JOURNAL OF APPLIED PHYSICS

VOLUME 88, NUMBER 12

15 DECEMBER 2000

Tensile stress evolution during deposition of VolmerWeber thin films

Steven C. Seel and Carl V. Thompsona)
Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge,
Massachusetts 02139

Sean J. Hearne and Jerrold A. Floro

Sandia National Laboratories, Albuquerque, New Mexico 87185-1415

Received 1 August 2000; accepted for publication 21 September 2000

A simple model is presented that predicts the kinetics of tensile stress evolution during the
deposition of thin films that grow by the VolmerWeber mechanism. The generation of a tensile
stress was attributed to the impingement and coalescence of growing islands, while concurrent stress
relaxation was assumed to occur via a microstructure-dependent diffusive mechanism. To model the
process of island coalescence, finite element methods were employed and yielded average tensile
stresses more consistent with experimental observations than those predicted using previously
reported analytical models. A computer simulation was developed that models the process of film
growth as the continuous nucleation of isolated islands, which grow at a constant rate to impinge
and coalesce to form a continuous polycrystalline film. By incorporating the finite element results
for stress generation and a microstructure-dependent stress relaxation model, the simulation
qualitatively reproduced the complex temperature-dependent trends observed from in situ
measurements of stress evolution during the deposition of Ag thin films. The agreement includes
simulation of the decreasing stress relaxation rate observed during deposition at increasing
temperatures. 2000 American Institute of Physics. S0021-89790101501-8

I. INTRODUCTION

in a significant energy reduction. Hoffman postulated that if

neighboring islands are within close proximity, they will
stretch toward each other to form a grain boundary to reduce
the interfacial energy at the expense of an associated strain
energy.3 Transmission electron microscopy TEM observations of films at various stages of growth, coupled with stress
measurements, supported the idea that tensile stress generation during the early stages of film growth was associated
with the process of impingement and coalescence of growing
islands.46
In these works, the maximum tensile stress measured
during deposition of high mobility materials was typically an
order of magnitude lower than that for low mobility materials. However, if low mobility materials were deposited at
higher temperatures they behaved much like high mobility
materials.7 Conversely, high mobility materials deposited at
lower temperatures exhibited large intrinsic stresses similar
to low mobility materials.8 These observations are consistent
with the existence of a diffusive stress relaxation mechanism
with a strong dependence on temperature. Since the microstructural scale during deposition can be on the order of 100
,6,9 diffusion distances along surfaces and grain boundaries
will be short so that significant stress relaxation may occur
during deposition.
In this study, we present a simple model for tensile stress
generation and relaxation that captures the kinetics of tensile
stress evolution during deposition of thin films that grow by
the VolmerWeber mechanism. For comparison with previous analytical models for tensile stress generation, we utilized finite element methods FEM to model the island coalescence process. By minimizing the sum of the positive
strain energy and associated reduction in interfacial energy,

Stresses are commonly generated during the growth of

thin films on substrates. The reliability and performance of
thin film devices depend on the properties and behavior of
the deposited films. Understanding the origins of stresses to
control the level of stress is of great importance for applications in microelectronics, magnetic storage, and microelectromechanical systems.
Measurements of stress during the early stage of deposition of polycrystalline thin films have been performed by
many researchers.1,2 For refractory materials with low adatom mobility, tensile stresses observed during deposition increased with film thickness, becoming approximately constant once the film became fully continuous. Tensile stresses
greater than 1 GPa were measured in refractory materials
such as W, Ti, and Cr during deposition at room temperature.
For fcc materials with high adatom mobility, tensile stresses
peaked in the early stages of film growth and decreased or
became compressive for thicker continuous films. The maximum in the tensile stress during deposition of fcc materials
such as Ag, Cu, and Au at room temperature was on the
order of 100 MPa.
For films that grow by the VolmerWeber mechanism,
crystallites of critical size nucleate on the surface of the substrate as isolated islands. As the islands grow in diameter,
they impinge to form a network of islands, eventually coalescing into a continuous polycrystalline film. As two islands
impinge and form a grain boundary at their intersection,
part of the free surface of each island is eliminated, resulting
a

Electronic mail: cthomp@mtl.mit.edu

0021-8979/2000/88(12)/7079/10/$17.00

7079

2000 American Institute of Physics

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7080

Seel et al.

