JOURNAL OF APPLIED PHYSICS
VOLUME 87, NUMBER 1
1 JANUARY 2000
Atomistic simulation of the vapor deposition of Ni/Cu/Ni multilayers:
Incident adatom angle effects
X. W. Zhou and H. N. G. Wadley
Department of Materials Science, School of Engineering and Applied Science, University of Virginia,
Charlottesville, Virginia 22903
共Received 8 March 1999; accepted for publication 21 September 1999兲
Molecular dynamics simulations have been used to explore the effects of incident adatom angle
upon the atomic scale structure of Ni/Cu/Ni multilayers grown by vapor deposition under controlled
incident atom energy conditions. For incident atom energies of 1 eV or less, increasing the incident
angle increased interfacial roughness, resulted in void formation in the nickel layer, and intermixing
at the interfaces between metal layers. The interfacial roughness that formed during low impact
energy oblique angle deposition was significantly reduced by substrate rotation during growth.
However, rotation had no beneficial effects upon interfacial mixing. The use of a higher incident
atom energy 共⭓5 eV/atom兲 resulted in flatter interfaces and eliminated voids under oblique
incidence conditions, but it also caused more severe interfacial mixing by an atomic exchange
mechanism. When low 共thermal兲 impact energies were used to deposit the first few monolayers of
each new metal layer, intermixing by the exchange mechanism during subsequent hyperthermal
energy deposition could be significantly reduced. Using this modulated incident energy growth
strategy, films with little interfacial roughness and intermixing could be grown over a wide range of
incident angles with or without substrate rotation. © 2000 American Institute of Physics.
关S0021-8979共00兲01301-3兴
I. INTRODUCTION
rate, the depositing atoms’ incident energy, their incident
angle, etc. These conditions are in turn controlled by many
other deposition method dependent processing parameters
共such as chamber pressure, target–substrate distance, bias
voltage, and so on兲. The selection of the best process and
optimization of the process conditions for depositing high
performance GMR multilayers requires a better understanding of the fundamental relationship between a deposited multilayer’s GMR controlling nanostructure and its deposition
conditions.
Molecular dynamics methods have been used to explore
the mechanisms of surface roughening during low temperature film growth35 and investigate the role of incident atom
energy upon surface modification.36 These studies indicated
that surface roughness increased with film thickness, and increases in incident atom energy promoted surface flattening.
Molecular dynamics simulations have also begun to identify
process conditions that are likely to result in good GMR
properties.37 For normal incident angle atom deposition, both
the interfacial roughness and the extent of intermixing at
interfaces were found to depend upon the incident atom energy. High 共e.g., above 5 eV兲 incident atom energies lowered
the atomic scale interfacial roughness but promoted interfacial mixing by an exchange mechanism. Conversely, very
low 共⬃0.1 eV兲 incident atom energies reduced intermixing,
but resulted in films with rough interfaces. The best tradeoff
between roughness and intermixing was found when an incident energy of 1–2 eV was used for deposition. However,
the model indicated that significant further improvement
could be achieved by using a modulated energy deposition
technique in which low 共⬃0.1 eV兲 incident energies were
used to deposit the first few monolayers of a new material
Metal multilayers consisting of pairs of 50–70 Å thick
ferromagnetic layers that sandwich a thin 共10–30 Å兲 copper
layer can exhibit giant magnetoresistance 共GMR兲.1–28 GMR
materials undergo large 共5%–80% or more兲 reductions in
their electrical resistance when a magnetic field is applied
because of changes in the spin dependent electron
scattering.1,3 GMR materials are becoming technologically
important; they have already been used for magnetic field
sensors in the read heads of hard disk drives,29–31 and are
being developed for nonvolatile magnetic random access
memories.31 These applications have spurred intense interest
in the design of economical deposition processes capable of
producing thermally stable GMR materials that exhibit high
GMR ratios 共defined as the maximum resistance change divided by the resistance at magnetic saturation兲 with low saturation magnetic fields.
GMR multilayers can be created by many physical vapor deposition techniques including molecuelar beam epitaxy,11–13,27 various sputter deposition
methods,6–8,14–28,32,33 as well as ion-beam and ion-beam assisted deposition.34 Each process results in multilayers with
widely different GMR behavior presumably resulting from
slightly different atomic scale structures. While much work
remains to fully determine the relationship between GMR
properties and atomic scale structure, it appears that the best
GMR properties are obtained when the roughness of the interfaces and the degree of mixing between the ferromagnetic
and the conductive layers are minimized. These structural
attributes of the films are strongly affected by deposition
conditions such as the substrate temperature, the deposition
0021-8979/2000/87(1)/553/11/$17.00
553
© 2000 American Institute of Physics
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
554
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
X. W. Zhou and H. N. G. Wadley
FIG. 1. Schematic illustration of an ion beam deposition system 共the translation energy of sputtered metal atoms increases with acceleration voltage of
the ion beam gun in the 50–1000 V range兲.
layer and higher energies 共e.g., 5 eV兲 were used for its
completion.
Energy modulation appears possible with most sputtering processes,38 but may be most easily implemented in ion
beam deposition systems. A schematic illustration of an ion
beam deposition system is shown in Fig. 1. An ion beam gun
is used to generate a high energy 共⬃0.1–2 keV兲 Ar⫹ , Kr⫹ ,
or Xe⫹ ion beam which is directed at a target 共e.g., copper or
permalloy兲. Ion bombardment of the target results in atom
sputtering from its surface. These metal atoms then propagate toward a substrate and are deposited on the substrate
surface. Recent molecular dynamics simulations of the sputtering of nickel by argon ions indicated that the average energy of the emitted nickel atoms increases from ⬃4 to ⬃13
eV as the argon ion energy is increased from 50 to 1000
eV.39 Controlling ion gun voltage therefore provides one
method for manipulating the adatom energy during deposition. Collisions with the background gas atoms can also reduce the energy of the sputtered atoms as they are transported from the target to the substrate. The probability of
共energy loss兲 collisions increases with the background pressure P B or the target–substrate distance d TS . The impact
energy of depositing atoms can therefore also be controlled
by selecting an appropriate combination of the ion beam gun
acceleration voltage, background pressure, and target–
substrate distance.
Both geometric constraints within multielement ion
beam deposition systems and gas phase scattering can make
it difficult to achieve normal 共␪⫽0°兲 impact angle deposition.
Experiments40 and recent molecular dynamics simulations of
thin film deposition41–43 have shown that increasing the fraction of deposited atoms with oblique incident angles leads to
a rougher growth surface. Collimation methods can be used
to control the angular dispersion, but it results in lower deposition rates and can be an undesirable source of contamination. In some deposition systems, substrate rotation is used to
reduce roughness and to improve surface thickness uniformity. However, this can also be a significant source of particle contamination. Here, the molecular dynamics model
used earlier to investigate energy dependent deposition37 has
FIG. 2. Atomic structure of Ni/Cu/Ni multilayers as a function of incident
angle ␪ at a fixed incident energy of 1.0 eV, grown with a stationary substrate: 共a兲 ␪⫽0°, 共b兲 ␪⫽20°, 共c兲 ␪⫽35°, 共d兲 ␪⫽40°, 共e兲 ␪⫽50°, and 共f兲
␪⫽55°.
been extended to explore the dependence of interfacial
roughness and intermixing on the flux incident angle and
substrate rotation. To simplify comparisons with previous
work, the growth of a model Ni/Cu/Ni multilayer in the
关111兴 direction has been analyzed.
II. COMPUTATION METHOD
Details of the molecular dynamics model used here have
been described in an earlier paper.37 Briefly, a Cu–Ni alloy
embedded atom method potential44 was used to calculate interaction forces between the atoms. This potential was cut off
at 4.95 Å which is larger than the third nearest neighbor
distance, yielding a realistic description of the fcc crystal. An
identical crystal geometry to that previously employed37 was
used for the calculation. A nickel substrate was orientated so
that its x, y, and z edges were aligned in the 关 112̄兴 , 关111兴, and
关 11̄0 兴 directions, respectively. To achieve reasonable computation efficiency, the size of the substrate crystal was chosen to contain 120 (224̄) planes in the x direction, three 共111兲
planes in the y direction, and 12 共220兲 planes in the z direction 共see Fig. 2兲. The effects of the small crystal size were
minimized by using periodic boundary conditions in the x
and z directions so that the crystal can be viewed as extending infinitely in these two directions. It should be noted that
the z length of the substrate cell was about 14.9 Å. While this
is approximately one fourth of the x length of the cell, it is
well above the cutoff distance of the potential.
By imposing a free boundary condition at the top 共y兲
surface and using a fixed boundary condition for the bottom
共y兲 surface, growth in the y direction could be simulated.37
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
Multilayer deposition was simulated by alternatively depositing about 20 Å of copper followed by about 20 Å of nickel.
The incident atom energy E i as well as its incident angle ␪
共defined as the angle between the incoming direction of adatoms and the surface normal兲, as shown in Figs. 1 and 2,
were simply implemented by assigning an appropriate velocity vector to each atom. To uniquely define a three dimensional incident direction relative to a stationary substrate, ␪
was initially set in the x – y plane. For some of the simulations, the effect of substrate rotation was examined. In such
cases, the azimuth angle ␺ 共see Fig. 1兲, was varied 共at the
prescribed rotation rate兲 during deposition while the substrate remained fixed. It should be pointed out that the use of
a relatively small thickness 共⬃15 Å兲 in the z direction with
the periodic boundary condition would underestimate the
significance of the shadowing and prevent the formation of
long wavelength roughness along this direction. Because we
only analyzed the roughness in the x direction, the
roughness–rotation trends are likely to be unaffected by this
approximation. As in previous work, all simulations were
conducted at a fixed substrate temperature of 300 K and a
fixed deposition rate of 10 nm/ns. The use of this high deposition rate was necessary in order to deposit an appropriately
thick sample within the computational time constraint of molecular dynamics.41 Under a high rate deposition condition,
thermal diffusion events on the microsecond time scale are
not captured in the simulation. The simulation results are
therefore strictly valid for low temperature deposition processes where thermally activated diffusion is negligible.
To quantify the dependence of the interfacial roughness
upon incident angle, a roughness parameter needs to be defined. In experiments, the surface roughness parameter r 2 is
usually defined as the average deviation of the surface profile
from the mean surface height over a selected area or length
scale of micrometer.32 However, problems arose when using
the average deviation of surface atoms from the mean surface height to calculate r 2 for a given atomic configuration.
For instance, different sections of a nonplanar surface are
usually associated with different crystallographic planes of
different atomic densities. Hence, the deviation from the
mean surface height averaged over all surface atoms essentially measures different sections of the surface with different
weighing factors. Because the mean height lies somewhere
between the highest and lowest point on the surface profile,
r 2 may also incorrectly specify a higher roughness for a surface containing one high terrace than for a 共rougher兲 surface
composed of many asperities 共as long as their heights are
within the height difference of the high and low terraces兲. It
has been shown that the surface roughness of atomic configurations is better represented by a roughness parameter r 1
defined as the average aspect ratio of surface asperities.37 For
surfaces with a sinoidal roughness, the r 1 parameter can be
scaled to an equivalent value of r 2 if the average width of
asperities is known.37
III. INCIDENT ANGLE EFFECTS
A. Atomic configurations
To investigate the effect of incident atom direction upon
atomic scale structure, the deposition of Ni/Cu/Ni multilay-
X. W. Zhou and H. N. G. Wadley
555
FIG. 3. Interfacial roughness as a function of incident angle ␪ at a fixed
incident energy of 1.0 eV. 共a兲 copper on nickel interface and 共b兲 nickel on
copper interface.
ers was simulated using incident atom angles from 0° to 60°
and a fixed incident atom energy of 1.0 eV. Representative
examples of the atomic configurations for selected incident
angles are shown in Fig. 2, where dark spheres represent
nickel atoms while the lighter ones represent copper atoms.
It is apparent that the roughness of the interfaces and intermixing in these multilayers was sensitive to the deposition
angle. Oblique angle deposition produced high interfacial
roughness 共especially at the copper on nickel interface兲.
When the incident angle was increased to about 40° and
beyond, voids also formed during the deposition of nickel
layers. The size of these voids increased with incident angle,
and extended through the entire nickel layer thickness and
into the copper layer for the highest incident angles. Figure 2
also indicates that mixing of the two materials at their interfaces increased with incident angle. As in our earlier study,37
copper atoms tended to mix in the nickel layers more frequently than vice versa.
B. Interfacial roughness
Roughness parameters were calculated for each incident
angle of deposition and the results are shown for copper on
nickel and nickel on copper interfaces in Fig. 3. Results for a
rotated substrate are also shown and will be discussed later.
It can be seen that as the incident angle was increased, the
roughness of both copper on nickel and nickel on copper
interfaces increased. Both the absolute value of the roughness and its dependence upon incident angle were higher for
the copper on nickel interface than vice versa. Because the
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
556
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
X. W. Zhou and H. N. G. Wadley
copper on nickel interface was always the last interface
formed in the system shown in Fig. 3, it could be conjectured
that its higher roughness was caused by an increase of surface roughness with increasing thickness.35 To clarify this,
additional simulations were conducted by first depositing a
nickel layer on the copper substrate followed by a copper
layer and a second nickel layer. Again, the copper on the
nickel interface was rougher. Also, the growth of the Ni/
Cu/Ni multilayers shown in Fig. 3 was continued to form a
second nickel on copper interface. The results revealed that
while the second nickel on copper interface was rougher than
the first one 共due to an increase of roughness with thickness兲,
it was still smoother than the copper on nickel interface. The
higher roughness of the copper on nickel interface was therefore a material effect and did not depend upon the order of
the deposition sequence. Figure 3 also indicates that the
roughness of the copper on nickel interface was particularly
sensitive to the incident angle when ␪⬎40°. By comparing
with Fig. 2, it can be seen that this coincides with the beginning of the formation of voids in the nickel layer.
C. Roughening mechanisms
Molecular dynamics simulations of deposition at normal
incidence indicated that adatoms with low incident energies
frequently attached to the top and the sidewalls of surface
asperities.37 When all atoms have the same incident angle,
shadowing can then exclude the incoming atoms from parts
of the valleys between the asperities, resulting in an enhancement of the surface roughness.37 This phenomenon is an
usual explanation for the high surface roughness and void
content during oblique angle deposition of pure metal films,
and has been analyzed for two dimensional growth.43
Complexities are introduced when growth occurs in
three dimensions and/or different materials 共with different
mobilities兲 are sequentially deposited. To investigate the
evolution of interfacial roughness and void morphology during oblique angle deposition, additional simulations were
conducted using larger crystals to allow detailed observations of the three dimensional growth process. Figure 4
shows representative stages of the growth of a multilayer
deposited with 1.0 eV adatoms incident at an angle of 55°.
To simplify visualization of the internal voids, parts of the
crystal have been removed after the simulation.
During the initial stage of deposition, atoms were randomly deposited on the surface and formed small surface
asperities. Under collimated oblique angle condition, these
asperities blocked the incoming flux of adatoms, resulting in
an unreachable shadowed region behind each asperity. While
further deposition caused the unshadowed regions to continuously grow, growth in shadowed region was slower because it required the surface migration of atoms from neighboring regions. Because a high deposition rate was used for
the simulations, the vertical deposition rate exceeded the surface 共diffusion兲 transport rate. The shadowed region then became large valleys such as the region with a profile R x in
Fig. 4共b兲. The valley width along the x direction 共in which
shadowing occurred兲 scaled with (h 1 – h 2 ) tan(␪), where h 1
and h 2 are the heights of the two walls bounding the shad-
FIG. 4. Three dimensional illustration of void formation during oblique
angle deposition at an incident angle of 55° and an incident energy of 1.0
eV: 共a兲 t⫽0 ps, 共b兲 t⫽350 ps, 共c兲 t⫽450 ps, and 共d兲 t⫽550 ps.
owed region in the x direction. Because the shadowed region
was initially nucleated on a flat surface, h 1 – h 2 was small
and hence, the width of the valley in the x direction was also
small 共see the surface profile R x 兲. Extension of the valley in
the z direction was seen to accompany the growth in its depth
when the flux incident angle ␪ 共in the x – y plane兲 was large.
This was caused by the lateral growth 共in the z direction兲 of
the shadowing asperities due to the long range atomic attractions between surface atoms and nearby vapor atoms. The
surface valley then became a local trench whose depth increased as the deposition time increased. At time
t⫽450 ps, Fig. 4共c兲, continued deposition of atoms on the
trench lip caused a bulge at the trench opening. Atoms eventually became sufficiently close that the attractive forces between surface and vapor atoms caused a pinch-off of the
trench, resulting in the trapping of a void in the film at
t⫽550 ps, Fig. 4共d兲. Figure 4共d兲 shows that the void formed
was extended in the z direction, but was relatively narrow in
the x direction. Although surface roughness developed in the
z direction due to the randomness of atom arrivals, no deep
valleys were observed in this direction 关see the surface profile R z in Fig. 4共b兲兴.
Increasing the incident angle increased the significance
of shadowing and reduced the probability that surface
trenches would be filled. Consequently, surface roughness
and void size increased as the incident angle was increased.
Since valleys can be filled by the migration of surface atoms,
surface roughness and void content can be reduced by thermally activated surface diffusion when temperature is high
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
FIG. 5. Intermixing as a function of incident angle ␪ at a fixed incident
energy of 1.0 eV: 共a兲 copper mixing in nickel and 共b兲 nickel mixing in
copper.
or deposition rate is low. If the energy of the depositing
atoms is high, the shadowed regions can also be filled by
impact induced diffusion mechanisms.45,46 This is the dominant mechanism of surface transport active in the simulations
reported here. Because the extent of surface migration depends on atom mobility and the mobility of copper atoms is
greater than that of nickel atoms,47 the roughness on the
nickel surface is less likely to be flattened. As a result, both
roughness and the incident angle dependence of the roughness were higher for the copper on nickel interface than for
the nickel on copper interface, and voids preferentially developed in the nickel layer.
D. Interfacial mixing
Figure 2 indicates significant mixing of copper and
nickel within about 3–4 monolayers 共ML兲 of the interfaces.
It is also apparent that copper was mixed in the nickel layer
much more frequently than nickel was mixed in the copper
layer. The degree of mixing can be quantified by a probability p of finding a copper 共nickel兲 atom surrounded by at least
8 nickel 共copper兲 atoms.37 The parameter p as a function of
incident angle was determined for the mixing of copper in
nickel and nickel in copper, and is shown in Fig. 5. The
results clearly indicate that copper was dissolved in nickel to
a much greater extent than nickel in copper. This asymmetry
in mixing was retained over the entire incident angle range
X. W. Zhou and H. N. G. Wadley
557
FIG. 6. Probability of impact exchange between an adatom and a surface
atom as a function of incident angle ␪: 共a兲 11.0 eV nickel adatom impacting
on flat copper surface and 共b兲 24.0 eV copper adatom impacting on flat
nickel surface
studied. The degree of mixing was seen to remain roughly
constant up to an incident angle of about 40°, and then began
to increase.
E. Intermixing mechanisms
In earlier work for normal incident angle deposition at
low temperature and high rate,37 mixing was found to be
predominantly accomplished by an adatom impact exchange
mechanism. Energetic impacting atoms were able to penetrate into a surface by displacing surface atoms. The ensuing displacement cascade often resulted in an atom within
the surface being ejected onto the top of the surface. The
probability of exchange between an impacting adatom and
an atom in the surface was deduced for this normal incident
angle process and found to depend upon the impact energy
and the surface perfection at the impact site.37 For an 11 eV
incident energy, about 50% of impacting nickel atoms exchanged with a copper atom in a perfect 共flat兲 copper surface.
Because the binding between nickel atoms is stronger, impacting copper atoms must have higher energies 共⬃24 eV兲 to
achieve a 50% probability of exchange with a nickel atom in
a flat nickel surface. When impacts occurred at defective
regions of the surface 共e.g., at a surface ledge兲, the energies
required for the exchange were reduced significantly. This
mechanism reconciled the observation that copper dissolution in the nickel layer was much more significant than vice
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
558
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
X. W. Zhou and H. N. G. Wadley
FIG. 7. Time evolution of atomic structure of a copper surface during an 11.0 eV nickel adatom impact at a low incident angle of 10°: 共a兲 t⫽0.00 ps, 共b兲
t⫽0.15 ps, 共c兲 t⫽0.30 ps, and 共d兲 t⫽0.50 ps.
versa, and explained why mixing was enhanced when very
low energy deposition conditions 共which resulted in defective rough surfaces兲 were used.
To more fully understand the incident angle effect on
interfacial mixing, the probability of exchange for an 11.0
eV nickel atom impacting on a flat copper surface and a 24.0
eV copper atom impacting on a flat nickel surface was deduced as a function of incident angle ␪ between 0° and 60°.
Results for nickel impacts with a copper surface and copper
impacts with a nickel surface are shown in Figs. 6共a兲 and
6共b兲, respectively. The results in Fig. 6 indicate that the exchange probability was high and more or less independent of
␪ for small angles of incidence, and rapidly diminished at
higher 共␪⬎20°兲 incident angles.
To investigate the exchange mechanism at off-normal
incident angle, stop-action picture series were used to reveal
the detailed atomic process during an 11.0 eV nickel atom
impact with a flat copper surface. One example of the impact
process for ␪⫽10° is shown in Fig. 7, where the black ball
represents the impacting nickel atom, and the blue, yellow,
turquoise, and red balls are used to show the four copper
surface atoms that underwent the largest displacement during
the impact. For clarity, only atoms 共light balls兲 of three
atomic planes are displayed in Fig. 7. It can be seen from
Fig. 7 that the impacting atom penetrated into the surface. It
then displaced both the yellow and turquoise atoms into the
crystal and squeezed both the blue and red atoms out of the
surface, Fig. 7共b兲. Figure 7共c兲 shows that an impacting atom
with a momentum component in the x direction enabled the
yellow atom to acquire a significant momentum component
in a direction close to a lattice jump path at 0.30 ps. The
yellow atom then jumped to take the original site of the blue
atom while the nickel impacting atom took the original site
of the yellow atom at 0.50 ps, Fig. 7共d兲. This resulted in the
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
FIG. 8. Interfacial roughness as a function of incident energy E i at a fixed
incident angle of 50°: 共a兲 copper on nickel interface and 共b兲 nickel on copper
interface.
ejection of the blue atom upwards onto the top of the surface.
During this process, the turquoise and the red atoms relaxed
to their original surface site.
Similar stop-action picture series analysis for the impact
of a flat copper surface by an 11.0 eV nickel atom at larger
incident angles 共e.g., ␪⫽60°兲 indicated that the nickel impacting atom was less likely to penetrate the surface. Instead,
it usually skipped on the surface, resulting in biased
diffusion.45 Clearly, exchange was significant at low incident
angles, and the peak exchange rate occurred when the incident direction is slightly off the surface normal which provided the lateral momentum for a lattice jump. The rapid
decline in exchange probability at high incident angle was
the result of a transition from adatom exchange to adatom
skipping in the direction of incidence.45
The impacts analyzed above at first sight suggest that the
intermixing should decrease with incident angle. However,
the opposite trend was observed: Fig. 5 shows that intermixing increased with incident angle. This occurred because at
large incident angles, the surface roughness was very high.
While the nominal incident angle 共defined as the angle between the incident direction and the average normal of a
macroscopic scale surface兲 was large, the average local incident angle had decreased because a large fraction of adatoms
now impacted the sides of valleys on the rough surface. In
addition, the exchange probability increased significantly
with surface roughness which, at low incident energy, increased with ␪. The net effect was an intermixing that increased with incident angle.
X. W. Zhou and H. N. G. Wadley
559
FIG. 9. Intermixing as a function of incident energy E i at a fixed incident
angle of 50°: 共a兲 copper mixing in nickel and 共b兲 nickel mixing in copper.
IV. INCIDENT ENERGY DEPENDENCE
The simulations above were conducted with an incident
energy of 1.0 eV and the results suggested that both interfacial roughness and intermixing were minimized with a near
normal incident angle deposition. This significantly constrains the design of a vapor deposition technology. An earlier study37 indicates that high adatom incident energies result in smoother surfaces. To evaluate its significance for
oblique deposition, the interfacial roughness and the mixing
probability as a function of incident energy for ␪⫽50° were
calculated and the results are shown in Figs. 8 and 9. Figure
8 indicates that the roughness initially decreased with incident energy for both interfaces. This trend continued for the
copper on nickel interface up to at least 6 eV as seen in Fig.
8共a兲. The roughness of the nickel on copper interface, Fig.
8共b兲, exhibited a minimum at an incident energy around 3 eV
and then increased with incident energy. The apparent increase in roughness of the nickel on copper interface for
incident energy above 3.0 eV was an anomalous consequence of the mixing of copper in the nickel layer which
created an increasingly diffuse interface as the energy increased. A detailed examination of the simulations indicated
that the initial decrease in roughness of both interfaces resulted from impact energy induced surface diffusion which
promoted step flow growth and the transport of atoms into
the shadowed regions.37 Because copper atoms are more mobile than nickel atoms, the copper surface was flatter than the
nickel surface and the nickel on copper interface was therefore smoother. Figure 9 summarizes the interfacial mixing
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
560
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
X. W. Zhou and H. N. G. Wadley
FIG. 11. Interfacial roughness as a function of substrate rotation rate at a
fixed incident energy of 1.0 eV and a fixed incident angle of 50°.
FIG. 10. Atomic structure of Ni/Cu/Ni multilayers as a function of substrate
rotation rate at a fixed incident energy of 1.0 eV and a fixed incident angle
of 50°: 共a兲 0.001 turns/ML, 共b兲 0.003 turns/ML, 共c兲 0.009 turns/ML, and 共d兲
0.196 turns/ML.
observations. It can be seen that both copper mixing in nickel
and nickel mixing in copper increased continuously with incident energy over the entire incident energy range studied.
This resulted from an increased exchange probability at
higher incident energy and is consistent with the results obtained for the normal incident angle deposition.37
These results indicate that while high energy deposition
improves interfacial roughness, it also results in significant
interlayer mixing. Conversely, low energy deposition minimizes intermixing, but it results in rough interfaces. Clearly,
the use of a high energy alone during high incident angle
deposition is unlikely to produce high quality GMR
multilayer structures.
V. SUBSTRATE ROTATION
High interfacial roughness and intermixing during oblique angle deposition have been correlated with shadowing
effects. Changing the direction of the incoming flux with
respect to the crystal by substrate rotation during growth
should significantly reduce shadowing and hence offers the
promise of improved interfacial roughness for oblique angle
deposition at a sufficiently low energy to induce no intermixing.
A. Rotation rate behavior
To explore rotation effects, a series of molecular dynamics calculations were conducted at a fixed incident atom
angle of 50° and an incident energy of 1.0 eV using simulated substrate rotation rates that varied from 0.001 turns/ML
to 0.200 turns/ML. Selected atomic configurations are shown
in Fig. 10. It can be seen that for low energy oblique deposition, increasing the substrate rotation rate reduced the interfacial roughness and eliminated the formation of voids in
the nickel layer. The dependence of interfacial roughness on
substrate rotation rate is summarized in Fig. 11. The interfa-
cial roughness is seen to initially decrease rapidly as the
substrate rotation rate was increased from 0 to about 0.01
turns/ML. It then became insensitive to the rotation rate beyond about 0.01 turns/ML.
Substrate rotation was not found to improve intermixing,
共Figs. 5 and 9兲. The beneficial effect of substrate rotation
upon roughness is especially significant for high incident
angle and low energy deposition conditions. Figure 3 shows
the improvement in interfacial roughness by a substrate rotation rate of 0.157 turns/ML for various incident angles at a
low incident energy of 1.0 eV. The combined effect of substrate rotation and incident energy upon roughness can be
seen in Fig. 8. It shows that the beneficial effects of substrate
rotation became progressively smaller as the incident energy
was increased. It also shows that the benefits for a nickel on
copper interface were significantly smaller than those for a
copper on nickel interface. These occurred because the probability of forming surface asperities during stationary deposition was reduced by either a high incident energy or deposition of high mobility copper. Rotation was therefore of
most benefit when other processes for diminishing roughness
were the least effective.
VI. DISCUSSION
The results above indicate that interfacial roughness and
intermixing of Ni/Cu/Ni multilayers increased with the incident angle of adatoms during deposition. Interfacial roughness was inherited from the surface roughness that formed as
a result of shadowing at oblique angles of incidence. Because the probability for impact induced exchange was
higher on a rougher surface, the intermixing was correlated
with the surface roughness and increased with incident angle.
The surface roughness and the extent of intermixing also
depended on the activation energies for surface diffusion of
each element deposited. For copper, which has lower activation energies for surface diffusion than nickel,47 the surface
shadowed regions were readily filled through impact induced
atom diffusion. As a result, copper surfaces 共and thus nickel
on copper interfaces兲 were relatively smooth compared to
nickel surfaces 共and therefore copper on nickel interfaces兲.
For similar reasons, the formation of voids occurred first in
the nickel layer as the incident angle was increased to above
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
about 40°. It was also apparent that copper mixing into the
nickel layer was much more significant than vice versa. This
arose because the cohesive energy of copper atoms is lower
than that of nickel atoms. As a result, a nickel impacting
atom is more likely to penetrate a copper surface and eject a
copper atom onto the top of the copper surface than the reverse case where a copper atom impacts a nickel surface.
Increasing the incident atom energies was a successful
strategy for flattening the interfaces grown under oblique
angle conditions. However, it also caused significant additional intermixing by an atomic exchange mechanism. To
reduce this mechanism of intermixing, incident energies
must be kept below about 1.0 eV, but this results in rough
interfaces especially at oblique angles of incidence. No
single energy deposition exists that results in the deposition
of the high quality interfaces when the incident angle of the
flux is high. Substrate rotation during low energy deposition
was found to reduce shadowing effects and resulted in flat
interfaces. Voids in the nickel layer were also eliminated
when the substrate rotation rate exceeded about 0.01
turns/ML 共3.6°/ML兲. However, rotation did not significantly
reduce intermixing for oblique angle depositions, 共Figs. 5
and 9兲. While the combination of interfacial roughness and
intermixing obtained using substrate rotation during low energy oblique angle depositions was improved, the result was
still far from perfection 共see, for example, Fig. 10兲.
One other growth strategy can be explored. In earlier
work37 it was shown that a modulated energy deposition
method significantly reduced interfacial roughness without
causing intermixing for a normal incident angle deposition.
In this approach, the first few monolayers of a new element
were deposited at a low incident energy and the remainder of
that material was then deposited at a high incident energy.
The low initial energy prevented intermixing while the high
energy resulted in surface flattening. To investigate the potential benefit of energy modulation during oblique angle
FIG. 12. Atomic structure of Ni/Cu/Ni multilayers as a function of E h for
modulated energy deposition at a fixed E 1 of 0.1 eV and a fixed incident
angle of 50°, grown with a stationary substrate: 共a兲 E h ⫽2.0 eV, 共b兲 E h
⫽3.0 eV, 共c兲 E h ⫽4.0 eV, and 共d兲 E h ⫽5.0 eV.
X. W. Zhou and H. N. G. Wadley
561
FIG. 13. Atomic structure of Ni/Cu/Ni multilayers as a function of E h for
modulated energy deposition at a fixed E 1 of 0.1 eV and a fixed incident
angle of 50°, grown with substrate rotation at 0.157 turns ML: 共a兲 E h
⫽2.0 eV, 共b兲 E h ⫽3.0 eV, 共c兲 E h ⫽4.0 eV, and 共d兲 E h ⫽5.0 eV.
deposition, simulations were performed using a low energy,
E 1 共of 0.1 eV兲, for the first ⬃5 ML of each element deposited, and various high energies E h 共between 2.0 and 5.0 eV兲
for the remainder of that element. Atomic configurations as a
function of E h are shown in Figs. 12 and 13 for stationary
and rotating substrates, respectively, using a fixed incident
FIG. 14. Interfacial roughness as a function of E h for modulated energy
deposition at a fixed E 1 of 0.1 eV and a fixed incident angle of 50°: 共a兲
copper on nickel interface and 共b兲 nickel on copper interface.
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
562
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
FIG. 15. Intermixing as a function of E h for modulated energy deposition at
a fixed E 1 of 0.1 eV and a fixed incident angle of 50°: 共a兲 copper mixing in
nickel and 共b兲 nickel mixing in copper.
angle of 50° and a rotation rate of 0.157 turns/ML. This
strategy clearly improved both interfacial roughness and intermixing. The functional dependence of interfacial roughness and intermixing upon E h were calculated and are plotted in Figs. 14 and 15. It can be seen that increasing the
value of E h significantly improved both interfacial roughness
and intermixing. Furthermore, the beneficial effects of substrate rotation were most significant at relatively low E h values. Above an E h of about 3.0 eV, rotation had little benefit
for roughness or intermixing. These observations are a result
of the effectiveness of the modulated energy strategy at
eliminating surface roughness even at high angles of incidence.
To determine the incident angle range over which the
beneficial effects of modulated energy deposition can be
achieved, additional simulations were conducted using incident angles between 0° and 60°, an E 1 of 0.1 eV, and an E h
of 5.0 eV, for both stationary and rotating substrates. The
results are shown in Figs. 16 and 17. They indicate that very
low levels of both interfacial roughness and intermixing
were present over the entire incident angle range explored.
They also show that no beneficial effect of substrate rotation
was detectable.
These results provide significant guidance for the design
and optimization of a growth approach for GMR multilayers.
Today, two competing deposition techniques are used for
synthesizing the high performance GMR multilayers: rf magnetron 共or diode兲 sputtering and ion beam deposition. In rf
sputtering systems, sputtered atoms must be transported
X. W. Zhou and H. N. G. Wadley
FIG. 16. Interfacial roughness as a function of incident angle ␪ for modulated energy deposition with E 1 ⫽0.1 eV and E h ⫽5.0 eV: 共a兲 copper on
nickel interface and 共b兲 nickel on copper interface.
through a plasma to reach the substrate. Implementation of
the modulated energy deposition strategy could be achievable by manipulation of the rf power.38 Modulated energy
deposition using ion beam deposition can be implemented by
manipulation of ion beam gun voltage. Since the results
above indicate that high quality interfaces can be achieved
over wide range of impact angles, the energy modulation
strategy eliminates the need for substrate rotation 共unless it is
needed to improve thickness uniformity兲 and flux collimation, which are both troublesome sources of particle contamination. It also extends the incident angle range 共and thus the
design flexibility兲 of a multielement deposition system where
high quality interfaces can be created.
VII. CONCLUSIONS
Molecular dynamics simulations have been used to explore incident atom angle and energy effects during the
growth of 共111兲 Ni/Cu/Ni multilayers from the vapor phase.
The study has revealed that:
共1兲For low incident energies of 1 eV or less, shadowing
during high incident angle deposition led to a significant increase in the interfacial roughness, and to the formation of
voids across the entire thickness of nickel layers.
共2兲 Significant intermixing occurred at nickel on copper
interfaces. It was correlated with the interfacial roughness
and increased with incident atom angle.
共3兲 Substrate rotation reduced the interfacial roughness
obtained during low energy, high incident angle deposition,
but had little effect upon intermixing.
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
J. Appl. Phys., Vol. 87, No. 1, 1 January 2000
X. W. Zhou and H. N. G. Wadley
563
5
FIG. 17. Intermixing as a function of incident angle ␪ for modulated energy
deposition with E 1 ⫽0.1 eV and E h ⫽5.0 eV: 共a兲 copper mixing in nickel
and 共b兲 nickel mixing in copper.
共4兲 High quality interfaces with minimal interfacial
roughness and intermixing were achievable using an energy
modulation strategy during multilayer growth. When a low
incident energy of 0.1 eV was used for the growth of the first
⬃5 ML of each new element and a higher energy 共above 4.0
eV兲 was used for the remainder of that element, high quality
interfaces could be deposited with or without substrate rotation over the entire 共0°–60°兲 deposition angle range explored.
共5兲 The implementation of a modulated energy deposition strategy during either sputtering or ion beam deposition
would allow the use of uncollimated fluxes from targets positioned even at acute angles with respect to the substrate,
providing significant deposition reactor design flexibility.
ACKNOWLEDGMENTS
The authors are grateful to the Advanced Research
Projects Agency 共A. Tsao, Program Manager兲 and the National Aeronautics and Space Administration for support of
this work through NASA Grants Nos. NAGW 1692 and
NAG-1-1964.
P. M. Levy, J. Magn. Magn. Mater. 140–144, 485 共1995兲.
M. D. Stiles, Phys. Rev. B 48, 7238 共1993兲.
3
W. H. Butler, X. G. Zhang, D. M. C. Nicholson, and J. M. MacLaren,
Phys. Rev. B 52, 13399 共1995兲.
4
M. N. Baibich, J. M. Broto, A. Fert, F. N. V. Dau, F. Petroff, P. Eitenne,
G. Creuzet, A. Friederich, and J. Chazelas, Phys. Rev. Lett. 61, 2472
共1988兲.
1
2
G. Binasch, P. Grunberg, F. Saurenbach, and W. Zinn, Phys. Rev. B 39,
4828 共1989兲.
6
E. E. Fullerton, D. M. Kelly, J. Guimpel, and I. K. Schuller, Phys. Rev.
Lett. 68, 859 共1992兲.
7
M. L. Yan, W. Y. Lai, Y. Z. Wang, S. X. Li, and C. T. Yu, J. Appl. Phys.
77, 1816 共1995兲.
8
T. L. Hylton, K. R. Coffey, M. A. Parker, and J. K. Howard, J. Appl.
Phys. 75, 7058 共1994兲.
9
S. Z. Hua, D. S. Lashmore, L. Salamanca-Riba, W. Schwarzacher, L. J.
Swartzenruber, R. D. McMichael, L. H. Bennett, and R. Hart, J. Appl.
Phys. 76, 6519 共1994兲.
10
Z. J. Yang and M. R. Scheinfein, IEEE Trans. Magn. 31, 3921 共1995兲.
11
R. Clarke, D. Barlett, F. Tsui, B. Chen, and C. Uher, J. Appl. Phys. 75,
6174 共1994兲.
12
K. P. Wellock, B. J. Hickey, J. Xu, M. J. Walker, and N. Wiser, J. Magn.
Magn. Mater. 140–144, 585 共1995兲.
13
W. F. Egelhoff, Jr. and M. T. Kief, IEEE Trans. Magn. 28, 2742 共1992兲.
14
S. Honda, S. Ohmoto, R. Imada, and M. Nawate, J. Magn. Magn. Mater.
126, 419 共1993兲.
15
R. J. Pollard, M. J. Wilson, and P. J. Grundy, J. Magn. Magn. Mater. 146,
L1 共1995兲.
16
H. Zhang, R. W. Cochrane, Y. Huai, M. Mao, X. Bian, and W. B. Muir, J.
Appl. Phys. 75, 6534 共1994兲.
17
T. R. McGuire, J. M. Harper, C. Cabral, Jr., and T. S. Plaskett, J. Appl.
Phys. 76, 6601 共1994兲.
18
J. D. Kim, A. K. Petford-Long, J. P. Jakubovics, J. E. Evetts, and R.
Somekh, J. Appl. Phys. 76, 6513 共1994兲.
19
D. H. Mosca, F. Petroff, A. Fert, P. A. Schroeder, W. P. Pratt, Jr., and R.
Laloee, J. Magn. Magn. Mater. 94, L1 共1991兲.
20
S. S. P. Parkin, Z. G. Li, and D. J. Smith, Appl. Phys. Lett. 58, 2710
共1991兲.
21
H. Sato, Y. Kobayashi, Y. Aoki, R. Loloee, and W. P. Pratt, Jr., J. Magn.
Magn. Mater. 140–144, 567 共1995兲.
22
R. J. Highmore, W. C. Shih, R. E. Somekh, and J. E. Evetts, J. Magn.
Magn. Mater. 116, 249 共1992兲.
23
K. Kagawa, H. Kano, A. Okabe, A. Suzuki, and K. Hayashi, J. Appl.
Phys. 75, 6540 共1994兲.
24
K. Meguro, S. Hirano, M. Jimbo, S. Tsunashima, and S. Uchiyama, J.
Magn. Magn. Mater. 140–144, 601 共1995兲.
25
H. Sakakima, M. Satomi, K. Onaka, and S. Yamamoto, J. Magn. Magn.
Mater. 145, L268 共1995兲.
26
S. Hossain, D. Seale, G. Qiu, J. Jarratt, J. A. Barnard, H. Fujiwara, and M.
R. Parker, J. Appl. Phys. 75, 7067 共1994兲.
27
K. Noguchi, S. Araki, T. Chou, D. Miyauchi, Y. Honda, A. Kamijima, O.
Shinoura, and Y. Narumiya, J. Appl. Phys. 75, 6379 共1994兲.
28
X. Bian, J. O. Strom-Olsen, Z. Altounian, Y. Huai, and R. W. Cochrane,
J. Appl. Phys. 75, 7064 共1994兲.
29
J. Daughton, J. Brown, E. Chen, R. Beech, A. Pohm, and W. Kude, IEEE
Trans. Magn. 30, 4608 共1994兲.
30
C. Tsang, R. E. Fontana, T. Lin, D. E. Heim, V. S. Speriosu, B. A.
Gurney, and M. L. Williams, IEEE Trans. Magn. 30, 3801 共1994兲.
31
J. L. Simonds, Phys. Today 26 共April 1995兲.
32
J. C. S. Kools, J. Appl. Phys. 77, 2993 共1995兲.
33
T. C. Anthony, J. A. Brug, and S. Zhang, IEEE Trans. Magn. 30, 3819
共1994兲.
34
B. Vavra 共private communication, 1997兲.
35
C. L. Kelchner and A. E. DePristo, Surf. Sci. 393, 72 共1997兲.
36
C. M. Gilmore and J. A. Sprague, Phys. Rev. B 44, 8950 共1991兲.
37
X. W. Zhou and H. N. G. Wadley, J. Appl. Phys. 84, 2301 共1998兲.
38
H. N. G. Wadley et al., Mater. Res. Soc. Symp. Proc. 538, 323 共1999兲.
39
X. W. Zhou and H. N. G. Wadley 共unpublished兲.
40
J. A. Thornton, J. Vac. Sci. Technol. 12, 830 共1975兲.
41
X. W. Zhou, R. A. Johnson, and H. N. G. Wadley, Acta Mater. 45, 1513
共1997兲.
42
R. W. Smith and D. J. Srolovitz, J. Appl. Phys. 79, 1448 共1996兲.
43
L. Dong, R. W. Smith, and D. J. Srolovitz, J. Appl. Phys. 80, 5682 共1996兲.
44
S. M. Foiles, Phys. Rev. B 32, 7685 共1985兲.
45
X. W. Zhou and H. N. G. Wadley, Surf. Sci. 431, 42 共1999兲.
46
X. W. Zhou and H. N. G. Wadley, Surf. Sci. 431, 58 共1999兲.
47
C. L. Liu, J. M. Cohen, J. B. Adams, and A. F. Voter, Surf. Sci. 253, 334
共1991兲.
Downloaded 04 Apr 2005 to 128.143.35.28. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp