, which gives the image of the surface topography or morphology at the atomic level
[49,50]. The atomic force microscopy is used to study the atomic structure on crystal surfaces,
thin films of semiconductor, metals, etc…
Figure 16: Shows a schematic diagram for Atomic Force Microscopy.
III.3.4. Magnetic force microscopy (MFM)
The variation of the magnetic force interaction between a magnetic tip and the sample, which is
measured by mounting a magnetic tip at the end of an AFM-style cantilever, yields the magnetic
force image of the magnetic material. The tip for MFM is different from the AFM, which has a
ferromagnetic material at the top of the tip [51]. The tip is scanned several tens or hundreds of
nanometers above the sample, avoiding contact. Magnetic field gradients exert a force on the
magnetic moment at the tip, and monitoring the tip/cantilever response gives the magnetic force
image. To enhance the sensitivity, most MFM instrument oscillates the cantilever near its
PiezoScanner
Phase, Amplitude Adjustor
OpticalDetector
O ptica lSource
C an tileverSam ple
Piezo
Tip
Display
Computer and Feedback
Controller
27
resonant frequency (around 100 kHz) with a piezoelectric element. Both hard and soft magnetic
materials can be imaged by the magnetic force microscopy.
III.3.5. Magnetic Optic Kerr Effect (MOKE)
All magneto-optical properties of magnetic material can be described basically in terms of the
anisotropy induced in the optical parameters of the material by its magnetization.
For convenience of description, it is usual to classify the magneto-optic effects according to
whether the effect is observed in radiation transmitted through the material or in radiation
reflected from it. The former is termed the "Faraday Effect" and the latter the "Kerr Magneto-
Optic Effect". Both of these effects are usually subdivided into three categories as seen in Fig. 17,
according to the orientation of the magnetization relative to the plane of incidence of the radiation
and to the plane of the boundary surface of the sample [52]. For the longitudinal and polar
orientation, if the incident radiation is plane polarized in or perpendicular to the plane of
incidence, the reflected and transmitted radiation is generally polarized elliptically.
Figure 17: Shows the orientation of the magnetization.
Moreover, the rotation of the major axis of the ellipse and its ellipse are both proportional to the
magnetization of the material. In this case of the transverse orientation, the intensities of the
reflected and transmitted radiation are simply modulated by amounts proportional to the
magnetization. In the elementary treatment of the effect, the angel of rotation X of the major axis
of the ellipse is given as (X = K M d) and (X = KR M) for the Faraday and Kerr effects
respectively; M is the magnetization, d the thickness of the film and K & KR are constants. The
Kerr and Faraday rotation and ellipticities for a particular material depend on:
1- Orientation of the magnetization.
2- Thickness of the film.
3- State of polarization of incident light.
4- Angle of incidence.
MMM
28
III.3.6. X-ray Photoemission Spectrum (XPS)
X-ray photoelectron spectroscopy (XPS) technique is used to study the surface of the materials.
The excitation source in XPS is X-ray photons, in which the photons impinge on the surface of a
material and the photoelectrons are emitted from the surface atoms of the material, see Fig. 18.
These photoelectrons originate from discrete electronic energy levels associated with these atoms
that are unique and characteristic of the atoms. The emitted photoelectrons can in turn give
information about the local electronic environment. Since these energy levels can be dependent
on the oxidation state or the electronic state of the atoms. The advantage of the XPS technique is
the determination of the oxidation or valance states of atoms as well as the detection of their
presence. In this work the XPS is used to detect whether the aluminum layer is thick enough to
cover the bottom electrode without pinholes [53,54].
Figure 18: Schematic diagram for XPS radiation.
III.4. TMR device characterization methods
III.4.1. Characterization of the electrical properties of TMR devices
The electrical measurements carried out comprise I–V of the tunnel magnetoresistance junctions.
In both cases the junction is measured using the four points probe geometry as shown in Fig. 19.
With this method, measurement errors due to possible contact resistances are eliminated. All
measurements on tunnel junctions presented in this thesis were carried out at different
temperatures. For I–V measurements of the junctions, a computer controlled current is applied
and the voltage across the barrier is measured. The current apparatus used was a Keithley 224
programmable current source. The voltage is measured using a Keithley 2000 multimeter. The I–
V curve is normally taken by sweeping the current from a negative value to a positive value.
Typically I–V diagram are taken in the voltage range of (-500 to 500 mV). The I-V curves were
measured at different temperatures varied between 4-300K.
29
Figure 19: Shows the TMR devices. Top view shows the four points measurements and the cross section shows thestructure of the TMR devices.
III.4.2. Characterization of the magnetic properties of TMR devices
Magnetoresistance measurements were carried out using I–V characteristic apparatus described
above where the sample was surrounded by magnetic coils. The magnetic field sweep is
computer controlled by a calibration source model DIGISTANT type 6705. A field of up to
240kA/m can be applied. The field is measured with a digital teslameter model DTM 132, which
is recorded by the computer. During the magnetoresistance measurement, the junction is biased
with a constant current and the voltage is measured. The bias dependence of the
magnetoresistance is obtained by either measuring a magnetoresistance curve at various voltages,
or by measuring the resistance (V/I) as a function of the swept magnetic field (from positive to
negative and vice versa) at the situation of parallel and anti-parallel alignment of the
magnetization direction of the two ferromagnetic layers. The latter is only possible if the junction
has two well-defined states. The magnetoresistance effect is depending on the change of the
coercivity of the two layers. If the top electrode has a lower coercivety than the bottom electrode,
then its magnetization will start to rotate before the bottom electrode. So the resistance will
change as a function of the magnetization direction in the two electrodes. The magnetoresistance
ratio was measured at different temperatures varied between 4-300K.
Au
Al-OxidNiFe
Co
JunctionContact Pad Contact Pad
Cross Section AA
I
V
Current Source Bottom Electrode
Junction
Voltage Source
SiOx
Top Electrode
Top View
A A
SiO2
Si
SiOxSiOx
30
III.4.3. Realibility and breakdown of TMR devices
The low resistive tunnel magnetoresistance devices are interesting for many applications like read
heads and memory elements. A strongly reduced RC time constant will lead to a better dynamic
performance. Moreover, for an improved signal-to-noise ratio, a low resistance in combination
with higher working voltages (0.3 to 0.5 V) would be required. The physical limit will be the
dielectric breakdown voltage, which is typically 1 V for ferromagnetic tunnel-junctions. [55]. In
the breakdown measurements two methods were used. The first method, the device is stressed
using fixed voltage (0.3,0.4,0.5 volts, etc.…) and the breakdown time is recorded. Such
procedure require very long times, i.e. days or months, depending on the applied voltage and
junction area. In the second method a high voltage (> 1V) is generally applied. Therefore, the
breakdown is measured by applying different voltage ramp speed, in which the applied voltage
increases monotonically with time, and the breakdown voltage is recorded. The breakdown is
investigated with different ramp speed (dV/dt) as a function of the junction area.
III.4.4. Micromagnetic simulation of TMR devices based on solution of Landau-Lifshitz-Gilbert equation
Understanding how the domain walls move in a magnetic material while being driven by an
applied magnetic field is a problem common to many magnetic devices such as: transformers,
magnetic recording read/write heads and various other magnetic sensors. Micromagnetic
structure, such as that present in surface domain walls, can be extracted with standard methods
for the solution to the Landau-Lifshitz-Gilbert equation. Such methods have been given in the
literature by Brown [56], LaBonte [57], Aharoni [58], Hubert [59] and Schabes [60]. The
equilibrium magnetization configuration results from the minimization of the total system free
energy. The energy of a ferromagnetic system is composed of (a) the magnetocrystalline
anisotropy energy EK which describes the interaction of the magnetic moments with the crystal
field characterized by the constant Kv (erg/cm3); (b) the surface magnetocrystalline anisotropy
energy Eks which corrects the broken symmetry near surfaces in the interaction of the magnetic
moments with the crystal field which, characterized by the constant Ks erg/cm 2; (c) the
magnetostatic self-energy Es which arises from the interaction of the magnetic moments with the
magnetic fields created by discontinuous magnetization distributions both in the bulk and at the
surface; (d) the external magnetostatic field energy Eh which arises from the interaction of the
magnetic moments with any externally applied magnetic field; and (e) the magnetostrictive
31
energy Er which arises when mechanical stress (strain) is applied to a ferromagnetic material
which introduces effective anisotropy into the system characterized by Km erg/cm3. The
ferromagnetic equations are based upon the assumption that the bulk saturation magnetization Ms
(emu/cm3) is constant microscopically throughout the ferromagnet. The parameter Ms represents
saturation magnetization at room temperature. For most practical systems being considered (Fe,
Co or Permalloy), there is little deviation in Ms at room temperature from K=0 value. The value
of the magnetization vector M(r) at each point within the ferromagnet is the saturation
magnetization multiplied by the direction cosines, that is
M(r) = (Mx (r), My (r), Mz (r) ) = Msα (r) (4)
= Ms (α (r), β (r),γ (r) ).
The constraint equation implied by the constant magnetization assumption is α(r) =1.
The individual contributions to the energies in this continuum model are calculated by integrating
the energy expressions over the structure in question. The energy integrals below are integrated
over the appropriate dimension, dV. The external field energy Eh for an applied field of H0 is
simply given by Eh = ∫dV H0 . α . MS (5)
In order to calculate the magnetic microstructure of ferromagnets, the time evolution of a
magnetization configuration inside a ferromagnet, which is described by the Landau Lifshitz
Gilbert equation, should be evaluated. The Landau-Lifshitz-Gilbert equation has been examined
experimentally and theoretically [61,62,63] and found to yield an accurate description of the time
evolution for a magnetic moment of fixed magnitude in a magnetic field. This equation has the
following form dM/dt = (ω/(1+λ2)) . M . Heff +( (ω λ)/(1+ λ2) . M . M . Heff) (6)
Here, the gyromagnetic ratio ω = g ωe/2 is determined from the free electron value of ωe
and the spectroscopic splitting factor g= 2. The gyromagnetic ratio ω, the damping parameter λ
and the magnitude of the effective fields determine the time scales of interest. For time domain
simulations, we use the free electron gyromagnetic value of ω = 1.78 x 107 Oe sec-1. The
damping parameter λ is not well known. We have used values between 0.005 and 2.0 for λ and
found that for the calculation of equilibrium magnetization configurations in domain walls and
uniform ferromagnetic systems it is not important. The effective magnetic field on each magnetic
moment is determined from the total system energy Etot as
Heff = -dEtot /(d(Ms α)) (7)
32
III.4.5. Characterization of the TMR elements using MRAM architecture
The electrical and magnetic properties of MRAM arrays based on tunnel magnetic junctions were
investigated. The processing of the MRAM arrays is divided into three steps. The first step is the
preparation of the GaAs diode, which is important to select the TMR junction.
The second step is preparation of the tunnel magnetoresistance junctions, in which the structure
of the sample was NiFe 20nm/Al2O3 1.3nm/Co 15nm, the Al was oxidized by UV oxidation
method for one hour. The third step is pattering the MRAM arrays, in which the mask consists of
3x3 tunnel junction cells of different sizes (40-60µm2). During the patterning processes, the bit
lines and the word lines must be defined in which the bit lines were the bottom electrode 25-60
µm wide and 20 nm thick. The word line on the top of the arrays was separated from the junction
by a SiOx passivation layer (~ 200nm thick). The thickness of the Au word line is 150nm of
width 14-40 µm.
Figure 20: Array of 3x3 magnetic tunnel junctions.
Fig. 20 shows the MRAM array of 3x3 magnetic tunnel junctions. The arrows represent the
current directions through the storing information process (writing) at junction (a), while at (b)
the junction will receive only the field, which generated by the vertical current-line, but at (c) the
junction didn‘t receive any field at all. Fig. 21 shows a schematic cross section of full integration
an MRAM cell in which the diode and the tunnel junction of each cell are combined [64]. This is
a general configuration of a memory cell including a tunnel junction. The three terminal devices
on the bottom of the drawing represent the diodes, which are connected to the ground terminal
and the other contacts are connected to the bottom electrodes of the tunnel junctions. The
a
b c
Bottom Electrode
Top Electrode
Junction
33
schematic is for a stacked cell concept, where the tunnel junction and the diode are vertically
integrated on the top of each other is presented in Fig. 22.
Figure 21: Schematic cross section of an MRAM cell that combines a diode and a tunnel junction in each cell.
To fabricate the MRAM arrays, which are based on tunnel magnetic junctions a magnetic tunnel
junction is sputtered deposition on the top of the diode, which was a GaAs diode. The GaAs p-n
diode have doping levels of 10-17 cm-3 for both p and n regions and the diode with dimension
320µm2 and the magnetic tunnel junctions dimension varies between 1µm x 3µm and 4-6 µm x
10µm [65].
Figure 22: MRAM cell contains diode and tunnel junction.
34
IV. Results and Discussions
IV.1. Characterization of the substrate
IV.1.1. Surface roughness for Si/SiO2 wafer using AFM
The surface roughness is one of the critical parameters of the preparation the integrated magnetic
and semiconductor devices. In this work the used substrate was a Si wafer, which is thermally
oxidized to SiO2. This oxidized layer (SiO2) is important for the electrical insulation between the
devices on the same wafer. Before preparing the tunnel magnetoresistance devices the surface
roughness of the Si/SiO2 wafer (the substrate) was investigated by AFM. Fig. 23 shows two AFM
images with different thickness of a SiO2, the left image shows smooth surface with 50 nm layer
of SiO2, while the right image shows a rough surface and many small islands on the top of 1 µm
SiO2 layer.
Figure 23: Two AFM images: left SiO2 of 50nm, right SiO2 of 1µm.
Fig. 24 presents the relation between the thickness of the SiO2 and the surface roughness, which
shows that with increasing the thickness of the SiO2 the surface roughness is also increased. Fig.
25 depicts a system of Si/SiO2/Co/Al layers, which shows the effect of the surface roughness of
the substrate at preparation the tunnel magnetoresistance junctions. If the substrate is rough and
contains islands on the top, the Co/Al layers also have highly rough interfaces between the two
materials which generate a magnetic coupling (orange coupling). This leads to a reduction of the
magnetoresistance effect. In addition, these islands can create pinholes in the aluminum layer,
which is the main reason for frequently observed short circuits of the tunnel magnetoresistance
device [66]. According to this AFM investigation 50nm SiO2 layer was used in this work.
35
Figure 24: The surface roughness as a function of SiO2 thickness.
Figure 25: Schematic diagram of aluminum and cobalt layers on the top of Si/SiO2 wafer.
IV.2. Characterization of the magnetic materials
Magnetic films patterned into nanometer-scale can serve as a means of storing information [67].
