Materials Science Forum Vols. 539-543 (2007) pp. 3528-3533
online at http://www.scientific.net
© (2007) Trans Tech Publications, Switzerland
Atomic Assembly of Thin Film Materials
X. W. Zhou1,a, D. A. Murdick1,b, B. Gillespie1,c, J. J. Quan1,d, H. N. G.
Wadley1,e, R. Drautz2,f, and D. Pettifor2,g
1
Department of Materials Science and Engineering, 116 Engineer’s Way, Charlottesville, VA
22904-4745, USA
2
Department of Materials, University of Oxford, Parks Road, Oxford OX1 3PH, UK
a
b
c
d
[email protected], [email protected], [email protected], [email protected],
f
g
[email protected], [email protected], [email protected]
e
Keywords: Molecular dynamics, vapor deposition, interatomic potentials.
The atomic-scale structures and properties of thin films are critically determined by the
various kinetic processes activated during their atomic assembly. Molecular dynamics simulations
of growth allow these kinetic processes to be realistically addressed at a timescale that is difficult to
reach using ab initio calculations. The newest approaches have begun to enable the growth
simulation to be applied for a wide range of materials. Embedded atom method potentials can be
successfully used to simulate the growth of closely packed metal multilayers. Modified charge
transfer ionic + embedded atom method potentials are transferable between metallic and ionic
materials and have been used to simulate the growth of metal oxides on metals. New analytical bond
order potentials are now enabling significantly improved molecular dynamics simulations of
semiconductor growth. Selected simulations are used to demonstrate the insights that can be gained
about growth processes at surfaces.
Abstract.
Introduction
Many electronic and optical devices are constructed using nanoscale multilayers composed of
metals, metal oxides, or semiconductors. For instance, metal multilayers that use a pair of thin (~50
Å) ferromagnetic metal layers sandwiching a thin (~20 Å) conductive metal layer can exhibit giant
magnetoresistive (GMR) properties [1]. They are used for high performance readhead sensors for
hard disk drives [1]. Magnetic tunnel junction (MTJ) multilayers that use a pair of thin (~50 Å)
ferromagnetic metal layers sandwiching a thin (~20 Å) oxide layer exhibit tunneling
magnetoresistive (TMR) properties [2,3]. They can be used to build magnetic random access
memory (MRAM) [2,3]. Semiconductor thin films are widely used in cellular phones, global
positioning systems, light-emitting diodes, lasers, infrared detectors, and solar cells [4,5,6,7].
The performance of all these various devices critically depends on the atomic-scale structures of
the thin films. The conductive layer in the GMR multilayers and the oxide tunnel barrier layer in the
TMR multilayers must be continuous, have uniform thickness, and form sharp unmixed interfaces
with the adjacent ferromagnetic layers. It is a challenge to grow these layers, especially when their
thickness is required to be very thin. Semiconductor films must have a high crystallinity and a low
defect concentration. This can be difficult to achieve when a low growth temperature is utilized to
increase dopant concentrations. Because of these difficulties, optimization of the growth using
experimental approaches alone can be a prolonged process. Molecular dynamics (MD) simulations
of vapor deposition track the evolution of film structures by solving for the positions of deposited
and substrate atoms. They use Newton’s equations of motion and interatomic forces and can provide
accurate insights to help improve the growth processes.
There are two challenges to the MD approach. First, the material systems of interest involve
metallic, ionic, and covalent bonding. Each requires a different type of high fidelity interatomic
potential to calculate the corresponding interatomic forces. Secondly, growth occurs by random
All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the
written permission of the publisher: Trans Tech Publications Ltd, Switzerland, www.ttp.net. (ID: 128.143.168.167-08/12/06,17:15:48)
3529
THERMEC 2006
condensation of atoms on the surface under kinetically constrained conditions. Complex surfaces
and lattice defects are therefore encountered. Accordingly, the interatomic potential must accurately
predict properties of all the atomic configurations that can possibly form on or below a surface. The
newest MD approaches have begun to address these challenges, enabling the growth simulation to
be applied for a wide range of materials. Here we illustrate three developments by considering
growth simulations of the metallic multilayers, metal/metal oxide multilayers, and covalent
semiconductor thin films.
Growth of Giant Magnetoresistive Multilayers
Interatomic Potential. GMR multilayers are constructed only from metals. The embedded atom
method (EAM) potential initially developed by Daw and Baskes [8] has provided a sufficiently
good potential format for the metals that are used in the GMR multilayers. The EAM potentials for
16 metals and their alloys have been parameterized in our EAM potential database [9,10]. This
potential database was used in the growth simulation of the Co90Fe10/Cu /Co90Fe10 multilayer [10].
Results. The MD simulation approach [10,11,12] was used to grow a 5.8 nm Co90Fe10/1.5 nm
Cu/4.0 nm Co90Fe10 multilayer unit on a 1 nm Ni80Fe20 substrate at 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 use of the accelerated deposition rate enabled the films to be grown with existing computers. It
inhibited various surface diffusion processes as surface atoms were rapidly buried into the bulk of
the films. Because high temperature accelerates the diffusion, the simulated films can be better
related to the experimental films obtained at reduced temperatures.
The detailed atomic structure of the simulated multilayers is shown in Fig. 1(a), where Ni, Fe,
Co, and Cu atoms are marked by heavy gray, white, light gray, and dark balls, respectively. This
image clearly shows that the Cu-on-CoFe interface is relatively sharp, whereas the CoFe-on-Cu
interface is quite diffuse. Cu atoms are incorporated in subsequently deposited CoFe layer more
significantly than in the underlying CoFe layer. These observations from the simulated film have all
been verified by a three-dimensional atom probe (3DAP) experiment [10].
Fig. 1 MD simulated Co90Fe10/Cu/Co90Fe10 multilayers grown on a Ni80Fe20 substrate. (a) Constant
adatom energy; and (b) Modulated energy.
The results discussed above showed that MD simulations can provide reliable structural
information about the GMR multilayers. More importantly, these simulations enable the
mechanisms of structure formation to be revealed. Further MD simulations revealed that because Cu
has a larger atomic size and smaller surface energy than either Co or Fe, Cu tends to segregate onto
Materials Science Forum Vols. 539-543
3530
the CoFe surface to release the surface tension under equilibrium conditions. In addition, Co or Fe
atoms impacting a Cu surface are more likely to penetrate into the Cu surface, because the Cu lattice
interstices are large and the relatively weakly bonded Cu atoms can be easily displaced. This
penetration causes the exchange of the impacting Co or Fe atoms and the surface Cu surface atoms,
resulting in continuous Cu diffusion into the subsequently deposited CoFe layer. The Cu-on-CoFe
interface is relatively sharp because the bigger Cu atoms impacting a more strongly bonded CoFe
surface are much less likely to penetrate the lattice. The understanding of this mixing mechanism
indicates that a reduction in the impact energy of the Co and Fe adatoms during their deposition on
Cu can reduce the exchange probability and therefore sharpens the CoFe-on-Cu interface.
Simulations, however, indicated that a reduction of the adatom energy increased the surface and
interfacial roughness, because any nucleated surface asperities could not be flattened by the
energetic impacts. The tradeoff between interlayer mixing and interfacial roughness then defines an
intermediate adatom energy that optimizes the GMR properties, in good agreement with
experiments [13,14]. The analyses also indicated that due to the asymmetric interfacial structures,
using a lower adatom energy to deposit the CoFe layer on Cu than to deposit the Cu layer on CoFe
(interlayer energy modulation) is better than using a single adatom energy to deposit the entire
multilayer stack. This strategy has been found experimentally to improve the GMR properties [15].
Based on the insights revealed by simulations, the new idea of modulating the energy during
deposition of each layer (interlayer energy modulation) was explored. In this scheme, the first few
atomic layers of Cu (or CoFe) were deposited on CoFe (or Cu) using a low adatom energy to avoid
the impact-induced mixing. After the interface was buried well below the surface, the deposition
was shifted to higher adatom energy to flatten the surface. The result of the same multilayer
structure deposited using this modulated energy scheme is shown in Fig. 1(b). It can be seen that
compared to the result shown in Fig. 1(a), the interlayer mixing is greatly reduced while the
interfacial roughness remains almost unchanged. In this case, the simulations helped improve the
deposition process.
Growth of Magnetic Tunneling Junction Multilayers
Interatomic Potentials. Charges are induced on atoms as metals are oxidized and hence an ionic
component to a potential must be considered during simulations of metal oxides. For
heterostructures involving both metals and metal oxides, an additional complexity arises because
the charge on atoms varies from zero in local metal regions to the maximum value in the oxides.
Such problems can only be addressed by advanced variable charge potentials [16,17]. However,
early versions of variable charge potentials [16,17] did not incorporate the physics of electron
valence. As a result, their parameterization is overly constrained, and they can only be used for
binary systems involving oxygen and one metal element. More recent modified charge transfer ionic
potentials (CTIPs) [18] overcame this problem and can be applied for systems involving oxygen and
any number of metal elements. One such modified CTIP has been integrated with EAM for a
quinternary O-Al-Ni-Co-Fe system [19]. It has been applied to study the growth of an AlOx tunnel
barrier layer on a Ni65Co20Fe15 surface [20].
Results. The MD simulation approach [20] was used to deposit first a Ni65Co20Fe15 layer on an
initial (111) Ni65Co20Fe15 substrate surface and then an Al layer on a Ni65Co20Fe15 surface. Various
Al layer thicknesses were explored. Structures of two selected films, one with about five atomic Al
layers (~12 Å) and a second with only a single atomic Al layer (~2.5 Å) are shown on the left of Fig.
2(a) and 2(b) respectively, where white and gray balls represent Al and O atoms and the darker balls
refer to Ni, Co, and Fe atoms. Accelerated oxidation of the Al surface was simulated by introducing
an atomic oxygen vapor above the Al/Ni65Co20Fe15 multilayer using a high vapor temperature of
8000 K and a high vapor density of 0.0003 oxygen atoms/Å3 (~12 atmospheres). The corresponding
film structures after oxidation are shown on the right of Fig. 2(a) and 2(b), respectively.
3531
THERMEC 2006
Fig. 2 MD simulated AlOx layer grown on a Ni65Co20Fe15 substrate. (a) Oxidation of an ~12.0 Å
aluminum layer deposited on a Ni65Co20Fe15 underlayer; and (b) Oxidation of an ~2.5 Å
aluminum layer deposited on a Ni65Co20Fe15 underlayer.
The formation of amorphous AlOx layers is seen in Fig. 2, in agreement with experimental
observations [21]. Fig. 2(a) indicates that oxidation of the thicker aluminum layer results in the
formation of a continuous and flat AlOx film. However, when the aluminum layer thickness is
reduced to 2.5 Å, Fig. 2(b), the AlOx film is highly discontinuous even though the Al layer prior to
oxidation has been continuous and relatively smooth. Oxidation of the thin Al layer is hence found
to be the cause for the formation of holes in the AlOx layer that expose areas of the underlying
Ni65Co20Fe15 layer. Experiments revealed that the TMR properties initially improved as the
thickness of the Al layer prior to the oxidation was reduced from 50 Å [22,23], but they abruptly
disappeared when the Al layer thickness was reduced below ~8 Å. This was thought to arise from
the formation of discontinuous AlOx layers [23,24]. Simulations account for this observation. They
further suggest that the inhibition of the nucleation and growth of holes in the gradually oxidized
thin Al layer is the key to grow a continuous AlOx layer at reduced layer thickness.
Growth of Semiconductor Thin Films
Interatomic Potentials. The bonding of semiconductor materials requires angular-dependent
interatomic potentials. Recent analysis [25] has indicated that the best interatomic potential
approach capable of realistic simulations of the growth of the stoichiometric semiconductor
compound films using arbitrary vapor flux ratio and vapor particles (atoms or molecules) is the
bond order potential [26,27]. A parameterization of a GaAs bond order potential [28] has been
successfully used to simulate the growth of the GaAs films in the [001] direction [29].
Results. The MD simulation method [29] was used to grow GaAs films on an initial (001) GaAs
substrate from atomic Ga and molecular As2 vapor fluxes using an adatom energy ~0.17 eV/atom, a
vapor flux direction normal to the surface, a growth rate between 0.14 – 0.22 nm/ns, and various
As:Ga vapor flux ratios and temperatures, T. Selected atomic configurations of the simulated films
are shown in Fig. 3(a) – 3(d), where As and Ga atoms are represented by the dark and light balls,
respectively, and the substrate prior to the deposition is shown as the shaded area.
Fig. 3 indicates that at approximately the same flux ratio, increased temperature improves the
crystallinity and reduces the defects. At the low simulated temperature of 700 K, increasing the
As:Ga vapor flux ratio causes the extra As atoms to be incorporated in the films, resulting in a
decrease in the film crystallinity and an increase in the defect concentration. At the relatively high
simulated temperature of 1100 K, increasing the flux ratio causes the excess As adatoms to
Materials Science Forum Vols. 539-543
3532
evaporate, and hence is beneficial to maintain the stoichiometric composition, the high crystallinity,
and the low As vacancy concentration in the buried films, except that the topmost surface layer
became more As rich. Experiments indicated that at the high temperatures, the As dimers could not
remain stuck to an As-rich surface [30], and crystalline stoichiometric films could only be grown
under an As:Ga vapor flux ratio >> 1 [4,31,32,33]. At low temperatures, the As dimers were more
likely to stick to the surface [4,34,35,36,37] and the crystalline stoichiometric films could only be
grown at an As:Ga vapor flux ratio ≈ 1 [38,39]. These experimental observations for both low- and
high-temperature growth regimes were well predicted in Fig. 3.
Fig. 3 MD simulated growth of a GaAs layer as a function of temperature, T, and As:Ga vapor flux
ratio. (a) T = 700 K, As:Ga = 1.24; (b) T = 700 K, As:Ga = 2.97; (c) T = 1100 K, As:Ga =
1.20; and (d) T = 1100 K, As:Ga = 2.93.
Summary
Recent MD approaches for vapor deposition simulation have opened new windows into the growth
of a wide range of materials including metallic multilayers, ionic metal oxide multilayers, and
covalent semiconductor compound layers. The simulations can reveal insights to improve the
growth processes that cannot be obtained using experiments alone.
Acknowledgments
We are grateful to the Defense Advanced Research Projects Agency and Office of Naval Research
(C. Schwartz and J. Christodoulou, Program Managers) for support of this work through grant
N00014-03-C-0288. We also thank S. A. Wolf for numerous helpful discussions.
References
[1] G. A. Prinz: Science Vol. 282, (1998), p. 1660.
[2] R. S. Beech, J. Anderson, J. Daughton, B. A. Everitt and D. Wang: IEEE Trans. Magn. Vol. 32,
(1996), p. 4713.
[3] M. Durlam, P. J. Naji, A. Omair, M. DeHerrera, J. Calder, J. M. Slaughter, B. N. Engel, N. D.
Rizzo, G. Grynkewich, B. Butcher, C. Tracy, K. Smith, K. W. Kyler, J. J. Ren, J. A. Molla, W.
A. Feil, R. G. Williams and S. Tehrani: IEEE J. Sol. State Cir. Vol. 38, (2003), p. 769.
[4] Properties of Gallium Arsenide, edited by M. R. Brozel and G. E. Stillman, Vol. 16, 3rd ed.
(INSPEC, London,1996).
[5] S. Strite and H. Morkoc: J. Vac. Sci. Technol. B Vol. 10 (1992), p. 1237.
[6] S. J. Pearton, J. C. Zolper, R. J. Shul and F. Ren: J. Appl. Phys. Vol. 86, (1999), p. 1.
[7] I. Akasaki: IPAP Conf. Series Vol. 1 (2000), p. 1.
[8] M. S. Daw and M. I. Baskes: Phys. Rev. B Vol. 29, (1984), p. 6443.
3533
THERMEC 2006
[9] X. W. Zhou, R. A. Johnson and H. N. G. Wadley: Phys. Rev. B. Vol. 69, (2004), p. 144113.
[10] X. W. Zhou, H. N. G. Wadley, R. A. Johnson, D. J. Larson, N. Tabat, A. Cerezo, A. K.
Petford-Long, G. D. W. Smith, P. H. Clifton, R. L. Martens and T. F. Kelly: Acta Mater. Vol.
49, (2001), p. 4005.
[11] X. W. Zhou and H. N. G. Wadley: J. Appl. Phys. Vol. 84, (1998), p. 2301.
[12] W. Zou, H. N. G. Wadley, X. W. Zhou, R. A. Johnson and D. Brownell: Phys. Rev. B Vol. 64,
(2001), p. 174418.
