Full text of DOI:10.1557/JMR.2002.0394

Growth of electromigration-induced hillocks in
Al interconnects
J.A. Nucci, A. Straub, E. Bischoff, E. Arzt, and C.A. Volkerta)
Max-Planck-Institut fu¨r Metallforschung, Heisenbergstraße 3, D-70569 Stuttgart, Germany
(Received 6 June 2002; accepted 5 August 2002)
Electromigration-induced hillock growth in polycrystalline Al segments was
extensively investigated. Hillocks composed of columnar grains grew near the anode
by epitaxial Al addition at the interface between the Al and underlying TiN layer,
which pushed up the original Al film. The hillocks rotated away from their initial
(111) out-of-plane orientation in a manner consistent with the physical rotation of the
hillock surface. Wedgelike and rounded hillocks were observed, and their formation is
explained by the interaction between grain extrusion and grain growth. Trends
elucidated by review of both thermal- and electromigration-induced hillock studies can
be explained by the mechanisms identified in this work.
I. INTRODUCTION
or multiple grains and no voids or pores were found
underneath them. However, only thermal annealing
produces round hillocks surrounded by thin grooved
regions or moats,4,9 which are believed to result from
tensile stress relaxation during cooling. Electromigra-
tion-induced hillocks that appear extruded10–13 are likely
the early stages of whisker formation. At the later stages
of electromigration testing, the hillocks are larger and
exhibit more complex and fanciful shapes.13 Since elec-
tromigration involves longer range atom transport than
occurs during thermal cycling, it is not surprising that
electromigration-induced hillocks tend to be larger than
thermal hillocks.
Both electromigration- and thermal-induced hillocks ei-
ther nucleate as new, randomly oriented grains at grain
boundaries or triple points located on top of the Al
film3,5,6,14,15 or they grow from the lower Al inter-
face,2,8,10,11 which can occur epitaxially.2,10 In both cases,
the crystallographic orientations of the grains compos-
ing the hillocks deviates considerably from the (111) out-
of-plane texture typical of the surrounding grains.3,4,7,14,15
Links between hillock growth and lateral grain growth
have been identified for both thermal and electromigration-
induced hillocks. Gladkikh et al. found that electromigration-
induced hillock growth was associated with grain boundary
migration where the hillock border coincided with the
moving boundary.12 Several studies, including this one,
revealed that grain boundaries bordering hillocks often
tilted.9,12 In a recent article by Kim et al., grain boundary
migration was shown to enhance thermal hillock forma-
tion by at least a factor of four.16
Passing current through an interconnect generates a
net flux of atoms, typically in the direction of the electron
flow, which is called electromigration. Divergences in
this current-induced atom flux lead to regions of atom
accumulation and depletion, which in turn lead to the
local generation of mechanical stresses. Particularly large
stresses are created at interconnect segment ends, and it
is generally accepted that voids result from tensile
stresses generated at the cathode end while whiskers, and
more commonly, hillocks, form in response to compres-
sive stresses generated at the anode end. If these stresses
exceed a critical value,1 the resulting damage can lead to
the electrical failure of interconnects.
Hillocks and voids that compromise metallization re-
liability can also form as a result of the thermal stresses
generated by annealing a metal film or line that is rigidly
attached to a silicon substrate. Since the thermal expan-
sion coefficient of the metal greatly exceeds that of the
silicon, heating induces compressive stresses and thermal
hillock formation while cooling induces tensile stresses
and stress-induced void formation.
Both electromigration- and thermal-induced hillocks
in Al lines and films have been studied and their mor-
phologies and crystallographic orientations have been
reported in literature. At the early stages of electromi-
gration testing and after annealing, whiskers and fairly
simple shapes, such as rounded2–8 and wedge-shaped
hillocks4–6,9 have been reported. Hillocks with flat sur-
faces inclined to the film surface5,9–11 as well as those
with terraced surfaces7 have also been reported. For all
morphologies, the hillocks consist of either a single grain
Formation and growth mechanisms for hillocks are a
function of the active diffusion pathways and both the
availability and efficiency of these pathways depend on
the sample fabrication, geometry, and microstructure.
Thus, it is not surprising that various mechanisms,
a)Address all correspondence to this author.
e-mail: volkert@mf.mpg.de
J. Mater. Res., Vol. 17, No. 10, Oct 2002
© 2002 Materials Research Society
2727
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
such as grain boundary diffusion,4,6,17 coupled grain
Transmission electron microscopy (TEM) cross-
sectional samples of electromigration hillocks were pre-
pared using the FIB. A tungsten layer was initially
deposited on the sample surface in the FIB as protection
from further Ga ion beam damage. TEM lamellae ap-
proximately 20 ␮m long × 4 ␮m high × 250 nm thick
were cut from the desired region of the sample. The
lamellae were then transferred onto a TEM grid using a
micromanipulator under an optical microscope. The grids
were carbon coated to better secure the specimens.
Images were obtained using an energy filtering TEM at
120 kV with a LaB6 filament.
boundary sliding and interface diffusion,18 coupled
grain boundary diffusion and surface diffusion,9,19 dis-
location creep,4 and Nabarro-Herring creep,20 have all
been proposed to explain hillock formation and growth in
thin metal films and interconnects.
In this paper, we thoroughly examine electromigration-
induced hillock formation in unpassivated, polycrystal-
line aluminum interconnects. The goal of this study is to
elucidate which mechanisms contribute to and control
the hillock formation process. It will be argued that
subtle differences in the relative importance of the dif-
ferent mechanisms can lead to very different hillock
morphologies as well as to different values of the critical
stress for electromigration-induced damage formation.
III. RESULTS
The as-fabricated Al segments were of uniform width
and height and consisted of columnar grains. ␪-2␪ x-ray
scans of films from which the samples were fabricated
showed only a (111) peak, indicating that the films were
strongly (111) textured, and rocking curves showed that
the (111) direction lay within 5° of the film normal. FIB
and TEM cross sections of the samples revealed that
small (approximately 100 nm) TiAl3 precipitates formed
at the Al/TiN interface during annealing, probably by
reaction of Al with excess Ti in the TiN film. FIB analy-
sis revealed an average grain size slightly larger than the
film thickness for the as-deposited films and showed that
annealing for 1 h anneal at 400 °C produced grain growth
only in the 800-nm-thick films. For example, the as-
deposited and annealed 100-nm-thick films had an aver-
age grain size of 0.12 ␮m while the 800-nm-thick films
had an average grain size of 1.02 ␮m in the as-deposited
condition and 1.49 ␮m after annealing.
The FIB images also revealed a network of surface
grooves which plainly coincided with the grain bound-
aries in all but the annealed 800-nm-thick films. From
this observation, it is believed that the grooves formed at
the grain boundaries during or immediately following
deposition and did not move during subsequent anneal-
ing since they are replicated in the native oxide layer.
Therefore, in all but the 800-nm annealed segments, the
grooves marking the grain boundary positions before
electromigration testing were used to look for evidence
of grain growth during electromigration.
II. EXPERIMENTAL
The structures used for electromigration testing con-
sisted of 10-␮m-wide Al segments of various lengths
(from 6 to 26 ␮m) on a continuous 10-␮m-wide TiN/Ti
runner. They were fabricated from films of Ti (45 nm
thick), TiN (200 nm thick), and pure Al (100, 200, 400,
600, or 800 nm thick), which were sputter deposited onto
oxidized (100)Si wafers at 225 °C and then patterned into
electromigration test structures using electron beam li-
thography in a scanning electron microscope (SEM) and
reactive ion etching. After fabrication, the structures
were annealed in air at 400 °C for 1 h to stabilize the
grain structure. The samples were then electromigration
tested in a needle probe station at 225 °C in air using a
current density of 1 MA/cm2 for 4 to 50 h.
The sample microstructure and electromigration dam-
age were characterized using a number of methods. Fo-
cused ion beam (FIB) microscopy with a 30 keV Ga
beam was used to image the Al grain structure and to cut
cross sections through both undamaged segments and
hillocks. The cross sections were subsequently imaged in
the FIB or in a field emission scanning electron micro-
scope (FESEM) at 3 keV. The FESEM was also used to
closely examine hillock morphology during periodic in-
terruptions of electromigration testing. Atomic force
microscopy (AFM) was used to study the hillock mor-
phology and surface structure after testing was com-
pleted. The AFM was run in tapping mode with a
5–10-nm-radius, 10–15-␮m-high, etched silicon tip.
Global texture data was collected on unpatterned films
by ␪-2␪ x-ray scans in a diffractometer. Localized crys-
tallographic information within segments was gathered
using Electron backscatter diffraction (EBSD) in a SEM.
Grain and hillock orientations were monitored by manual
EBSD measurements taken from selected locations both
before electromigration testing and during periodic inter-
ruptions of electromigration tests. The EBSD determined
orientations are accurate to approximately 1°.
After several hours of electromigration testing at
1 MA/cm2, damage was observed in the longer segments,
as shown in Fig. 1. In this figure and those that follow,
the x direction is along the segment length (direction
of current flow) and the y direction is across the seg-
ment width. Al depleted from the cathode end of the
segments accumulated as hillocks at the anode ends.
While hillocks mostly formed at the segment edge, they
were also observed as a far as 10 ␮m from the anode.
That the shorter segments were not damaged is due to
the existence of the well-known “critical length” for
2728
J. Mater. Res., Vol. 17, No. 10, Oct 2002
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
electromigration damage formation.1 Upon further test-
material added to the hillock during each temperature
interval exhibited a different contrast in the FIB. The top
stripe corresponds to the original film thickness, and the
next three stripes correspond to the material added at
225 °C (second stripe from the top), 250 °C, and 275 °C
(bottom stripe). The different contrast stripes reveal
the progression of the hillock shape during growth and
show that growth proceeded by atom addition at the Al/
TiN interface. The cross-sectional FIB image in Fig. 3
provides additional support for hillock growth from the
Al/TiN interface; small TiAl3 particles, which were lo-
cated all along the Al/TiN interface, were pushed up-
wards as Al atoms were added to the interface during
hillock growth.
Channeling contrast in the FIB also revealed that the
atoms are epitaxially added to the growing hillock, as
shown in the cross-sectional FIB images in Figs. 3, 4(b)
and 5. The hillocks in Figs. 3 and 4 are composed of
single grains while the hillock in Fig. 5 is composed
of several columnar grains. In general, small hillocks
are single grained and larger hillocks are composed of
either a single, large grain or a few moderately sized,
columnar grains. Often, the grain boundaries bordering
the hillocks, such as those in Figs. 3 and 4(b), are tilted
away from the hillock. Grain boundaries within undam-
aged regions are generally vertical.
ing of the structure in Fig. 1, all of the longer seg-
ments shortened to roughly the critical length (13 ␮m at
1 MA/cm2) by cathode depletion, and electromigration
stopped. The total hillock volume was larger for longer
segments, as expected, since the amount of transported
material was also larger. Depletion generally occurred
uniformly at the cathode edge accompanied by the thin-
ning of individual grains near the cathode end.21 Addi-
tional studies that examine the influence of the sample
geometry and testing conditions on the critical product
for these samples are reported by Straub.22
Figure 2 shows a FIB microscope image of a hillock
formed directly at the anode end of a segment. This
sample was first tested at 225 °C, then 250 °C, and fi-
nally at 275 °C. Presumably because of a surface reac-
tion, such as Al oxidation or reaction of etch residues, the
That hillock growth can be accompanied by lateral
grain growth was confirmed by comparing the grain
boundary positions with the grooves at the film surface.
For example, in Fig. 4, grooves are evident at the hillock
surface while the grain boundaries are located at the edge
of the hillock, indicating that the grain boundaries
moved during hillock formation. However, hillock growth
was not always accompanied by grain growth, and no
FIG. 1. FIB image of 600-nm-thick Al segments after testing at 1 MA/
cm2 and 225 °C for 50 h.
FIG. 3. FIB image of a cut through a hillock in an 800-nm-thick Al
segment sectioned along the line length. The original Al/TiN interface
is marked by TiAl3 precipitates, which were pushed up during hillock
growth. Since the contrast between the hillock and the adjacent grain
is minimal, the boundary between them was drawn in for easier
viewing.
FIG. 2. FIB image of an as-grown hillock formed at the end of a
600-nm-thick Al segment. Material accumulation at the Al/TiN inter-
face pushed up the original Al film (indicated by the arrow). The three
layers under the original film were added to the hillock as a result of
sequential electromigration testing at 225, 250, and 275 °C.
J. Mater. Res., Vol. 17, No. 10, Oct 2002
2729
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
evidence of grain growth during electromigration was
8(b). The surrounding, rounded regions of the often
larger, Mode II hillocks indicate the extent of lateral
grain growth as hillock growth proceeded. Of the
36 hillocks investigated after 10 h of testing or less,
27 (75%) showed evidence of lateral hillock growth.
Hillocks tested for longer times displayed more varied
morphologies ranging from simple, rounded hillocks, as
in Figs. 2 and 5, to more unusual shapes, as shown in
Figs. 1 and 4.
The surface of the hillock shown in Fig. 7(a) was care-
fully examined by AFM. Figure 7(b) shows an AFM im-
age of this hillock surface, and Fig. 7(c) contains a line
scan across part of this region. The striations always lie
perpendicular to the apparent growth direction and result
from 5- to 10-nm-high steps on the sample surface. They
are not consistent with the appearance of surface slip
bands, since one would expect a series of roughly
found in either subcritical length segment, such as in the
three shortest segments in Fig. 1, or in undamaged re-
gions of the longer segments. Several FIB observations
suggest the possibility of grain growth near depletion
sites at the cathode; however the observations are not
conclusive.
FESEM images of hillocks tested for 10 h or less re-
vealed two distinct growth morphologies that may be
correlated to the absence or occurrence of lateral grain
boundary motion during hillock growth. Flat-topped,
wedge-shaped hillocks appeared to extrude from the seg-
ment at a constant angle (Mode I), as shown in Fig. 6. If
this flat defines the grain size at hillock formation, then
Mode I hillocks grow without lateral grain growth. Other
hillocks contained a flat region surrounded by a curved,
striated surface (Mode II), as shown in Figs. 7(a) and
FIG. 5. FIB cross section showing a hillock composed of several
columnar grains.
FIG. 4. FIB images of an electromigration-induced hillock in a 600-
nm-thick film. (a) Grooves on the film surface mark the original grain
boundary locations. (b) A cross section through the hillock shows that
lateral grain growth accompanied hillock growth. Grain growth did not
occur in the undamaged regions of the interconnect.
FIG. 6. SEM micrograph of a flat-topped, wedge-shaped hillock in an
800-nm-thick film after 10 h of electromigration testing. The flat,
which is indicated by the arrow, is part of the original Al surface, and
the grain appears to have extruded from the film without accompany-
ing lateral grain growth.
2730
J. Mater. Res., Vol. 17, No. 10, Oct 2002
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
straight, parallel steps resulting from slip activity on the
The crystallographic orientations of the hillocks dur-
ing electromigration were examined using EBSD. Map-
ping the anode ends of several segments prior to electron
microscope testing revealed predominantly (111) tex-
tured grains, consistent with the x-ray diffraction results.
However, EBSD analysis performed during periodic in-
terruptions of electromigration testing often showed clear
rotations in hillock orientation. In general, the hillocks
that grew by pure extrusion (Mode I) showed no evi-
dence of crystallographic rotation, presumably after an
initial rotation, whereas those which were accompanied
by lateral extension (Mode II) rotated away from a (111)
out-of-plane orientation during growth. For example, a
(111) pole figure in Fig. 8(a) shows crystallographic ori-
entations of the lower hillock in Fig. 8(b), measured after
4, 6, and 8 h of electromigration testing. The (111) di-
rection of the hillock rotated away from its original,
plane with the highest Schmid factor. However, they are
consistent with fracture of the native oxide, which has a
thickness of approximately 7 nm, based upon Auger
depth profiles for oxygen.
Several hillocks were cross sectioned using the FIB
and investigated in the TEM. The samples were exam-
ined using two beam conditions at damaged and intact
regions of the line for better observation of dislocations.
Similar low dislocation densities of approximately
1013/m2 were measured in both regions. No evidence of
activated slip systems was found.
FIG. 7. (a) SEM micrograph of a hillock formed at the line edge. (b)
AFM image of the region corresponding to the white box in (a). The
fine surface structure appears to result from fracture of the overlying
native oxide layer as the hillock grows. The dashed line denotes the
direction of the AFM line scan shown in (c), which reveals 5–10 nm
step heights, consistent with the native oxide thickness. The arrows
denote a 6.4-nm-high step.
FIG. 8. (a) (111) pole figure for a hillock whose orientation was
measured after 4, 6, and 8 h of electromigration testing. The dashed
great circle denotes the plane in which the hillock normal rotates. The
corresponding rotation axis, roughly parallel to the y axis of the line,
is shown as the cross in the pole figure. (b) The SEM image for this
hillock shows a rigid body rotation about roughly the same axis.
J. Mater. Res., Vol. 17, No. 10, Oct 2002
2731
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
approximately out-of-plane orientation about an axis par-
bottom Al/TiN interface indicates that Al atoms are
somehow transported from the grain boundaries into this
interface and added to the growing hillock. Lattice and
pipe diffusion are too slow to account for this transport.
This leaves two possible mechanisms: combined grain
boundary and interface diffusion18 or the glide of dislo-
cations from the grain boundaries to the Al/TiN interface,
where they are then annihilated.27
allel to the segment end. SEM images and AFM scans of
several hillock surfaces confirmed that the normal to the
flat region of the hillock surface often rotated in the same
direction and by roughly the same amount as the (111)
crystallographic direction.
IV. DISCUSSION
If the stresses generated by transport of atoms along
the grain boundaries were relieved by atomic diffu-
sion from the grain boundaries directly into the Al/TiN
interface, it is expected that the flux of atoms into the
Al/TiN interface would be largest where the normal com-
pressive stresses in the grain boundaries are largest.
However, there must be some threshold for addition and
removal of atoms from this interface, otherwise stresses
in Al films would relax to zero. It is generally known that
Al films sustain stresses at elevated temperatures over
extended periods of time, and this was verified specifi-
cally for the films used in this study using wafer curva-
ture measurements.
It is possible, although less likely, that dislocation
glide may be responsible for the observed hillock growth
process. Once sufficient stresses had been generated at
the anode end, dislocation glide would occur and lead to
a shape change of the grain. However, TEM revealed
similar, low dislocation densities in both damaged and
undamaged regions. Considerable dislocation activity
would be necessary to account for the large strains re-
quired for hillock growth and it is reasonable to expect
that slip planes or glide bands would be observed there.
However, no evidence of slip planes was seen in the
TEM images or from AFM of the hillock surface. Also,
the observed crystallographic rotations of grains during
hillock growth are not consistent with those predicted by
dislocation glide in single crystals deformed in con-
strained geometries.28
Once atoms find their way to the Al/TiN interface,
likely by combined grain boundary and interface diffu-
sion, they are epitaxially added to an Al grain, generating
stresses normal to the interface, as well as stresses in
the overlying oxide. These stresses can be relieved by
grain boundary sliding and fracture of the native oxide,
which leads to the formation of small, often wedge-
like extruded regions such as the Mode I hillocks
observed in this study. If atoms are nonuniformly added
to the bottom interface, then the hillocks would crystallo-
graphically and physically rotate as they grow, as
shown schematically in Fig. 9(b). The anode end of
the line before hillock formation is shown for compari-
son in Fig. 9(a). If lateral growth of the grain composing
the hillock occurs simultaneously, then the combina-
tion of extrusion and grain growth can lead to the for-
mation of rounded hillocks. The small, discrete steps
observed on the rounded hillock surfaces may result
The local accumulation of Al to form electromigration-
induced hillocks requires the relatively long range trans-
port of atoms. Grain boundaries are the dominant
diffusion path in polycrystalline Al lines at typical elec-
tromigration testing temperatures. In a study on bamboo
segments patterned from the same films as used in this
study, diffusion along the Al/TiN interface was shown to
be somewhat slower than grain boundary diffusion, and
lattice diffusion and dislocation core diffusion were ruled
out10 as contributing to damage formation. Thus, it is
concluded that the Al in these polycrystalline segments
is likely transported from one end of the segment to the
other along grain boundaries. Since the interface between
pure Al and its native oxide is known to be a poor dif-
fusion path,17 the Al/TiN interface is most likely the
main diffusion pathway for the bamboo segments manu-
factured from the same film stack. The Al/TiN interface
may also contribute to shorter range transport in the poly-
crystalline samples.
Given the stress gradient induced by electromigra-
tion,23,24,25 it is not surprising that depletion occurs at
the cathode end and hillocks form near the anode end
of the segments. That compressive stress is relieved by
hillock formation, rather than uniform thickening of
the film, is shown to be energetically favorable except
for very small film stresses.18 The fact that hillocks
are occasionally observed as far as 10 ␮m from the
anode is presumably due to local variations in the stress
created by microstructural-induced divergences in the
electromigration flux. These local variations in the stress
are superimposed on the macroscopic stress gradient
along the segment and presumably determine the speci-
fic sites for hillock formation. The source of these
flux divergences is a combination of the variations in
grain boundary diffusivity due to misorientations be-
tween adjacent grains26 and the component of the elec-
tric current along the grain boundary. Although the
film is (111) textured, the deviations from perfect
(111) out-of-plane texture and the random in-plane
texture are sufficient to expect variations in grain bound-
ary diffusivities approaching those of random grain
boundaries.
During electromigration, atoms are transported along
the grain boundaries and presumably accumulate in the
grain boundaries at the anode end of the segment. The fact
that material is added to the growing hillocks at the
2732
J. Mater. Res., Vol. 17, No. 10, Oct 2002
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
from a repeated process of grain boundary sliding and
advancing hillocked area to keep the total grain boundary
area as small as possible.30 However, this idea is not
consistent with the fact that many of the grain boundaries
at the edge of hillocks are inclined [See Figs. 3 and 4(b)]
since a vertical grain boundary has a smaller area than an
inclined boundary.
accompanying native oxide fracture above the grain
boundary followed by grain boundary migration, as
depicted in Fig. 9(c). This picture is consistent with
AFM observations of the hillock surfaces. If atoms
are added under several grains, then multigrain hillocks
may form, with or without accompanying grain growth.
It is difficult to visualize the formation of rounded
hillocks from a single grain without simultaneous
grain growth, since the necessary deformation of the
overlying film would lead to nonuniform rotations and to
subgrain formation, for which there is no evidence in
this study.
Hillock growth was accompanied by grain growth for
the majority of hillocks studied, and grain growth may be
caused by a number of factors. Hillock formation likely
reduces local stresses in the grains composing the
hillock, which creates a chemical potential driving force
for growth of these grains at the expense of their more
highly stressed neighbors.29 It has also been argued
that the grain boundaries will be pushed ahead of the
It is easy to imagine that changes in the relative im-
portance of the possible diffusive pathways can lead to
different hillock formation mechanisms and morpholo-
gies. In particular, if interface diffusion is very slow or
the incorporation of atoms at the bottom and top inter-
faces is effectively turned off, then excess atoms accu-
mulated in the grain boundaries have no place to go to
relieve the stresses. It is reasonable to expect that the
overlying native oxide may fracture above a grain bound-
ary or a triple line and Al may extrude out to form a
hillock on top of the original film. This is exactly
the hillock morphology observed in both thermal and
electromigration hillock studies of Al films deposited
directly on SiO2. In contrast, in many samples with an
underlying Al/refractory metal interface, both electromi-
gration- and thermal-induced hillocks often form by ma-
terial addition at this interface, which is presumably a
more efficient diffusion pathway than the Al/SiO2 inter-
face. While most thermal hillock studies were conducted
on Al films deposited directly onto oxide, most electro-
migration studies were performed on Al deposited onto a
refractory metal layer. However, the trend of hillock
formation from the upper or lower Al interface likely
depends more upon the Al interfaces present than on the
nature of the induced stress. As further evidence of
the importance of the Al interfaces, another study
showed that changing the upper interface from Al/Al2O3
to a pure Al surface changed the dominant stress relaxa-
tion mechanism altogether.17 When ultrahigh vacuum
(UHV) fabrication and in situ thermal cycling were used
to suppress native oxide formation, grain boundary com-
pressive stresses were relieved by Al diffusion out of the
grain boundaries and onto the sample surface. The same
films annealed under non-UHV conditions formed
hillocks.
The correlation between hillock shape and grain
growth is supported by other studies of hillocks.
Rounded, often single-grained hillocks bounded by in-
clined grain boundaries have also been reported in the
literature when hillock growth and grain boundary mo-
tion occurred simultaneously.9,12 Wedge-shaped hillock
formation in the absence of lateral grain growth has been
reported by Wang et al.4 for a 0.1-␮m-grain size film, in
which the grain boundaries were pinned by oxygen im-
purities. Thermal cycling of these films deposited on
TiN produced hillocks consisting of approximately
200 grains, and the Al was added epitaxially to each
grain from the Al/TiN interface. Despite the differences
in film composition and grain size between the samples
FIG. 9. (a) Schematic diagram of three grains at the anode end of an
Al interconnect segment. The dashed lines represent the build-up of
compressive stresses at the grain boundary prior to hillock formation.
(b) Schematic diagram of Mode I hillock growth. The original grain,
labeled 1, is initially rotated (step 2) and then extruded at constant
angle without lateral grain growth (steps 3–5) as a result of uniform Al
addition at the Al/TiN interface and grain boundary sliding. (c) Sche-
matic diagram of Mode II hillock growth. The original grain forms a
rounded hillock by an iterative process of nonuniform Al addition at
the Al/TiN interface, grain boundary sliding (even numbered steps),
and lateral grain boundary motion (odd numbered steps).
J. Mater. Res., Vol. 17, No. 10, Oct 2002
2733
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
used for this study and Wang’s, the processes of epitaxial
hillock growth without lateral grain growth produced
wedge-shaped hillocks in both cases.
interface diffusion. The epitaxial Al addition at the Al/
TiN interface under a grain or grains pushes the original
material up, leading to grain boundary sliding, native
oxide fracture, and extrusion. If Al is nonuniformly
added at the interface, both physical and crystallographic
rotation of the hillock occurs about roughly the same
axis. The two basic hillock morphologies observed are
proposed to result from hillock growth either without, or
more commonly with, accompanying lateral grain
growth. Wedgelike extrusions are alleged to form when
the grain boundaries bordering hillocks remain stationary
and the grain extrudes at a constant angle by grain bound-
ary sliding. In contrast, rounded hillocks are proposed
to form by an iterative process of grain boundary migra-
tion and grain boundary sliding. The same hillock shapes
would be expected in alloyed Al interconnects, pro-
vided the addition of the impurity does not affect
the mechanisms involved in hillock growth. All of these
mechanisms, with the possible exception of native oxide
fracture, are most likely also involved in grain depletion
and thinning at the cathode end of segments.
These observations were used to explain trends regard-
ing both electromigration and thermal hillocks and in
particular, to reveal the importance of the Al interface
character on hillock formation and growth. While these
results do not elucidate a single critical event or critical
stress necessary for hillock formation they do make clear
that the many different hillock morphologies observed
for both thermally and electromigration-induced hillocks
likely reflect subtle changes in the nature of the various
mechanisms involved in hillock formation.
The mechanisms and processes involved in hillock
formation may be the same ones that determine the criti-
cal stress for damage formation during electromigration.
The magnitude of this critical stress is approximately
200 MPa.25 It is tempting to attribute the critical stress to
dislocation glide, since extensive modeling of thin films
during thermal cycling has attributed stresses of this
magnitude to the generation or glide of dislocations in
the small film dimensions.31 Other mechanisms, such as
creep and other diffusion controlled processes, have
much smaller thresholds. However, the idea that dislo-
cation glide is the controlling mechanism for hillock for-
mation in these lines can be ruled out based on the fact
that there is no evidence of slip bands, either in the
grains, as determined by TEM, or at their surface, as
measured by AFM, and on the fact that the crystallo-
graphic rotations of grains during hillock growth are not
consistent with those predicted by dislocation glide in
single crystals deformed in constrained geometries.28
Given that dislocation activity is likely ruled out, an-
other event with a threshold is necessary to explain the
critical stress for hillock formation. One possibility is
that atom addition at the Al/TiN interface is difficult and
requires that a critical chemical potential or stress be
exceeded.32 This would mean that the critical stress for
hillock formation would be extremely sensitive to the
exact nature of this interface. Another possibility is frac-
ture of the native oxide, which occurs at least once during
hillock formation. However, estimates of the fracture
stress, based on simple geometric considerations and as-
suming a grain boundary normal stress of a few hundred
MPa, indicate that the fracture stress must be fairly large,
on the order of a few GPa. What is clear, in any case, is
that just as samples have different hillock morphologies,
they will also have different critical stresses for hillock
formation.
ACKNOWLEDGMENTS
The authors gratefully acknowledge Ilan Blech for
valuable technical discussions, Birgit Heiland for TEM
sample preparation, Danielle Cantarutti for AFM analy-
sis, and Sabine Ku¨hnemann for FESEM work. This
project was partially support by the Deutsche For-
schungsgemeinschaft Leibniz program.
V. CONCLUSION
Hillock growth in pure Al segments on continuous
TiN runners was studied during electromigration. As a
result of the careful, periodic observation of the hillock
morphology, crystallographic orientation and surface
structure during electromigration testing, various proc-
esses and mechanisms associated with hillock formation
and growth were identified. Under the bias of the electric
current, Al atoms are transported by grain boundary dif-
fusion toward the anode end of the segment, where com-
pressive stresses normal to grain boundaries are
generated. The atoms are then transported from the grain
boundaries to the bottom interface, either by dislocation
glide or more likely, by combined grain boundary and
REFERENCES
1. I.A. Blech, J. Appl. Phys. 47, 1203 (1976).
2. H. Mori, H. Okabayashi, and M. Komatsu, Thin Solid Films 300,
25 (1997).
3. A. Buerke, H. Wendrock, T. Koetter, S. Menzel, K. Wetzig, and
A.V. Glasow, in Materials Reliability in Microelectronics IX, ed-
ited by C.A. Volkert, A.H. Verbruggen, and D.D. Brown (Mater.
Res. Soc. Symp. Proc. 563, Warrendale, PA, 1999), p. 109.
4. P. Wang, J. Hwang, A. Chuang, and F-S. Huang, Thin Solid Films
358, 292 (2000).
5. F. Ericson, N. Kristensen, and J-A. Schweitz, J. Vac. Sci. Tech-
nol. B 9, 58 (1991).
6. D. Gerth, D. Katzner, and M. Krohn, Thin Solid Films 208, 67
(1992).
2734
J. Mater. Res., Vol. 17, No. 10, Oct 2002
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35

J.A. Nucci et al.: Growth of electromigration-induced hillocks in Al interconnects
7. B.C. Martin, C.J. Tracy, J.W. Mayer, and L.E. Hendrickson, Thin
20. E. Glickman and M. Nathan, Microel. Eng. 50, 329 (2000).
21. R.A. Augur, R.A.M. Wolters, W. Schmidt, A.G. Dirks, and
S. Kordic, J. Appl. Phys. 79, 3003 (1996).
Solid Films 271, 64 (1995).
8. D-K. Kim, B. Heiland, W.D. Nix, E. Arzt, M.D. Deal, and
J.D. Plummer, Thin Solid Films 371, 278 (2000).
9. F.Y. Genin and W.J. Siekhaus, J. Appl. Phys. 79, 3561 (1996).
10. J. Bo¨hm, C.A. Volkert, R. Mo¨nig, T.J. Balk, and E. Arzt, J. Elect.
Mater. 31, 45 (2002).
22. A. Straub, Ph.D. Dissertation, University of Stuttgart, Stuttgart,
Germany (2000).
23. I.A. Blech and C. Herring, Appl. Phys. Lett. 29, 131 (1976).
24. M.A. Korhonen, P. Borgeson, K.N. Tu, and C-Y. Li, J. Appl.
Phys. 73, 3790 (1993).
11. J. Proost, L. Delaey, J. D’Haen, and K. Maex, J. Appl. Phys. 91,
9108 (2002).
25. P.C. Wang, G.S. Cargill, I.C. Noyan, and C-K. Hu, Appl. Phys.
12. A. Gladkikh, Y. Lereah, E. Glickman, M. Karpovsky, A. Palevski,
and J. Shubert, Appl. Phys. Lett. 66, 1214 (1995).
13. C. Witt, Ph.D. dissertation, University of Stuttgart, Stuttgart, Ger-
many (2000).
14. H. Takatsuija, K. Tsujimoto, K. Kuroda, and H. Saka, Thin Solid
Films 343–344, 461 (1999).
Lett. 72, 1296 (1998).
26. O.V. Kononenko, V.N. Matveev, and D.P. Field, J. Mater. Res.
16, 2124 (2001).
27. M.S. Jackson and C-Y. Li, Acta. Metall. 30, 1993 (1982).
28. R.E. Reed-Hill, Physical Metallurgy Principles, 2nd ed. (Litton,
Monterey, CA, 1973), p. 195.
15. R.A. Schwarzer and D. Gerth, J. Elect. Mater. 22, 607 (1993).
16. D-K. Kim, W.D. Nix, R.P. Vinci, M.D. Deal, and J.D. Plummer,
J. Appl. Phys. 90, 781 (2001).
17. C.Y. Chang and R.W. Vook, Thin Solid Films 228, 205 (1993).
18. P. Chaudhari, J. Appl. Phys. 45, 4339 (1974).
19. L.M. Klinger, L. Levin, and E.E. Glickman, in Materials Reliabil-
ity in Microelectronics V, edited by A.S. Oates, W.F. Filter,
R. Rosenberg, A. Lindsay Greer, and K. Gadepally (Mater. Res.
Soc. Symp. Proc. 391, Pittsburgh, PA, 1995), p. 271.
29. C.V. Thompson, J. Mech. Phys. Solids 44, 657 (1996).
30. F.Y. Genin, J. Appl. Phys. 77, 5130 (1995).
31. W.D. Nix, Metall. Trans. 20A, 2217 (1989).
32. E. Arzt, M.F. Ashby, and R.A. Verrall, Acta. Metall. 31, 1977
(1983).
J. Mater. Res., Vol. 17, No. 10, Oct 2002
2735
http://journals.cambridge.org
Downloaded: 13 Mar 2015
IP address: 138.251.14.35