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

the equilibrium configuration resulting from island coalescence was determined as a function of island radius. The
magnitude of the average tensile stress calculated using FEM
were compared to the analytical model for island coalescence
stress presented by Nix and Clemens.10
We also performed in situ thin film stress-thickness measurements using a wafer-curvature based technique for Ag
deposited on oxidized silicon substrates. Films were deposited at different temperatures to study the kinetics of stress
evolution during deposition. TEM and scanning electron microscopy SEM were used to correlate the microstructure
with the measured stress-thickness product at a given thickness. Qualitative comparisons were made between the experimental results and the model for tensile stress generation
and relaxation. In addition, we have developed computer
simulations that model the process of film growth from the
initial stages of island nucleation, through island growth and
impingement, to film continuity. Our goal is to reproduce the
qualitative trends observed experimentally for tensile stress
evolution during film deposition, using the FEM results for
stress generation and the proposed microstructure-dependent
stress relaxation model.
II. TENSILE STRESS GENERATION

Following Nix and Clemens,10 a model for calculating

the tensile stress generation associated with island impingement and coalescence was examined. Consider an array of
two-dimensional islands with hemispherical shapes that coalesce to form a surface with a cycloid shape, as shown in Fig.
1a. The cusps at the surface of the islands were treated like
cracks that allowed for a cracklike analysis similar to the
Griffith criterion. The resulting expression for the average
stress, , from this analysis of island impingement and coalescence was given by10

1 2 s gb
E
1
r

1/2

where is Poissons ratio, E is Youngs modulus s is the

surface energy of the island, gb is the grain boundary energy, and r is the radius of the island. An expression for the
height of the grain boundary resulting from coalescence, also
called the zipping distance or z 0 , for a hemispherical island
was approximately given by10
z 0

36 1 1 3 2 s gb
E

1/4

r 3/4.

Using typical values for silver given in Table I, a grain radius of 100 results in a zipping distance of 64 and an
average stress of 6.8 GPa, while a grain radius of 1000
gives a zipping distance of 360 and an average stress of
2.2 GPa.
The NixClemens model provides an intuitive understanding of how tensile stresses can be generated during
deposition, along with simple analytical expressions for calculating stress. However, the model predicts tensile stresses
that are significantly higher than those observed in experiments. While observed stress levels may be mitigated by
relaxation processes, we examined the accuracy of the Nix

FIG. 1. a Schematic of the island impingement and coalescence process

resulting in tensile stress generation. The dashed lines represent the hemispherical island at impingement, while the solid lines represents the cycloid
surface of the island after coalescence through grain boundary zipping to a
height z 0 . b FEM model of island coalescence represented by a twodimensional element under plane strain conditions. The x axis represents the
island-substrate interface where traction was imposed. The y axis is an axis
of symmetry along which sliding was allowed. The arrowed lines are the
series of displacement that represent zipping to a height z 0 of an island of
radius r.

Clemens analytical approach by performing FEM calculations using the commercial software ADINA. An island was
represented by a two-dimensional element under plane strain
conditions and with perfect traction at the island-substrate
interface, as shown in Fig. 1b. The plain strain condition
imposed in the xy plane implies that impingement is occurring between two infinitely long cylinders with semicircular
cross sections. A series of displacements were imposed along
the surface to a height z 0 to mimic the zipping process. For a
given island radius, the positive strain energy from FEM
modeling and the associated reduction in interfacial energy
were calculated as a function of the zipping distance. The
sum of these two energies represents the change in energy of
the system and the negative-valued minimum corresponds to
the equilibrium value of the zipping distance. For comparison with the values calculated from the NixClemens model,
the zipping distance and average stress versus island radius
from the FEM modeling are shown in Fig. 2. From the FEM
model with plane strain conditions and traction at the islandsubstrate interface, the zipping distance decreases with the
island radius raised to the 0.675 power, compared to the 0.75
dependence in the NixClemens model, while the average
stress has an exponential dependence of 0.814 on the island radius, compared to an inverse square root dependence

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

Seel et al.

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

7081

TABLE I. Materials properties for Ag at room temperature.

Material

Youngs modulus,a
E GPa

Poisson ratio,a

Surface energy,b
s J/m2

Grain boundary
energy,b gb J/m2

Ag

87.3

0.354

1.5

0.47

G. Simmons and H. Wang, Single Crystal Elastic Constants and Calculated Aggregate Properties: A Handbook, 2nd ed. M.I.T. Press, Cambridge, MA, 1971.
b
L. E. Murr, Interfacial Phenomena in Metals and Alloys Addison-Wesley, Reading, MA, 1975.

in the NixClemens model. The material properties for Ag

used in the calculation of Fig. 2 for both the FEM results and
the NixClemens model are shown in Table I.
The NixClemens model for tensile stress generation resulting from island coalescence utilizes an analysis originally
used to describe surface roughening.11 The geometry of the
island coalescence and surface roughening problems are not
completely analogous which may cause the NixClemens
model to overestimate the stress. Our FEM model of island
coalescence represents island zipping in a straightforward
manner and predicts smaller stress more consistent with experimental observations.
During the growth of a real film, impingement of a periodic array of islands does not occur as described before. At
low substrate coverage, island impingement will typically
involve an island that has not yet impinged with any other
islands. To model the first coalescence of an island using
FEM, an island was represented as a plane strain twodimensional element with either traction at the island-

FIG. 2. Comparison of FEM calculations with results from the Nix

Clemens model, showing a the zipping distance, z 0 , divided by the island
radius, r, and b the average stress as a function of the island radius.

substrate interface island i in Fig. 3a or with sliding at

the island-substrate interface island iii in Fig. 3a. The
zipping due to the single coalescence was represented by
displacements imposed along only one side of the island.
The equilibrium zipping distance Fig. 3b and average
stress Fig. 3c versus island radius resulting from a single

FIG. 3. a Schematic of island coalescence showing i the first and ii the

second coalescence of an island with traction at the island-substrate interface, and iii the first and iv the second coalescence of an island with
sliding at the island-substrate interface. b The equilibrium zipping distance, z 0 , divided by the island radius, r, and c the average stress as a
function of the island radius from FEM modeling of a two-dimensional
element under plane strain conditions for the cases shown in a.

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7082

Seel et al.

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

coalescence are shown for both the cases of traction and

sliding at the island-substrate interface. If traction was imposed at the substrate, the coalescing island stretched towards the forming grain boundary to eliminate part of its
surface. Since island movement was constrained due to traction, tensile stresses were generated in the island. If sliding
was allowed along the substrate, the island moved its center
towards the forming grain boundary during zipping, resulting
in a very small, compressive mean stress. However, without
traction, no load transfer between the island and substrate
can occur, so the stress will not result in a substrate curvature.
At higher substrate coverage, an impinging island most
likely will have already coalesced with another island forming what will be referred to as an island cluster. The second
coalescence of an island was modeled using a plane strain
element with one edge pinned and displacements imposed
along the opposite edge to mimic zipping to a height z 0 . For
the case with traction at the island-substrate interface, the
second coalescence see island ii in Fig. 3a approximately doubled the average stress in the island, as shown in
Fig. 3c. If sliding at the island-substrate interface is allowed, the consequence of subsequent impingements is less
obvious. If the entire island cluster is able to slide, then island coalescence will result in a slightly compressive stress,
as shown previously. However, sliding of an island may become inhibited after a few coalescence events, as discussed
in the following paragraph. If sliding was allowed along the
island-substrate interface but with one edge of the island
pinned, the second coalescence see island iv in Fig. 3a
generates tensile stresses similar in magnitude to the first
coalescence of an island with traction, as shown in Fig. 3c.
The mechanism responsible for island sliding must be
examined more closely to understand how load transfer between the film and substrate can occur even with interface
sliding. When islands coalesce, large shear stresses are generated near the grain boundary due to the displacements resulting from zipping. These shear stresses can drive the
movement of dislocation-like entities along the filmsubstrate interface away from the grain boundary.12 Island
sliding will result if the dislocation can travel across the
entire island. As the substrate coverage increases, most islands have coalesced with two or more islands. The movement of the dislocation-like entities across the entire island
will be opposed by shear stresses from previous coalescence
events. Consequently, we would expect that sliding on an
island will become inhibited after the island has coalesced
with a few other islands. If island sliding is inhibited, subsequent coalescence of an island will generate tensile stresses
in the film, as shown in Fig. 3c. Since traction now exists
between the island and the substrate, the tensile stress generation will results in a substrate curvature. If perfect sliding
occurs at the island-substrate interface, no load transfer can
occur between the film and substrate and consequently no
substrate curvature will be measured.
A potentially more accurate representation of the geometry of island coalescence would use an axially symmetric
coalescence analogous to contacting two spheres. However,
the axisymmetric representation conflicts with the boundary

FIG. 4. Proposed stress relaxation mechanism involving the diffusion of

atoms across the surface of the island and along grain boundaries to the
strained regions near the grain boundary. Since surface diffusion and grain
boundary diffusion occur in series, either process may be rate limiting.

condition of perfect traction at the island-substrate interface.

The islands attempt to zip towards each other in the plane of
the substrate but are prevented from doing so by the traction
with the substrate. Even though the geometries are different,
the axisymmetric model with sliding was found by FEM to
give quantitatively similar average stresses resulting from
island impingement compared to the plane strain case with
sliding. Since both the traction and sliding cases can be
treated with the plane strain geometry, the plane strain finite
element results were implemented in the modeling of island
impingement.

III. STRESS RELAXATION

Tensile stress generation resulting from island coalescence occurs due to the localized displacements at the grain
boundary. One possible mechanism by which the tensile
stress can relax is through transport of matter to the strained
region within the grain boundary as shown schematically in
Fig. 4. A fast diffusion path for atoms is along the island
surface, which acts as a source of atoms, and down the grain
boundary. Since both surface and grain boundary diffusion
are required in series, either diffusive mechanism may be
rate limiting. With or without sliding at the film-substrate
interface, the stress relaxation rate, , will take the form1315

C0
exp Q/kT ,
h3

where C 0 is a material-dependent, temperature- and stressindependent constant, is the average stress in the film, h is
the film thickness, k is Boltzmanns constant, T is temperature in Kelvin, and Q is the activation energy for the ratelimiting diffusive process, either grain boundary or surface
diffusion. Implicit in this expression is that the grain size
scales with the film thickness. Unlike the uniform strains
associated with lattice mismatch or thermal expansion mismatch, the strains created by island coalescence result from
localized surface displacements due to island zipping. Consequently, matter diffusing to the grain boundaries can relax
all of the tensile stress generated by island coalescence.
Since the diffusion distance along the surface and grain
boundary is very short, this diffusive process may be an important stress relief mechanism even at low temperatures.

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

Seel et al.

7083

FIG. 5. a Stress-thickness vs nominal film thickness for Ag thin films

deposited at 2 /s on oxidized Si substrates. Inset is an enlargement of the
initial 200 . b The same data replotted as stress-thickness divided by the
nominal film thickness.

IV. IN SITU FILM STRESS MEASUREMENTS

Ag films were grown by electron-beam evaporation in an

ultra-high vacuum base pressure of 11010 Torr, deposition pressure of 5109 3108 Torr. The 100-m-thick
Si 001 substrates had either a native oxide or a thermally
grown oxide similar results were obtained for both types of
substrates. The substrates were prepared using a solvent
clean followed by a sulfuric acid-hydrogen peroxide etch
prior to loading in the vacuum chamber. The substrate was
outgassed at 350 C for 1 h in vacuo at a pressure below 5
1010 Torr. Prior to deposition, the substrate was allowed
to equilibrate at the desired growth temperature for at least 1
h. The temperature during deposition was monitored using a
thermocouple metal bonded to a witness wafer, and the
deposition rate was controlled using a quartz crystal monitor.
Wafer curvature was measured in situ using a sensitive
multibeam optical deflection technique.16 If the thickness of
a film is much less than the substrate thickness, the product
of the average film stress and film thickness can be related to
the substrate curvature using Stoneys equation, which is
commonly used to interpret substrate curvature
measurements.17,18 The stress-thickness was determined
from substrate curvature measurements during growth of Ag
at a constant deposition rate of 2 /s at various substrate
temperatures Fig. 5a. For comparison, we have replotted
the data in Fig. 5b as the stress-thickness product divided
by the nominal film thickness. Since the sensitivity of the

FIG. 6. Sequence of plan-view TEM images of Ag films at nominal thicknesses of a 120, b 160, and c 260 .

curvature measurements is related to the stress-thickness

product, the calculated values in Fig. 5b at very small film
thicknesses tend to be noisy.
In order to study the evolution of microstructure, films
having a laterally graded thickness were grown by moving a
shutter across the wafer during deposition. Ex situ plan-view
TEM micrographs and SEM images were then used to characterize the microstructure as a function of the nominal film
thickness. Figure 6 shows a typical sequence of TEM micrographs at different nominal film thicknesses of Ag deposited
at 30 C and at a rate of 2 /s. Consistent with previous
studies,2,6 the maximum in the tensile stress-thickness for all
deposition temperatures coincided with the late channel stage
when the fractional substrate coverage was greater than 0.95.

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7084

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

FIG. 7. Sequence of simulated structures of polycrystalline film formation

by a VolmerWeber mechanism, under conditions of continuous nucleation
and a constant radial growth velocity.

V. SIMULATION OF THIN FILM GROWTH

The growth of thin films by the VolmerWeber mechanism involves the nucleation and growth of individual islands that impinge and coalesce to form a continuous film as
shown in a sequence of simulated structures in Fig. 7. Thin
film growth was modeled in a two-dimensional simulation
by tracking the nucleation and growth of circles, which represent the intersection of islands with the substrate. For simplicity, the film-substrate interface energy was assumed to
equal the substrate surface energy so that the nucleated islands have hemispherical shapes i.e., the island height
equals the radius of the island. Nucleation was assumed to
occur continuously during deposition at a constant rate, so

Seel et al.

