Acta mater. 49 (2001) 4005–4015
www.elsevier.com/locate/actamat
ATOMIC SCALE STRUCTURE OF SPUTTERED METAL
MULTILAYERS
X. W. ZHOU1†, H. N. G. WADLEY1, R. A. JOHNSON1, D. J. LARSON2, N. TABAT2,
A. CEREZO3, A. K. PETFORD-LONG3, G. D. W. SMITH3, P. H. CLIFTON4, R. L.
MARTENS5 and T. F. KELLY5
1
Department of Materials Science and Engineering, University of Virginia, Charlottesville, VA 22903,
USA, 2Recording Head Operations, Seagate Technology, Minneapolis, MN 55435, USA, 3Department of
Materials, University of Oxford, Oxford OX1 3PH, UK, 4Seagate Technology, Research and Development,
Londonderry BT48 0BF, UK and 5Department of Materials Science and Engineering, University of
Wisconsin, Madison, WI 53706, USA
( Received 25 May 2001; received in revised form 21 June 2001; accepted 24 July 2001 )
Abstract—A combined theoretical and experimental approach has been used to study nanoscale
CoFe/Cu/CoFe multilayer films grown by sputter deposition. Such films have applications in sensors that
utilize the giant magnetoresistance effect, for example, read heads in high-density information storage devices.
Atomistic simulations based on a molecular dynamics approach and an alloy form of the embedded atom
method have been developed to accurately model the sputter deposition of the CoFe/Cu/CoFe multilayers.
The simulations show that relatively flat interfaces are formed because of the energetic deposition conditions.
However, significant intermixing at the CoFe-on-Cu interface, but not at the Cu-on-CoFe interface, was
observed. An abnormal Fe depletion zone is also revealed at the CoFe-on-Cu interface. A three-dimensional
atom probe method has been used for a nanoscale chemical analysis of the films. It provided direct verification
of the simulations. The simulations have then been used to understand the mechanism responsible for the
formation of the intermixing defects observed in the multilayers. A novel deposition technique is proposed
which reduces both interfacial mixing and Fe depletion by controlling the incident adatom energies.  2001
Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.
Keywords: Magnetoresistive effects; Multilayers; Surfaces & interfaces; Atom probe; Molecular dynamics
1. INTRODUCTION
When ferromagnetic layers (such as CoFe) are separated by a non-magnetic spacer layer (such as Cu) of
an appropriate thickness (for example, approximately
2 nm), their magnetic moments will align antiferromagnetically in a zero magnetic field [1]. Spin-dependent conduction electron scattering [2, 3] then results
in a relatively high electrical resistance. If a sufficiently large magnetic field is applied to reverse the
magnetic moment of one of the ferromagnetic layers,
a ferromagnetically aligned sandwich can be created.
This alignment reduces the spin-dependent conduction electron scattering, and causes a decrease in the
electrical resistance of the film—an effect known as
giant magnetoresistance (GMR) [2, 3]. The GMR
effect was first observed in vapor deposited Fe/Cr/Fe
sandwiches in 1988 [4]. Since then, many other
† To whom all correspondence should be addressed. Tel.:
+1-804-982-5672; fax: +1-804-982-5677.
E-mail address: [email protected] (X. W. Zhou)
material systems have been shown to exhibit the
effect,
including
the
Co/Cu/Co
[5,
6],
NiFe(Co)/Cu/NiFe(Co) [7, 8], and NiCo/Cu/NiCo
systems [9]. The significant technological importance
of GMR materials as magnetic field sensors arises
from their very high sensitivity to an external magnetic field when fabricated as submicron devices.
Because of this, GMR multilayers are now widely
used as read-head sensors in hard disk drives, and are
also being actively explored for a new form of magnetic random access memory [10].
GMR materials have been synthesized using
numerous vapor deposition techniques including
sputter-deposition [7], ion beam deposition [11], and
molecular beam epitaxy [4]. These studies have
shown that the magnetotransport properties are highly
sensitive to the method and conditions of growth.
Experiments have indicated that a significant decrease
in the GMR ratio occurs when the average spacing
between the ferromagnetic layers is changed, even by
as little as one monolayer [5]. High-quality GMR
multilayers must therefore have a precisely controlled
1359-6454/01/$20.00  2001 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.
PII: S 1 3 5 9 - 6 4 5 4 ( 0 1 ) 0 0 2 8 7 - 7
4006
ZHOU et al.: SPUTTERED METAL MULTILAYERS
layer thickness. In addition, if the layers are comparably rough in amplitude and wavelength to the spacer
layer thickness, Néel coupling occurs and it becomes
more difficult to reverse the magnetic moment of the
layers. Furthermore, the alloying of one layer by the
metal atoms of another causes an increase in spin
independent scattering [12] and a loss of local magnetic alignment [13]. Relatively smooth, unmixed
interfaces therefore appear desirable [7, 14]. The
intrinsic properties of the material system are also
important, together with other nanoscale structural
features, including interfacial impurities [15], texture
[16], any buffer layer material [17], the presence of
surfactants during growth [18], the grain
size/morphology of each layer, various lattice defects
(vacancies, voids, dislocations and twins) [19], and
perhaps residual stress. Of these, the atomic scale perfection across the multilayer interfaces appears to
most critically affect the GMR properties [20].
Although the importance of these various effects is
clear, the very small scale of the defects have made
it difficult to make a direct link between a given effect
and the change in magnetic properties. Results
obtained from different investigations are often contradictory. In experimental studies of Fe/Cr/Fe multilayers, for instance, Petroff et al. found that increasing the interfacial roughness and the interfacial
mixing both increased the GMR ratio [21], while
Belien et al. indicated that a lower interfacial roughness and reduced interfacial mixing result in a higher
GMR ratio [22]. The contradictory nature of these
results most likely arises because the atomic scale
structures of the deposited films are not well understood. In the present work, atomistic simulations
based on the molecular dynamics (MD) approach
have been used to investigate growth mechanisms and
their control in order to create improved interfaces.
This has been combined with direct atomic-scale
characterization of sputter-deposited layers using the
three-dimensional atom probe (3DAP), which has
allowed experimental verification of the results of
the simulation.
reconstruction is simulated by tracking the positions
of atoms using Newton’s equations of motion. This
correctly identifies the detailed atomic structures of a
deposited film, and reveals many of the mechanisms
active throughout the deposition process. However,
the accuracy of the simulation depends upon the fidelity of the interatomic potentials used to calculate
the interatomic forces. The embedded atom method
(EAM) potential initially developed by Daw and
Baskes [23] has provided a good potential format for
the metal atoms that are used in the GMR multilayers.
The EAM potential assumes that the crystal energy
is the sum of a pairwise potential and an energy
required to embed an atom into a local medium with
a given electron density. Because this embedding
energy depends on the local “crowding” (atom
density), EAM accounts for the local environment
dependence of the potential, and therefore has been
successfully used to solve a variety of problems
involving surface structures and defect states. The
potentials used in the present work are listed in
Appendix A.
Initial Ni80Fe20 substrate crystals with a lateral size
of 120 {224} planes in the x direction and 16 {220}
planes in the z direction were used in the MD simulations (see Fig. 1). Periodic boundary conditions were
used to extend the lateral dimensions of thin films.
The growth was simulated by injecting adatoms to
the top xz surface from random starting positions far
above at a frequency that gives rise to the desired
deposition rate. To account correctly for the relaxation of crystal size under the influence of the internal
stress between different layers, a Lagrangian form of
Newton’s equations of motion [24] was used to solve
for the evolution of atomic structures. To prevent the
2. ATOMISTIC SIMULATION OF ATOMIC
STRUCTURE
Many factors contribute to the structure of a vapordeposited multilayer film, including layer composition, growth temperature and deposition energy. To
explore the effects of these factors within a computer
model, a highly predictive atomistic simulation
method must be used. The method must not only produce reliable results, but must also not require preexisting knowledge of the film growth mechanisms.
In our work, a MD approach is used to provide an
accurate description of the way in which atoms in
the vapor phase arrive at a substrate surface and then
subsequently rearrange themselves.
In a MD simulation, the transition of atoms from
vapor to a solid surface and the subsequent surface
Fig. 1. MD simulated NiCo/Cu/NiCo multilayer.
ZHOU et al.: SPUTTERED METAL MULTILAYERS
crystal shift during energetic impacts, the bottom
region of the substrate was fixed. Damping forces
were applied to the atoms in a subsurface region to
adjust the atom velocities so that a near constant temperature was maintained in this region. During deposition, this isothermal region was advanced upwards
in a way that left several free monolayers at the surface. A 5.8 nm Co90Fe10/1.5 nm Cu/4.0 nm Co90Fe10
multilayer unit was simulated using an adatom energy
of 3.0 eV, a normal adatom incident angle, a substrate
temperature of 300 K, and a deposition rate of 1
nm/ns. The multilayer stack employs similar
materials to those used to produce the “spin valve”
structures used for magnetic field sensors. Because
the intent is to explore the atomic scale structure of
the interfaces, the layer thickness was somewhat arbitrarily chosen and the multilayer structure is not one
that would necessarily give rise to the best GMR
properties. An adatom energy of 3.0 eV was used
because a direct simulation Monte Carlo study indicated that this is the typical energy during sputter
deposition at chamber pressures between 1 and 10
mTorr [25]. The use of normal adatom incidence and
substrate temperature of 300 K also mimics the typical low temperature GMR sputter deposition processes where adatoms impact the growth surface with
a cosine distribution peaking at normal incidence. To
complete the simulation within a realistic time frame,
an accelerated deposition rate of 1 nm/ns is used in
the MD simulation. This unrealistically high deposition rate reduces the time during which atoms can
diffuse on the surface before they are embedded into
the bulk. Fortunately, this approximation of high
deposition rate does not significantly affect the results
because surface diffusion is negligible at the low substrate temperature range of interest here [26].
The detailed atomic structure of the simulated
multilayers is shown in Fig. 1, where Ni, Fe, Co and
Cu atoms are marked green, yellow, blue and red,
respectively. It is interesting to note that the Cu-onCoFe interface appears relatively sharp, whereas the
CoFe-on-Cu interface is quite diffuse. A significant
number of Cu atoms are seen within the CoFe layer,
several atomic planes from the nominal CoFe-Cu
interface. This indicates that Cu incorporation in the
subsequently deposited CoFe layer is much greater
than in the previously deposited CoFe layer.
To quantify the atomic distributions near the interfaces, a composition vs depth profile is needed. To
calculate this, a small box was defined and placed at
the bottom of the layers. The atomic fraction of Cu,
Co and Fe in the box was calculated by dividing the
number of atoms of each species by the total number
of atoms in the box. Composition–depth profiles were
mapped out (by continuously moving the box across
the multilayer in the growth direction), as shown in
Fig. 2. The thickness of the box was chosen to be
about one atomic layer (0.2 nm), and the lateral
dimension of the box was such that it contained a
total number of atoms per sample of N⬇100. If the
4007
Fig. 2. Composition vs thickness profiles for the simulated
NiCo/Cu/NiCo multilayer.
mean composition in the box satisfies a binomial distribution, then the standard deviation for any calculated composition C is [27]
s⫽
冪
C(1⫺C)
N
(1)
The (1s) error bar is therefore 5.0% at C = 50 at.%.
For the composition range of 5–10 at.%, this error bar
is further reduced to between 2.2 and 3%. Calculated
composition profiles for Co, Cu and Fe are shown in
Fig. 2 with solid, dashed and bold lines, respectively.
The statistical fluctuations are clearly visible, for
example in the Fe profile within the CoFe layer.
The composition–depth profiles of Fig. 2 give a
quantitative measure of the chemical sharpness of the
interfaces. Consistent with Fig. 1, Fig. 2 indicates that
the Cu-on-CoFe interface is much sharper than the
CoFe-on-Cu interface. If measured between 10 and
90% Cu, the interfacial widths are about 0.33 and
1.44 nm for the Cu-on-CoFe and the CoFe-on-Cu
interfaces, respectively. To further explore the distribution of various species (especially the ferromagnetic atoms) across the interfaces, the Co/Fe ratio was
calculated along the growth direction. The results of
this calculation, which provide a good indication of
the relative Co and Fe distributions across interfaces,
are included in Fig. 2 as a shaded region. Compared
to a nominal Co/Fe ratio of 9, Fig. 2 shows a region
of abnormally higher Co/Fe ratio at the CoFe-on-Cu
interface indicative of Fe depletion near this interface.
However, there are fewer Co and Fe atoms in this
region than within the body of the CoFe layer, so
that the errors on the Co/Fe ratio are larger. We have
therefore applied a two-sample T-test to determine the
significance of the observed variation in the Co/Fe
ratio. This test is used to determine at what confidence level the difference between the means of two
distributions is statistically significant. Since it is
reasonable to assume that a variation in the Co/Fe
ratio may exist in the plane of the layers, three separate composition profiles were measured at three different lateral regions as shown in Fig, 1. Each profile
4008
ZHOU et al.: SPUTTERED METAL MULTILAYERS
was divided into three intervals: the Cu-on-CoFe
interface, the CoFe-on-Cu interface and a CoFe
region that has the average data of the upper and
lower CoFe layers. Mean and variance values for the
Co/Fe ratio were calculated for each interval in each
profile. Two-sample T-tests were then performed to
determine the level of confidence at which a significant difference in means could be stated. Table 1
shows the results comparing the CoFe-on-Cu and the
Cu-on-CoFe interfaces to the CoFe layer. All three
profiles shown in Table 1 indicate a higher Co/Fe
ratio at the CoFe-on-Cu interface than in the bulk
CoFe, with two profiles showing this difference at a
significant confidence level of ⬎90%. When combining all data profiles, we find that this difference is
significant at the 97% confidence level. The difference between the Cu-on-CoFe interface and the bulk
CoFe is not obvious as two profiles show a lower
Co/Fe ratio at the Cu-on-CoFe interface than the bulk
CoFe, while the other one shows the reverse. An
additional conclusion that can be drawn from the
combined MD data calculation is that the Co/Fe ratio
at the CoFe-on-Cu interface is higher than that at the
Cu-on-CoFe interface at a confidence level of 98%.
3. EXPERIMENTAL MEASUREMENT OF ATOMIC
STRUCTURE
Experimental characterization of the physical and
chemical nature of interfaces in nanoscale multilayers
is a challenging task, due to the very fine scale of
any roughness or intermixing sufficient to generate
changes in magnetic properties. What is required is a
technique which has ultra-high spatial resolution in
all three dimensions. Transmission electron
microscopy (TEM) [28], traditionally used to study
these materials, cannot in most cases chemically
identify single atoms and can only provide average
structure information through the specimen thickness.
To overcome this problem, a 3DAP technique [27,
29] has been used that enables the three-dimensional
atomic distributions in a sample to be reconstructed.
The basic concept of the 3DAP experiment is
shown in Fig. 3. The sample consists of a needle with
an end radius of 50–100 nm, which is held at cryogenic temperature in an ultra-high vacuum system.
Under an applied electric field, pulsed field evaporation is used to ionize the atoms on the sample sur-
face and evaporate them one by one. The evaporated
ions are accelerated in a radial direction by the electric field until they impact on a time- and positionsensitive detector. This allows both the impact position and the flight time of the ion to be measured.
The 3DAP acts as a point projection microscope,
allowing the original position of each atom on the
specimen surface to be deduced, whilst the flight time
is used to determine the chemical identity of the ions.
As more and more atoms are removed from the sample, sample material is gradually eroded away, and
the third dimension of the measured image is built
up from the sequence of ion detection. The 3DAP
technique has a depth resolution of a single atomic
layer and sub-nanometer lateral resolution, the latter
being poorer because of small aberrations in the trajectories of the ions in the near-surface region. To
obtain atomic-layer resolution for the interfaces, a
new technique has been developed so that the sample
can be prepared with the interfaces normal to the sample axis and the region of interest positioned within
100 nm of the needle apex [30, 31]. This new development enables a 3DAP analysis of atomic features
across the interfaces of GMR multilayers. To prepare
3DAP samples, small posts (approximately 4 µm
square and 100 µm long) are fabricated on a standard
(001) silicon substrate. TEM results have indicated
that growth of multilayers on posts of this size does
not significantly affect the morphology of the layers
[32].
The multilayer structure investigated by 3DAP
consisted of a basic structure as follows: 5.0 nm
Ni82Fe18/4.0 nm Co90Fe10/3.0 nm Cu/4.0 nm Co90Fe10
(layer thicknesses are nominal). To induce a ⬍111⬎
texture in the layers during growth, a seed layer was
first deposited on the Si wafer. Ten repeats of the
multilayer structure were then deposited on the seed
layer, followed by a 50 nm Ni82Fe18 cap to protect
the structure during subsequent sample preparation
procedures. The deposition was carried out using DC
magnetron sputtering at a base pressure of about
1×10⫺6 Pa. The 3DAP analysis was performed in an
energy-compensated optical position-sensitive atom
probe, with an ultra-high vacuum of about
5×10⫺9 Pa, a pulse fraction of 15% and a pulse repetition rate of 1500 Hz. The specimen temperature
was held at 70–80 K during field-ion imaging and
atom probe analyses.
Table 1. MD simulated composition distribution: mean Co/Fe ratio ⬍r⬎, its standard deviation s, confidence level C1 for the CoFe-on-Cu interface
to differ from the bulk CoFe, and confidence level C2 for the Cu-on-CoFe interface to differ from the bulk CoFe
Region 1
⬍r⬎
CoFe-on-Cu interface
Bulk CoFe
Cu-on-CoFe interface
Confidence level, C1
Confidence level, C2
13.80
9.63
9.27
Region 2
s
12.30
2.97
3.00
93%
32%
⬍r⬎
12.57
8.67
8.30
Region 3
s
3.79
2.08
0.74
⬎99%
75%
⬍r⬎
10.00
9.85
11.45
s
2.23
4.18
3.56
21%
74%
ZHOU et al.: SPUTTERED METAL MULTILAYERS
Fig. 3. 3DAP approach for atomic scale characterization.
A 3DAP reconstruction of a section of the stack is
shown in Fig. 4. As in Fig. 1, the Ni, Fe, Co and
Cu atoms are marked green, yellow, blue and red,
respectively. The Cu layer thickness was found to be
thinner than the nominal thickness. A remarkable
similarity is found between the simulated image in
Fig. 1 and the experimental image in Fig. 4. For
instance, the roughness (amplitude and wavelength)
for both the Cu-on-CoFe interface and the CoFe-onCu interface seen in Fig. 4 are very close to those in
Fig. 1. As in the simulated image, the Cu-on-CoFe
interface in Fig. 4 is relatively sharp while the CoFeon-Cu interface is diffuse, and the Cu atoms are
mixed in the CoFe layer. This experimental observation verifies that Cu mixing in the subsequent CoFe
layer is much more significant than in the CoFe layer
deposited previously.
To quantify the 3DAP data, the same sampling
method described above was used to analyze the variation of Co, Cu, and Fe concentration along the
growth direction. To highlight the statistical variations of the data, measurements were performed for
various regions on several samples. The results of the
composition profiles for two of the samples are shown
in Figs 5(a) and (b), respectively. Again, a remarkable
Fig. 4. 3DAP measured NiCo/Cu/NiCo multilayer.
4009
4010
ZHOU et al.: SPUTTERED METAL MULTILAYERS
3DAP data profiles, we found that the difference is
significant at a confidence level of 99%. Additionally, Table 2 suggests a lower Co/Fe ratio at the Cuon-CoFe interface than in the CoFe bulk. We also
found from the analyses of all 3DAP data that the
Co/Fe ratio at the CoFe-on-Cu interface is higher
than that at the Cu-on-CoFe interface at a confidence level of 93%.
4. MECHANISMS FOR FORMATION OF INTERFACE
IMPERFECTIONS
Fig. 5. Composition vs thickness profiles for the 3DAP measured NiCo/Cu/NiCo multilayer.
similarity is found between the simulated results in
Fig. 2 and the experimental results in Fig. 5. It can
be seen from both Figs 2 and 5 that the Cu, Co, and
Fe composition profiles are asymmetric across the
two interfaces on either side of the Cu layer. The
compositions undergo a sharp transition across the
Cu-on-CoFe interface, but the transition across the
CoFe-on-Cu interface is much more diffuse. Quantitatively, Fig. 5(a) shows interfacial widths of about
0.3 and 0.95 nm, respectively, for the Cu-on-CoFe
and the CoFe-on-Cu interfaces, and Fig. 5(b) shows
interfacial widths of about 0.5 and 1.2 nm, respectively, for the Cu-on-CoFe and the CoFe-on-Cu interfaces. These result in estimates of mean width and
standard deviation of 0.4±0.14 and 1.08±0.18 nm,
respectively for the Cu-on-CoFe and the CoFe-onCu interfaces.
Similar to the calculation for the MD profiles, a
determination of the significance of the variation in
Co/Fe ratio over the profiles calculated from the
3DAP data was performed using two-sample T tests.
The results obtained from five 3DAP composition
profiles (including the two shown in Fig. 5) that
encompass different regions of the multilayer are
listed in Table 2. It can be seen from Table 2 that
all five profiles indicate a higher Co/Fe ratio at the
CoFe-on-Cu interface than in the CoFe bulk, with
three of the profiles showing a confidence level of
above 90% for this difference. When combining all
Elementary thermodynamic arguments indicate that
a material A cannot wet the surface of a material B
if B wets A. When wetting does not occur, flat interfaces are difficult to form in nanoscale multilayers
because of the tendency for the nonwetting layer to
grow by an island mechanism. As a result, GMR
multilayers with all interfaces flat cannot be obtained
under kinetically unconstrained (thermal equilibrium)
growth conditions. Previous MD simulations have
indicated that the use of energetic (several eV)
adatom impact induces atomic jumps, resulting in a
flattening of the surface asperities present during
deposition by island growth [25]. This promotes
flatter interfaces, and appears to account for the observation that multilayers grown by magnetron sputtering [7] and ion beam deposition processes [11]
(where adatoms have an average impact energy of 苲3
eV or above) often have superior GMR properties to
those grown by a low impact energy molecular beam
epitaxy method [4].
However, energetic adatom impacts can cause
other phenomena to occur, in particular atomic mixing. Figs 1, 2, 4 and 5 indicate that in sputter
deposited CoFe/Cu/CoFe multilayers, the Cu-onCoFe interface is relatively sharp, while the CoFe-onCu interface is quite diffuse. MD analyses of metal
multilayer growth by energetic atom deposition have
indicated that this interfacial mixing can occur by an
impact induced exchange event [25]. In this process,
the energetic adatom penetrates the surface sufficiently so that it is embedded in the substrate and
ejects a substrate atom onto the surface. The exchange
probability is lower on a flat surface than on a rough
surface, and increases with increasing impact energy.
The question raised within the present work is why
there should be an asymmetry in the intermixing. In
order to address this question, the MD method was
used to calculate the exchange probabilities during a
large number of random impacts by 10 eV Cu, Co
and Fe atoms on flat (111) surfaces of fcc Cu and Co
crystals. Because the intent is to determine the relative exchange probabilities among different species,
a high energy of 10 eV was chosen for study so that
significant exchanges can be observed on the ideally
flat surface. The results indicate that Co and Fe atoms
impacting on a Cu surface have an exchange probability of 82 and 54%, respectively, while Cu atoms
of similar energy impacting on a Co surface have an
ZHOU et al.: SPUTTERED METAL MULTILAYERS
4011
Table 2. 3DAP measured composition distribution: mean Co/Fe ratio ⬍r⬎, its standard deviation s, confidence level C1 for the CoFe-on-Cu
interface to differ from the bulk CoFe, and confidence level C2 for the Cu-on-CoFe interface to differ from the bulk CoFe
Sample 1
⬍r⬎
CoFe-on-Cu
interface
Bulk CoFe
Cu-on-CoFe
interface
Confidence level,
C1
Confidence level,
C2
Sample 2
s
⬍r⬎
Sample 3
s
⬍r⬎
Sample 4
s
⬍r⬎
Sample 5
s
⬍r⬎
s
11.30
0.93
17.72
10.50
10.10
2.05
12.55
6.97
17.40
9.53
9.94
6.55
1.91
2.14
11.39
5.37
3.93
2.16
8.49
6.16
3.86
4.24
9.18
6.72
3.36
2.42
9.85
5.37
3.33
5.70
93%
92%
81%
59%
98%
89%
⬎99%
74%
94%
78%
exchange probability of only 6%. The higher
exchange probability between Co adatoms and Cu
surface atoms therefore leads to a diffuse CoFe-onCu interface. Conversely, because Cu adatoms are
less likely to exchange with Co or Fe surface atoms,
the Cu-on-CoFe interface is relatively sharp
(unmixed).
Since the melting point of Cu is lower than that
of Co or Fe, it might be expected that incident atoms
of whatever type would be more likely to exchange
with surface Cu atoms than with surface Co or Fe
atoms. However, a further driving force for thermal
diffusion induced exchange would be provided by
any tendency for surface segregation. To investigate
this further, Cu and Co crystals with a (111) surface
were simulated. A monolayer of Cu, Co, or Fe was
placed either on the surface or one monolayer below
the surface of these crystals. After relaxing the crystals by energy minimization based on a conjugate
gradient method, the average energy change (⌬E)
associated with moving atoms from subsurface to
surface lattice positions was calculated. The values
of ⌬E for Cu segregation to the surface of Co, Co
segregation to the surface of Cu, Fe segregation to
the surface of Cu, and Fe segregation to the surface
of Co are ⫺0.22, 0.19, 0.00, and ⫺0.20 eV per
atom, respectively. It can be seen that Cu and Fe
both tend to segregate to the surface of Co (both
processes reduce the energy).
Both the atomistic simulation and the 3DAP data
above indicate that in the Co90Fe10/Cu/Co90Fe10
multilayer, the Fe concentration is lower than the
nominal value at the CoFe-on-Cu interface. While it
is unclear if this Fe depletion has any effect on the
GMR properties, several mechanisms have been
identified that could be the cause of this effect. During deposition of the CoFe layer on top of a given
Cu plane, a deposited Co atom exchanges more frequently with a Cu atom in that plane than a deposited
Fe adatom. As a result, the fraction of deposited Co
atoms that will occupy that plane will be greater than
in the bulk of the CoFe layer. A relatively Fe depleted
(with respect to Co) region is then created at the
CoFe-on-Cu interface. In addition, Fe tends to segregate on the surface of Co. A continuous Fe migration
to the surface of the CoFe layer through thermally
activated diffusion would also result in a Fe depleted
zone at the CoFe-on-Cu interface and a Fe enhancement at the Cu-on-CoFe interface. At the high deposition rate (or the short time scale) simulated, thermally activated diffusion is negligible. However, any
thermally activated diffusion that can exist at the
realistic low deposition rate should be reflected by the
3DAP data. Evidence for the Fe enhancement is less
strong than for the Fe depletion described previously.
This is because both impact induced exchange and
thermal diffusion induced exchange cause the Fe
depletion at the CoFe-on-Cu interface, but only the
thermal diffusion can cause the Fe enhancement at
the Cu-on-CoFe interface. It is likely that insufficient
bulk diffusion exists during room temperature sputter
deposition for Fe surface segregation to occur, and
the Fe depletion effect seen at the CoFe-on-Cu interface is believed to be mainly caused by the higher
probability of impact induced exchanges with Cu surface atoms.
5. GROWTH PROCESS IMPROVEMENT
MD simulations provide compelling evidence that
increasing the adatom energy flattens interfaces, but
induces interlayer mixing at interfaces when a high
tendency exists for the underlying material to segregate to the surface of the material being deposited.
While lowering the adatom energy can sharpen the
chemical boundary of the interface, it is
accompanied by increased interfacial roughness. As
a result, better GMR properties are obtained from
materials grown with an intermediate adatom
energy [7, 11]. Detailed analyses of the asymmetric
interfacial structures and their formation mechanisms indicate that the use of different adatom
energies to deposit the Cu and the CoFe layers may
be more beneficial than the use of a single
“optimum” energy for deposition of all layers. In
particular, the use of a lower adatom energy for
4012
ZHOU et al.: SPUTTERED METAL MULTILAYERS
deposition of the CoFe layer on Cu than for the Cu
on CoFe layer (interlayer energy modulation) can
effectively sharpen the CoFe-on-Cu interface. This
strategy has been experimentally implemented and
found to improve the GMR ratio [33]. Even better
results may be possible by modulating the energy
during deposition of each layer (intralayer energy
modulation). To investigate this idea, a simulation
was conducted where the first few monolayers of
Cu (or CoFe) were deposited on CoFe (or Cu) using
a low adatom energy to avoid impact induced mixing. After the interface was well buried below the
surface, higher adatom energy was used to flatten
the surface. The result of simulating the growth of
the 5.8 nm Co90Fe10/1.5 nm Cu/4.0 nm Co90Fe10
multilayer on a 1 nm Ni80Fe20 substrate using this
modulated energy scheme is shown in Fig. 6. It can
be seen that compared to the result shown in Fig.
1, the interlayer mixing is greatly reduced while the
interfacial roughness remains almost unchanged. A
composition depth profile was also calculated, and
is shown in Fig. 7. The interfaces are seen to be
very sharp, and the Fe depletion at the CoFe-onCu interface due to the exchange mechanism is also
removed. The interfacial widths are about 0.33 and
0.24 nm, respectively, for the Cu-on-CoFe and the
CoFe-on-Cu interfaces. The two interfaces on either
side of the Cu layer are hence much narrower and
more symmetric.
An intralayer energy modulated deposition has not
been tested experimentally. This is because current
sputter deposition and ion beam deposition systems
are not designed for an independent control of adatom
Fig. 6. A NiCo/Cu/NiCo multilayer achieved using a modulated energy deposition method.
Fig. 7. Composition vs thickness profiles obtained using the
modulated energy deposition.
energy within the range required for the energy
modulation. Nevertheless, the perspective provided
by the combination of simulation and experiment
immediately leads to a direction for modifying the
deposition parameters to improve multilayer properties.
6. CONCLUSIONS
An integrated approach combining atomistic
simulations and three-dimensional atom probe
experiments has been developed to study atomic
scale structures of nanoscale metal multilayers. The
application of this approach to the sputter deposition
of nanoscale CoFe/Cu/CoFe multilayers revealed
the following:
1. Energetic adatom impact with the growth surface can cause exchange between the adatom
and underlying atoms. The exchange probability
is much higher when the underlying material is
Cu rather than Co and Fe. Consequently, the
Cu-on-CoFe interface is much sharper than the
CoFe-on-Cu interface, and Cu mixing in the
subsequently deposited CoFe layer is much
more significant than in the pre-deposited
CoFe layer.
2. During deposition of the CoFe layer on Cu, the
Co adatoms exchange more frequently with
underlying Cu atoms than the Fe adatoms. As a
result, a relatively Fe depleted (with respect to
Co) region is created at the CoFe-on-Cu interface.
3. Adatom energy is a key parameter for the deposition of GMR multilayers. Low adatom energies
reduce the exchange probability and hence the
interlayer mixing. However, high adatom energies
are essential to flatten the surface. An interlayer
modulated energy deposition can produce lower
combination of interfacial roughness and interlayer mixing than a single “optimum” energy
deposition. An intralayer modulated energy depo-
ZHOU et al.: SPUTTERED METAL MULTILAYERS
4013
REFERENCES
17. Dei, T., Nakatani, R. and Sugita, Y., Jpn. J. Appl. Phys.,
1993, 32, 1097.
18. Chopra, H. D., Hockey, B. J., Chen, P. J., Egelhoff, W.
F., Wuttig, M. and Hua, S. Z., Phys. Rev., 1997, B55, 8390.
19. Schulthess, T. C., Butler, W. H., Zhang, X. G. and Nicholson, D. M. C., Phys. Rev., 1997, B56, 8970.
20. Parkin, S. S. P., Phys. Rev. Lett., 1993, 71, 1641.
21. Petroff, F., Barthelemy, A., Hamzic, A., Fert, A., Etienne,
P., Lequien, S. and Creuzet, G., J. Magn. Magn. Mater.,
1991, 93, 95.
22. Belien, P., Schad, R., Potter, C. D., Verbanck, G., Moshchalkov, V. V. and Bruynseraede, Y., Phys. Rev., 1994,
B50, 9957.
23. Daw, M. S. and Baskes, M. I., Phys. Rev., 1984, B29,
6443.
24. Parrinello, M. and Rahman, A., J. Appl. Phys., 1981, 52,
7182.
25. Zhou, X. W. and Wadley, H. N. G., J. Appl. Phys., 1998,
84, 2301.
26. Yang, Y. G., Johnson, R. A. and Wadley, H. N. G., Acta
mater., 1997, 45, 1455.
27. Miller, M. K., Cerezo, A., Hetherington, M. G. and Smith,
G. D. W., Atom Probe Field Ion Microscopy. Oxford University Press, Oxford, 1996.
28. Portier, X. and Petford-Long, A. K., Phys. D: Appl. Phys.,
1999, 32, R89.
29. Cerezo, A., Godfrey, T. J., Sijbrandij, S. J., Warren, P. J.
and Smith, G. D. W., Rev. Sci. Instrum., 1998, 69, 49.
30. Larson, D. J., Wisman, B. D., Villieux, R. J., Martens, R.
L., Gribb, T. T., Erskine, H. F., Kelly, T. F. and Tabat,
N., Microsc. Microanal., 2001, 7, 24.
31. Larson, D. J., Clifton, P. H., Tabat, N., Cerezo, A., PetfordLong, A. K., Martens, R. L. and Kelly, T. F., Appl. Phys.
Lett., 2000, 77, 726.
32. Larson, D. J., Martens, R. L., Kelly, T. F., Miller, M. K.
and Tabat, N., J. Appl. Phys., 2000, 87, 5989.
33. Hylton, T. L., Coffey, K. R., Parker, M. A. and Howard,
J. K., J. Appl. Phys., 1994, 75, 7058.
34. Wadley, H. N. G., Zhou, X. W., Johnson, R. A. and Neurock, M., Prog. Mater. Sci., 2001, 46, 329.
35. Johnson, R. A., Phys. Rev., 1989, B39, 12554.
1. Parkin, S. S. P., Bhadra, R. and Roche, K. P., Phys. Rev.
Lett., 1991, 66, 2152.
2. Butler, W. H., Zhang, X. G., Nicholson, D. M. C. and
MacLaren, J. M., J. Magn. Magn. Mater., 1995, 151, 354.
3. Levy, P. M., J. Magn. Magn. Mater., 1995, 140-144, 485.
4. Baibich, M. N., Broto, J. M., Fert, A., Nguyen Van Dau,
F., Petroff, F., Etienne, P., Creuzet, G., Friederich, A. and
Chazelas, J., Phys. Rev. Lett., 1988, 61, 2472.
5. Honda, S., Ohmoto, S., Imada, R. and Nawate, M., J.
Magn. Magn. Mater., 1993, 126, 419.
6. Parkin, S. S. P., Li, Z. G. and Smith, D. J., Appl. Phys.
Lett., 1991, 58, 2710.
7. Kools, J. C. S., J. Appl. Phys., 1995, 77, 2993.
8. Meguro, K., Hirano, S., Jimbo, M., Tsunashima, S. and
Uchiyama, S., J. Magn. Magn. Mater., 1995, 140-144, 601.
9. Bian, X., Strom-Olsen, J. O., Altounian, Z., Huai, Y. and
Cochrane, R. W., J. Appl. Phys., 1994, 75, 7064.
10. Prinz, G. A., Science, 1998, 282, 1660.
11. Schmeusser, S., Rupp, G. and Hubert, A., J. Magn. Magn.
Mater., 1997, 166, 267.
12. Butler, W. H., Zhang, X. G., Nicholson, D. M. C. and
MacLaren, J. M., J. Magn. Magn. Mater., 1995, 151, 354.
13. Nicholson, D. M. C., Butler, W. H., Zhang, X. G.,
MacLaren, J. M., Gurney, B. A. and Speriosu, V. S., J.
Appl. Phys., 1994, 76, 6805.
14. Nozieres, J. P., Speriosu, V. S., Gurney, B. A., Dieny, B.,
Lefakis, H. and Huang, T. C., J. Magn. Magn. Mater.,
1993, 121, 386.
15. Wellock, K. P., Hickey, B. J., Xu, J., Walker, M. J. and
Wiser, N., J. Magn. Magn. Mater., 1995, 140-144, 585.
16. Pollard, R. J., Wilson, M. J. and Grundy, P. J., J. Magn.
Magn. Mater., 1995, 146, L1.
EAM potentials developed for atoms of a single
element generally cannot be applied to alloys. The
potential cut-off distances fitted for individual
elements are often not consistent. By normalizing
the EAM potentials and introducing an EAM alloy
model, we have developed a procedure to generalize
the EAM potentials and their cut-off distance [34].
This generalized EAM potential has been extended
to enable calculations of alloys with any combination of 16 metals (Cu, Ag, Au, Ni, Pd, Pt, Al, Pb,
Fe, Mo, Ta, W, Mg, Co, Ti, and Zr). These potentials are well fitted to basic material properties such
as lattice constants, elastic constants, bulk moduli,
vacancy formation energies, sublimation energies,
and heats of solution. The use of molecular dynamics for the simulation of GMR multilayer assembly
from alloy vapors reported here has now become
possible due to the recent development of this
potential database.
The derivation of integrated EAM potentials for a
variety of elements and their alloys has been
described in detail previously [34], and hence only
the results are listed here. In EAM, the total energy
E of the crystal can be expressed as
sition was predicted to have a further potential to
reduce the interfacial roughness and interlayer
mixing and to remove the Fe depletion zone at
CoFe-on-Cu interfaces.
The limitations of past experimental techniques
for the direct analysis of atomic scale microstructural features combined with a limited understanding of atomic assembly processes have been responsible for a slow convergence upon the ideal
processes for synthesizing GMR multilayers. The
emergence of a reliable combination of computer
modeling and atomic level experiments provides an
integrated approach to the design of these (and
potentially many other) nanostructured materials. It
permits not only the identification of atomic scale
structure, but also the prediction of compositions
and processing routes needed to achieve desired
properties. This combined approach may be critical
to the successful synthesis of many of the complex
devices of future interest.
Acknowledgements—The University of Virginia is grateful to
the Defense Advanced Research Projects Agency for its support through the Virtual Integrated Prototyping program
(program managers: A. Tsao, D. Healey and S. Wolf). This
research was partially sponsored by the US National Science
Foundation (RLM and TFK). We would also like to thank Dr
B. D. Wisman and H. F. Erskine at Seagate for their assistance
with this research and Prof. B. Cantor at Oxford University for
provision of laboratory facilities.
APPENDIX A
4014
ZHOU et al.: SPUTTERED METAL MULTILAYERS
Table A1. EAM parameters
Cu
re
fe
re
a
b
A
B
␬
l
Fn0
Fn1
Fn2
Fn3
F0
F1
F2
F3
h
Fe
Ag
2.556162
1.554485
22.150141
7.669911
4.090619
0.327584
0.468735
0.431307
0.86214
⫺2.176490
⫺0.140035
0.285621
⫺1.750834
⫺2.19
0
0.702991
0.683705
0.921150
⫺2.191675
2.891814
1.106232
15.539255
7.944536
4.237086
0.266074
0.386272
0.425351
0.850703
⫺1.729619
⫺0.221025
0.541558
⫺0.967036
⫺1.75
0
0.983967
0.520904
1.149461
⫺1.751274
Fe
re
fe
re
a
b
A
B
␬
l
Fn0
Fn1
Fn2
Fn3
F0
F1
F2
F3
h
Fe
2.885034
1.529021
21.319637
8.086176
4.312627
0.230728
0.336695
0.420755
0.841511
⫺2.930281
⫺0.554034
1.489437
⫺0.886809
⫺2.98
0
2.283863
0.494127
1.286960
⫺2.981365
Mo
2.481987
1.885957
20.041463
9.818270
5.236411
0.392811
0.646243
0.170306
0.340613
⫺2.534992
⫺0.059605
0.193065
⫺2.282322
⫺2.54
0
0.200269
⫺0.148770
0.391750
⫺2.539945
E⫽
Au
Ta
2.728100
2.723710
29.354065
8.393531
4.476550
0.708787
1.120373
0.137640
0.275280
⫺3.692913
⫺0.178812
0.380450
⫺3.133650
⫺3.71
0
0.875874
0.776222
0.790879
⫺3.712093
冘
1
f (r ) ⫹
2i,j,i⫽j ij ij
2.860082
3.086341
33.787168
8.489528
4.527748
0.611679
1.032101
0.176977
0.353954
⫺5.103845
⫺0.405524
1.112997
⫺3.585325
⫺5.14
0
1.640098
0.221375
0.848843
⫺5.141526
冘
Fi(ri)
Ni
Pd
2.488746
2.007018
27.984706
8.029633
4.282471
0.439664
0.632771
0.413436
0.826873
⫺2.693996
⫺0.066073
0.170482
⫺2.457442
⫺2.70
0
0.282257
0.102879
0.509860
⫺2.700493
W
3.196291
0.544323
7.132600
10.228708
5.455311
0.137518
0.225930
0.5
1.0
⫺0.896473
⫺0.044291
0.162232
⫺0.689950
⫺0.90
0
0.122838
⫺0.226010
0.431425
⫺0.899702
Al
2.771916
2.336509
34.108882
7.079952
3.775974
0.449644
0.593713
0.413484
0.826967
⫺4.099542
⫺0.754764
1.766503
⫺1.578274
⫺4.17
0
3.474733
2.288323
1.393490
⫺4.174332
Mg
2.740840
3.487340
37.234847
8.900114
4.746728
0.882435
1.394592
0.139209
0.278417
⫺4.946281
⫺0.148818
0.365057
⫺4.432406
⫺4.96
0
0.661935
0.348147
⫺0.582714
⫺4.961306
(A1)
2.750897
1.595417
22.770550
7.605017
4.056009
0.385412
0.545121
0.425578
0.851156
⫺2.320473
⫺0.421263
0.966525
⫺0.932685
⫺2.36
0
1.966273
1.396717
1.399758
⫺2.362609
Pt
2.886166
1.392302
20.226537
6.942419
3.702623
0.251519
0.313394
0.395132
0.790264
⫺2.806783
⫺0.276173
0.893409
⫺1.637201
⫺2.83
0
0.929508
⫺0.682320
0.779208
⫺2.829437
Co
Ti
2.505979
1.975299
27.206789
8.679625
4.629134
0.421378
0.640107
0.5
1.0
⫺2.541799
⫺0.219415
0.733381
⫺1.589003
⫺2.56
0
0.705845
⫺0.687140
0.694608
⫺2.559307
冘
fj(rij)
(A2)
j,j⫽i
with fj(rij) the electron density at the site of atom i
arising from atom j at a distance rij away.
Alloy EAM potentials can be constructed from
elemental EAM potentials if the potentials are normalized [35] and unified cutoff functions are used. To
Zr
2.933872
1.863200
25.565138
8.775431
4.680230
0.373601
0.570968
0.5
1.0
⫺3.203773
⫺0.198262
0.683779
⫺2.321732
⫺3.22
0
0.608587
⫺0.750710
0.558572
⫺3.219176
冋 冉 冊册
冉 冊
冋 冉 冊册
冉 冊
A·exp ⫺a
ri ⫽
3.499723
0.647872
8.906840
8.468412
4.516486
0.134878
0.203093
0.425877
0.851753
⫺1.419644
⫺0.228622
0.630069
⫺0.560952
⫺1.44
0
0.921049
0.108847
1.172361
⫺1.440494
3.199978
2.230909
30.879991
8.559190
4.564902
0.424667
0.640054
0.5
1.0
⫺4.485793
⫺0.293129
0.990148
⫺3.202516
⫺4.51
0
0.928602
⫺0.981870
0.597133
⫺4.509025
fit such an EAM potential set, the generalized pair
potentials were chosen to have the form
i
where fij represents the pair energy between atoms i
and j separated by rij, and Fi stands for the embedding
energy to embed an atom i into a local site with electron density ri. ri can be calculated using
Pb
f(r) ⫽
r
⫺1
re
r
⫺␬
re
r
B·exp ⫺b ⫺1
re
⫺
20
r
1 ⫹ ⫺l
re
1⫹
20
(A3)
where re is the equilibrium spacing between nearest
neighbors, A, B, a, b are four adjustable parameters,
and ␬, l are two additional parameters for the cut
off [35].
The electron density function is taken with the
ZHOU et al.: SPUTTERED METAL MULTILAYERS
same form as the attractive term in the pair potential
with the same values of b, and l, i.e.,
冋 冉 冊册
冉 冊
fe exp ⫺b
f(r) ⫽
1⫹
r
⫺1
re
r
⫺l
re
density range rnⱕr⬍ro. For a smooth variation of
the embedding energy, these equations are required
to match values and slopes at their junctions. These
equations are listed as follows:
(A4)
20
4015
冘 冉 冊
3
F(r) ⫽
Fni
i⫽0
i
r
⫺1 , r⬍rn, rn ⫽ 0.85re
rn
(A6)
Using the alloy model [35], the pair potential between
different species a and b is constructed as
fab(r) ⫽
冉
冊
1 fb(r) aa
fa(r)
f (r) ⫹ b fbb(r)
a
2 f (r)
f (r)
(A5)
To have embedding energy functions that can work
well over a wide range of electron density, we have
used three equations to separately fit to different electron density ranges, r⬍rn, rnⱕr⬍ro and roⱕr. By
using rn = 0.85re and ro = 1.15re where re is the
equilibrium electron density, we can ensure that all
equilibrium properties can be fitted in the electron
冘冉 冊
3
F(r) ⫽
Fi
i⫽0
i
r
⫺1 , rnⱕr⬍ro, ro ⫽ 1.15re
re
(A7)
冋 冉 冊 册冉 冊
F(r) ⫽ Fe 1⫺ln
r
re
h
·
r h
, roⱕr
re
(A8)
With this model, the parameters needed to define all
the 16 metals are listed in Table A1.