At present, magnetic random access memories (MRAM), such devices are under development,
which are based on tunnel magnetoresistance (MR) elements. To compete against existing
memories, a new memory technology with high-density cells is required. Performance of these
MR elements will depend critically on the switching characteristics from one state to other.
Understanding the switching behavior of these elements as a function of its size is essential to
push the density limits of the storing technology. As recording densities increase, element sizes
decrease and therefore, the individual grains (ferromagnetic particles) that form each element
have to be smaller. The switching characteristics depend sensitively on the geometry of the
element (length, width and film thickness) and the material properties (exchange constant,
crystalline anisotropy, etc.) [68,69]. In this part, a systematic study of magnetic switching
characteristics for different magnetic films is presented. The magnetic films are patterned into
0 200 400 600 800 1000
0.2
0.4
0.6
0.8
1.0
Sur
face
Rou
ghne
ss (
nm)
SiO2 Thickness (nm)
Al
SiO2Co
Si
36
arrays in nanometer-scale with different lateral shapes; different widths of 100nm, 200nm,
600nm and aspect ratios (length/width) = 1.5, 2, 3, and 4.
IV.2.1. Characterization of the Cobalt thin film
IV.2.1.1. Magnetic switching characteristics of thin film Cobalt in nanometer-scalepatterned arrays of elements
Fig. 26 shows the switching behavior of cobalt film 15nm thick at aspect ratios of 1.5, 2, 3 and 4
of 100 nm width as a function of applied magnetic field. The elements with aspect ratio of 1.5
and 2 show low remanence and distorted hysteresis curves, while the elements with higher aspect
ratio of 3 and 4 do not display such behavior. This distortion in the hysteresis curve is due to the
magnetization behavior, which suffers some disturbance at small aspect ratio; this behavior will
be shown in the section summarizing the MFM results. The magnetization behavior at small
aspect ratio has no preferred orientation to rotate and the magnetic flux closes on itself leaving
low remanence (vortex state). While increasing the aspect ratio provides a well-defined
anisotropy axis, which hinders flux closure through vortex formation. At widths 100 nm and 200
nm, the magnetization vortices are observed at aspect ratios of 1.5 and 2, while at 3 and 4 the
magnetization vortices are not observed. However, elements of width 600nm display a weak
tendency for vortex formation at aspect ratio of 1.5, which completely disappear at higher aspect
ratios [70]. This means that there is a strong relation between the magnetization vortices, the
aspect ratio and the element width. Fig. 27 shows the micromagnetic simulation image of the
magnetization behavior at vortex state, in which the magnetization has no preferred direction to
rotate. Fig. 28 shows the magnetization behavior curves and the MFM images at different aspect
ratio of 1.5, 2 and 3 with width =200 nm. Fig. 28 shows at (a1 and b1) the magnetization behavior
at aspect ratio of 1.5, in which the hysteresis curve is distorted and the MFM image shows that,
most of the elements (the dark area) have no magnetic image (switched) i.e. most elements are
found in a vortex state. The early switching behavior of these low aspect ratio elements is due to
the majority of elements entering a vortex state. Fig. 28 shows at (a2 and b2) an aspect ratio of 2, a
few elements are found in a vortex state, this explains the higher remanence and sharper
hysteresis curve [71]. Fig. 28 shows at (a3 and b3) an aspect ratio of 3 in which no elements in a
vortex state at H applied = 0 Oe are presented. All elements of this size are aligned along their
easy axis. Fig. 29 displays the switching field as a function of the aspect ratio for different widths
of 100 nm, 200 nm, and 600 nm. The relatively flat behavior of the 600nm elements, with respect
37
to aspect ratio, reflects the stability of these elements to vortex formation. This means that the
switching behavior of the elements is strongly dependent on the aspect ratio i.e. elements of
width 100nm up or less than 200nm, have a higher propensity for vortex formation. This explains
the curved behavior in the figure. In the Stoner-Wohlfarth single domain model [72], switching
characteristics of a single domain element are simply determined by their geometric factors and
magnetization (M). For rectangular thin film elements with the switching field in the easy axis
direction, Hc can be approximated to a flat ellipsoid and expressed as:
Hc = t/w f (e), and e = (1 – W2 / L2 )1/2 ... (8)
Where L, W and t are the element length, width and thickness respectively, and L ≥ W>> t; f (e)
contains complete elliptic integrals, which is function of (e) only. For a fixed aspect ratio L/W the
single domain Hc should be inversely proportional to the element width W. Compared with our
results, this single domain model is do not follow that for low aspect ratio elements. This again
suggests that these elements have a vortex state.
Figure 26: Magnetization as a function of the applied magnetic field at different aspect ratios of 100nm width.
-150 -100 -50 0 50 100 150
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W = 3
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W = 4
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W = 1.5
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W = 2
M/M
s
Applied Magnetic Field µ0H (mT)
38
Figure 27: Magnetization behavior at vortex state.
(a1) (b1)
(a2) (b2)
-100 -50 0 50 100
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W= 2
M/M
s
Magnetic Field µ0H (mT)
-100 -50 0 50 100
-1.0
-0.5
0.0
0.5
1.0 L/W= 1.5
Hs
M/M
s
Magnetic Field µ0H (mT)
39
(a3) (b3)
Figure 28: Magnetization as a function of the applied magnetic field for 200 nm width (a1-3), the MFM images forarrays with different aspect ratio for 200 nm width (b1-3).
Figure 29: Switching field as a function of the aspect ratio (length/width) for different widths.
IV.2.1.2. Magnetic switching characteristics of thin film Cobalt at different temperatures
With the macroscopic method, SQUID magnetometer, the magnetization was measured as a
function of the applied magnetic field at various temperatures. The specimens are arrays of
elements of the cobalt film 15nm thick of dimensions 100nm x 150nm and 100nm x 400nm, and
each array contains 108 elements. Fig. 30 shows the magnetization curves, which obtained by
measuring the magnetic moment of the sample as a function of external magnetic field at
-100 -50 0 50 100
-1.0
-0.5
0.0
0.5
1.0
Hs
L/W= 3
M/M
s
Magnetic Field µ0H (mT)
1.5 2.0 2.5 3.0 3.5 4.0
-20
-10
0
10
20
30
40
100 nm 200 nm 600 nm
Sw
itchi
ng F
ield
µ0H
(mT
)
Aspect ratio (%)
40
different temperatures 400K, 300K and 4K. Fig. 30 (left) shows the magnetization curves of the
elements have an aspect ratio of 1.5. Again the distorted magnetization curves are observed, in
which with decreasing the temperature the distortion in the magnetization curve is more less than
at higher temperature, while the switching field value does not change dramatically within this
range of the temperature. Fig. 30 (right) shows the magnetization curves of the elements have an
aspect ratio of 4, in which the switching field shows temperature dependence i.e. with increasing
the temperature the switching field is decreased and vice versa. Fig. 31 shows the switching field
as a function of aspect ratio (Length/Width = 1.5, 2, 3 and 4) for 100nm width with different
temperatures. It is shown that the highest aspect ratio has the highest switching field and with
decreasing the temperature, the switching field of the elements increases for larger aspect ratio
[73]. These results might be due to the magnetic anisotropy, which has temperature dependence
i.e. the magnetic anisotropy decreased with increasing the temperature and increase with
decreasing the temperature. Also the magnetic anisotropy depends on the aspect ratio of the
element i.e. with decreasing the aspect ratio, the magnetization has no preferred ordination to
rotate and the magnetic anisotropy decreased i.e. the magnetic anisotropy is small at aspect ratio
of 1.5 than 4 which leads to a small changes of the switching field at different temperature at
aspect ratio of 1.5 than 4. By contrast at small aspect ratio of 1.5 the magnetic anisotropy was
very small, which leads to low temperature dependence while with increasing the aspect ratio the
magnetic anisotropy increase and show temperature dependence. This explains at low aspect
ratios there are no changes at the switching field as a function of the temperature compared with
large aspect ratio.
41
Figure 30: Shows the magnetization as a function of the applied magnetic field for two arrays with dimension of100nm x 150nm (left) and 100nm x 400nm at different temperatures.
-150 -100 -50 0 50 100 150-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
400 K
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
4 K
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
300 K
M/M
s
Applied Magnetic Field µ0H (mT)
-100 -50 0 50 100-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
300 K
M/M
s
Applied Magnetic Field µ0H (mT)
-100 -50 0 50 100-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
400 K
M/M
s
Applied Magnetic Field µ0H (mT)
-150 -100 -50 0 50 100 150-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
(c)
(b)
(a)
4 K
M/M
s
Applied Magnetic Field µ0H (mT)
42
Figure 31: Shows the switching field as a function of the aspect ratio at different temperatures.
IV.2.1.3. Images for Cobalt thin film patterned element arrays by AFM and MFM microscopes
Fig. 32 shows the MFM (left) and AFM (right) images for cobalt 15nm thin film which were
patterned into element arrays of nanometer-scale. The images show the AFM and MFM for
elements with width = 200 nm at different aspect ratios of 1.5 at (a), of 2 at (b) and of 3 at (c).
Fig. 32 (right) depicts the AFM images, which shows sharp and well-defined shapes of the
elements, which are arranged in regular arrays. The spots on the top are residual resist from
processing during electron beam lithography. Due to the dimensions of the elements, the
magnetization lies predominantly on the film plane. Furthermore, because the elements are twice
as long as they are wide, the “easy” axis (or the preferred axis where the magnetization would
orient) is more or less parallel to the long dimension (horizontal). The MFM images (left) for
cobalt elements show a rich and complex magnetic structure behavior. The elements in Fig. 32a-c
show bright and dark areas which can be regarded to the north and south poles of these nano-
magnets. Elements with a single distinct bright/dark pair are known as “single domains”, while
those that have more complex structures are commonly referred to as “multidomains” [74]. At (a)
the MFM image shows an array of elements with dimension 200nm x 300nm (aspect ratio of 1.5)
in which a few elements have single domain magnetization and the rest of the elements haven’t
any magnetic image (this means the elements are switched) which indicate that most of the
elements are found in a low moment vortex state. At (b) the MFM image shows elements with
1.5 2.0 2.5 3.0 3.5 4.0
-20
0
20
40
60
c 4Kb 300 K
c
b
a
a 400K
Sw
itchi
ng F
ield
µ0H
(m
T)
Aspect ratio (L/W) for Width=100 nm
43
dimension 200nm x 400nm (aspect ratio of 2), which have a few elements are found in a vortex
state, these elements are clearly switched. At (c) the MFM image shows elements with dimension
200nm x 600nm (aspect ratio of 3) which have no elements in a vortex state at H _applied = 0
(Oe). Also the south poles of all elements are on their one side while the north poles of all
elements in the other side, which means that the magnetic moment of all elements are in the same
direction. Fig. 33 shows the difference between the multidomain and single domain, the top Figs.
show elements at (a, b) have multidoamin, which contains two or three domains. Also due to the
nonuniformity of the magnetization vector elongated elements, the magnetization vector tends to
be tangential to the edges of the magnetic element to prevent the creation of free poles (and
increased demagnetization energy) [75]. At bottom Fig. (a) element shows the single domain
which has two poles south and north. While at (b) the image shows an element didn’t present any
magnetic image, which indicate that the element is switched.
(a)
(b)
44
(c)
Figure 32: MFM (left) and AFM (right) images with width 200nm at aspect ratio of 1.5 at (a), 2of at (b) and 3 of at(c).
Figure 33: MFM images for single domain (bottom) and multidomain (top).
a
b
aba
a
b
a
b
45
IV.2.1.4. Magnetic switching simulation of cobalt thin film at different aspect ratios
A micromagnetic model utilizing the Landau-Lifshiz-Gilbert equation has been used to
investigate the switching field variation in patterned film elements. A patterned film element is
modeled by discretizing it into a two dimensional arrays of tetragonal cells.
Each cell has a size of 10nmx10nmxfilm thickness. To study the magnetic switching properties,
an external field is always applied along the length direction of the film element in the film plane.
The amplitude of the field is varied step-wise with 2 Oe field steps. The switching field is
investigated for patterned Co thin film 15nm thickness as a function of aspect ratio of 1.5, 2, 3
and 4 for width 100nm, 200nm and 600nm. Considering at this thickness, the crystalline structure
of Co film usually is ffc. Cubic crystalline anisotropy is therefore assumed with anisotropy
constant of k1=-3.5x105 erg/cm3 (K2 is assumed to be zero). The crystalline orientation for each
discretization cell is assumed to be three-dimensional (3D) random [76]. Fig. 34-a, b and c shows
the switching field as a function of experimental and modeling aspect ratio using LLG formula
for 100nm, 200nm, 600nm width. The switching field of the experimental values is different from
the modeling values. The difference can be explained as, when one applies an external field to a
particulate medium each particle is sensitive not only to this field but also to the field created by
the nearest neighbor particles, in which this creation field is exist only at real particle.
Domain structures of real samples on the other hand, are not uniform, instead they display some
magnetization variation especially near sample edges. Fig. 35 shows the schematic diagram for
the real and ideal magnet [77], in which the arrows represent the direction rotation of the
magnetization which is different at the ideal and the real magnet due to the demagnetization
effect at the real magnet.
46
Figure 34: Shows the switching field experimental and modeling values for Co 15nm thick as a function of the aspectratio of width 100nm at (a), 200nm at (b) and 600nm at (c).
1.5 2.0 2.5 3.0 3.5 4.0
-15
0
15
30
45
60
75(a)
i
j
j Modelingi Expermintal
Sw
icth
ing
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
1.5 2.0 2.5 3.0 3.5 4.0
-5
0
5
10
15
20
25
30
35
(b)
i
j
j Modeling
i Expermintal
Sw
itch
ing
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
1.5 2.0 2.5 3.0 3.5 4.0
9
12
15
18
21
(c)
i
jj Modelingi Expermintal
Sw
itch
ing
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
47
Figure 35: Schematic drawing for the real and the ideal magnet.
IV.2.2. Characterization of Nickel-Iron thin film
IV.2.2.1. Magnetic switching characteristics of Nickel-Iron thin film in nanometer-scalepatterned arrays of elements
The AGM results for NiFe layer thin film patterned into nanometer-scale arrays using electron
beam lithography are presented, the dimension of the sample was 3.5x3.5mm which have 108
elements. Fig. 36-top shows the hysteresis loop for NiFe thin film, the dimension of each element
was 100nm x 150nm of thickness 2nm, 3nm and 4nm. It is shown that with decreasing the
thickness of the film the magnetic moment decreases, which lead to decrease the switching field
of the sample. For a fixed aspect ratio (L/W) the switching field Hc should be inversely
proportional to the element width and directly proportional to the thickness according to the
Stoner-Wohlfarth eq. (8). Fig. 36 bottom shows the hysteresis loop for 600nm width with
different aspect ratio at fixed thickness 3nm, which shows that with decreasing the aspect ratio,
the magnetic moment decreased [78].
Fig. 37 shows the results of the magnetic switching field of NiFe thin film as a function of aspect
ratio of 1.5, 2, 3 and 4 at element widths 100nm, 200nm and 600nm of thickness 2nm, 3nm and
4nm. It is shown that the switching field of the element at 2 nm thickness shows the lowest
switching field value than thickness of 3nm and 4nm. The switching field value is decreased with
increasing the width from 100nm to 600nm [79]. So in contrast the switching field decreased
with increasing the width and decreased with decreasing the thickness of the film (become more
sensitive to the magnetic field). These results are very important in designing the MRAM matrix
and magnetic field sensors.
Real magnet Ideal magnet
48
Figure 36: The magnetic moment as a function of the applied magnetic field at different thickness for dimension 100x 150nm (top), and at different aspect ratios of 600nm width (bottom).
-8 -6 -4 -2 0 2 4 6 8
-2.0x10-5
-1.0x10-5
0.0
1.0x10-5
2.0x10-5 100 nm x150 nm
2 nm 3 nm 4 nm
M (e
mu)
Applied Magnetic Field µ0H (mT)
-6 -4 -2 0 2 4 6
-3.0x10-5
-2.0x10-5
-1.0x10-5
0.0
1.0x10-5
2.0x10-5
3.0x10-5
NiFe 3nm
0.6x1.2 0.6x2.4 0.6x0.9
M (
emu
)
Applied Magnetic Field µ0H (mT)
49
(a)
(b)
(c)
Figure 37: The switching field as a function of aspect ratio for NiFe thin film of widths 100nm, 200nm, and 600nm ofthickness 2nm, 3nm and 4nm.
1.5 2.0 2.5 3.0 3.5 4.00
1
2
3
4
5
6
k
j
ik 4nmj 3nmi 2nm
Sw
itch
ing
Fie
ld µ
0H (
mT
)
Length/Wdith (Width=200 nm)
1.5 2.0 2.5 3.0 3.5 4.0-10123456789
k 4nmj 3nm
k
j
i
i 2nm
Sw
itchi
ng
Fie
ld µ
0H (
mT
)
Length/Width (width = 100nm)
1.5 2.0 2.5 3.0 3.5 4.00
1
2
3
4
k
j
i
k 4nmj 3nmi 2nm
Sw
itchi
ng F
ield
µ0H
(m
T)
Length/Width (Width=600 nm)
50
IV.2.2.2. Images of Nickel-Iron thin film patterned element arrays by AFM and MFMmicroscopes
Fig. 38-a, b shows the MFM image (left) and AFM image (right) for NiFe layer of dimensions at
(a) 600nm x 2400nm with 2nm and at (b) 600nm x 1800nm 4nm thickness. The MFM images
were taken in the absence of magnetic field H = 0, with decreasing the thickness (2nm, 3nm) the
magnetic moment is decreased, in which it is very low signal to be detected by MFM. The
600nmx2400nm image with 2nm thickness shows a very low contrast. This can be observed
within the interior regions is similar to the background areas outside of the patterns. For the
dimension of 600nmx1800nm and thickness of 4nm image shows a strong contrast which can be
observed within the interior regions to the background area outside the patterns compared with
dimension of 600nmx2400nm and thickness of 2nm image, which means that with increasing the
element thickness the magnetic moment increases, then the element has a contrast and a stable
magnetization direction along the length of the element [80]. The MFM results show the same
behavior as AGM results with decreasing the thickness the magnetic moment decrease.
(a)
(b)
Figure 38: Shows the MFM (left) and AFM (right) images for NiFe thin film at different geometry.
51
IV.2.2.3.Magnetic switching simulation of Nickel-Iron thin film at different geometry
The micromagnetic program using LLG equation was used to study the magnetization behavior
and switching field for NiFe thin film at different widths 100nm, 200nm and 600nm for different
aspect ratio of 1.5, 2, 3 and 4 with thickness 2nm, 3nm, 4nm. By comparing the simulation value
of NiFe thin film at different aspect ratio with the experimental value it shows that at 2nm, 3nm
and 4nm thickness the switching experimental and modeling value of the small aspect ratio are
close compared with the large aspect ratio. The modeling switching value has the same behavior
like experimental switching value, which means with increasing the aspect ratio, the switching
field value increase. At 600nm width, the modeling and experimental values are close compared
with 200nm and 100nm widths, which means that with decreasing the dimension the structure is
complex and the difference between the modeling and the experimental is high while at large
dimension the difference is small [81]. This difference between the modeling and the measured
value is due to the demagnetization created by the nearest neighbor particle.
1.5 2.0 2.5 3.0 3.5 4.0
0
2
4
6
8
10
12
14 (a) 100nm Exp. 100nm Mod. 200nm Exp. 200nm Mod. 600nm Exp. 600nm Mod.
Sw
itchi
ng
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
52
Figure 39: The switching field as a function of the aspect ratio for thickness 2nm at (a), 3nm at (b) and 4nm at (c).
IV.3. Characterization of the barrier
IV.3.1. Barrier thickness variations
The barrier thickness determines the resistance of the magnetic tunnel junctions. For optimal
reproducibility on wafer level, the barrier thickness must be controlled almost on an atomic level.
Important barrier properties such as the thickness and the potential height can be obtained fairly
well from the results of electric transport measurements using the theoretical models for spin
1.5 2.0 2.5 3.0 3.5 4.002468
101214161820
(b) 100nm Mod. 100nm Exp. 200nm Exp. 200nm Mod. 600nm Exp. 600nm Mod.
Sw
itchi
ng
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
1.5 2.0 2.5 3.0 3.5 4.0
02468
10121416182022
(c) 100nm Exp. 100nm Mod. 200nm Exp. 200nm Mod. 600nm Exp. 600nm Mod.
Sw
itchi
ng
Fie
ld µ
0H (
mT
)
Aspect ratio (length/width)
53
dependent transport in magnetic tunnel junctions. Fig. 40 shows the difference between an ideal
and the more realistic tunnel barrier. The realistic barrier has thickness variations due to the
growth of the bottom electrode and the initial roughness of the substrate. Thickness variations
will strongly influence the tunnel current distribuation on the junction area, since the tunneling
probability is exponentially dependent on the barrier thickness. Thus it may be possible that
almost all the electron transport in a tunnel junction takes place in a small fraction of the total
junction area [82].
Figure 40: Schematic drawing of an ideal barrier at (a) and a realistic barrier at (b).
In literature several oxidation processes have been described. All processes can yield tunneling
barriers, but having different barrier properties. The density of the oxide in the barrier seems to
be the key issue and therefore, a thin Al layer is deposited first and then oxidized, either (in-situ)
by a glow discharge or in oxygen ambient, or (ex-situ) in air (natural oxidation). An optimum
oxidation state has to be achieved as too thin aluminum film can result in some oxidation of the
ferromagnetic underlayer, which has a strong implication on the spin-polarization of this layer. A
too thick aluminum film will weaken the spin-dependent tunneling due to scattering in the
remaining aluminum film. For practical purposes the most important parameter is the resistance
of the micro-scale tunnel junction itself. To understand the relation between the barrier height and
the barrier thickness, Simmon’s model was used for a junction of 1 µm2 area as shown in Fig. 41.
The resistance (Ω) was calculated as a function of two parameters, the barrier thickness (from 0.5
to 2 nm) and the barrier height 0.5 to 3 eV. This model is widely used to compare the different
barrier types. Another parameter which has an effect the barrier height is the oxidation method in
which the aluminum by one oxidation method is completely oxidized while by other oxidation
method is not complete [83].
(a) (b)
54
Figure 41: Junction resistance (Ω) for 1 µm2 area as a function of the two parameters in the model: barrier heightand thickness.
IV.3.2. Structural characterization of the barrier using (XPS)
Investigating structure, composition and thickness of the barrier is a direct way, which lead to
more information and understanding of the relation between the MR effect and the barrier
properties. The main topic of the barrier structure is the composition of the A12O3 (i.e. if the
oxidation time is not sufficient, this leads to leftover Al and if the oxidation time is too high, this
leads to formation the bottom electrode oxide). One of the methods used, which characterize the
barrier, is X-ray Photoemission Spectrum (XPS). The samples have the same thickness of the
bottom electrode, which was 20 nm cobalt film while the aluminum thickness changed in each
sample, which were 0, 1, 2 and 3 nm thick. Fig. 42 shows four XP-spectra in the Co 2p core level
region of the Co film without aluminum on top at (A), and with an aluminum layer on top of
thickness 1 nm at (B), 2 nm at (C) and 3 nm at (D), respectively. All the samples were oxidized
by natural oxidation for 1 month. The spectra of the single Co film clearly exhibit the cobalt
oxide peak at a binding energy of 782 eV and the metallic Cobalt 2p3/2 peak, perceptible as a
shoulder at 779 eV. The cobalt oxide peak is already drastically reduced in the spectrum at (B).
This result is a strong evidence that only 1 nm aluminum layer is sufficient to suppress nearly or
completely the formation of Co oxide. The spectra with 2 nm aluminum at (C) and 3nm
aluminum at (D) exhibit no Co oxide peak anymore. The above results show that less than 2 nm
0.5
0.8
1.1 1.
4 1.7 2.
0
0.50.60.70.80.91.01.11.21.31.41.51.61.71.81.92.02.12.22.32.42.52.62.7
2.82.93.0
1 .00E-01
1.00E+00
1.00E+01
1.00E+02
1.00E+03
1.00E+04
1.00E+05
1.00E+06
1.00E+07
1.00E+08
1.00E+09
1.00E+10
1.00E+11
1.00E+12
1.00E+13
1.00E+14R
esis
tan
ce
Thickness (nm)
Barrier Height (eV)
55
aluminum is sufficient to cover the Co surface without pinholes and to suppress the Co oxide
formation [84].
Figure 42: Co 2p spectra of an air-oxidized polycrystalline Co film without Al layer at (A), Air-oxidized Co/Aloverlayer structures with 1 nm Al at (B), 2 nm at (C) and 3 nm at (D) on the top of 20 nm Co.
IV.4.Charactarization of TMR devices
IV.4.1. Magnetic switching characteristics of NiFe/Al2O3/Co trilayers in nanometer-scalepatterned arrays of elements
In this part the results of magnetic trilayers patterned into arrays using electron beam lithography
(each array contains 108 elements) are presented. Fig. 43 shows the magnetic switching as a
function of external magnetic field measured by AGM method with different aspect ratio of 1.5,
2, 3 and 4 for trilayers of NiFe 20nm/Al2O3 0.8nm/Co 15nm of width of 200nm. The barrier
thickness of 0.8nm was prepared by ultraviolet radiation assisted in oxygen for 1 hour oxidation
time under oxygen pressure of 100 mbar. From Fig. 43 it is shown that with increasing the aspect
ratio the magnetization curve against the applied magnetic field is changed. Fig. 43 a-d shows the
switching field for both NiFe and Co layers at different aspect ratio of 1.5, 2, 3 and 4
respectively. It is shown clear that the switching field for both NiFe and Co layers is changed
with increasing the aspect ratio. The switching field of a free magnetic single domain stripe is
directly proportional to the magnetization and the thickness of the magnetic film and inversely
810 800 790 780 770
Co 2p 1/2
Cobalt Oxid
Co 2p3/2
Binding Energy [eV]
(A)
(C)
(D)
(B)
100
200
300
400
56
proportional to the width of the stripe. The non distinguishable switching field for both soft layer
(I) and hard layer (II) as a function of the aspect ratio due to the magnetostatic coupling (dipole-
dipole interaction) between the two magnetic layers and the magnetization vortex for the Co
single layer which is strong at small aspect ratio see Fig. 43-a (inside curve). The inside curve
shows the magnetization curve of Co 15nm single layer in which the distortion in the curve (due
the magnetization vortex) leads to distortion in the trilayers curve for the same aspect ratio . The
experimental results on the switching fields of the TMR devices are summarized in Fig. 44 in
which the switching field of the soft layer and the hard layer increase with increasing the aspect
ratio. This behavior for soft and hard layers due to the dipole-dipole interaction between the
layers which provides a well pronounced change of magnetization of the two layers. The
resulting forces per volume are stronger for the smallest elements [85]. This finding can be
considered in designing MRAM matrix based on TMR elements.
(a)
-100 -50 0 50 100
-1.0
-0.5
0.0
0.5
1.0
200nmx300nm
(II)
(I)
M/M
s
Magnetic Field µ0H (mT)
-100 -50 0 50 100
-1.0
-0.5
0.0
0.5
1.0
Co single layer
L/W= 1.5
Hs
M/M
s
Magnetic Field µ0H (mT)
57
(b)
(c)
-150 -100 -50 0 50 100 150
-1.0
-0.5
0.0
0.5
1.0
(I)
(II)
200nm x 400nm
M/M
s
Magnetic Field µ0H (mT)
-60 -40 -20 0 20 40 60
-1.0
-0.5
0.0
0.5
1.0 200nm x 600nm
(II)
(I)
M/M
s
Magnetic Field µ0H (mT)
58
(d)
Figure 43: Shows the AGM results for NiFe/Al2O3/Co trilayers with different aspect ratios of 200 nm width.
Figure 44: Dependence the switching field on the aspect ratio for the soft layer (NiFe) and the hard layer (Co).
-60 -40 -20 0 20 40 60
-1.0
-0.5
0.0
0.5
1.0 200nm x 800nm
(II)
(I)
M/M
s
Magnetic Field µ0H (mT)
1.5 2.0 2.5 3.0 3.5 4.0
-60
-45
-30
-15
0
15
30j
ij Co layeri NiFe layer
Sw
itchi
ng
Fie
ld µ
0H (
mT
)
Length/Width ratio ( width= 200 nm)
59
IV.4.2. Images of NiFe/Al2O3/Co trilayers patterned element arrays by AFM and MFM microscopes
Fig. 45 shows the AFM (right) and MFM (left) images for NiFe 20nm/Al2O3 0.8nm /Co 15nm,
elements of 200nm width with different aspect ratio. Ideally, the MFM measures the net
magnetization of the two magnetic layers. It is shown clearly that all elements are patterned along
the element length. Fig. 45-a shows the MFM image (left) for dimension of 200nmx300nm in
which some of the elements have the south pole on the left side while in other elements on the
right side, which means the magnetization of the elements have random distribution at aspect
ratio of 1.5. Fig. 45 shows at (b and c) the effect of increasing the aspect ratio of 2 and 3 on the
MFM images, in which the south poles start to have one direction in some elements at
200nmx400nm and increased at 200nmx600nm in which most of the elements have south poles
in one direction. This means at H= 0 Oe the magnetic moments of all elements start to be in one
direction [86], which means that the magnetization doesn’t suffer any distortion at large aspect
ratio. The MFM result is compatible with the AGM result, in which the magnetization curves is
dependence of the aspect ratio and in accordance with dipole-dipole interaction. The dipole-
dipole interaction is strong interaction at small aspect ratio and the magnetization vortex for Co
single layer which appear at small aspect ratio and disappear with increasing the aspect ratio.
(a)S N N S
60
(b)
(c)
Figure 45: AFM and MFM images for NiFe 20nm/Al2O3 0.8nm/Co 15nm trilayers at 200nm width and at aspectratio of 1.5 at (a), of 2 at (b) and of 3 at (c).
IV.4.3. Magnetic Optic Kerr Effect (MOKE) for NiFe/A2O3/Co trilayers
The MOKE loops for a NiFe 15nm single film and a Co 10nm single film are shown in Fig. 46-a
and 46-b, respectively. They depict simple square loops, in which the switching fields of the
single films differ from the Co 10nm/A12O3 1.3nm/NiFe 15nm trilayers as shown in Fig. 46-c.
The MOKE loop for trilayers clearly shows two distinct switching fields, in which the small step
in the magnetization curve corresponds to the switching of the NiFe electrode (soft layer), the
large step corresponds to switch the Co electrode (hard layer), as can be readily deduced from
the ratio of the magnetic moments of the respective layers. In the trilayers curve, the switching
field of the NiFe electrode is larger and the switching field of the Co electrode is smaller than the
N SN S
S N N S
61
switching field in case of single film [87]. This indicates that there is weak coupling between the
ferromagnetic electrodes in the trilayers. This weakness of the coupling suggests that the
insulating barrier is almost pinhole free which supports the XPS results on similar samples.
(a) (b)
(c)
Figure 46: MOKE measurement for NiFe single layer at (a), Co single layer at (b), and Co 10nm/A12O3 1.3nm/NiFe15nm trilayers at (c).
IV.4.4. Magnetic switching simulation for NiFe/Al2O3/Co trilayers at different aspect ratios
The magnetic switching of the trilayers (NiFe/Al2O3/Co) was determined by micromagnetic
modeling based on Landau-Lifshiz-Gilbert equation with different aspect ratios. Fig. 47 shows
the magnetic switching value (modeling and experimental) of the trilayers (NiFe 20nm/Al2O3
0.8nm/Co 15nm) as a function of the aspect ratio, in which the modeling value is not equal to the
measured value, this is due to the sensitivity of the element which is not only to the external
magnetic field but also to the field created by the nearest neighbor particles, in which this
-4 -2 0 2 4
Mag
netiz
atio
n (a
.u.)
Applied Magnetic Field µ0H (mT)
-4 -3 -2 -1 0 1 2 3 4
Mag
netiz
atio
n (a
.u.)
Applied Magnetic Field µ0H (mT)
-4 -3 -2 -1 0 1 2 3 4
Co
NiFe
Mag
net
izat
ion
(a.u
.)
Applied Magnetic Field µ0H (mT)
62
creation field is exist only at real particle. From the previous results of NiFe and Co single layer it
is shown clearly that the switching field modeling and experimental values are different which
lead to the difference between the switching field modeling and experimental values for the
trilayers.
Fig. 48 shows the micromagnetic modeling images of the magnetization behavior for soft layer
(NiFe) and hard layer (Co). In each layer the magnetization at the opposite ends of the element
are antiparallel to each other [88]. When the two layers started to switch, the observed
configurations by micromagnetic simulation show that, when the hard layer (Co) start to switch a
deviation from homogeneous magnetization occurs only at the layer boundary while at switching
the soft layer (NiFe) a deviation from homogeneous magnetization occurs towards the center of
the layer.
Figure 47: Switching field value, experimental and modeling, as a function of aspect ratio for NiFe/Al2O3/Cotrilayers.
Figure 48: The magnetization configuration of soft and hard layers using micromagnetic modeling.
1.5 2.0 2.5 3.0 3.5 4.0
-60-40-20
020406080
100120140 Soft layer Exp.
Soft layer Mod. Hard layer Exp. Hard layer Mod.
Sw
itchi
ng F
ield
µ0H
(m
T)
Aspect ratio (length/width)
Soft Layer (NiFe) Hard Layer (Co)
63
IV.4.5. Electrical and magnetic properties of the TMR devices at different oxidationmethods
The first measurement of the TMR devices is always the electrical measurements I-V that gives
information about the quality of the barrier. The device was measured using the four
measurements points. The nonlinearity of the I-V curves indicate that the barrier is pinhole-free
(pinholes would lead to a short circuit between the two electrodes). The left parts of Fig. 49-a, b
and c presents the typical I-V curves at room temperature for three junctions. At (a) the junction
area was 10x10 µm. It was oxidized by natural oxidation (ex-situ) at room temperature for
oxidation times of 96 hours. The junction structure was NiFe 20nm/Al2O3 1.3nm/Co 15nm. At
(b) the junction was oxidized by natural oxidation (in-situ) inside the chamber filled by pure
oxygen under pressure of 130 mbar for 12 hours. The junction structure was FeMn 12nm/CoFe
6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm and the junction area was 2x4 µm [64]. At (c) the junction
was oxidized by UV oxidation method for 1 hour under pressure of 100 mbar. The junction
structure was NiFe20nm/ Al2O3 1.3nm/Co15nm and the junction area was 10x4 µm. In all three
cases the characteristic shows a non-linear behavior, which is similar to the theoretical I-V curves
calculated by Simmon (1963) [11]. The non-linearity is caused by the decrease in the effective
barrier height with increasing applied voltage. By fitting such measured I-V curves with
Simmon’s expression the correlated values for the barrier height and thickness could be obtained.
The I-V curves have less pronounced non-linearity. This result is consistent with the calculation
of Simmons model. In the right part of Fig. 49-a, b and c the measurements of the resistance as
function of the external applied magnetic field are shown. Clearly the states with parallel and
anti-parallel magnetization vectors of the hard magnetic Co and the soft magnetic NiFe are
separated. The measured resistance values at parallel state (Rp) and antiparallel state (Ra) lead to
MR ratio of 6.4 % at natural oxidation (ex-situ) at room temperature at (a), an MR ratio of 18.7%
with natural oxidation (in-situ) at room temperature at (b) and an MR ratio of 20% at room
temperature with ultraviolet oxidation at (c). At Fig. 49-a and c the two ferromagnetic electrodes
with different coercivity of NiFe and Co which used in a tunnel junction as top and bottom
electrodes. At a high positive field, the magnetization of both electrodes NiFe and Co are parallel,
when the field is lowered and raised in the negative direction, the magnetization of the electrode
with the smallest coercivity of NiFe will rotate first. The net moment of both layers is lowered, or
become zero if the saturation magnetization of both layers is identical. The NiFe electrode will
rotate till the magnetization in the two electrodes are antiparallel, which leads to maximum MR
64
ratio. Then, with increasing the negative field, the magnetization of the electrode with highest
corecivity Co will switch, and again a parallel alignment which leads to minimum MR ratio. In
the second case at Fig. 49-b the switching from parallel to antiparallel by exchange coupling of
one of the layers to an antiferromagnet was obtained, which is called exchange-biasing [90]. The
magnetization of the electrode will only rotate when the exchange-anisotropy field is reached
while the other free electrode has a low coercive field. At low positive field the free layer will
switch to antiparallel alignment. This big changes between the parallel and antiparallel of the two
electrodes leads to a large value change in the magnetoresistance. Further experimental findings
that the failure rate of devices on a substrate is fairly large, 30% of the samples showed tunnel
behavior of MR effect at natural oxidation ex-situ while at in-situ more junctions show MR
effect. In addition, the samples degraded with time where after weeks short circuits were
observed. While, no changes of the resistance or the MR ratio were observed even after two years
the junction prepared with UV oxidation method. With comparing the oxidation methods for the
same electrode thickness of NiFe and Co and for the same Al thickness 1.3nm the MR ratio was
6.5% with natural oxidation ex-situ, 8% with natural oxidation in-situ and 20% with UV
oxidation method at room temperature.
-10 -8 -6 -4 -2 0 2 4 6 8 10-1
0
1
2
3
4
5
6
7Ra
Rp
(a)MR= 6.4%T= 300KEx-situ oxid.
MR
rat
io (%
)
Applied Magnetic Field µ0H (mT)-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3
-0.9
-0.6
-0.3
0.0
0.3
0.6
0.9(a)
dI/d
V (
mA
/V)
Tun
nel c
urr
ent (
mA
)
Applied Voltage (V)
0
1
2
3
4
5
6
7
-0.4 -0.2 0.0 0.2 0.4-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3(b)
dI/d
V (
mA
/V)
Tun
nel C
urre
nt (
mA
)
Applied Voltage (V)
0.57
0.58
0.59
0.60
0.61
0.62
-20 -10 0 10 20
0
3
6
9
12
15
18
21 Ra
Rp
MR= 18.7%T= 300KIn-Situ
(b)
MR
rat
io (%
)
Applied Magnetic Field µ0H (mT)
65
Figure 49: I-V curves (left) and the MR ratio as a function of applied magnetic field (right) for different oxidationmethods. At (a) natural oxidation method (ex-situ), the junction has area of 10x10 µm and structure of NiFe20nm/Al2O3 1.3nm/Co 15nm. At (b) natural oxidation method (in-situ), the junction has area of 2x4 µm and structureof FeMn 12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm. At (c) UV oxidation method, the junction has area of10x4 µm and structure of NiFe20nm/AlOx1.3nm/Co15nm.
IV.4.6. Electrical and magnetic properties of the TMR devices at different temperatures
Fig. 50-a shows the temperature effect on the I-V curves within temperature range 290K-4K. At
(a) the junction has a size of 180µm2 and a structure of NiFe 20nm/Al2O3 1.3nm/Co 15nm, the
sample was prepared by ultraviolet oxidation for 1hour. The I-V curves were measured by
applying sweeped voltage ±300 mV on the junction, which is sufficient enough to observe the
nonlinearity of the tunnel curve. With decreasing the temperature, the tunnel current decreased.
This result is in agreement with Strattons theory [91], who explains the relation between the
conductance and the temperature. The tunneling current at constant voltage can be expressed as
I(T)= I (0) (1+1/6π2C2K2B T2) (9)
where I (T) and I(0) are the tunneling currents at temperature T and 0 K, respectively. KB is
Boltzmann‘s constant and C is a constant which depends on the barrier height and thickness.
-20 -10 0 10 20-5
0
5
10
15
20 Ra
Rp
MR= 20%T= 300KUV oxid.
(c)
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
-0.4 -0.2 0.0 0.2 0.4-3
-2
-1
0
1
2
3(c)
dI/d
V (
mA
/V)
Tun
nel C
urre
nt (
mA
)
Applied Voltage (V)
4
5
6
7
8
66
Figure 50: I-V curves at different temperatures at (a) and I (T)/I (0) vs. T2 at (b).
Fig. 50-b shows that the I (T)/I (0) values of the junction is roughly proportional with T2 in the
whole temperature range measured which is the same behavior as explained before by Strattons
theory. Fig. 51-a shows the MR ratio curves as a function of the external magnetic field at
different temperatures for a junction with a size of 180µm2 . The structure was NiFe 20nm/Al2O3
1.3nm/Co 15nm. The sample was prepared by ultraviolet oxidation for 1hour. The MR ratio was
12% at room temperature, which increased to 19.9 % with decreasing the temperature down to
77K. When the temperature was decreased to 4K, no changes in the MR ratio were observed. Fig.
51-b shows the magnetoresistance ratio as a function of the external magnetic field at different
temperatures. The junction has 24µm2 area and a structure of FeMn 12nm/CoFe 6nm/Al2O3
1.3nm/CoFe 2nm/NiFe 8nm, which was prepared by natural oxidation (in-situ) in oxygen under
pressure 130mbar for 12 hours. The junction shows the temperature effect on the MR ratio, which
is the same behavior as explained before. Fig. 52 summarized the MR ratio as a function of
different temperatures for samples prepared by (a) UV oxidation method and (b) natural
oxidation method (in-situ) for the above two samples. There are two reasons for the dependence
of the MR ratio on temperature; the first one was explained by Zhang et al. [92] who put forward
a theory which explains the dependence of the conductivity for spin up and spin down and
therefore the magnetoresistance effect on temperature with magnon generation and absorption in
the electrodes. The other explanation is due to the traps in the barrier, which offer sites located in
the barrier with energies below the barrier height. Electrons which tunnel in two steps through the
barrier via the trap have their spin information randomized. With increasing temperature, traps
become more important in conduction; their activation energy is lower than for thermionic
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6-6
-4
-2
0
2
4
6
4k25k110K270K
(a)
290 k
Tun
nel C
urre
nt (
mA
)
Applied Voltage (V)0 20000 40000 60000 80000
0.0
0.2
0.4
0.6
0.8
1.0 (b)
I(T
) / I
(0)
T2 (K2)
67
emission. At low temperature the traps will be filled with electrons and the electron will tunnel
from the first electrode to the second electrode through the barrier as direct tunneling in one step
and not through the traps which leads to high MR ratio.
.
Figure 51:Magnetoresistance curves at different temperature for junction oxidized by UV oxidation method and hasa structure of NiFe 20nm/Al2O3/Co 15nm at (a), and natural oxidation method has a structure of FeMn 12nm/CoFe6nm/Al2O3/ CoFe 2nm/NiFe 8nm at (b).
-20 -15 -10 -5 0 5 10 15 20
0
3
6
9
12
15
18
99K
167K225K
300K
MR
rat
io (%
)
Applied Magnetic Field µ0H (mT)
-20 -10 0 10 20
0
5
10
15
20
25
30
(b)
300K 200K 171K 91K
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
68
Figure 52: shows the MR ratio as a function of the temperature for a junction oxidized by UV oxidation method hasa structure of NiFe 20nm/Al2O3/Co 15nm at (a) and natural oxidation method (in-situ) has a structure of FeMn12nm/CoFe 6nm/Al2O3/ CoFe 2nm/NiFe 8nm at (b).
IV.4.7. Further Parameters which effect on the tunnel magnetoresistance devices
There are different parameter, which effect on the tunnel magnetoresistance devices like bias,
temperature, magnetization direction and current direction, were investigated in this work.
(a)- For bias dependence, the tunnel magnetoresistance effect was measured as a function of the
applied voltage. With studying several junctions, it was found that the MR ratio dropped by a
factor two at applied voltage in the order of 200-500mV. The bias dependence of tunnel
magnetoresistance is not yet understood completely. Fig. 53 shows at (a1 and b1) the MR ratios
as a function of the applied voltage. At a1 the junction has a structure of NiFe 20nm/Al2O3
1.3nm/ Co 15nm and area of 10x3 µm which was prepared by UV oxidation for 1 hour, the MR
shows 10.1 % at 10mV and 4.81% at 600mV. At b1 the junction has a structure of FeMn
12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm and area of 2x3µm which prepared by natural
oxidation method (in-situ) for 12 hours, the MR shows 16.7% at 10mv while at 400mv shows
7.2%. Fig. 53 shows at (a2 and b2) similar results of bias dependence for the same junctions a1
and b1 at 91K. The bias dependence effect has been explained in the literature by different
phenomenon: like excitations of magnons [93,94] or phonons [93]. Inelastic tunneling process via
defect [93,95], electron interactions at the electrode-barrier interface [96,97,98]. The bias
dependence of the tunnel magnetoresistance devices can change from barrier to other depending
on the quality of the barrier, which is explained with phenomenological model [95]. This model
suggests that, there are localized defect states in the barrier. Excitation of electrons from these
50 100 150 200 250 300
12
14
16
18
20 (a)
MR
rat
io (
%)
Temperature (K)
50 100 150 200 250 300
16
18
20
22
24
26
28
(b)
MR
rat
io (
%)
Temperature (K)
69
defect states either thermally or by hot electron impact, creates states available for a two-steps
tunneling channel, as shown in Fig. 54. Therefore there are two tunneling processes, direct
tunneling through the barrier and indirect tunneling (two tunneling channel) through the defects.
In these two tunneling channels the electrons tunnel from the first electrode to one of the
localized defects while the second tunneling channel, the electron tunnel from the localized
defect to the second electrode. The tunneling in the two-step is spin independent because the
electrons lose its spin memory during the tunneling, which leads to a decrease in the MR ratio.
The MR of this spin independent current can be expressed as:
MR = ((Ipara –Ianti )/(Ianti))/(1+I2/Ianti ) (10)
Where I2 is the current carried by the two-step tunneling. Ipara and Ianti are spin dependent currents
carried out by direct tunneling when magnetic moments are parallel and antiparallel, respectively.
It is clear that the ratio of the spin-dependent and spin-independent currents determines the MR
ratio. With increasing the bias, the probability of the electrons which can tunnel through the
defects in two-step channel is increased. This leads to a decrease in the MR ratio and explains the
bias dependence of tunnel magnetoresistance devices.
-20 -10 0 10 20-2
0
2
4
6
8
10
12
-20 -10 0 10 20
0
3
6
9
12
15
18
600 mV300 mV100 mV
(a1) 10 mV
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
400 mV300 mV100 mV
(b1)
10 mV
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
70
Figure 53: The MR ratio as a function of an applied magnetic field at (a1and b1) for different applied voltage atroom temperature and low temperatures at (a2 and b2).
Figure 54: Schematic view of the two-step tunneling via defect states.
(b)- Temperature dependence has explained before at IV.4.6.
(c)-Magnetization direction of the junction as a function of the external magnetic field direction
is one of the parameters, which effect on the tunnel magnetoresistance devices. Fig. 55-a shows
the MR ratio at easy axis (i.e. the magnetization direction is parallel to the external magnetic
field) and hard axis (i.e. the magnetization direction perpendicular to the external magnetic field)
of a junction prepared by UV oxidation method that has a structure of NiFe 20nm/Al2O3 1.3nm/
Co 15nm and area of 10x3 µm. The MR ratio shows 10.1% at easy axis and 8.7% at hard axis.
The magnetization at easy axis rotates first at the electrode which has low coercivity while the
Barrier
Metal 1 Metal 2
Step 1 Step 2
Conduction band
Trap States
-20 -10 0 10 20
0
4
8
12
16
20
24
-20 -10 0 10 20-2
0
2
4
6
8
10
12
14
16
400 mV300 mV100 mV
(b2)
T= 91K
10 mV
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
400 mV300 mV100 mV
T= 91K(a2) 10 mV
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
71
magnetization at the second electrode still has the opposite direction till the magnetization in the
two electrodes are antiparallel. This leads to maximum MR ratio. The magnetization at hard axis
is difficult to rotate completely and the magnetization in the two electrodes is not completely
antiparallel which leads to a decrease in the MR ratio. Fig. 55-b shows the MR ratio of a junction
prepared by natural oxidation method (in-situ) for 12 hours under oxygen pressure of 130 mbar.
The junction has a structure of FeMn 12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm and the
junction area was 2x3µm. The MR ratio shows 16.8% at easy axis and 3.8% at hard axis.
Because all layers are sputtered in the same direction, therefore the magnetization in all layers are
parallel. At easy axis the magnetization of the soft layer will rotate while at the pinned layer not
and the magnetization of the two layers will be antiparallel which leads to maximum MR. At hard
axis the soft layer NiFe which is the top electrode will rotate and the second layer CoFe will
rotate while the bottom electrode will stay in the same direction because the antiferromagnetic
pinned the magnetization in the other direction. Therefore the magnetization will not be
antiparallel at easy axis which leads to decrease the MR ratio.
Figure 55: MR ratio as a function of applied magnetic field at easy and hard axis for two junctions. At (a) preparedwith UV oxidation and has a structure of NiFe 20nm/Al2O3 1.3nm/ Co 15nm. At (b) prepared with natural oxidation(in-situ) and has a structure of FeMn 12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm.
(d)- One of the parameter, which effects on the tunnel magnetoresistance devices, which was
observed during the experimental work, is the current direction. Fig. 56-a shows a junction
prepared by UV oxidation method and has a structure of NiFe 20nm/Al2O3 1.3nm/Co 15nm and
-20 -10 0 10 20
0
2
4
6
8
10
12
-20 -10 0 10 20
0
2
4
6
8
10
12
14
16
18
hard axis(a) easy axis
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
easy axis
(b)
hard axis
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
72
junction area of 30µm2, in which the current direction changed the MR ratio at positive current
the MR ratio is 10.1% while at negative current is 9.8%. For the other junction at (b) the structure
was FeMn 12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm and has a junction area of 2x3µm,
the MR as a function of applied voltage at room temperature and at low temperature shows bias
dependence direction. The difference at MR ratio as a function of the current direction is
consistence with the band-structure effect as explained in literature [99]. Since the tunneling
current comes mostly from electrons near to the Fermi level of the positive electrode, which are
closest in energy to the barrier top, variation in their tunneling probability as a function of bias is
a mapping of the negative electrode‘s density of states above its Fermi level. Difference in the
electrode/barrier interface cannot be ruled out as another possible cause of this correlation [3].
Till now the MR ratio as a function of the current direction is not completely understood and
needs more research to define which parameters plays a role in this effect (i.e. band structure or
the quality of the barrier).
Figure 56: MR ratio as a function of applied magnetic field with different current direction for junction preparedwith UV oxidation and has a structure of NiFe 20nm/Al2O3 1.3nm/ Co 15nm at (a). MR ratio as a function of appliedvoltage for junction prepared with natural oxidation (in-situ) at different temperatures and the junction has astructure of FeMn 12nm/CoFe 6nm/Al2O3 1.3/ CoFe 2nm/ NiFe 8nm at (b).
-20 -10 0 10 20-2
0
2
4
6
8
10
12
-450 -300 -150 0 150 300 450
9
12
15
18
21
24
negative current(a) positive current
MR
rat
io (
%)
Applied Magnetic Field (mT)
i
jj = 91k
(b) i = 300K
MR
rat
io (
%)
Applied Voltage (mV)
73
IV.4.8. Realibility and breakdown of TMR devices
In the breakdown measurements different junction area varied between 4-600 µm² were
investigated. All junctions have the same electrode thickness of Ni80Fe20 20nm/Al2O3 1.3nm/ Co
15nm, which was prepared by UV oxidation for 1 hour. Fig. 57 a and b shows the I-V curves for
a single junction of 60µm² and 45µm² respectively. The breakdown occurs during a voltage ramp
measurement. At (a) the breakdown occurs at 1.83V while, at (b) the breakdown occurs at 2.1V.
Biasing the junction at a slightly lower voltage (1 V) leads to a strong increase in the lifetime of
the device, which means several days or more [100]. After breakdown, the I-V characteristic is
found to be nearly ohmic. The resistance of the junctions after breakdown is typically in order of
20-130 Ω, which depends on the junction area. At (c) the junction area as a function of the
breakdown voltage shows that with increasing the junction area the breakdown voltage decrease.
The breakdown in the barrier leads to a decrease in the magnetoresistance effect from 14% to
0.8% as shown in Fig. 57-d the junction has a structure of NiFe 20nm/Al2O3 1.3nm/Co 15nm and
area of 3x10µm. The resistance also decreased after breakdown from 1964 Ω to 18 Ω at low bias.
The breakdown as a function of the junction area can be explained by the probability of finding a
weak point in the oxide. In the small junction area the probability to find a weak point in the
oxide is less than in the larger area. This is explained in the schematic diagram of Fig. 58 which
shows the defects in the barrier. At smaller junction like (B) junction there are no defects, while
at larger junctions like (A) and junction (C) there are many defects depending on the junction
area. Therefore, by decreasing the junction size the probability of the junction breakdown will
decrease. This observation has a great importance in the MRAM technology application, where
the ultimate goals design a large array of reliable sub-micron sized junctions.
74
Figure 57: I–V curves for two-junction areas at (a) 60µm2 and at (b) 45µm2. Breakdown voltages of series ofNiFe/Al2O3/Co tunnel junctions as a function of junction area at (c). MR ratio as a function of applied magneticfield before and after breakdown at (d).
0 50 100 150 200
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1
0.00
0.02
0.04
0.06
0.08
-25 -20 -15 -10 -5 0 5 10 15 20 25
0
3
6
9
12
15
0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4
0.000
0.001
0.002
0.003
0.004
0.005
(c)Bre
akdo
wn
Vol
tage
(V
)
Junction area (µm2)
(a)
Breakdown
Tun
nel C
urre
nt (
A )
Applied Voltage ( V )
(d)
before breakdown
after breakdown
MR
rat
io (
%)
Applied Magnetic Field µ0H (mT)
(b)
Breakdown
Tun
nel C
urre
nt (
A)
Applied Voltage (V)
75
Figure 58: Schematic diagram for the defects in the barrier.
IV.4.9. MRAM array of tunnel magnetoresistance elements characterization
The interest in magnetic random access memory (MRAM) devices has been reached recently
with a possibility to use spin tunnel junctions as the memory cell element. The MRAM cells,
which used as a memory array, are needed to be selectable by the diode. Each tunnel junction can
be connected with one diode with which the current will flow only through the single junction.
By turning the diode on or off, the bit is selected for reading. Writing information into the
MRAM cells is quite different from reading them. The cell is written by a magnetic field, which
is generated by current flows through a word line. These words line are densely packed in the
core of the memory with each word line covering many different bits [101].
Figure 59: Mask for MRAM array.
Junction
Bottom electrode
Top electrode
Second ElectrodeFirst Electrode Barrier
A
B
C
Defects
76
Figure 60: Shows several memory elements, selected out of the array, also the bit line current and the word linecurrent for each memory element.
In the magnetic tunnel junction structure, the direction of polarization of the hard layer is used to
store the information. The resistance of the memory tunnel junction is either low or high
dependent on the relative polarization, parallel or antiparallel, of the free layer with respect to the
hard layer. Fig. 59 shows the mask for TMR junctions of the top and bottom electrode. The
presented junction has a structure of NiFe 20nm/Al2O3 1.3nm/Co15 nm with different areas. The
junctions were prepared by ultraviolet oxidation for an hour under pressure 100mbar. Fig. 60
shows the picture for MRAM mask in which the bit lines was the bottom electrode of 25µm wide
and 20 nm thick. The word line on top was separated from the junction by a SiO2 passivation
layer ≈ 200nm thick. The thickness of the Au word line is 150nm of 14 µm width.
Figure 61: Resistance as a function of applied magnetic field for a tunnel magnetic junction.
Word line current
Bit line current
Junction
-20 -10 0 10 20
1250
1300
1350
1400
Res
ista
nce
(Ω
)
Applied Magnetic Field µ0H (mT)
77
Fig. 61 shows the resistance as a function of the applied magnetic field for a junction of MRAM
cells with area 4 x 10µm, the MR ratio shows 13% at room temperature and the resistance was
1.25kΩ. The MRAM memory circuit consists of magnetic tunnel junctions, which are integrated
with 0.25µm CMOS technology [102]. The magnetic tunnel junction is added after preparing the
CMOS circuitry. In this work two different ways are used to characterize the MRAM cell, which
is integrated with a diode. The first one is sputter depositing the magnetic tunnel junction on the
top of the diode. The second , which is used in this work, is connecting the magnetic tunnel
junction in series (by bonding) with the diode, which was GaAs diode. The GaAs p-n diode has a
doping level of 10-17 cm-3 for both p and n regions. The diode dimension is 320µm2 and the
magnetic tunnel junctions dimension varies between 1 x 3µm and 4-6 x 10µm. Fig. 62 shows the
I-V curves for the MTJ, diode, diode + resistance and the device (MTJ + diode). The MTJ is
connected in series with a diode, to separate the diode characteristics from the MTJ
characteristics; the diode was connected with a resistor, which is the bottom electrode in order to
define the I-V of the junction after connection with the diode. For GaAs the voltage needed to run
the diode is 1 V, at this value a lot of current starts to flow through the GaAs diode. The I-V
characteristics of this compound device were measured at zero field.
Figure 62: I-V curves for the MTJ and the diode.
0.0 0.4 0.8 1.2 1.6 2.0-2.0x10-4
0.0
2.0x10-4
4.0x10-4
6.0x10-4
8.0x10-4
1.0x10-3
MTJ diode diode+R diode+MTJ
Tun
nel
Cur
rent
(A
)
Applied Voltage (V)
78
For the writing process there are two possibilities, the first one by applying a large current
through the word line and top electrode to generate a magnetic field sufficient to switch the hard
layer which is chosen to store the information. The second possibility is applying a sufficient
large external magnetic field to switch the hard layer. For the writing an external magnetic field
was applied of 5mT which was enough to switch the hard layer (Co=15nm).
Figure 63: Shows the expected writing and reading timing diagram.
Fig. 63 shows the writing processing in which, if a certain current was applied to write “1” or
“0”, it is expected to appears across the memory element storage/sense line to read them by the
sense output voltage Vs [29]. The Figs. 64 and 65 show the reading signal after writing, in these
figures two different currents were applied. The first one through the word line in order to
generate small magnetic field sufficient to switch the soft layer and the second one through the
junction in order to measure the changes in the resistance through the junction after and before
switched the soft layer which was in two states at low state “1” and high state “0”. The input
signal was in two states the first one from +50mA to –50mA and the second one from 0 to –
50mA as shown in Figs. 64-a and 65-a, while Figs. 64-b and 65-b show the output signal in the
two-directions (±). With comparing the top and the bottom Figs. (i.e. the input and the output) it
is shown clear that with changing the input signal, a strong response at the pulse sense output is
appeared across the memory element storage / sense line.
Read ProcessWrite Process
1
0
1
0
Word currentIw
Sense OutputVs
79
The output signal is shown in Fig. 65-b is measured at 100 mV which shows a very strong output
response compared with the output Fig. 64-b which measured at 1.3 mV. From these two curves
it is concluded that at higher voltage 100mV show more reliable readout than at 1.3mV because
at 100mV produces more response signal than the lower voltage 1.3 mV. The input signal is
changed compared with Fig. 64 to define the memory cell ability to read at any signal and with
different direction word current. The memory cell for reading and writing was tested at input
word current from few mA-100mA for different times and shows reliability for readout process
in both directions positive and negative. As comparing the previous Fig. 63 with the results in
Figs. 64 and 65 show a good agreement.
(a)
(b)
Figure 64: Input and output signals for readout process at +50>-50mA (left) and at 0> -50mA (right).
2 4 6 8 10
12
14
16
18
20
22
24
26
-40
-20
0
20
40
60
Cur
ren
t (m
A)
Time (arb.units)
2 4 6 8 10 12 14 16 18 20 22 24 261.20
1.25
1.30
1.35
1.40
Vol
tage
(m
V)
Time (arb.units)
80
(a)
(b)
Figure 65: Input and output signals for readout process.
In the next Fig. 66 shows the read curves after writing in the two directions positive and negative
writing. The figure shows the output voltage from the memory cell as a function of the current
through the bit line. The cell can read from a few mA and become more saturated from 25mA till
100mA. The difference between the positive and negative changes is due to the changes in the
magnetic single at the MR loop, which is shown in Fig. 61. This is a disadvantage for the sample,
which has non-homogeneity in the MR loop. This is due to the anisotropy shape, which plays a
very important role in the MR loop. The MR loop needed to be sharp in the switching which is
determined by (a combination of) (a) the shape anisotropy, (b) include exchange coupling.
2 4 6 8 10
12
14
16
18
20
22
24
26
-40
-20
0
20
40C
urre
nt (
mA
)
Time (arb.units)
5 10 15 20 25102
103
104
105
106
107
Vol
tage
(m
V)
Time (arbit.units)
81
Figure 66: Read curves after writing in two directions positive and negative.
Finally it is concluded that MRAM arrays can quite well be used in non-volatile memory
applications, which can read and write based on Si technology, but it needs a lot more time for
research before being ready for large-scale fabrication.
-100 -50 0 50 100
1.28
1.30
1.32
1.34
1.36
1.38
Pos. Write Neg. Write
Vol
tage
(m
V)
Current (mA)
82
Summary
The aim of this work is developing the ferromagnetic/insulator/ferromagnetic (TMR)
devices for magnetic data storage and magnetic field sensors. Therefore the switching field of the
element as a function of the size which is one of the important parameters for designing arrays of
MRAM was studied. In addition the characterization of the magnetic materials used were
investigated, the oxidation methods which were used to prepare the TMR devices, the barrier
characterization, the electrical and magnetic properties at different oxidation methods and
different temperatures, realiability and breakdown properties of the TMR devices, and MRAM
array characterization based on TMR elements were investigated.
From the switching characteristics and magnetization behavior of the magnetic materials
used it is shown that:
[a]- For Co single layer of 15nm thickness a trapped magnetization vortices appear in
structures with low aspect ratios (length/width) of 1.5 and 2 which disappeared at aspect ratio >
3. It was found that the magnetization vortices of these patterned elements are also strongly
dependent on the width of the element. With narrower line-width the presence of magnetization
vortices is more pronounced. The switching field and magnetic moment of the arrays which were
measured by (SQUID) magnetometer at different temperatures (4, 300 and 400 K) show
significant temperature dependence. The switching field of the elements increased as the
temperature decreased while at the same temperature it increased with increasing the aspect ratio.
The micromagnetic simulation results show an agreement with the experimental results.
[b]- For NiFe single layer the switching characteristics of the NiFe as a function of the
film thickness (2, 3 and 4 nm) showed that with increasing film thickness the switching field
increases as the total magnetic moment increases. The switching field increased with increasing
the aspect ratio for the same thickness.
[c]- The switching behavior of magnetic-tunnel junctions consisting of trilayers NiFe
20nm/ Al2O3 0.8nm/ Co 15nm shows a relation with the size of the elements. The switching field
for the soft layer and the hard layer increases with increasing the aspect ratio, this is due to the
83
dipole-dipole interaction which is very strong at small aspect ratio and also due to the
magnetization vortex which appear at small aspect ratio for Co single layer and disappeared at
large aspect ratio.
The barrier thickness plays an important role in the preparation of the tunnel
magnetoresistance devices. For optimal reproducibility on wafer level, the barrier thickness must
be controlled on an atomic level. Important barrier properties such as the thickness and the
potential height can be obtained fairly well from the results of the electric transport
measurements using the theoretical models of the spin dependent transport in the magnetic tunnel
junctions.
The barrier characterization results were:
[a] The XPS results showed that, 1nm thin aluminum film is sufficient to cover the bottom
electrode, which then leads to large magnetoresistance.
[b] The MOKE results showed a weak coupling between the two electrodes at 1.3nm
Al2O3 which indicate that the barrier is pinhole free.
The tunnel magnetoresistive junctions which were prepared with different oxidation
methods like natural oxidation in air (ex-situ) or natural oxidation in pure oxygen (in-situ) or
ultraviolet radiation assisted oxidation in oxygen has the following characterization:
[a] The oxidation times for UV are conveniently short, less than 20 min is sufficient to
completely oxidize a thin film of 1.3 nm Al.
[b] The resistance values of the junctions change from one oxidation method to the other for
the same barrier thickness and the same junction area.
[c] The largest MR ratio, which was obtained with the UV radiation assisted oxidation
method, is 20% at room temperature.
[d] The UV assisted process is much more reliable than all the others. At UV radiation
assisted oxidation in oxygen 90% from the TMR junctions show MR values at least 10% at
300 K, which is much higher than for natural oxidation in-situ and ex-situ.
84
[e] No changes of the resistance or the MR ratios were observed even after two years with
junctions prepared with UV radiation assisted oxidation in oxygen while the junctions
prepared with natural oxidation show intrinsic breakdown i.e. short circuit after few weeks.
The tunnel magnetoresistance showed a bias and temperature dependence irrespective of
the applied oxidation method. The reasons for temperature and bias dependence is partly given by
Zhang et al. [92] who put forward a theory which explains the dependence of the conductivities
for spin up and spin down and thus the magnetoresistance effect as function of temperature and
bias with magnon generation absorption in the electrodes.
The other explanation that is due to the presence of the trap states in the barrier at which the
electron doesn’t tunnel directly from electrode to the other in one step but the tunneling occurs in
two or even more steps. In the first step the electron tunnels from the first electrode to one of the
trap states in the barrier and in the second step the electron tunnels from the trap state to the
second electrode. In this mechanism the electrons lose some of their spin information. The
number of such trap states depends on the quality of the barrier, which changed from one
oxidation method to the other i.e. i.e. with controlling the quality of the barrier the bias and
temperature dependence can be neglected.
The dielectric breakdown of patterned magnetic tunnel junctions was also studied. For
junctions with a barrier of 1.3 nm and thinner, an almost immediate breakdown was observed
when the applied voltage approached 2.1 V. For junctions, which have large areas the probability
to breakdown is much higher than for junctions with smaller areas. Our conclusion for this is the
probability to find a weak point in the oxide is much higher at large area than at smaller areas. At
the moment of breakdown a single short circuit is created in the barrier, leading to a low
resistance path in the barrier and a strong decrease in the magnetoresistance ratio. The time
dependent breakdown probability is voltage dependent. These investigations indicate that ultra
small junctions have the most promising features for applications of TMR as storage elements.
From MRAM characterization it is concluded that the MRAM arrays based on tunnel magnetic
junctions can quite be used as nonvolatile memory cells, which can read and write the digital
information depending on Si technology, but it needs long time for research before starting to use
in a fabrication scale.
85
References
[1] P. Gruenberg, R. Scheiber, Y. Pang, M. B. Brodsky, and Sowers, Phys.
Rev. Lett. 57, 2442 (1986).
[2] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Phys. Rev. Lett.
74, 3273 (1995).
[3] S. S. P. Parkin, K. P. Roche, M. G. Samant, P. M. Rice, and P. B. Beyers, J.
Appl. Phys. 85, 5828 (1999).
[4] J. P. Jan, in Solid State Physics 5, 96 (1957).
[5] T. R. Mc Guire, and R. I. Potter, IEEE Trans. Mag. 11, 1018 (1975).
[6] B. Dieny, V. S. Sperioso, S. Metin, S. S. P. Parkin, B. A. Gurney, P.
Baumgart, and D. R. Wilhait, J. Appl. Phys. 69, 4774 (1991).
[7] M. Julliere, Phys. Lett. A 54, 225 (1975).
[8] A. M. Bratkovsky, Phys. Rev. B 56, 2344 (1997).
[9] J. C. Fischer and I. Giaever, J. Appl. Phys. 32, 172 (1961).
[10] S. R. Pollack and C. E. Morris, J. Appl. Phys. 35, 1503 (1964).
[11] J. G. Simmons, J. Appl. Phys. 34, 2581 (1963).
[12] P. Zeemann, phil. Mag. 43,226 (1897).
[13] R. Meservey and P.M. Tedrow, Phys. Rev. 238, 173 (1994).
[14] C. T. Tanaka, J. Nowak, and J. S. Modera, J. Appl. Phys. 81, 5515 (1997).
[15] J. J. Sun, R. C. Sousa, T. T. P. Galvao, V. Soares, T. S. Plaskett, and P. P.
Freitas, J. Appl. Phys. 83, 6694 (1998).
[16] H. Tsuge, and T. Mitsuzuka, Appl. Phys. Lett. 71, 3296 (1997).
[17] R. Behrisch, “Sputtering by Particle Bombardment” I Topics Appl. 2, 47
(1981).
[18] A. S. Penfold and J. A. Thornton, U.S.Patent, 3, 884, 793 (1975).
[19] P. D. Davidse and LM. Maissel, Transactions of the 3rd International Vacuum
Congress, Stuttgart, J. Appl. Phys. 37, 574 (1966).
[20] N. F. Mott, Trans. Faraday Soc., 39 (1940).
[21] R. S. Beech, J. Anderson, J. Daughton, B. A. Everitt, and D. Wang, IEEE
Trans. Mag. 32, 4713 (1996).
[22] J. S. Moodera, L. R. Kinder, J. Appl. Phys. 79, 4724 (1996).
[23] Y. Lu, X. W. Li, G. Xiao, R. A. Altman, J. Gallagher, A. C. Marley, K. P.
86
Roche, and S. S. P. Parkin, J. Appl. Phys. 83, 6515 (1998).
[24] R. C. Sousa, J. J. Sun, V. Soares, P. P. Freitas, A. Kling, M. F. da Silva, and
J. C. Soares, Appl. Phys. Lett. 73, 3288 (1998).
[25] W. Oepts Ph.D. thesis.
[26] Integrated Circuit Engineering Corporation, Study “Memory 1997”
[27] M. Dox, semiconductor international (1997).
[28] R. Scheuerlein, W. Gallagher, S. S. P. Parkin, A. Lee, S. Ray, R. Robertazzi,
and W. Reohr, ISSCC 2000 / Feb (2000).
[29] Z. H. Wang and Y. Nakanura, IEEE Trans. Mag. 32, 2 (1996).
[30] T. Zhu and R. Swanson, MMM conference (1999).
[31] L. V. Melo, L. M. Rodrigues, and P. P. Freitas, IEEE Trans. Mag. 33, 5
(1997).
[32] T. Kobayashy, “Solid-State Sensors and their Applications in Consumer
Electronics and Home Applications, “Tech Digest, Transducers’85, 8 Int.
Conf. On solid-state Sensors and Actuators (1985).
[33] M. Baumler, and L. J. Olsson, Proc. Of a Battelle Europe Conf. 1989, P.77.
[34] R. Coehoorn, S. R. Cumpson, J. J.M. Ruigrok, and P. Hidding, IEEE Trans.
Mag. 35, 2586 (1999).
[35] J. J. M. Ruigrok, E. A. Draaisma, and H. W. Van Kesteren, Philips J. Rev.
51, 21 (1998).
[36] S. Kumagai, N. T. Miyozaki and T. Miyozaki, J. Appl. Phys. 36, 1498
(1997).
[37] S. Kumagai, T. Yooi, and T. Miyozaki, J. Magn. Magn. Mater. 166, 71
(1995).
[38] T. Miyozaki, and N. Tezuka, J. Magn. Magn. Mater. 151, 403 (1995).
[39] T. Miyozaki, N. Tezuka, and S. Kumagai, Physica B 240 (1997).
[40] T. G. S. M. Rijks, W. J. M. de Jonge, W. Folkerts, J. C. S. Kools, and R.
Coehoorn, Appl. Phys. Lett. 65, 916 (1994).
[41] N. Cabrera, N. F. Mott, Rep. Progr. Phys. 12, 163 (1949).
[42] R. W. Hannah, J. Appl. Phys. 34, 1793 (1963).
[43] N. Cabrera, Phil. Mag. 40, 175 (1949).
[44] D. W. Abraham et al.(to be published), J. Slonczewski (unpublished work).
87
[45] H. Zijlstra, Rev. Sci. Inst.41, 1241 (1970).
[46] W. Roos, K. A. Hempel, C. Voight, H. Dedericks, and R. Schippan, Rev. Sci.
Inst.51, 612 (1980).
[47] H. M. Rosenberg, Low Temperature Solid State Physics, Oxford, London,
Chap. 6 (1963).
[48] R. C. Jaklevic, B. D. Josephson, Phys. Lett. 1, 251 (1962).
[49] G. Binnig, C. F. Quate, and Ch. Gerber Phys. Rev. Lett. 56, 930 (1986).
[50] A. Kikukawa, S. Hoseaka, Y. Honda, and S. Tanaka Appl. Phys. Lett. 61,
2607 (1992).
[51] A. Hubert and R. Schaefer, Mag. Domains (Springer, Heidelberg) (1998).
[52] Hartman. U, Appl. Phys. Lett , 51, 374 (1987).
[53] M. Polcarova, A. R. Lang, Appl. Phys. Lett. 1, 13 (1962).
[54] J. Kwo, G. K. Wertheim, M. Gurvitch, and D. N. E. Buchanan, IEEE Trans.
Mag., 19, 795 (1983).
[55] W. Opets, H. J. Verhagen, W. J. M. de Jonge, and R. Coehoorn, J. Appl.
Phys. 73, 2363 (1998).
[56] W. F. Brown, A. E. La Bonte, J. Appl. Phys. 36, 1380 (1965).
[57] A. E. La Bonte, J. Appl. Phys. 40, 2450 (1969).
[58] A. Aharoni, J. Appl. Phys. 46, 914 (1975).
[59] A. Hubert, Phys. Stat. Sol. 38, 699 (1970).
[60] M. E. Schabes, HN Bertram, J. Appl. Phys. 64, 1347 (1988).
[61] A. Aharoni, Phil. Mag. 26, 1473 (1972).
[62] M. R. Scheinfein and JI. Blue, J. Appl. Phys. 69, 7740 (1991).
[63] A. Hubert, Phys. Stat. Sol. 32, 519 (1969).
[64] H. Boeve, J. De Boeck, IMEC, Privat Commenication.
[65] H. Boeve, C. Bruynseraede, Jo Das, K. Dessein, Go Borgns, and J. De
Boeck, to be published at IEEE Trans. Mag. (2000).
[66] H. Boeve, (NM)2 project report (1999).
[67] R. M. H. New, R. F. W. Pease, R. L. White, R. M. Os good, and
K. Babcock, J. Appl. Phys. 79, 5851 (1996).
[68] S. Tehrani, E. Chen, M. Durlam, T. Zhu, and H. Goronkin, IEDM, Technical
Digest 96, 193 (1996).
88
[69] J. F. Smyth, S. Schtlz, D. R. Fredkin, T. Hoehler, I. R. McFaydlin, D. P.
Kern, and S. A. Rishton, J. Appl. Phys. 63, 4237 (1988).
[70] Y. Otani, S. Gu Kim, T. Kohda and, K. Fukamichi, IEEE Trans. Mag. 34,
1090 (1998).
[71] E J. Shi, S. Tehrani, T. Zhu, Y. F. Zheng and J. G. Zhu, Appl. Phys. Lett. 74,
2527 (1999).
[72] E. C. Stoner and E. P. Wohlfarth, Philos. Trans. R. Soc. London, Ser. A
240, 559 (1984).
[73] B. D. Cullity, “ Introduction to magnetic materials” (1972).
[74] R. M. H. New, R. F. W. Pease, and R. L. White, IEEE Trans. Magn.
31, 3805 (1995).
[75] M. Ledermann, D. R. Fredkin, R. O’Barr, S. Schultz, and M. Ozaki,
J. Appl. Phys. 75, 10 (1994).
[76] G. J. Parker and C. Cerjan J. Appl. Phys. 87, 9 (2000).
[77] Yu. Lu, P. L. Trouilloud, D. W. Abraham, R. Koch, J. Slon Czewski, s.
Brown, J. Bucchignano, E. O ‘ Sullivan, R. A. Wanner, W. J. Gallagher, and
S. S. P. Parkin J. Appl. Phys. 85, 5267 (1999).
[78] R. D. Mc Michael, M. J. Donahue, IEEE Trans. Mag. 33, 4167 (1997).
[79] E. Y. Chen, S. Tehrani, T. Zhu, M. Durlam, and H. Goronkin “Sub Micron
Spin Valve MRAM Cell”, J. Appl. Phys. 81, 3992 (1997).
[80] T. Pokhil, D. Song, and J. Nowak, J. Appl. Phys. 87, 1 (2000).
[81] Y. F. Zheng, J. G. Zhu, J. Appl. Phys. 81, 5471 (1997).
[82] F. Bardou, Europhys. Lett. 39, 239 (1997).
[83] H. Boeve, W. Oepts, E. Girgis, J. Schelten, R. Coehoorn, J. De Boeck,
and G. Borghs, ISSCC conference. (1999).
[84] H. Kohlstedt, and C. Daniels, private communication.
[85] J. Shi, T. Zhu, M. Durlam, E. Chen, S. Tehrani, Y. F. Zheng, and
J. G. Zhu, IEEE Trans. Mag. 34, 997 (1998).
[86] R. D. Gomez, M. C. Shih, R. M. H. New, R. F. W. Pease, R. L. White, J.
Appl. Phys. 80, 342 (1996).
[87] P. Rottlaender, H. De Geronkel, H. Kohlsteht, E. Girgis, J. Schelten and
P. Gruenberg, J. Magn. Magn. Mater. 210, 251 (2000).
89
[88] D. R. Fredkin, T. R. Kohler, IEEE Trans. Mag. 26, 1518 (1990).
[89] T. C. Schlthess and W. H. Butler, J. Appl. Phys. 87, 9 (2000).
[90] A. Yelon, M. Frankcombe, and R. Hoffman “in Physics of Thin Films:
Advances in Research and Development” 6, 205 (1971).
[91] R. Strotton, J. Phys. Chem. Solids 23,1177 (1962).
[92] S. Zhang et al, Phys. Rev. Lett. 79, 3744 (1997).
[93] A. M. Bratkovsky, Appl. Phys. Lett. 72, 2334 (1998).
[94] E. Y. Tsymbal and D. G. Pettifor, Phys. Rev. B 58, 432 (1998).
[95] J. Zhang and R. M. White, J. Appl. Phys. 83, 6512 (1998).
[96] S. T. Chui, Phys. Rev. B 55, 5600 (1997).
[97] X. Zhang, B. Z. Li, G. Sun, and F. C. Pu, Phys. Rev. B. 56, 5484 (1997).
[98] M. van Kampen, R. J. M. van de Veerdonk, and A. A. Smits,
unpublished results.
[99] J. Nowak, D. Song, and E. Murdock, J. Appl. Phys. 87, 9 (2000).
[100] W. Oepts, H. J. Verhagen, R. Coehoorn, and W. J. M. de Jonge,
J. Appl. Phys. 86, 7 (1999).
[101] R. C. Sousa, P. P. Freitas, V. Chu, and J. P. Conde, IEEE Trans. Mag.
35, 5 (1999).
[102] M. Durlam, P. Naji, M. Deherrera, S. Tehrani, G. Kerszkoweski, and
K. Kyler, ISSCC 2000/ Feb (2000).
90
Zusammenfassung
In der Wissenschaft und der Technologie, sowie im täglichen Leben spielt der
Magnetismus eine wichtige Rolle. Die Anwendungen des Magnetismus in der gegenwärtigen
Technologiegesellschaft kann in der Fähigkeit der magnetischen
Materialien gesehen werden, Informationen zu speichern. Dieses umfaßt allgemein zu nutzende
und verwendete Geräte, z.B.: Floppy-Disketten, Festplatten, Optische Speicherelemente und
magnetische Streifen auf Kreditkarten. Neue Fortschritte in diesem Bereich sind in vielerlei
Hinsicht durch die Verbesserungen in den magnetischen Materialien voran getrieben worden.
Eine spezielle Kategorie vielversprechender künstlich gebildeter Materialien sind die
sogenannten magnetischen Multilayers oder Superlattices, bestehend abwechselnd aus
magnetischen Schichten mit unterschiedlichen Eigenschaften und einigen nicht magnetischen
metallischen Schichten (Fe/Cr) mit typischen Dicken in Nanometerbereich. Es wurde festgestellt,
daß große Änderungen des Widerstandes eintraten, wenn ein magnetisches Feld angelegt wurde.
Dieser Effekt ist das Resultat der gegenseitigen Ausrichtung der Magnetisierungs- Richtungen
der magnetischen Schichten als Funktion des angelegten Feldes. Da die Elektronenstreuung in
solchen Systemen spinabhängig ist, entsteht der beobachtete Effekt des Magnetowiderstandes.
Die Entdeckung dieses Effektes führte zu einer enormen Zunahme der Bemühung der
wissenschaftlichen Forschung an überlagerten magnetischen Filmen und damit zur Entdeckung
des Riesenmagnetowiderstand (GMR) Effektes bei vielen weiteren Kombinationen abwechselnd
magnetischer/nicht magnetischer Schichtsysteme (z.B. Co/Cu/NiFe). Wesentlich für den
Riesenmagnetowiderstandseffekt ist die Fähigkeit von zwei ferromagnetischen Schichten, ihre
Magnetisierungsrichtung separat zu drehen. Dieser Effekt existiert ungeachtet der Richtung in der
ein Strom durch das Schichtsytem fließt. Es kann zwischen die zwei Konfigurationen
unterschieden werden: Einerseits kann der Strom parallel zur des Schichtfläche fließen (CIP),
anderseits senkrecht zu des Schichtfläche (CPP), wobei die Elektronen durch alle Schichten der
mehrschichtigen Struktur fließen. Der gefundene Wert für die CPP Konfiguration war sogar
größer als der in der CIP Konfiguration. Vor kurzem richtete sich die Entwicklung der
Magnetowiderstands-Bauelemente auf die Tunnelkontakte, die aus zwei ferromagnetischen
91
Metallschichten bestehen, welche durch eine sehr dünne Isolierungschicht getrennet sind. Der
Elektronentransport erfolget senkrecht zur Fläche der Schichten und wird gemessen, indem an
den Elektroden über die Isolatorschicht eine Spannung anlegt wird. Der Magnetowiderstand
basiert auf einer spinabhängigen Tunnelwahrscheinlichkeit, bedingt durch eine energetische
Aufspaltung der Energiebänder, verursacht durch den Spin der Elektronen. Dieser
Magnetwiderstand ist direkt proportional zu den Polarisationsstärken der Elektronen an den zwei
Isolator / Metallgrenzflächen, die zu der Polarisation im Inneren der zwei magnetischen Filme
unterschiedlich sein können.
Gegenwärtig bekommt der Tunnelmagnet-Widerstand (TMR) eine größere Bedeutung, da
er ihn in Mikro- und Submikro- Dimensionen strukturiert werden kann. Dies wird in der
Hochtechnologieindustrie bei zwei Hauptanwendungen, einerseits als Mittel der Speicherung von
Information und anderseits als Sensor genutzt, um magnetisch gespeicherte Informationen
auslesen zu können.
In dieser Arbeit werden die Resultate der Experimente zum spinabhängigen Transport in
den strukturierten mehrschichtigen Systemen dargestellt.
Der erste Teil der Arbeit behandelt die Charakterisierung der magnetischen Materialien mit
unterschiedlichen Techniken. Im zweiten Teil werden Auswirkungen auf die Herstellung und die
Entwicklung der Tunnelmagnetwiderstands- Einheiten mit unterschiedlichen
Oxidationsmethoden dargestellt. Diese sind die natürliche Oxidation, die thermische Oxidation
und die von ultravioletter Strahlung unterstützte Oxidation in Sauerstoff. In dieser Arbeit werden
die Resultate der ultravioletten Oxidationsmethode ausgiebig diskutiert, weil sie weltweit neu
sind.
Der dritte Teil, behandelt die zukünftige Anwendbarkeit der Magnetowiderstandkontakte
als RAMS (MRAM). Bevor jedoch diese Strukturen in den kommerziellen Bereich gelangen
können, ist ein gutes fundamentales Verständnis ihrer Eigenschaften und ihrer Grenzen
erforderlich. Das stellt das Ziel der Forschung dar, die in dieser Arbeit beschrieben wird.
Die Tunnelwahrscheinlichkeit eines Elektrons durch eine Isolatorbarriere ist abhängig von der
Elektronendichte. Die Dichte am Fermi- Niveau ist ungleich für “Spin-oben“ und “Spin-unten“
Elektronen. Dies ist das Resultat der Austauschwechselwirkung in den magnetischen Schichten.
92
Die Tunnelwahrscheinlichkeit der Spin-oben und Spin-unten Elektronen ist mithin verschieden.
Das Resultat ist ein spinpolarisiertes Tunneln, vorausgesetzt, es gibt keine Spinflips. Die
wirksame Dichte der Zustände für die tunnelnden Elektronen mit Spin-oben und Spin-unten in
beiden Elektroden, D1 oben, D1 unten und D2 oben, D2 unten, werden in Abbildung (1) gezeigt.
Der Tunnelstrom für die parallele und antiparallele Magnetisierung kann ausgedrückt werden
durch:
Ip α D1↑ D2 ↑ +D1 ↓ D2 ↓
Ia α D1↑ D2 ↓ + D1 ↓ D2 ↑
Die wirksamen D↑ und die D↓ sind nicht die realen Dichten der Zustände in den Ferromagneten,
weil die tunnelnden Elektronen durch die Grenzfläche zwischen Elektrode und Isolator
beeinflußt werden. Die Asymmetrie in der wirksamen Dichte der Zustände der Auf- und Ab-
Elektronen wird durch die Polarisation P an den Grenzflächen beschrieben:
P= (D↑ – D↓)/ (D↑ + D↓)
Das Magnetwiderstands-Verhältnis ist damit:
∆R/R= 2P1P2/ (1-P1P2)
P1 und P2 sind die Polarisation der tunnelnden Elektronen an der ersten und zweiten Elektrode.
Das TMR-Verhältnis kann prinzipiell unendlich groß werden, wenn beide Materialien eine 100 %
vollständige Polarisation aufweisen.
Abbildung (1): Schema des spinabhängigen Tunnels: Die ähnliche Dichte der Zustände für eine paralleleAusrichtung der Magnetisierungs- Richtung der ferromagnetischen Schichten (obere Abbildung) führt zu einergrößeren Leitfähigkeit, verglichen mit der antiparallelen Ausrichtung.
Eine große Anzahl der Proben wurde mit elektrischen und magnetischen Messtechniken
vorbereitet und characterisiert. Die benutzten Substrate sind thermisch oxidierte
e-EF
M1 M2
EF
M1 M2
93
Silizium- Wafer (100), da diese (vor der Oxidation) eine geringe Oberflächenrauhigkeit besitzen.
Die typische Oxiddicke schwankt zwischen 50 und 150 nm. Diese Oxidschicht ist für die
elektrische Isolation zwischen den Kontakten auf dem gleichen Wafer erforderlich. Für diese
Oxidschichten erhöht sich die Oberflächenrauheit etwas. Sie ist zur Oxiddicke proportional. Die
untere magnetische Schicht ist aus unterschiedlichen Gründen die kritischste:
(a) Die Oberflächen-rauigkeit dieser Schicht muß klein sein.
(b) Für die Strukturierung ist es erforderlich, daß die Dicke dieser Schicht mindestens 10 nm
beträgt.
(c) Das aufgebrachte Aluminium sollte eine benetzende Eigenschaft auf der Oberfläche haben.
Für die Aufbringung der drei Schichten von Co/Al2O3/NiFe wurde die Methode des Magnetron-
Sputterns verwendet. Mit den drei Targets: Kobalt, Aluminium und Permalloy nacheinander
wurden die Schichten gesputtert, ohne das Vakuum zu brechen. Der Basisdruck betrug 2 x 10-6
mbar. Während der kurzen Sputterzeit muß das Substrat mit einem wassergekühlten
Substrathalter auf 22 ° C gehalten werden. Nach dem Sputtern des Co-Films von 10-20 nm folgt
ein dünner
Aluminum-Film von 0,8-2 nm. Die Aluminiumschicht wurde entweder an Luft bei
Raumtemperatur (ex-situ), d.h. außerhalb der Sputteranlage oder in-situ, d. h. innerhalb der
Sputteranlage in Sauerstoff, ebenfalls bei Raumtemperatur oxidiert, auch durch thermische
Oxidation (in-situ), bei einer Temperatur von 60-90° C oder durch Plasmaoxidation (in-situ), und
durch ultraviolette Strahlung unterstützte Oxidation in einer Sauerstoffatmosphäre (in-situ)
oxidiert.
Schließlich wird ein Dünnfilm von NiFe aufgesputtert. Die weitere Bearbeitung der
Wafer mit diesen drei Schichten erfolgt in einem Reinraumlabor.
Mit optischer Lithographie werden die Querschnitte der Kontakte definiert und
ausschließend geätzt, indem eine Ionenstrahlätzung (IBE) durchführt wird. Eine Argonenergie
von 250 eV wurde verwendet, um die Seitenwandbeschädigung so klein wie möglich zu halten.
Die Ätzung wird gestoppt, sobald die Co-Schicht erscheint. Nach einem weiteren
lithographischen Prozeß wird die Struktur definiert und mit Ionenstrahl geätzt. Dann wird der
strukturierte Wafer mit einer 150 nm dicken SiO2- Schicht bedeckt, welche die Zuleitungen
isoliert. Danach erfolgt eine 300 nm dicke Gold-Aufdampfung auf der Oberseite nach einem
weiteren lithographischen Prozeß, womit die Zuleitungen erzeugt werden. Die elektrischen
94
Messungen erfolgten dann an vier Punkten zwischen Goldkontaktauflagen und der
Grundelektrode. Systematisch wurden die Kontaktquerschnitte innerhalb der Grenzen von 1 bis
600 µm2 verändert.
Zur Vorbereitung der Tunnelmagnetwiderstands-Elemente wurden die Materialien der
zwei Elektroden und der Barriere mit unterschiedlichen Techniken untersucht. Die
Schaltungseigenschaften für die verwendeten Materialien (NiFe, Co und Co/Al2O3/NiFe) sind
wichtig, wenn man Reihen von MRAMs und Sensoren des magnetischen Feldes entwirft. Diese
Parameter wurden mit verschiedenen Techniken, wie alternativem Gradientmagnetometer (AGM)
bei Raumtemperatur und SQUID bei unterschiedlichen Temperaturen untersucht. Die Proben für
AGM und SQUID wurden mittels Elektronenstrahllithographie vorbereitet (Substrat 3.5 mm x
3.5 mm) und in Arrays von 107 Elementen mit Elementgrößen von 100 x 150 nm bis 6000 x 2400
nm strukturiert.
Der Abstand zwischen den Elementen schwankt zwischen 0,5 und 5 µm, abhängig von
der Geometrie der Elemente und der Dicke des Materials. Die Morphologiedes Strukturen wurde
mit dem RasterKraftmikroskop (AFM) und das Magnetisierungsverhalten mit dem
Magnetokraftmikroskopie (MFM) untersucht. Die Abmessungen der Proben sind die gleichen
wie bei AGM und SQUID. Ein Problem ist die Schaltcharakteristik, die zwischen der
magnetischen Kopplung der Elektroden besteht, welche mit dem Magnetooptischen Kerreffekt
(MOKE) untersucht wurde. Die Größe der Proben waren 10 mm x 10 mm. Sie wurden mit
MOKE nach dem Sputtern und vor der Lithographie gemessen.
Hochwertige dünne Barrieren haben keine Löcher, die zu Kurzschlüssen führen können.
Solche Löcher können mit der Röntgenstrahl-Photoemissions-Spektroskopie (XPS) ermittelt
werden. Nach dem Sputtern der Kobalt-Grundelektrode wurde eine sehr dünne
Aluminiumschicht auf gesputtert und mit XPS untersucht, ob sich während der Oxidation des
Aluminiums ein Oxid der Grundelektrode gebildet hat.
Die Schalteigenschaften und die Magnetisierungsstrukturen von 15 nm dicken Kobalt-
Einzelschichten, bestehend aus Arrays unterschiedlicher Breiten (100 nm, 200 nm und 600 nm)
wurden untersucht. Diese Arrays wurden durch Elektronenstrahllithographie mit
unterschiedlichem Längenverhältnis (Länge / Breite = 1.5, 2, 3 und 4) geschrieben. Der Effekt
95
der Breite und das Längen-zu-Breite-Verhältnis wurden systematisch mit AGM, AFM, MFM
und mit der mikromagnetischen Simulation, die auf der Lösung der Landau-Lifshitz-Gilbert-
Gleichung basiert, studiert. AGM und MFM zeigen, daß aufgefangene Magnetisierungs-
Turbulenzen in den Strukturen mit niedrigen Längenverhältnissen L/W = 1.5 und 2, aber nicht in
den Strukturen mit hohem Längenverhältnis L/W>3 erscheinen. Es wurde gefunden, daß die
Magnetisierungs- Turbulenzen dieser Elemente stärker von der Breite des Elements abhängig
sind. Bei kleinerer Linienbreite ist das Vorhandensein der Magnetisierungs-Turbulenzen offenbar
schon vorhanden. Das Schaltfeld und das magnetische Moment der Arrays wurden mit einem
DC-SQUID-Magnetometer bei unterschiedlichen Temperaturen gemessen (4K, 300K, 400K).
Die Schaltfelder zeigen eine Temperaturabhängigkeit. Dieses Feld der Elemente wird erhöht mit
dem gleichen Längenverhältnis (Länge / Breite), während die Temperatur verringert wird. Ein
magnetisches Kraftmikroskop des Typs Digital Instruments 3000 (MFM) wurde benutzt, um
magnetische Bilder der remanenten Zustände bei Raumtemperatur zu erhalten. Die
mikromagnetischen Simulationsresultate stimmen mit den experimentellen Resultaten gut
überein.
Die Schalteigenschaften und des Magnetisierungsverhalten von NiFe wurden als
Funktion der Filmschichtdicke studiert. Sie zeigen, daß sich bei Zunahme der Filmdicke das
Schaltfeld erhöht, da sich das gesamtmagnetische Moment erhöht und es schwer ist, die
Magnetisierung zu drehen, wenn die (Dimensionen) Abmaße eines Elements sehr klein werden.
Das Schaltverhalten von magnetischen Tunnelkontakten, bestehend aus Trilayern NiFe 20 nm /
AlOx 0.8nm / Co 15nm verschiedener Längenverhältnisse (Länge zu Breite)=1.5, 2, 3, 4 und bei
unterschiedlichen Breiten von 100 nm, 200 nm und 600 nm wurde untersucht. Die beobachteten
Resultate zeigen die Relation zwischen Schaltfeld und Größe der Elemente. Das Schaltfeld für
die weiche magnetische Schicht erhöht sich bei der Zunahme des Längenverhältnisses, während
das Schaltungsfeld der harten magnetischen Schicht bei Zunahme des Längen-Breiten-
Verhältnisses verringert wird. Mit dem Verringern des Verhältnisses (L/W) wird die
magnetostatische Kopplung zwischen den zwei ferromagnetischen Schichten erhöht.
Die Barrierendicke spielt eine wichtige Rolle für die Vorbereitung der
Tunnelmagnetwiderstands-Einheiten. Für optimale Reproduzierbarkeit auf dem Wafer muß die
Barrierendicke bis auf eine Atomlage genau gesteuert werden. Wichtige Barriereneigenschaften,
wie die effektive Dicke und Höhe, können aus den elektrischen Transportmessung mit den
theoretischen Modellen zum Tunneln ermittelt werden.
96
Durch die XPS-Messung wurde festgestellt, daß ein 1 nm dicker Aluminiumfilm genügt,
um die Grundelektrode zu bedecken.
Die Tunnel-Magnetwiderstände werden mit unterschiedlichen Oxidationsmethoden,
natürliche Oxidation in Luft (ex-situ), natürliche Oxidation im reinen Sauerstoff (in-situ) oder
UV unterstützte Oxidation in Sauerstoff hergestellt.
Diese haben folgende Eigenschaften:
(A) - Die Oxidationszeiten bei des UV Methode sind üblicherweise kurz. Bereits weniger als 20
Minuten vermutlichgenügen um einen Film von 1.3 nm Aluminium vollständig zu oxidieren.
(B) - Die Widerstandswerte der Kontakte unterscheiden sich von einer Oxidationsmethode zur
anderen bei gleicher Barrierendicke und gleicher Kontaktgeometrie.
(C) - Der Widerstand der Kontakte ist umgekehrt proportional zur Kontaktfläche. Dieses zeigt
die Homogenität des Prozesses.
(D) - Mit den präparierten Proben wurde der größte Magnetowiderstand (MR) von 20% mit der
UV-Oxidations-Methode bei Raumtemperatur erreicht.
Der UV unterstützte Prozeß ist viel zuverlässiger. Ein MR- Wert von mindestens 10% bei 300
Kelvin wurde erreicht, und dies mit einer Ausbeute von 90%, die wesentlich höher ist als bei
natürlicher Oxidation, in-situ oder ex-situ. Es wurden selbst nach zwei Jahren keine Änderungen
des Widerstandes oder des Magnet-Widerstands-Wertes mit dieser UV-Oxidation beobachtet.
Dagegen zeigen die Kontakte, mit natürlicher Oxidation hergestellt, nach wenigen Wochen ein
intrinsisches “Breakdown”.
Der Tunnelmagnetwiderstand zeigte eine Vorspannungs- und Temperaturabhängigkeit,
die bei unterschiedlichen Oxidationsmethoden verschieden groß war.
Dafür gibt es mindestens zwei Gründe. Nach Zhang[92] ist die inelastische magnetische
Elektronenstreuung nicht zu vernachlässigen. Die Erzeugung und Vernichtung von Magnonen ist
mit einem Spinflip verbunden. Durch solche Prozesse verlieren die tunnelnden Elektronen ihr
Spingedächtnis, d. h. es gibt einen Elektronentransport durch die Barriere, der nicht mehr von der
gegenseitigen Richtung der Magnetisierungen abhängt. Da dieser Beitrag temperaturabhängig ist
und auch spannungsabhängig sein wird, kann damit die Beobachtung zumindest qualitativ erklärt
werden.
97
Für nicht perfekte Barrieren, d. h. für Barrieren mit Gitterdefekten oder größeren Baufehlern
öffnet sich ein weiterer Kanal für den Elektronentransport von Defekt zu Defekt. Solche
Hopping- und Mikrotunnelprozesse sind mit erhöhter Wahrscheinlichkeit mit
Spinrichtungsänderungen gekoppelt und können dadurch den Magnetwiderstandseffekt
verringern. Die Verringerung ist um so größer, je mehr Transportkanäle mit Spinflip es gibt. Die
Experimente zeigen, daß Defekte in den Isolatorschicht die Eigenschaften des magnetischen
Tunnelkontakts erheblich beeinflussen.
Der elektrische Durchschlag von gesputterten magnetischen Tunnelkontakten, die UV
unterstützt in O2 oxidiert wurden, war Gegenstand weiterer Untersuchungen. Solche Werte
stellen interessante technologische Spezifikationen für eine Realisierung von MRAMs und von
magnetischen Leseköpfen dar.
In den Kontakten mit einer Barriere von 1.3 nm und kleiner wird fast ein sofortiger
Zusammenbruch beobachtet, wenn die angelegte Spannung annähernd 2.1 V beträgt. Bei
Kontakten mit größerer Fläche ist die Wahrscheinlichkeit zum Durchbruch größer als bei
kleineren. Die Wahrscheinlichkeit, einen schwachen Punkt im Oxid zu finden, ist größer bei
großen als bei kleinen Querschnitten. Ein Durchbruch erzeugt einen Defekt in der Barriere, führt
zu einem niedrigen Widerstand in der Barriere und zu einer starken Abnahme des
Tunnelwiderstands-Verhältnisses. Die Wahrscheinlichkeit für einen Durchbruch ist von der
angelegten Spannung abhängig.
Für zukünftige Anwendungen werden extrem kleine Kontakte als MRAMs gefordert.
Es wird festgestellt, daß für die Anwendung der Tunnelmagnetkontakte die bisher
benutzte MRAM-Matrix als Permanentspeicher zum Lesen und Schreiben genutzt werden kann.
Allerdings ist noch viel Entwicklungsarbeit nötig, bevor eine Massenfertigung beginnen und eine
Markteinführung stattfinden kann.
98
Acknowledgment
I am very grateful to Prof. J. Schelten, Institute for Thin Film and Ion Technology,
Forschungszentrum Juelich, Germany, for proposing the point of research and his supervision of
the present work. His continuous guidance, advice and valuable comments throughout the
discussion of the results are much appreciated.
The author would like to thank Prof. P. Grünberg, for his helpful discussion, IFF,
Forschungszentrum Juelich, Germany,
The author wishes to express his deep gratitude to H. Boeve, Dr. J.De Boek and Prof. G.
Borghs, IMEC, Leaven, Belgium for their cooperation during the measurements of the MRAM
carried out there.
My deep gratitude to Dr. S. Taherani, Dr. H. Gronkin, Dr. J. Shi and J. Jansesk, MRAM
group, Motorola labs. Physical Science Research Lab., AZ, USA, for their cooperation and kind
support during my stay at Motorola. Their help to finish the research work “Characterization and
micromagnetic modeling of different magnetic materials patterned at nonmeter-scale arrays” are
much appreciated.
The author is very grateful to the scientific and technical staff of the lithography
department: Dr. R. Lehmann, Dr. M. Pabst, Dr. A. van der Hart, F. Schroteler, J. Zilikens, A.
Pracht, M. Noon, P. Bochum, U. Kurz, Dr. M. Siegel, A. Steffen, Institute for Thin Film and Ion
Technology, Forschungszentrum Juelich, Germany, for their kind help throughout this work.
The author would like to thank Dr. W. Oepts and Prof. R. Coehoorn, Philips Research
Labs., Eindhoven, Netherlands for their helpful discussion of the breakdown measurements.
Finally, I am thankful to the Egyptian Government, “Edaret El-Behsat” for offering me
the chance to have one of its fellowships to finish the present work.