that the added number of new nuclei per unit time per unit
exposed substrate area was constant. The nucleated islands
grow with a constant radial growth velocity proportional to
the deposition rate.19 At early times, most of the islands have
impinged with less than two other islands and therefore have
formed at most one grain boundary. As islands grow larger
and more impingements occur, triple points are formed
where grain boundaries meet within an ensemble of three
islands or grains. Eventually, all grain boundaries terminate
at triple points when the film is fully continuous. The assumed growth conditions of continuous nucleation at a constant rate and a constant radial growth velocity result in a
continuous film composed of grains with a JohnsonMehl
structure.20
In the simulation, once two islands impinged, the equilibrium zipping distance and stresses within the island were
calculated using the FEM results for the plane strain geometry with traction at the island-substrate interface. The case
with sliding at the island-substrate interface will be considered later. For each new coalescence, the average stress calculated using the FEM results was simply added to any existing average stress in the island. The principal of
superposition states that two strains may be combined by
direct superposition, with the order of application having no
effect on the final strain of the body. However, for subsequent coalescence events, the calculation of the strain energy, used to determine the equilibrium zipping distance and
the corresponding average stress, neglected the existing
stresses and strains in the island. Once coalescence between
two islands occurred, further lengthening of the grain boundary due to island growth was assumed to generate no additional stress. However, continued deposition was specified to
occur epitaxially so that new material inherited the stress
of the underlying layer.21
During deposition, islands with different radii will impinge and behave differently from the symmetric cases considered previously. In the simulation, the energy minimizing
z 0 was determined for the coalescence of islands with dissimilar sizes using the FEM calculations of strain energy for
different zipping distances as a function of island radius. If
two islands with dissimilar sizes impinged, the boundary of
the smaller island was found to zip more than if it had coalesced with an island of the same size. Consequently, the
stress in the smaller island impinging on a larger island was
greater than if it had coalesced with an island of the same
size. The opposite trends are true for the larger island.
Stresses generated by island coalescence were assumed
to be relaxed by a microstructure-dependent diffusive stress
relaxation mechanism, similar to that described by Eq. 3.
The proposed expression for stress relaxation is the most
appropriate for continuous films since the stresses are approximately equibiaxial and the grains have formed boundaries on all sides. However, after a single coalescence, an
island has only one grain boundary and the stresses are not
equibiaxial. Nonetheless, the form of the stress relaxation
rate given by Eq. 3 captures the origin of the driving force
for relaxation and the microstructural dependence of the
mechanism. The stress relaxation rate given by Eq. 3 was
used in the simulation with the film thickness term replaced

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

Seel et al.

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

7085

FIG. 8. Fractional substrate coverage vs film thickness from SEM images

during various stages of deposition of Ag at different temperatures. The
lines are from simulations of microstructural evolution. For each deposition
temperature, a scaling factor was determined that gave reasonable agreement between the simulated microstructure and the measured coverage vs
nominal film thickness.

by the island height, and using the average stress in the island. This implementation implies that stress relaxation in
each individual island can be treated independently. These
approximations are not expected to significantly affect the
results of the simulation.
The simulation of stress evolution during thin film deposition is based upon the nucleation and growth simulation
coupled with the FEM results for stress generation and the
analytical model for the microstructure-dependent stress relaxation mechanism. The experimental inputs to the simulation were the deposition rate and temperature, which both
affect the rates of stress generation and relaxation. The
physical dimensions of the simulation are unitless so a scaling factor must be defined that relates the microstructural
dimensions of the real film to the simulated structure. From
SEM images of discontinuous Ag films, the fractional substrate coverage versus film thickness was measured for all
deposition conditions. For each deposition temperature, a
scaling factor was determined that gave reasonable agreement between the simulated microstructure and the measured
coverage versus nominal film thickness, as shown in Fig. 8.
This scaling factor influences both the island-size-dependent
stress generation model and the microstructure-dependent
stress relaxation model. In addition, a diffusivity for the rate
limiting diffusive process of the stress relaxation mechanism
must be supplied. An activation energy of approximately 0.3
eV and a pre-exponential factor of an appropriate magnitude
were chosen to best match the simulation to the experimentally measured stress-thickness versus thickness curves at
different deposition temperatures. For each deposition temperature, results were averaged over approximately 2500
island/grains from ten simulations using different random
starting seeds. Assuming traction at the island-substrate interface, the average stress-thickness versus thickness and average stress versus thickness from simulations of Ag films
deposited at different temperatures are shown in Fig. 9.
In another set of simulations, the equilibrium zipping
distance and average stress were calculated using the FEM
results with sliding at the island-substrate interface. Contrary
to the case with traction, two different types of coalescence

FIG. 9. Simulation of Ag thin films with traction at the film-substrate interface during deposition at 2 /s at different temperatures showing, a stressthickness product vs nominal film thickness, and b average stress vs nominal film thickness. The simulated microstructure was formed by continuous
nucleation at a constant rate and using a constant radial growth velocity
proportional to the deposition rate. Stress generation was modeled using the
FEM approach, while stress relaxation was assumed to occur via a
microstructure-dependent diffusion mechanism.

events can occur. We have assumed that the first few coalescence events for a given island will occur with sliding since
the movement of dislocation-like entities will be relatively
unopposed. Increasing the number of coalescence events before sliding is inhibited will delay the onset of the rise of the
tensile stress until larger thickness. If a coalescing island can
slide easily, the average stress resulting from coalescence is
slightly compressive see Fig. 3c. However, without any
regions of traction, there is no load transfer from the island
to the substrate and no substrate curvature will result. Subsequent coalescence events were assumed to generate tensile
stress since sliding is inhibited within an island, as described
in the previous section. These tensile stresses were again
assumed to be relaxed by the proposed diffusive mechanism.
Otherwise, the simulations were run under exactly the same
conditions as for the case with island-substrate traction. Under conditions of island-substrate sliding, simulations of the
average stress-thickness versus thickness and average stress
versus thickness were performed for Ag films deposited at
different temperatures as shown in Fig. 10. For the simulations shown in Fig. 10, islands with fewer than four impingements were assumed to slide, although values of two or three
impingements gave qualitatively similar results.
VI. DISCUSSION

The measured stress-thickness versus thickness curves

exhibited four distinct regimes at different nominal film

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7086

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

FIG. 10. Simulation of Ag thin films with sliding at the film-substrate interface during deposition at 2 /s at different temperatures showing, a
stress-thickness product vs nominal film thickness, and b average stress vs
nominal film thickness. Islands with fewer than four impingements were
assumed to slide, while subsequent coalescence events produced tensile
stresses. Otherwise, the simulations were performed under the same conditions as those shown in Fig. 9.

thicknesses. During early deposition times, no appreciable

stress-thickness was measured up to a critical thickness,
which increased with increasing deposition temperature as
seen in Fig. 5a. From TEM micrographs, the island density
was initially very high but decreased with increasing film
thickness, indicating the presence of island coarsening by
surface diffusion and/or grain growth. If neighboring islands
have different radii then the gradient in chemical potential
may be sufficient to drive island coarsening by surface diffusion, especially at higher temperatures. However, if islands
are approximately the same size or once islands are large, the
gradient will be smaller and impinging islands will form
grain boundaries rather than undergoing coarsening. The
simulation of film growth does not include island coarsening
so all island impingements resulted in stress generation. In
the simulation with traction at the island-substrate interface,
the onset of an appreciable stress-thickness occurred at a
much smaller film thickness inset of Fig. 9a than was
observed in experiments inset of Fig. 5a.
Even once numerous grain boundaries had formed, as
seen in Fig. 6a for a 120 Ag film deposited at 30 C, the
measured tensile stress-thickness in the film was still very
low. Because the stress relaxation mechanism is strongly dependent on the island size, the tensile stresses generated as a
result of the initial coalescence are expected to be relaxed
very quickly. However even with a microstructure-dependent
stress relaxation mechanism, the simulated stress-thickness

Seel et al.

for the case of island-substrate traction was appreciable even

before a nominal film thickness of 100 , as seen in the inset
of Fig. 9a. If sliding along the island substrate was assumed, the first four coalescence events for a given island
generated no measurable stress, as described in the previous
section. This had the effect of retarding tensile stress generation in the early stages of film growth, as shown in the inset
of Fig. 10a. Since multiply-impinged islands were assumed
to support tensile stresses, the average stress increased dramatically as the film approached a percolated structure. For a
Ag film deposited at 30 C, percolation occurred around 175
which approximately coincided with the significant increase in the tensile stress-thickness at the same film thickness as seen in Fig. 5a. In the simulation, the greater the
number of coalescence events that occurred before islandsubstrate sliding was inhibited, the larger the thickness at
which an appreciable tensile stress-thickness occurred.
The second feature of the stress-thickness versus thickness measurements to be considered is the magnitude of the
maximum tensile stress-thickness and the film thickness at
which the maximum occurred. From Fig. 5a, the magnitude
of the maximum tensile stress-thickness decreased and the
film thickness coinciding with the maximum increased with
increasing deposition temperature. For all deposition temperatures, the maximum tensile stress-thickness occurred
when the film was in the late channel stage just before continuity, as confirmed using SEM. Since continuity occurred
at larger film thicknesses with increasing deposition temperature, the average grain size at continuity must similarly be
larger at higher deposition temperatures. Based on the FEM
results in Fig. 3c, the magnitude of the maximum tensile
stress should decrease with increasing grain size. However,
predicting the relative magnitudes of the maximum stressthickness with deposition temperature also requires consideration of the influence of stress relaxation.
The simulation, which accounted for both stress generation and relaxation, reproduced the experimental trends of a
decreasing maximum tensile stress-thickness that occurred at
larger film thicknesses with increasing deposition temperature. In addition to the experimental variables of deposition
temperature and rate, two other parameters had to be supplied to simulate the stress evolution during thin film deposition. Since the dimensions of the simulation are unitless,
the island or grain size scaling of the simulation must be
determined by matching the fractional substrate coverage
versus film thickness from the simulation to results from
SEM images of Ag films at different nominal thickness, as
shown in Fig. 8. The island/grain size scaling influences both
the stress generation see Fig. 3c and the stress relaxation
rate see Eq. 3, and roughly determines the film thickness
at which the maximum in the tensile stress-thickness occurs,
since the film thickness at continuity generally scales with
the grain size. The second parameter required for the simulation was the diffusivity for the rate-limiting process for
stress relaxation. With an activation energy of 0.3 eV and an
appropriate pre-exponential factor, the simulation could be
used to predict stress-thicknesses within a factor of 2 of the
experimental measurements for all deposition temperatures.
If an activation energy of 0.6 eV was used, as reported by

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

Kobrinsky and Thompson22 for diffusive stress relaxation in

continuous Ag films, the simulation produced the same
qualitative trends with deposition temperature, but using a
value of 0.3 eV gave better agreement with the experimental
results.
The third feature of the stress-thickness versus thickness
curves to be considered is the observed temperature dependence of the negative slope region immediately following
film continuity. Since the zipping mechanism for tensile
stress generation becomes inactive once the films achieve
continuity, the subsequent decrease in the stress-thickness is
presumably due in part to stress relaxation. As can be seen
from Fig. 5, the slope of the curves just after continuity
suggests that the rate of stress relaxation decreased with increasing deposition temperature. Even though diffusion increases rapidly with increasing temperature, the proposed
stress relaxation mechanism in Eq. 3 is also strongly dependent on the microstructural scale which is strongly temperature dependent as well. As shown in Fig. 8, the film
thickness coinciding with continuity increased with increasing deposition temperature. Therefore, films deposited at
higher temperatures had slower relaxation rates after continuity because of the strong thickness dependence of the
stress relaxation rate. As shown in Figs. 9 and 10, the simulation reproduces the lower postcontinuity stress relaxation
rates observed at higher deposition temperatures using the
same activation energy 0.3 eV, which also correctly reproduces the maximum tensile stress-thickness dependence with
deposition temperature.
One last feature of the experimentally measured curves
that has not been addressed here is the compressive stresses
observed at large film thicknesses. Compressive stresses during island-type growth are typically attributed to the Laplace
pressure induced by surface stresses.23,24 In support of the
compressive stress model, reduced lattice parameters for
very small islands have been measured for various fcc
metals.25 As an island grows in diameter, the lattice parameter attempts to change according to the size-dependent
Laplace pressure. If island-substrate sliding occurs easily at
low substrate coverage, the compressive strain in the island
is not imposed on the substrate and will not result in substrate curvature, which may explain why compressive
stresses were not measured at very small nominal thicknesses
of Ag. At larger thicknesses once the film reaches a percolated structure, interfacial shear is no longer effective for
island sliding. As the island grows larger, the substrate exerts
forces on the island which prevent it from adjusting its lattice
parameter thereby creating stress in the island. While the
tensile stresses resulting from island coalescence are relaxed
by diffusion of matter to the grain boundaries, the same diffusive mechanism cannot fully relax the compressive stress
imposed by the substrate. Consequently, once the tensile
stresses near the grain boundaries have relaxed, the compressive stress will still be present and may explain the compressive stress observed in thicker continuous films.
VII. SUMMARY AND CONCLUSIONS

We have presented a simple model for tensile stress generation and relaxation that predicts the kinetics of the intrin-

Seel et al.

7087

sic tensile stresses evolution during deposition of polycrystalline thin films that grow by the VolmerWeber
mechanism. For comparison with previous analytical models
of tensile stress generation, we utilized FEM to model the
island impingement and coalescence process. By minimizing
the sum of the positive strain energy and the associated reduction in interfacial energy, we determined the equilibrium
configuration resulting from island coalescence as a function
of island radius. The magnitude of the average tensile stress
calculated using FEM is more consistent with experimental
measurements than the stresses calculated using the Nix
Clemens crack-closure model.
From additional FEM modeling, the traction imposed at
the island-substrate interface was found to strongly affect
both the stress in the film and the measurable substrate curvature resulting from island coalescence. When islandsubstrate sliding was allowed, two types of island coalescence behavior were considered. At low substrate coverage,
island sliding is relatively unconstrained, and coalescence
produces very small, slightly compressive stresses. Furthermore, load transfer between the film and substrate cannot
occur if perfect sliding is assumed, and no curvature evolution will result. At higher substrate coverage, the movement
of the dislocation-like entities may be opposed by shear
stresses from previous coalescence events. If sliding of an
island is inhibited, tensile stresses will results from subsequent coalescence events.
We have also performed in situ wafer curvature measurements during deposition of Ag thin films deposited on
oxidized silicon substrates. Films were deposited at different
temperatures to study the kinetics of stress evolution during
deposition. TEM and SEM were used to correlate the evolving microstructure with features of the stress-thickness
curves.
Computer simulations were developed to model the process of film growth through continuous nucleation of isolated
islands that grow to impingement, and eventually form a
continuous polycrystalline film. Using the FEM results for
stress generation and a microstructure-dependent stress relaxation model, the simulation reproduced the qualitative
trends observed experimentally. Comparisons of measurements and the simulation suggest that the delayed onset of a
measurable stress-thickness until a relatively large nominal
film thickness is reached can be attributed, at least in part, to
sliding at the island-substrate interface. The magnitude of the
measured maximum stress-thickness decreased, and occurred
at a larger film thickness, with increasing deposition temperature, consistent with the stress generation model for a
grain size that increases with increasing deposition temperature, as observed in TEM micrographs. The measurements of
the slopes of the measured stress-thickness curves at different temperatures show that once a film is continuous, the
stress relaxation rate decreased with increasing deposition
temperature. Although the proposed stress relaxation mechanism is thermally activated, the slower relaxation rate after
continuity reflects the strong film-thickness dependence of
the stress relaxation mechanism. While a compressive stress
component could be added to the simulation to account for
the compressive stresses measured at large thicknesses, our

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7088

model for tensile stress generation and relaxation, in its current form, captures the general characteristics of the experimentally measured tensile stress evolution during deposition
of Ag thin films.
ACKNOWLEDGMENTS

The authors would like to thank M. J. Kobrinsky for

useful insights and Paul Kotula for TEM. This work was
supported by the NSF through Contract No. DMR-9710139.
Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States
Department of Energy under Contract No. DE-ACO494AL85000.
K. Kinosita, Thin Solid Films 12, 17 1972.
R. Koch, J. Phys.: Condens. Matter 6, 9519 1994.
3
R. W. Hoffman, Thin Solid Films 34, 185 1976.
4
K. Maki, Y. Nakjima, and K. Kinosita, J. Vac. Sci. Technol. 6, 622
1969.
5
J. D. Wilcock, D. S. Cambell, and J. C. Anderson, Thin Solid Films 3, 13
1969.
1
2

Seel et al.

J. Appl. Phys., Vol. 88, No. 12, 15 December 2000

R. Abermann and R. Koch, Thin Solid Films 66, 217 1980.

G. Thurner and R. Abermann, Thin Solid Films 192, 277 1990.
8
D. Winau, R. Koch, A. Fuhrmann, and K. H. Rieder, J. Appl. Phys. 70,
3081 1991.
9
D. W. Pashley, and M. J. Stowell, J. Vac. Sci. Technol. 3, 156 1966.
10
W. D. Nix and B. M. Clemens, J. Mater. Res. 14, 3467 1999.
11
C.-H. Chiu and H. Gao, Int. J. Solids Struct. 30, 2983 1993.
12
H. Gao, L. Zhang, W. D. Nix, C. V. Thompson, and E. Arzt, Acta Mater.
47, 2865 1999.
13
J. F. Turlo, Ph.D. thesis, Stanford University, Stanford, CA, 1992.
14
M. D. Thouless, Acta Metall. Mater. 41, 1057 1993.
15
M. J. Kobrinsky and C. V. Thompson unpublished.
16
J. A. Floro, E. Chason, and S. R. Lee, Mater. Res. Soc. Symp. Proc. 406,
491 1996.
17
G. G. Stoney, Proc. R. Soc. London, Ser. A 82, 172 1909.
18
P. A. Flinn, D. S. Gardner, and W. D. Nix, IEEE Trans. Electron Devices
ED34, 689 1987.
19
C. V. Thompson, J. Mater. Res. 14, 3164 1999.
20
H. J. Frost and C. V. Thompson, Acta Metall. 35, 529 1987.
21
F. A. Doljack and R. W. Hoffman, Thin Solid Films 12, 71 1972.
22
M. J. Kobrinsky and C. V. Thompson, Appl. Phys. Lett. 73, 2429 1998.
23
R. Abermann and R. Koch, Thin Solid Films 129, 71 1985.
24
R. C. Cammarata, Prog. Surf. Sci. 46, 1 1994.
25
C. R. Henry, Cryst. Res. Technol. 33, 1119 1998.
6
7

Downloaded 21 Oct 2005 to 133.11.199.17. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp