Superconformal electrodeposition of silver in submicrometer features

Article abstract of DOI:10.1149/1.1490357

The generally of the curvature-enhanced accelerator coverage (CEAC) model of superconformal electrodeposition is demonstrated through application to superconformal filling of fine trenches during silver deposition from a selenium-catalyzed silver cyanide electrolyte. The CEAC mechanism involves (i) increase of local metal deposition rate with increasing coverage of a catalytic species adsorbed on the metal/electrolyte interface and (ii) significant change of local coverage of catalyst (and thus local deposition rate) in a submicrometer features through the changing area of the metal/electrolyte interface. Electrochemical and X-ray photoelectron experiments with planar electrodes (substrates) are used to identify the catalyst and obtain all kinetic parameters required for the simulations of trench filling. In accord with the model, the electrolyte yields optically shiny, dense films, hysteretic current-voltage curves, and rising current-time transients. Experimental silver deposition in trenches from 350 down to 200 nm wide are presented and compared with simulations based on the CEAC mechanism. All kinetics for the modeling of trench filling come from the studies on planar substrates. The results support the CEAC mechanism as a quantitative formalism for exploring morphological evolution during film growth.

Full text of DOI:10.1149/1.1490357

Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
C423
0
013-4651/2002/149͑8͒/C423/6/$7.00 © The Electrochemical Society, Inc.
Superconformal Electrodeposition of Silver
in Submicrometer Features
,z
T. P. Moffat,* B. Baker,* D. Wheeler, J. E. Bonevich, M. Edelstein, D. R. Kelly,
L. Gan, G. R. Stafford,* P. J. Chen, W. F. Egelhoff, and D. Josell
National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA
The generality of the curvature-enhanced accelerator coverage ͑CEAC͒ model of superconformal electrodeposition is demon-
strated through application to superconformal filling of fine trenches during silver deposition from a selenium-catalyzed silver
cyanide electrolyte. The CEAC mechanism involves ͑i͒ increase of local metal deposition rate with increasing coverage of a
catalytic species adsorbed on the metal/electrolyte interface and ͑ii͒ significant change of local coverage of catalyst ͑and thus local
deposition rate͒ in submicrometer features through the changing area of the metal/electrolyte interface. Electrochemical and X-ray
photoelectron experiments with planar electrodes ͑substrates͒ are used to identify the catalyst and obtain all kinetic parameters
required for the simulations of trench filling. In accord with the model, the electrolyte yields optically shiny, dense films, hysteretic
current-voltage curves, and rising current-time transients. Experimental silver deposition in trenches from 350 down to 200 nm
wide are presented and compared with simulations based on the CEAC mechanism. All kinetics for the modeling of trench filling
come from the studies on planar substrates. The results support the CEAC mechanism as a quantitative formalism for exploring
morphological evolution during film growth.
©
2002 The Electrochemical Society. ͓DOI: 10.1149/1.1490357͔ All rights reserved.
Manuscript submitted November 26, 2001; revised manuscript received February 8, 2002. Available electronically June 21, 2002.
Superconformal electrodeposition of copper in the Damascene
process for microelectronic fabrication is enabling a new generation
of integrated circuits. The use of copper interconnections has per-
mitted faster clock speeds, enhanced reliability, and lower process-
ing cost. Central to the success of the electrodeposition process is its
ability to yield void and seam-free filling of high aspect ratio fea-
tures. Empirically, such superfilling of trenches and vias with copper
results from more rapid metal deposition ͑growth͒ at the bottom of
the feature than at the sidewalls. Early models of superfilling of
copper in trenches assumed location-dependent growth rates derived
filling or void formation, and creation of overfill bumps, for filling
of trenches between approximately 350 and 100 nm wide and 500
nm deep.
8
-11
The CEAC mechanism as described is quite generic. Indeed, as
noted above, it has recently been used to explain superconformal
filling of fine features by iodine-catalyzed CVD of copper. Similarly,
there is no fundamental reason that superconformal filling by elec-
trodeposition should be limited to either copper or acid sulfate based
chemistries. This paper therefore examines the link between super-
conformal silver filling of fine features on patterned substrates and
hysteresis during current-voltage studies on planar substrates. The
catalyst in the additive-containing silver-cyanide electrolyte that was
used for these studies was identified using X-ray photoelectron spec-
troscopy ͑XPS͒. This paper presents two principal results: ͑i͒ it pre-
dicts superfilling conditions using only a CEAC-based model with
all kinetic parameters obtained from current-voltage and current-
time studies of deposition on flat electrodes, and ͑ii͒ it demonstrates
superconformal filling of trenches in the electrolyte under conditions
consistent with the predictions. While the electrical resistivity of
bulk silver is only slightly lower than that of bulk copper at room
temperature, the effective electrical resistivity of silver in fine fea-
tures like narrow lines has been noted to be substantially less than
that of similar copper lines due to surface scattering effects.16
1
,2
from diffusion-limited accumulation of only an inhibiting species.
Such models were unable to predict several key experimental
3
-6
observations. Understanding of the superfill phenomena has im-
proved substantially in the past two years. Electrolytes for the study
of superconformal electrodeposition of copper have been fully
disclosed.6 A correlation between efficacy for superconformal fill-
ing of fine features, hysteresis in cyclic current-voltage studies, and
,7
6
recrystallization of deposits has also been demonstrated. In addi-
tion, the curvature-enhanced accelerator coverage ͑CEAC͒ mecha-
nism has been used to quantitatively predict superconformal copper
8
-11
12
deposition in trenches.
and vias by electrodeposition as well as
1
3,14
by surfactant-catalyzed chemical vapor deposition ͑CVD͒.
The CEAC mechanism involves the gradual accumulation of a
metal-deposition-rate-accelerating species at the surface of the
growing metal. If there is a deposition-rate-inhibiting species, it is
presumed to be weakly bound and displaced, or altered, by the cata-
lyst. Superfill can occur without an inhibiting species ͑see Ref. 13,
Experimental
Details for chemical studies of planar specimens.—
Electrodeposition on freshly polished silver electrodes was exam-
ined in a commercial plating bath ͑Techni-Silver E, Lot no. f-3-0388
and F-1-11-4049, both from Technic, Inc. ͒ that are stated to contain
.34 mol/L KAg͑CN͒ and 2.3 mol/L KCN and two additional pro-
prietary components, a surfactant ͑i.e., wetting agent͒ and a bright-
ener or catalyst. Electrochemical experiments were performed using
a standard cell with three silver electrodes serving as the reference,
counter, and working electrodes, respectively. ͑All potentials are ref-
erenced to the immersed Ag wire and thus are cited as overpoten-
tials. The open-circuit potential corresponds to Ϫ0.670 V vs. the
saturated calomel electrode.͒ The silver working electrode was pol-
ished with a 320 grade SiC paper immediately before each experi-
a
14͒ but only if adsorption of the catalyst leads to significantly faster
0
2
deposition than occurs in its absence. All adsorbed species are pre-
sumed to remain on or float at the metal/electrolyte interface during
deposition ͑consistent with more recent work that has explicitly
1
5
quantified the effect of potential dependent additive consumption ͒.
Thus, local compression ͑or expansion͒ of adsorbed accelerator oc-
curs with the reduction ͑or increase͒ of surface area that accompa-
nies growth at points of high positive ͑or negative͒ curvature. Mod-
els based on the CEAC mechanism have been shown to yield
predictions that agree well with experimental results for supercon-
formal electrodeposition of copper over a wide range of processing
conditions. This includes a period of conformal growth, bottom-up
a
ment. The electrode was then masked with electroplater’s tape ͑3M ͒
leaving an exposed electrode area of 0.97 cm . The deposition reac-
tion was examined through cyclic current-voltage (i-␩) experiments
2
initiated from the open-circuit, or reversible, potential, swept to a
a
*
Electrochemical Society Active Member.
E-mail: thomas.moffat@nist.gov
Product names are included only for accuracy of experimental description. They do
z
not imply NIST endorsement.
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C424
Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
particular negative overpotential and then swept back. Deposition
was also investigated through studies of current-time transient i(t)
experiments at constant voltage performed over a similar range of
potentials.
The evolution of the chemistry of the growth interface was ex-
amined by XPS. For these experiments, 100 nm thick electron beam
evaporated copper films on silicon were used as substrates. Silver
was deposited at Ϫ0.35 V for various periods of time in the electro-
lyte. Upon emersion, the electrodes were rinsed with water and
transferred to ultrahigh vacuum ͑UHV͒ within 30 min. Binding en-
ergy data were referenced to internal copper and gold standards and
the values are reported relative to Cu (2p3/2) ϭ 932.7 eV,
Cu (3p3/2) ϭ 75.1 eV, and Au (4f7/2) ϭ 84.0 eV.
Details for filling studies with patterned specimens.—The super-
filling capacity of the electrolyte was examined by electroplating
silver in submicrometer trenches at different overpotentials. Speci-
mens for the filling studies were fabricated as follows. Electron
beam lithography was used to pattern trenches into polymethyl-
methacrylate that had been spin-coated on silicon substrates. All
trenches were Ϸ0.41 ␮m deep, with widths from 350 to 200 nm,
corresponding to aspect ratios ͑height/width͒ from Ϸ1.2 to 2. Each
specimen consisted of six groups of Ϸ100 ␮m long trench arrays,
with each array consisting of parallel trenches, including one of each
aspect ratio. A silver seed layer was deposited by electron beam
evaporation immediately prior to electrodeposition. Further details
on the patterning and seed layer deposition process can be found
Figure 1. A cyclic current-voltage study of the additive-containing electro-
lyte reveals hysteretic behavior. Sweep rate was 1 mV/s. Similar behavior is
also evident in the voltage dependence of the sampled current density data
͑data points at 20 and 1200 s taken directly from i(t) studies in Fig. 4͒. The
1200 s data, which corresponds to saturated or fully catalyzed behavior, is
seen to be consistent with the return sweep of the i-␩ curve. The 20 s data is
taken as a measure of the deposition rate for an uncatalyzed surface.
6
elsewhere. A relatively thick 130 nm seed was used in an attempt to
reduce electrolyte-induced damage of the patterned poly͑methyl
methacrylate ͑PMMA͒ substrate; this corresponds to coverage of
Ϸ10 nm on the sidewalls of the trenches. Nonetheless, pinholes or
voids in the seed layer often resulted in distortion of the underlying
PMMA resist due to reaction with the alkaline electrolyte. Interpre-
tation of images is subject to this concern.
while the Ag (3d5/2) photoelectron peak is at 368.1 eV, suggesting
the formation of silver selenide at the interface. The accumulation of
selenium during silver deposition is shown in Fig. 2b. The ratios of
the time-dependent integrated intensities of the selenium L M M
3
45 45
Following electrodeposition, the patterned wafers were cross sec-
tioned and mechanically polished to 0.5 ␮m grit. Residual polishing
damage was removed from the specimen surfaces using an argon ion
flux prior to imaging on a field-emission scanning electron micro-
scope ͑SEM͒. A subset of specimens was also prepared for imaging
by transmission electron microscope ͑TEM͒.
Auger and the Ag (3d5/2) photoelectron peaks are plotted with the
current transient data in Fig. 2c. There is a clear correlation between
the silver deposition rate ͑i.e., current density͒ and the selenium
coverage. The saturation selenium coverage is estimated to be Ϸ0.3
to 0.4 of a monolayer relative to a clean silver surface. ͓This value
was derived by taking the sensitivity factor for the Se L M M to
3
45 45
be 1.29 relative to 2.25 for the Ag (3d5/2), the values taken or
derived from standard spectra in Ref. 19 and a 1.5 nm escape depth
for the Ag (3d5/2), given its Ϸ1188 eV kinetic energy.͔
Studies with Planar Substrates
Kinetics of the deposition reaction experiments on planar
substrates.—An example of the i-␩ characteristics of the additive-
containing electrolyte is shown in Fig. 1. The solid curve was ac-
quired at the 1 mV/s sweep rate. Individual data points extracted
from i(t) data curves obtained at the indicated voltages are also
included. Hysteretic behavior is clearly evident in the cyclic i-␩
curve ͑a much smaller hysteresis loop observed with an additive-
free electrolyte not shown here is associated with increasing surface
area as the associated deposits were optically rough͒. Because elec-
trodeposits produced in the additive-containing electrolyte were
dense and near specular in nature, the hysteresis visible in Fig. 1 is
not caused by changing surface area. Rather, it reflects actual accel-
eration of the deposition rate caused by accumulation of a catalytic
species on the surface during the cycle ͑note that this conclusion is
supported by the results in Fig. 2c where the current density, after
increasing steadily, saturates at a value close to that on the return
sweep of the i-␩ curve at the same voltage, i.e., the t ϭ 1200 s,
It is well known that cyanide is strongly adsorbed on the silver
electrode and that reduction of silver cyanide complexes is a kineti-
cally slow process involving the formation-decomposition of an ad-
1
7,18
sorbed complex.
The results in Fig. 2 indicate that selenium acts
as an effective catalyst for the reduction of such complexes. This
might occur either through displacement or disruption of the ad-
sorbed cyanide or silver cyanide complex that forms in its absence.
Cyclic i-␩ sweeps for several sweep ranges are summarized in
Fig. 3a. Fits to the data, derivation of which is discussed later, are
shown in Fig. 3b. A switching potential below Ϫ0.4 V yields satu-
ration prior to completion of the return sweep at Ϸ800 s. This is
consistent with the time constant that characterizes the i(t) data
shown in Fig. 4a. Simulations of the i(t) data, using the same ki-
netics used to model the i-␩ data in Fig. 3, are shown in Fig. 4b
͑and also discussed later͒. The saturation evident in Fig. 3a and 4a
was also apparent when cyclic i-␩ sweeps were conducted multiple
times; current densities on both the outward and return sweeps es-
sentially traced the return sweep established during the first cycle
͑not shown͒.
␩
ϭ Ϫ0.35 V data point on Fig. 1͒. The current densities observed
in the i(t) experiments at t ϭ 20 s, as recorded on Fig. 1, reflect
deposition on a still uncatalyzed metal surface in the additive-
containing electrolyte as they lie quite close to the curve obtained
during cyclic current-voltage studies in an additive-free solution ͑not
shown͒.
Kinetics of the deposition reaction—Modeling experiments with
planar substrates.—The electrochemical experiments for selenium-
catalyzed silver deposition can be simulated using the CEAC model.
The i-␩ and i(t) data in Fig. 3a and 4a, respectively, are modeled
using the generalized reversible current-overpotential equation
Examination by XPS of electrodeposited films grown at Ϫ0.35 V
for 1000 s revealed the accumulation of selenium, or some deriva-
tive thereof, on the surface as shown in Fig. 2a. High resolution
spectra reveal the selenium L M M Auger line is at 177.7 eV
3
45 45
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
C425
Figure 3. ͑a͒ Slow sweep cyclic current-voltage studies of deposition on
planar substrates reveal hysteresis that increases with the voltage sweep
range. Note the saturation on the return sweeps for larger sweep range.
Sweep rate 1 mV/s. ͑b͒ Simulations obtained using kinetic parameters opti-
mized to simultaneously fit both i-␩ data in Fig. 3a and i(t) data in Fig. 4a.
2
limited deposition rate, i Ϸ 18 mA/cm , is determined by diffusion
L
across the boundary layer ͑thickness ␦͒, free convection, and silver
ion concentration in the bulk electrolyte according to the relation-
ship i ϭ FDAgCAg /␦. A diffusion coefficient of DAg ϭ 7.6
L
Ϫ6
2
Figure 2. ͑a͒ A survey spectrum for a sample after 1000 s of silver deposi-
tion at Ϫ0.35 V. ͑b͒ The selenium L M M Auger signal as a function of
deposition time. The intensities are scaled by the areas under the associated
Ag (3d5/2) photoelectron peaks. ͑c͒ The correlation between acceleration of
the deposition rate and accumulation of the selenium catalyst on the silver
surface is apparent for deposition at Ϫ0.35 V. The XPS data are delineated as
solid circles with the dashed line being only a guide to the eye.
ϫ 10 cm /s was calculated based on a boundary layer thickness
Ϫ4
3
3
45 45
␦ ϭ 135 ␮m and CAg ϭ 3.4 ϫ 10 mol/cm ͑the latter specified
by the electrolyte manufacturer͒. Note that F ϭ 96485 C/mol, R
ϭ 8.314 J/mol K, and T ϭ 293 K.
The selenium catalyst is assumed to adsorb according to Lang-
muir kinetics
d␪
i
ϭ kЈC ͑1 Ϫ ␪͒
͓2͔
Se
dt
i
␣͑␪͒F
i͑␪,␩͒ ϭ i ͑␪͒
ͩ
1 Ϫ
ͪ
ͫ
exp
ͩ
Ϫ
␩
ͪ
o
iL
RT
In Eq. 2, kЈ gives the kinetics of selenium accumulation onto the
metal surface for selenium concentration in the adjacent electrolyte
͑
1 Ϫ ␣͑␪͒͒F
i
Se
C
. This concentration is related to the bulk concentration CSe by
Ϫ exp
ͩ
␩
ͪ
ͬ
͓1͔
RT
equating the adsorbing selenium flux to that diffusing across the
boundary layer
The term (1 Ϫ i/i ) equals the ratio of the silver ion concentration
at the interface relative to the bulk concentration ͑i.e., C /CAg
L
i
i
DSe ͑CSe Ϫ C ͒
Se
͒
i
Ag
kЈC ͑1 Ϫ ␪͒ ϭ
͓3͔
Se
reflecting the concentration gradient across the boundary layer. This
deviates from unity at large overpotentials. The exchange current
⌫
␦
where ⌫ is the saturation coverage of the catalyst. Using Eq. 2 and 3
it can be shown that the parameters DSe , ␦, and ⌫ affect catalyst
accumulation only through the ratio DSe /␦⌫. For diffusion-limited
density i (␪) and the transfer coefficient ␣͑␪͒ are modeled as lin-
o
early dependent on the normalized surface coverage of the catalyst
␪, defined in terms of the maximum coverage ⌫. The transport-
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C426
Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
and 4a. Using the functional forms of ␪ thus obtained, kЈ(␩), i (␪),
o
␣
͑␪͒, and DSe were obtained by optimizing the fit of the curves
predicted by Eq. 1 to the data. In fitting the i-␩ data, the adsorption
rate constant kЈ, which was found to increase with overpotential, is
held at its maximum value for the entire return sweep to model
obstruction of any inhibiting layer’s healing by the adsorbed
catalyst. The kinetic parameters thus obtained are kЈ(␩)
4
6
2
3
ϭ 5.2 ϫ 10 ϩ 5.1 ϫ 10
␩ (cm /mols),
io(␪) ϭ 0.32
2
ϩ 0.86␪ (mA/cm ), and ␣(␪) ϭ 0.17 ϩ 0.11␪. Inclusion of a
term proportional to ␩ for the interface kinetics was found to give
2
a slightly better fit than inclusion of a term proportional to either ␩
3
or ␩ . The diffusion coefficient of the catalyst that optimizes the fit
Ϫ6
2
to the data is DSe ϭ 4.5 ϫ 10 cm /s. Figures 3b and 4b show the
fits to the data in Fig. 3a and 4a, respectively, obtained with these
kinetic parameters. The root-mean-square deviation between the
2
data and the fits is 0.5 mA/cm , quite small considering that data
from two different types of experiments, including thirteen curves,
eight of them cyclic, and all having inflections, have been fit with
just seven parameters. Note that the fitting parameters just obtained
from these studies on planar substrates fully define the kinetics that
are used in the modeling of trench filling that follows.
Studies with Patterned Substrates
Filling of trenches-modeling.—In the CEAC model describing
deposition on patterned ͑more generally, nonplanar͒ substrates, Eq.
1
describing the evolution of surface coverage of catalyst on the
surface of a flat substrate is modified to
d␪
i⍀Ag
i
ϭ
␬␪ ϩ kЈC ͑1 Ϫ ␪͒
͓4͔
Se
dt
F
The first term on the right reflects the change of the local coverage
due to changing surface area. This term is proportional to the
Figure 4. ͑a͒ Current-time studies of deposition on planar substrates at dif-
ferent overpotentials. Current-densities at 20 and 1200 s were used in Fig. 1.
␪
local interface velocity, v, equal to i(␪,␩)⍀Ag /F (⍀Ag
͑b͒ Simulations obtained using kinetic parameters optimized to simulta-
3
ϭ 10.3 cm /mol), and the local curvature ␬ of the interface. The ␬
neously fit both i-␩ data in Fig. 3a and i(t) data in Fig. 4a.
dependence leads to increasing ͑decreasing͒ local coverage on con-
cave ͑convex͒ surfaces during growth. This term dominates at the
bottoms of superfilling features. In computational modeling based
on this equation, where area reduction would have led to catalyst
coverage exceeding unity, the additional catalyst was deleted to keep
the coverage at unity. This is equivalent to either presuming local
incorporation of the catalyst within the metal deposit or desorption
or deactivation of a molecule that is no longer deposition-rate-
enhancing. Simulations of trench filling in the selenium catalyzed
electrolyte involved an explicit solution of the diffusion equations
that govern silver and catalyst transport through the electrolyte to
the evolving surface of the filling trench as well as boundary condi-
tions appropriate for the local catalyst coverage. A code using the
level-set method for interface tracking, with all kinetics from the
studies conducted on planar substrates, was implemented. Further
accumulation (DSe /␦⌫ Ӷ kЈ), the time constant for accumulation
scales with ␦⌫/C D . For accumulation limited by interface kinet-
Se Se
ics (DSe /␦⌫ ӷ kЈ), the time constant for accumulation scales with
1/kЈCSe . The time constant obtained by fitting the experimental data
when either of these limiting conditions dominates, thus specifies
only the value of the appropriate product, not the values of the
individual parameters DSe , CSe , ␦, or ⌫. This is particularly signifi-
cant in the present work where CSe is not known due to commercial
proprietary restrictions. Accordingly, kinetic parameters consistent
with the data in Fig. 1 have been determined by fixing C ϭ 1
Se
Ϫ8
3
Ϫ10
2
ϫ 10 mol/cm , and ⌫ ϭ 7.6 ϫ 10
mol/cm , the latter corre-
sponding to a 0.33 site occupancy on Ag͑111͒. The diffusion coef-
ficient DSe and interface kinetics kЈ(␩) are permitted to vary in the
optimization. The fixed values CSe and ⌫, though expected to indi-
10
details can be found elsewhere.
The feature filling predictions are shown in Fig. 5. Superconfor-
mal growth is achieved when the catalyst coverage, while still low
on the sidewalls, becomes concentrated on the bottom of the trench
by area reduction, resulting in bottom-up filling in Fig. 5a and e-k. If
the aspect ratio of the trench is too great, or the associated variation
of deposition rate along the surface profile too small, the sidewalls
impinge before the bottom of the trench can escape ͑Fig. 5b-d and
6
vidually be of the correct order of magnitude, are not necessarily
correct. They are fixed only to limit the parameter space for data
fitting to the most relevant parameters. Data where crossover condi-
tions (DSe /␦⌫ Ϸ kЈ) dominate would be better fit by adjusting all
parameters. However, it is unlikely that a substantially more mean-
ingful picture would be obtained due to the resulting large number
of fitting parameters, the quality of the data, and the simplicity of the
1͒, creating a vertical seam or void in the center of the trench. The
transition between superfilling and seam formation is predicted to be
a function of potential with the fill/fail boundary, i.e., the highest
aspect ratio trench that can be filled without seam formation, mov-
ing from an aspect ratio of Ϸ1 at ␩ ϭ Ϫ0.15 V to an aspect ratio as
high as Ϸ2.25 at ␩ ϭ Ϫ0.48 V and then back down to Ϸ2 at ␩
ϭ Ϫ0.69 V. It should be noted that there is no reason to believe
that these are the highest aspect ratio features that can be filled from
this electrolyte system, as it is unlikely that this composition is
optimized for superconformal deposition. A slight canting of the
model ͑which neglects consumption͒. Similarly, the value of DSe
obtained will only be as correct, individually, as the other param-
eters. The functional form of the interface kinetics kЈ(␩) obtained,
though scaled by a constant, should more closely reflect the true
voltage dependence, being the only voltage dependent quantity.
i
Using Eq. 3 to eliminate CSe , Eq. 2 can be numerically inte-
grated to obtain ␪ ͓t,DSe ,kЈ(␩)͔ for the experimental conditions
used to obtain each of the i-␩ and i(t) data curves shown in Fig. 3a
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
C427
Figure 5. Simulations of silver deposition in trenches with aspect ratios of
.0, 1.5, 2.0, and 2.25 ͑left to right͒ for indicated overpotentials. All simula-
1
tions are for trenches 0.5 ␮m deep. Trenches are predicted to fill supercon-
formally for aspect ratios up to, and including, 1 at Ϫ0.15 V, 2.25 at Ϫ0.48
V, and 2 at Ϫ0.69 V. Insufficient differentiation of deposition rate on the
bottom and sidewalls is predicted to result in seam formation in b, c, d,
and 1.
trench sidewalls, such that the trenches are wider at the top than at
the bottom, will enable geometric leveling that will increase the
nominal aspect ratio of the features that can be filled robustly.
Figure 7. SEM images of trenches filled at overpotential ␩ ϭ Ϫ0.325 V
͑
top to bottom͒. The formation of voids in the finer features is not consistent.
6
,20
The variability is likely associated with the visible roughness of the surface.
All three groups come from the same specimen.
Filling of trenches-experiments.—Silver deposition into trenches
.41 ␮m deep and between approximately 350 and 200 nm in width
0
was examined using the selenium-catalyzed electrolyte. As shown in
Fig. 6, filling improves as ␩ goes from Ϫ0.375 to Ϫ0.480 V, with
trenches of aspect ratios up to 1 or 1.5 filling at the lower voltage
aspect ratios that fill at these two overpotentials, both in value and
trend with overpotential, are consistent with the predictions in
Fig. 5.
͑
Fig. 6b and a, respectively͒ and trenches with aspect ratio up to 1.5
and 2 filling at the latter ͑Fig. 6d and c, respectively͒. The highest
It should be noted that when ␩ was increased to Ϫ0.585 and
Ϫ0.69 V, voids formed even at the tops of the trench with aspect
ratio of 1 ͑not shown͒. This contrasts with the simulations, where
filling is predicted for aspect ratios up to 2 over this voltage range.
These unexpected voids are believed to be related to the roughness
of the impinging side wall deposits ͑Fig. 7͒. The roughness of the
metal deposits is due, in part, to the roughness of the lithographi-
cally defined sidewalls ͑see Fig. 7͒ caused by attack in the alkaline
electrolyte. Additional roughening at the larger overpotentials is
likely associated with the operation at the current limit ͑see Fig. 3͒.
These conclusions are supported by the fact that surface roughness
was much reduced, and no voids were formed during filling of vias
patterned in ‘‘hard’’ oxides/nitrides, that were not attacked by the
2
1
electrolyte. Further work will be required in order to address the
impact of roughness evolution on feature filling.
Close examination of the largest feature imaged in Fig. 6c re-
veals grain contrast; parallel twin boundaries pass diagonally across
the trench indicating that a single grain spans its cross section. As
this area originally included a polycrystalline seed layer, this is un-
likely to have formed during the growth process. This suggests that
a postdeposition recrystallization process has occurred analogous to
that reported for electrodeposited copper films. The effect is more
clearly revealed in the TEM micrograph of a cross sectioned speci-
Figure 6. SEM images of trenches filled at overpotentials ␩: ͑a and b͒
Ϫ0.375 V and ͑c and d͒ Ϫ0.480 V. Trenches from two groups, showing
filling behavior representative of that observed in all six groups on each
specimen, are shown from each specimen. Aspect ratios of the features are
approximately 1.1, 1.5, and 2 from left to right in each of the four groups
shown. Arrows indicate the trenches that contain voids.
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C428
Journal of The Electrochemical Society, 149 ͑8͒ C423-C428 ͑2002͒
strates. The accumulation of the catalyst was shown to lead to en-
hancement of the silver deposition rate. As given by Eq. 1, the
enhancement of the deposition rate expressed as v(␪ ϭ 1)/v(␪
ϭ 0) is rather small. Ignoring silver ion depletion effects, it goes
from approximately 6 to 14 as the overpotential ␩ goes from Ϫ0.1
to Ϫ0.3 V at room temperature. This contrasts with the behavior
observed for copper electroplating where rate enhancements ranging
8
-11
from 30 to 300 are possible.
This is manifest in the much smaller
hysteresis for cyclic current-voltage studies with silver ͑Fig. 3a͒ as
compared to that exhibited in similar experiments with copper8 and
void-free filling of lower aspect ratio features than are achievable
with copper. Both of these observations are predicted quantitatively
by the CEAC model. Experiments examining the filling of 350-200
nm wide trenches were compared with shape-change simulations
based on the CEAC mechanism. With all kinetics used for modeling
of feature filling obtained from the experiments on planar substrates,
experimental and modeling results for trench filling were found to
be in reasonable agreement. The generality of the CEAC model has
thus been demonstrated by extension to a chemical system that is
distinct from that originally used to develop the physical model.
It is noteworthy that the CEAC mechanism also offers a quanti-
tative formalism for understanding the morphological evolution of
an arbitrary surface profile and thus explains the widely known
brightening action evoked by selenium addition to silver cyanide
electrolytes.
,9
Figure 8. Montaged TEM images of trenches filled at overpotential ␩
ϭ Ϫ0.480 V. The twins across the central feature indicate there is no seam,
clearly demonstrating the superconformal filling of the trench with silver.
men in Fig. 8. Of particular significance are the large grains, includ-
ing twins, that fill the two larger trenches.
Recrystallization
The spontaneous recrystallization of copper at room temperature
is a key attribute in the successful implementation of copper elec-
trodeposition in Damascene processing because it results in a sub-
stantial decrease ͑ϳ20%͒ of the electrical resistivity. If silver depos-
its do in fact recrystallize, they offer the prospect of a slightly lower
resistivity. The electrical resistances of unpatterned silver deposits
from the selenium catalyzed electrolyte were measured using a con-
ventional four-point probe in order to examine this matter in more
References
1. P. C. Andricacos, C. Uzoh, J. O. Dukovic, J. Horkans, and H. Deligianni, IBM J.
Res. Dev., 42, 567 ͑1998͒.
2. A. C. West, J. Electrochem. Soc., 147, 227 ͑2000͒.
2
detail. The silver films were grown galvanostatically at 10 mA/cm
3
. J. Reid and S. Mayer, in Advanced Metallization Conference 1999, p. 53, M. E.
Gross, T. Gessner, N. Kobayashi, and Y. Yasuda, Editors, MRS, Warrendale, PA
͑2000͒.
to a thickness of 1 ␮m. The substrates were comprised of a 100 nm
thick copper seed-layer supported on silicon ͑100͒ wafers with a 2.5
nm thick chromium adhesion layer. Following silver deposition the
films were rinsed with water and transferred to the probe station
within a few minutes, and the resistance was monitored for a period
of up to 8 days.
Measurements begun within minutes of film deposition revealed
a total decrease in resistance of ϳ1.8% over two days. Continued
monitoring over the following six days revealed no further change.
This resistance drop is much smaller than that associated with re-
crystallization of copper films, and seemingly inconsistent with the
large grains evident in Fig. 6 and 8. Because specimen preparation
for cross sectioned specimens involved heating the specimens to
cure the supporting epoxy layer, resistance measurements were also
made both before and after annealing specimens in flowing argon at
4. T. Ritzdorf, D. Fulton, and L. Chen, in Advanced Metallization Conference 1999, p.
01, M. E. Gross, T. Gessner, N. Kobayashi, and Y. Yasuda, Editors, MRS, War-
1
rendale, PA ͑2000͒.
5
. E. Richard, I. Vervoort, S. H. Brongersma, H. Bender, G. Beyer, R. Palmans, S.
Lagrange, and K. Maex, in Advanced Metallization Conference 1999, p. 149, M. E.
Gross, T. Gessner, N. Kobayashi, and Y. Yasuda, Editors, MRS, Warrendale, PA
͑
2000͒.
6
7
8
9
. T. P. Moffat, J. E. Bonevich, W. H. Huber, A. Stanishevsky, D. R. Kelly, G. R.
Stafford, and D. Josell, J. Electrochem. Soc., 147, 4524 ͑2000͒.
. P. Taephaisitphongse, Y. Cao, and A. C. West, J. Electrochem. Soc., 148, C492
͑2001͒.
. T. P. Moffat, D. Wheeler, W. H. Huber, and D. Josell, Electrochem. Solid-State
Lett., 4, C26 ͑2001͒.
. D. Josell, D. Wheeler, W. H. Huber, and T. P. Moffat, Phys. Rev. Lett., 87, 016102
͑
2001͒.
10. D. Wheeler, D. Josell, and T. Moffat, Submitted.
1. D. Josell, D. Wheeler, W. H. Huber, J. E. Bonevich, and T. P. Moffat, J. Electro-
1
150°C for 30 min. The annealed specimens exhibited a resistance
chem. Soc., 148, C767 ͑2001͒.
drop of Ϸ10%, similar in size to the resistivity drops exhibited by
1
2. D. Josell, D. Wheeler, and T. P. Moffat, Electrochem. Solid-State Lett., 5, C49
6
recrystallizing copper films. This implies that recrystallization of
͑
2002͒.
the silver specimens occurs during preparation of cross-sectioned
SEM and TEM specimens, and can be induced by very modest
annealing temperatures.
13. D. Josell, D. Wheeler, and T. P. Moffat, Electrochem. Solid-State Lett., 5, C44
͑2002͒.
1
4. D. Josell, D. Wheeler, and T. P. Moffat, in Advanced Metallization Conference
2
3
001, A. J. McKerrow, Y. Shacham-Diamand, S. Zaima, and T. Ohba, Editors, p.
63, MRS, Warrendale, PA ͑2002͒.
Discussion and Conclusion
1
5. T. P. Moffat, D. Wheeler, and D. Josell, Unpublished work.
The curvature-enhanced accelerator coverage ͑CEAC͒ model has
been extended to describe superconformal electrodeposition of sil-
ver from an electrolyte containing a selenium catalyst or brightener.
Features with aspect ratios larger than 0.5 ͑height/width͒ were seen
to be filled during silver electrodeposition ͑Fig. 6 and 8͒, the defi-
nition of superconformal filling. Catalyst accumulation at the metal/
electrolyte interface affects the silver deposition kinetics, enhancing
the silver deposition rate, and manifesting as hysteretic current-
voltage curves and rising current transients in studies on planar sub-
16. M. Hauder, J. Gstottner, W. Hansch, and D. Schmitt-Landsiedel, Appl. Phys. Lett.,
8, 838 ͑2001͒.
7
1
1
1
7. R. Y. Bek and N. A. Rogozhnikov, J. Electroanal. Chem., 447, 109 ͑1998͒.
8. H. Baltruschat and W. Vielstich, J. Electroanal. Chem., 154, 141 ͑1983͒.
9. Handbook of X-Ray Photoelectron Spectroscopy, C. D. Wagner, W. H. Rigg, L. E.
Davis, J. F. Moulder, and G. E. Muilenberg, Editors, Perkin Elmer Corp., Physical
Electronics Division, Eden Prairie, MN ͑1978͒.
2
2
0. D. Varadarajan, C. Y. Lee, A. Krishnamoorthy, D. J. Duquette, and W. N. Gill, J.
Electrochem. Soc., 147, 3382 ͑2000͒.
1. D. Josell, B. Baker, C. Witt, D. Wheeler, and T. P. Moffat, J. Electrochem. Soc.,
Submitted.
Downloaded on 2015-03-11 to IP 128.218.248.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).