[13] J. C. S. Kools: J. Appl. Phys. Vol. 77, (1995), p. 2993.
[14] S. Schmeusser, G. Rupp and A. Hubert: J. Magn. Magn. Mater. Vol. 166, (1997), p. 267.
[15] T. L. Hylton, K. R. Coffey, M. A. Parker and J. K. Howard: J. Appl. Phys. Vol. 75, (1994), p.
7058.
[16] A. K. Rappe and W. A. Goddard: J. Phys. Chem. Vol. 95, (1991), p. 3358.
[17] F. H. Streitz and J. W. Mintmire: Phys. Rev. B Vol. 50, (1994), p. 11996.
[18] X. W. Zhou, H. N. G. Wadley, J. -S. Filhol and M. N. Neurock: Phys. Rev. B. Vol. 69, (2004),
p. 35402.
[19] X. W. Zhou and H. N. G. Wadley: J. Phys.: Condens. Matter Vol. 17, (2005), p. 3619.
[20] X. W. Zhou and H. N. G. Wadley: Phys. Rev. B Vol. 71, (2005), p. 54418.
[21] L. F. Li, X. Y. Liu and G. Xiao: J. Appl. Phys. Vol. 93, (2003), p. 467.
[22] W. Zhu, C. J. Hirschmugl, A. D. Laine, B. Sinkovic and S. S. P. Parkin: Appl. Phys. Lett. Vol.
78, (2001), p. 3103.
[23] J. S. Moodera, E. F. Gallagher, K. Robinson and J. Nowak: Appl. Phys. Lett. Vol. 70, (1997),
p. 3050.
[24] J. H. Lee, H. D. Jeong, H. Kyung, C. S. Yoon, C. K. Kim, B. G. Park and T. D. Lee: J. Appl.
Phys. Vol. 91, (2002), p. 217.
[25] D. A. Murdick, X. W. Zhou and H. N. G. Wadley: Phys. Rev. B Vol. 72, (2005), p. 205340.
[26] D. G. Pettifor, M. W. Finnis, D. Nguyen-Manh, D. A. Murdick, X. W. Zhou and H. N. G.
Wadley: Mater. Sci. Eng. A Vol. 365, (2004), p. 2.
[27] D. G. Pettifor, M. W. Finnis, D. Nguyen-Manh, D. A. Murdick, X. W. Zhou and H. N. G.
Wadley: Mater. Sci. Eng. A Vol. 365, (2004), p. 2.
[28] D. A. Murdick, X. W. Zhou, H. N. G. Wadley, D. Nguyen-Manh, R. Drautz and D. G. Pettifor:
submitted to Phys. Rev. B (2005).
[29] D. A. Murdick, X. W. Zhou and H. N. G. Wadley: submitted to Phys. Rev. B (2005).
[30] C. T. Foxon and B. A. Joyce: Surf. Sci. Vol. 64, (1977), p. 293.
[31] M. Pristovsek, S. Tsukamoto, A. Ohtake, N. Koguchi, B. G. Orr, W. G. Schmidt and J. Bernholc: Phys. Status Solidi B Vol. 240, (2003), p. 91.
[32] J. R. Arthur: J. Appl. Phys. Vol. 39, (1968), p. 4032.
[33] J. R. Arthur: Surf. Sci. Vol. 43, (1974), p. 449.
[34] K. Mahalingam, N. Otsuka, M. R. Melloch, J. M. Woodall and A. C. Warren: J. Vac. Sci.
Technol. B Vol. 9, (1991), p. 2328.
[35] E. S. Tok, J. H. Heave, J. Zhang, B. A. Joyce and T. S. Jones: Surf. Sci. Vol. 374, (1997), p.
397.
[36] M. Kaminska, E. R. Weber, Z. Liliental-Weber, R. Leon and Z. U. Rek: J. Vac. Sci. Technol. B
Vol. 7, (1989), p. 710.
[37] A. Suda and N. Otsuka: Surf. Sci. Vol. 458, (2000), p. 162.
[38] M. Missous and S. O’Hagan: J. Appl. Phys. Vol. 75, (1994), p. 3396.
[39] V. Avrutin, D. Humienik, S. Frank, A. Koeder, W. Schoch, W. Limmer, R. Sauer and A.
Waag: J. Appl. Phys. Vol. 98, (2005), p. 23909.