Gold superfill in submicrometer trenches: Experiment and prediction

Article abstract of DOI:10.1149/1.2128765

Bottom-up deposition of gold in fine trenches, also called superfill, was recently demonstrated using a submonolayer coverage of preadsorbed, deposition-rate-accelerating lead followed by gold electrodeposition. The present study has used experiments on planar substrates to quantify the effect of Pb adsorption on the Au deposition rate and the rate at which the adsorbed Pb was consumed during the Au deposition process. The values obtained have been incorporated into the curvature enhanced accelerator coverage mechanism of superfill where they were used to quantitatively predict the nonconformal, bottom-up deposition observed during filling of the patterned features. The results indicate the potential of the process for damascene interconnect fabrication in GaAs, GaN, and related technologies.

Full text of DOI:10.1149/1.2128765

Journal of The Electrochemical Society, 153 ͑1͒ C11-C18 ͑2006͒
C11
0
013-4651/2005/153͑1͒/C11/8/$20.00 © The Electrochemical Society, Inc.
Gold Superfill in Submicrometer Trenches: Experiment
and Prediction
z
D. Josell, D. Wheeler, and T. P. Moffat*
National Institute of Standards and Technology, Metallurgy Division, Gaithersburg, Maryland 20899-8554,
USA
Bottom-up deposition of gold in fine trenches, also called “superfill,” was recently demonstrated using a submonolayer coverage
of preadsorbed, deposition-rate-accelerating lead followed by gold electrodeposition. The present study has used experiments on
planar substrates to quantify the effect of Pb adsorption on the Au deposition rate and the rate at which the adsorbed Pb was
consumed during the Au deposition process. The values obtained have been incorporated into the curvature enhanced accelerator
coverage mechanism of superfill where they were used to quantitatively predict the nonconformal, bottom-up deposition observed
during filling of the patterned features. The results indicate the potential of the process for damascene interconnect fabrication in
GaAs, GaN, and related technologies.
©
2005 The Electrochemical Society. ͓DOI: 10.1149/1.2128765͔ All rights reserved.
Manuscript submitted June 7, 2005; revised manuscript received August 26, 2005.
The bottom-up “superfill” process for fabrication of copper
interconnects is now the standard for state-of-the-art interconnect
been proposed, all involving an adsorbed intermediate AuCN or
1
AuCN− layer.
been viewed directly by scanning tunneling microscopy as well as
22,26,27
Such an adsorbed cyanide intermediate has
2
8
fabrication within the silicon-based semiconductor industry. How-
ever, gold is used to form a variety of ohmic and Schottky contacts
for semiconductor technologies based on, for example, gallium
2
9
by vibrational spectroscopy.
Like Pb, adsorbed Tl also increases the Au deposition rate, mak-
ing related research potentially relevant. In particular, the decreased
coverage of CN-containing species on the Au surface in the presence
2
3
arsenide and gallium nitride as well as for chip-level wire bonding;
hence, an interest exists in a process for fabricating gold intercon-
nects. A bottom-up, “superfill” process for Au that would be appro-
priate for fabricating interconnects in damascene processing was
2
5
of adsorbed Tl has been detailed. It has also been seen that the use
of Tl additive decreases the hardness and lowers the impurity con-
3
0
4
tent of Au deposits. The impact of Tl adsorbate on the Au deposi-
recently demonstrated for Au superfill in fine trenches, as was an
3
1
5
tion rate has been noted to increase with temperature.
alternative process. The present paper uses measurement techniques
It was on the basis of the CEAC mechanism that the electrolyte-
and models first developed to understand the related copper superfill
process in order to elucidate the relevant kinetics of the new Au
superfill process. The kinetics obtained are then used to quantita-
tively predict the Au superfill process.
4
additive system was selected for the first Au superfill study. That
work, while disclosing a process for Au superfill, did not obtain the
kinetic parameters required for CEAC-based prediction of the pro-
cess itself. As such it was unable to provide quantitative comparison
of theory and experiment. This work attempts to do both.
While several mechanisms have been proposed to underlie the
superfill process during electrochemical deposition ͑ECD͒ of
6
-9
copper,
the curvature-enhanced accelerator coverage ͑CEAC͒
8
,9
General Experimental Details
mechanism has been demonstrated to quantitatively predict super-
fill during electrodeposition of copper
8-12
13-16
and silver
and chemi-
The experiments detailed in this work utilized the same electro-
lyte ͑0.02 mol/L KAu͑CN͒2, 0.1 mol/L KOH, and 0.1 mol/L KCN
in 18 M⍀ water͒ and Pb͑NO ͒ additive used in the first Au super-
17
cal vapor deposition of copper. The CEAC mechanism predicts
that an electrolyte-additive system can yield superfill when two re-
quirements are met: ͑i͒ the additive adsorbs on the deposit surface
and accelerates the local deposition rate and ͑ii͒ the adsorbed addi-
tive remains on the surface of the deposit during deposition. Under
these circumstances, the CEAC predicts that the decreasing area of
the metal surface at the bottoms of filling features will lead to lo-
cally increasing coverage of the adsorbed accelerator, which will
lead to increased local deposition rate and bottom-up superfill. Pre-
dictive models based on the CEAC mechanism use kinetics obtained
from studies on planar substrates; no additional parameters are
needed for simulating feature filling or more general interface evo-
lution. Because superfill processes based on the CEAC mechanism
derive from surface segregation of adsorbates, they not only elimi-
nate void and seam formation but might also be expected to yield
high-purity, high-conductivity deposits needed for semiconductor
interconnects.
3
2
4
fill process. Electrolytes were used in 100-mL quantities that were
sparged with nitrogen or argon immediately prior to use in order to
remove oxygen that otherwise contributes a Pb coverage-dependent
current density that interferes with interpretation of current measure-
3
2
ments in terms of Au deposition rate. All depositions were con-
ducted at room temperature ͑ϳ23°C͒ under flowing nitrogen or
argon, and the platinum counter electrode was placed on the oppo-
site side of the beaker from the specimen.
Studies of kinetics were conducted on Si wafers coated with
80-nm Au using a 10-nm Ti, 10-nm Pd adhesion layer, in order from
the substrate, all deposited by electron-beam evaporation. The speci-
mens were masked with plater’s tape to expose a circular area of
2
0.98 cm . Studies of superfill in trenches were conducted on pat-
terned wafers with bilayer seeds of 3-nm Ti and 27-nm Au. Deposit
thicknesses on the sidewalls of the trenches were approximately
one-tenth these values.
Acceleration of the Au deposition rate from KAu͑CN͒ in the
2
presence of a surfactant adsorbate, the key to feature superfill ac-
cording to the CEAC mechanism, has been detailed for a variety of
Formalism of the CEAC Mechanism
18-25
additive-containing electrolytes;
erated Au deposition in the presence of adsorbed Pb.
these include studies of accel-
Quantitative modeling of superfill through the CEAC mechanism
requires establishing the forms and parameters of two principal
equations. The first equation quantifies the relationship between ad-
sorbate coverage and metal deposition rate ͑i.e., current density͒.
The second equation quantifies the evolution of the adsorbate cov-
erage. These two equations, with boundary conditions based on
physical understanding of conservation of metal and adsorbate at the
interface and transport within the electrolyte, allow for quantitative
prediction of feature filling.
2
2,24
The accel-
eration is presumed to arise from the impact of the adsorbate on the
rate-limiting step in kinetically controlled Au deposition from the
KAu͑CN͒ . A variety of mechanisms and rate-limiting steps have
2
*
Electrochemical Society Active Member.
E-mail: daniel.josell@nist.gov
z
Downloaded on 2015-03-29 to IP 152.14.136.77 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C12
Journal of The Electrochemical Society, 153 ͑1͒ C11-C18 ͑2006͒
*
Table I. Constants with kinetic parameters from fitting of current
transients on planar substrates as used to model filling of pat-
terned specimens.
i ͑␪͒
i ͑␪͒
i͑␪͒ =
͓5͔
*
1 +
iL
*
␪=0
+ ␪i␪=1
F, C/mol
9
6,485
1.2
where, for convenience, the shorthand i ͑␪͒ ϵ ͑1 − ␪͒i
2
i , mA/cm
has been defined.
L
␪
=0
2
2
i
i
, mA/cm
0.048
1.04
1 ϫ 10
2.8 ϫ 10
10.21
0.027
0.009
7 ϫ 10
2 ϫ 10
ϫ 10
For this study, the equation describing the evolution of adsorbate
coverage through area change, accumulation from the electrolyte
and consumption through incorporation in the deposit was written
␪
=1
, mA/cm
3
6
k , cm /mol s
+
4
k , 1/cm
−
3
d␪
⍀Au
⍀Au
⍀
, cm /mol
i
Pb
=
␪␬
i͑␪͒ + k C ͑1 − ␪͒ − k ␪
i͑␪͒
͓6͔
+
−
␦
␦
, cm ͑stagnant͒
, cm ͑agitated͒
dt
F
F
The first term, which imposes mass conservation during area change
on the moving interface, gives the CEAC mechanism its name. It
expresses the normalized rate of area change in terms of the local
curvature ␬ and the normal velocity v ͑where v equals the product
of the molar volume ⍀Au and current density i divided by Faraday’s
constant F for the monovalent Au ion in the electrolyte͒. It is this
term, which accounts for compression or dilation of the adsorbate
coverage during deposition on nonplanar surfaces ͑␬ ꢀ 0͒, that un-
derlies the superfill phenomenon. The second term assumes Lang-
2
−10
⌫
, mol/cm
o
3
−5
C
, mol/cm
Au
di
dt
−
5
6
2
ͯ
ͯ
,mA/cm s
drift
2
−5
−5
D
, cm /s
1.7 ϫ 10
1.7 ϫ 10
Au
2
DPb, cm /s
muir adsorption kinetics to give accumulation occurring at a rate
i
proportional to the concentration C of Pb in the adjacent electro-
lyte and the fraction of available sites on the surface of the deposit
Pb
The equation quantifying the relationship between the adsorbate
coverage and the local deposition rate is obtained from the product
of the concentration of Au ions in the electrolyte adjacent to the
interface, C iA u, and a linear combination of the electrochemical rate
constants for deposition on Au- and Pb-catalyzed surfaces, kAu and
kAuPb, weighted by the respective surface coverage of each in this
two-state system. In terms of the fractional Pb adsorbate coverage,
␪Pb, this is written
͑
rate constant k ͒. The third term invokes the loss, presumed in this
+
work to be through burial, of a constant fraction of adsorbate during
each increment of growth; it can be found in a previously published
33
model of adsorbate consumption for modeling of leveling. Recall-
ing the relationship between the growth velocity v and current den-
sity i, the constant k in Eq. 6 can be shown to be the fraction of
−
adsorbates that are buried per unit increase of deposit thickness.
Kinetics for Eq. 5 and 6 were obtained from the results of two
different types of experiments on planar substrates. The next section
details these experiments.
i
C
Au
i͑␪ ͒ = F
͓kAu͑1 − ␪ ͒ + k ␪ ͔
AuPb Pb
͓1͔
Pb
o
Au
Pb
C
For psuedo-steady-state transport across a boundary layer, the Au-
ion concentration at the surface of the planar deposit can be ex-
pressed in terms of the current by
Studies of Kinetics on Planar Substrates
Current increase during accumulation.— The first type of ex-
periment involved concurrent Pb adsorption and Au electrodeposi-
i
tion from Au-cyanide electrolytes containing different Pb͑NO3͒2
concentrations. These experiments were conducted at −0.95 V SCE
without agitation. They yielded rising current–time transients asso-
ciated with the gradual, concentration-dependent accumulation of
the rate-accelerating Pb adsorbate ͑see Fig. 1͒.
C
C
i͑␪͒
iL
Au
=
ͩ
1 −
ͪ
͓2͔
o
Au
where i is the transport-limited current ͑obtained from cyclic vol-
L
tammetry, not shown, and given in Table I͒. Also, ␪ has been used in
place of the more explicit ␪ , as it is henceforth. The rate constants
for deposition are given by
Current decay after derivatization.— The second type of experi-
ment began with simultaneous Pb accumulation/Au deposition as
described in the preceeding section, though for short, controlled pe-
riods, that was followed by transfer of the specimen to Pb-free elec-
trolyte for a longer period of Au deposition, again at −0.95 V SCE
without agitation. For most of these experiments, the Au-cyanide
electrolyte used for the first step had a 10 ␮mol/L concentration of
the Pb͑NO3͒2 additive, although some of the accumulation steps
were done with 8 ␮mol/L Pb͑NO3͒2 concentrations and are indi-
cated appropriately. Figure 2 shows current transients from speci-
mens that spent 0, 100, 200, and 300 s in the Pb-containing electro-
lyte prior to being transferred to the Pb-free electrolyte. The
accelerating deposition in the Pb-containing electrolyte is shown at
time less than zero. Deposition upon transfer to the Pb-free electro-
lyte was set at zero time for all specimens.
From Fig. 2, the longer the derivatization time in the Pb-
containing electrolyte, the faster the subsequent Au deposition in the
Pb-free electrolyte, consistent with higher coverage of the deposi-
tion rate-enhancing Pb adsorbate for longer derivatization time. The
overall agreement just prior to transfer and just after decay of the
insertion transient indicates that adsorbate coverage was generally
not affected by the transfer step. Falling transients, consistent with
consumption of the Pb adsorbate, are evident for derivatization pe-
riods of 200 and 300 s. Comparison to the current transients from
Pb
␣iF␩
o
k = k exp
ͩ
−
ͪ
͓3͔
i
i
RT
Because the analyzed measurements were taken at constant overpo-
tential ␩ and lacking information to the contrary, the transfer coef-
ficients ␣ have been presumed independent of the nature of the
i
surface, i.e., ␣Au = ␣AuPb ͓␩ = −0.14 V as all experiments were
conducted at −0.95 V vs a saturated calumel reference electrode
͑
SCE͒ with cyclic voltammetry, not shown, indicating −0.81 V SCE
as the equilibrium potential for reversible Au deposition/
dissolution͔. With this assumption, Eq. 1 can be rewritten in terms of
the partial currents for a Au surface ͑␪ = 0͒ and a Pb-saturated sur-
face ͑␪ = 1͒ to give
i
C
C
i͑␪͒
iL
Au
͓͑1 − ␪͒i + ␪i␪=1͔ =
␪=0
͓͑1 − ␪͒i + ␪i␪=1
␪=0
i͑␪͒ =
ͩ
1 −
ͪ
͔
o
Au
͓
4͔
Equation 4 can be rewritten to express the current density i as a
function of the adsorbate coverage ␪
Downloaded on 2015-03-29 to IP 152.14.136.77 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, 153 ͑1͒ C11-C18 ͑2006͒
C13
Figure 1. Current transients during Au deposition in electrolytes containing
various concentrations of the Pb͑NO ͒ additive ͑indicated͒ at a potential of
3
2
Figure 2. Current transients from specimens with 0, 100, 200, and 300-s Au
deposition in an electrolyte containing 10 ␮mol/L Pb followed by further Au
deposition in a Pb-free electrolyte. The accumulation transients in the Pb-
containing electrolyte, offset to allow comparison, are shown at times less
than zero. Depositions upon transfer to the Pb-free electrolyte begin at zero
time. Depositions were without agitation at a potential of −0.95 V SCE in
both electrolytes. The bold curves indicate the predicted behavior using the
kinetics in Table I. Data is split to facilitate viewing.
−
0.95 V SCE without agitation. The bold curves indicate the predicted be-
havior using the kinetics in Table I. Data is split to facilitate viewing.
specimens without preadsorbed Pb ͑0 s derivatization͒ makes clear
that the apparent increasing transient of the specimen derivatized for
1
00 s was dominated by the Pb-free baseline behavior.
The results of the current–decay transients, including those
shown in Fig. 2, are summarized in Fig. 3 and 4. Figure 3 plots the
average slope of each transient curve during the period between 100
and 300 s after transfer to the Pb-free electrolyte as a function of the
average current density during the same period. This period was
selected to minimize the impact of both initial specimen insertion
transient and long-term surface roughening on the results. Figure 3
includes data for specimens derivatized in 10 ␮mol/L Pb-containing
electrolyte, as in Fig. 2, and in 8 ␮mol/L Pb-containing electrolyte.
Figure 4 plots the average current as a function of the derivatization
time.
trolyte concentration Co appropriate to the experiment͒ and
simultaneously fitting the predictions to the rising transient data in
Pb
Fig. 1 and the decaying transient data in Fig. 2. The required time-
i
Pb
dependent interface concentration C required in Eq. 6 was ob-
tained from the bulk electrolyte concentration by equating the rate of
accumulation from the electrolyte onto the surface and the diffusive
flux of Pb through the boundary layer
o
i
D ͑C − C ͒
i
Pb Pb
Pb
⌫k+CPb͑1 − ␪͒ =
͓7͔
␦
Assuming that the Au deposition current density scales with Pb
coverage, Fig. 3 indicates the rate at which adsorbed Pb is consumed
during Au deposition, presumably through incorporation, as an im-
plicit function of the Pb coverage. Figure 4 is a measure of Pb
adsorbate coverage on the surface of each specimen after it has been
transferred from the Pb-containing electrolyte to the Pb-free electro-
lyte; consistently, the deposition rate ͑i.e., Pb adsorbate coverage͒ is
seen to increase with derivatization time and at a faster rate in the
higher concentration electrolyte.
the left side of Eq. 7 being the Langmuir accumulation term in Eq.
6 multiplied by the areal density ⌫ of sites available on the surface
for Pb adsorption. The value DPb in Table I has been equated to the
−
diffusion coefficient of the Au͑CN͒ ion, D , as previously used by
Au
2
0,22
other authors for Pb and Tl additives in similar electrolytes.
The
steady-state boundary-layer thickness ␦ in Table I was obtained from
the experimental limiting current density i using the expression
L
o
Au
FDAu
C
iL =
͓8͔
Determining Kinetics from Studies on Planar Substrates
␦
Kinetics for the feature filling simulations were obtained by fit-
ting the predictions of Eq. 5 and 6 to the transient data in Fig. 1 and
with the values of the other parameters found in Table I.
Fit to the accumulation data.— Predicted rising transient behav-
2. In this fitting, ␬ = 0 in Eq. 6 because the specimens are planar.
ior for the parameter values found in Table I and bulk catalyst con-
The predicted current–time transients for the data fitting were ob-
tained by evolving the deposition rate and surface coverage accord-
o
Pb
centrations C = ͑0, 1, 2, 3, 4, 6, and 10͒ ␮mol/L are overlaid
ing to Eq. 5 and 6 ͑with the initial adsorbate coverage ␪ and elec-
on the corresponding experimental data in Fig. 1. A 70-s time lag
o
Downloaded on 2015-03-29 to IP 152.14.136.77 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C14
Journal of The Electrochemical Society, 153 ͑1͒ C11-C18 ͑2006͒
count for the increasing current noted earlier for deposition in the
additive-free Au-cyanide electrolyte ͑ ͉di/dt͉drift
ϫ 10− mA/cm s as per Table I and Fig. 2, 0 µmol/L case͒.
=
5
2
6
Fit to the consumption data.— Predicted transients for Pb accu-
mulation followed by deposition in Pb-free electrolyte on the Pb-
derivatized specimens are overlaid on the corresponding experimen-
tal data in Fig. 2. These predictions were obtained using the
parameter values found in Table I, the accumulation portions, with
o
Pb
C
= 10 ␮mol/L, starting after the same 70-s time lag required to
fit the accumulation curves in Fig. 1; as would be expected, the
predicted accumulation curves are a poor fit to the experimental
accumulation curves at early times. The decay portions of the curves
o
Pb
were obtained using C = 0 ␮mol/L and initial Pb coverages ␪o
=
0, 0.17, 0.512, and 0.7 for the accumulation times of
0
, 100, 200, and 300s, respectively, equal to the final coverages
from the predicted accumulation curves. In modeling the decay tran-
sients, the boundary-layer thickness ␦ used in Eq. 7 was presumed to
ͱ
initially thicken with the depleted zone, i.e., ␦ = DAut, until it
Figure 3. The average slopes of the decay transients during the period
reached the steady-state value given in Table I, after which time it
was held constant. The concentration profile within the boundary
layer was presumed to be fully relaxed throughout this evolution
͑i.e., a pseudo-steady-state approximation͒, and the predicted decay
transients include the same constant current drift term noted earlier.
Importantly, although the predicted transients are shown starting
upon insertion, the kinetic parameters were obtained by fitting the
data only beyond 100 s, consistent with the summarized data in Fig.
1
00–300 s after transfer from the Pb-containing electrolyte to the Pb-free
electrolyte is plotted as a function of the average deposition current during
the same period. Data is included for specimens derivatized in 10 ␮mol/L
Pb-containing electrolyte ͑closed symbols͒, as in Fig. 2, and 8 ␮mol/L Pb-
containing electrolyte ͑open symbols͒. Derivatization times are indicated by:
crosses ͑0 s͒, squares ͑100 s͒, triangles ͑200 s͒, diamonds ͑300 s͒, and
circles ͑400+ s͒. A negative value indicates current density ͑deposition rate͒
decreases with time. The curve indicates the predicted behavior using the
kinetics in Table I.
3
, in order to avoid the uncetrtainty associated with the evolving
convective and diffusion fields.
Comparison to the data in Fig. 3 was facilitated by derivation of
a few equations. First, taking the time derivative of Eq. 5
between immersion and the inception of accumulation/deposition
was used to obtain better agreement between the predicted and ex-
perimental rising transient. This was necessary because no kinetic
parameter adjustment was capable of yielding the gradually increas-
ing deposition rates at the beginning of each rising transient ͑after
the current drop upon specimen insertion͒. This gradual increase
likely reflects the inception and initial evolution of the natural con-
*
−2
di͑␪͒
dt
d␪
i ͑␪͒
␪
=1
=
͑i − i␪=0
͒
ͫ
1 +
ͬ
͓9͔
dt
iL
which predicts the rate of current density change due to accumula-
tion or consumption of the accelerating adsorbate. As noted previ-
ously, this was presumed to occur slowly enough that associated
relaxation within the boundary layer could be assumed to be instan-
taneous. Equation 9 was modified by addition of the constant current
3
4
vection that underlies the boundary layer during deposition. A 70
s time scale for development of the natural convection would be
-
2
consistent with the time constant ͑␦ /DAu͒ required to develop the
concentration gradient in the 270-␮m-thick boundary layer. It is also
reasonable considering the small density gradient that underlies the
drift term ͉di/dt͉
noted earlier, which was presumed to also ap-
drift
ply to the electrolytes containing the dilute Pb͑NO ͒ additive
3
2
*
−2
2
di͑␪͒
dt
d␪
i ͑␪͒
di
ͯ ͯ
dt
natural convection for deposition rates Ͻ0.5 mA/cm and metal ion
␪=1
− i␪=0
=
͑i
͒
ͫ
1 +
ͬ
+
͓10͔
concentration of just 0.02 mol/L KAu͑CN͒ . The predicted accumu-
dt
iL
2
drift
lation transients all include a constant current density drift to ac-
The adsorbate evolution equation was expressed as a function of the
adsorbate coverage ␪ alone by substitution of Eq. 5 into Eq. 6 ͑with
o
␬
= 0 and C = 0͒ to obtain
Pb
*
−1
d␪
⍀
i ͑␪͒
Au *
=
− k−␪
i ͑␪͒
ͫ
1 +
ͬ
͓11͔
dt
F
iL
Using Eq. 11, Eq. 10 was also expressed as a function of the adsor-
bate coverage ␪ alone
*
−3
di
dt
⍀Au
i ͑␪͒
di
ͯ ͯ
dt
␪i ͑␪͒͑i − i␪=0
*
␪=1
͒
1 +
+
=
− k−
ͫ
ͬ
F
iL
drift
͓
12͔
With Eq. 11 and 12 both expressed as functions of ␪, the predicted
dependence of di/dt on i was readily evaluated using the parameters
from Table I. The resulting curve is overlaid on the experimental
data in Fig. 3.
The kinetic parameters i =0, i␪=1, k , and k , obtained from fitting
␪
+
−
the current transients in Fig. 1 and 2 and used to obtain all the
predicted curves, are given in Table I.
Experimental Feature Filling
Figure 4. The average current during the period between 100 and 300 s after
transfer to the Pb-free electrolyte was obtained from each curve in Fig. 4 and
is plotted as a function of the derivatization time.
General Details.— Depositions on patterned substrates were ac-
complished in three steps. The first step involved deposition in agi-
Downloaded on 2015-03-29 to IP 152.14.136.77 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, 153 ͑1͒ C11-C18 ͑2006͒
C15
Figure 6. Filling in a low-aspect-ratio trench on the same specimens shown
in Fig. 5.
Figure 5. ͑top͒ Trenches with three different widths with only the evapo-
rated Ti/Au seed. Filling of the same size trenches is shown after 100 s in the
agitated Pb-free electrolyte, 60 s in the stagnant 10 ␮mol/L Pb derivatization
electrolyte, and the indicated deposition times in the agitated Pb-free elec-
trolyte. All depositions were at −0.95 V SCE.
electrolyte was 60 s for all three plated specimens; the subsequent
deposition time in the Pb-free electrolyte ranged from 200 to 400 s.
The bottom-up filling associated with the superfill process is particu-
larly visible in the two wider trenches of Fig. 5. The enhanced
deposition at the corners of the very wide trench in Fig. 6 is also a
manifestation of superfilling. Details concerning the process for
4
tated Pb-free electrolyte at −0.95 V SCE. This step was primarily
used to ensure that a good seed existed for the superfill process. The
second step involved derivatization in stagnant Au-cyanide electro-
cross-section preparation can be found in a previous publication.
Figure 7 shows the currents recorded during filling of the pat-
terned specimens shown in Fig. 5 and 6; the current history for each
specimen has been normalized by the steady-state current at the end
of its initial Pb-free deposition step ͑from 0 to 100 s͒. The initial
deposition in the Pb-free electrolyte exhibits a gradual decay to
steady state for each Pb-free surface. The decay is likely related to
evolution of the boundary layer and associated gold complexes. Ac-
cumulation of Pb adsorbate during the 60-s derivatization step ͑be-
tween approximately 140 and 200 s on Fig. 7͒ manifests as the
rising transient after the decay of the insertion transient. The final
Au depositions exhibit a gradually rising current from a value that is
approximately four times larger than the value on the Pb-free sur-
lyte containing 10 ␮mol/L Pb͑NO ͒ for adsorption of the Pb ac-
3
2
celerator, concurrent with Au deposition, at −0.95 V SCE. The final
step involved the majority of the Au deposition and was conducted
in agitated Pb-free electrolyte at −0.95 V SCE. All immersions were
done at potential, and transfers between solutions were done as
quickly as possible to try to ensure that the specimens remained
wetted with the relevant electrolyte during the transfers; the elapsed
time between the emmersion starting and immersion ending each
transfer was nonetheless ϳ30 s due to the need to transfer beakers,
shift electrode connections, and adjust specimen position. Efforts
were also made to minimize carryover to the frequently refreshed,
additive-free solution.
The separation of the Pb derivatization and Au deposition be-
tween the second and third steps, respectively, though not complete
because there was concurrent Au deposition during the second step,
permitted agitation to reduce Au ion depletion during feature filling
without affecting the kinetics of the Pb adsorption process. In con-
trast, industrial processes for Cu superfill typically conduct the en-
tire deposition in a single electrolyte containing both metal ions and
additives. Deposition on specimens with preadsorbed accelerator
and zero accelerator concentration in the electrolyte has been previ-
ously demonstrated to yield CEAC-quantifiable superfill during both
12,15,35
Cu and Ag electrodeposition
that is indistinguishable from
superfill obtained through single electrolyte processes.
Experimental results.— Figure 5 shows sequential filling of
three trenches of different width, all approximately 0.2 ␮m deep.
Figure 6 shows sequential filling of a much wider trench. All images
with identical processing parameters come from the same specimen,
as each specimen included arrays of all the trench widths studied.
The derivatization time in the 10 ␮mol/L Pb͑NO ͒ -containing
Figure 7. The recorded deposition currents for the three specimens pictured
in Fig. 5 and 6, normalized by the steady-state current from the first deposi-
tion step in the Pb-free electrolyte.
3
2
Downloaded on 2015-03-29 to IP 152.14.136.77 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C16
Journal of The Electrochemical Society, 153 ͑1͒ C11-C18 ͑2006͒
Figure 8. Simulations of deposition in trenches with preadsorbed Pb coverage ␪ = 0.15 during Au deposition in agitated Pb-free electrolyte. Deposition is
o
shown in trenches with dimensions from Table II, corresponding to the experimental filling in Fig. 6.
face. The gradual increase of current during this last step is likely
due to increasing surface area ͑Fig. 5: note the curvature of the
surface between trenches after 400-s deposition͒.
deposit thickness after an additional 200 s of deposition on the ac-
celerated surface ͑compare the top two rows of either Fig. 5 or Fig.
6͒.
For the transients shown in Fig. 7 in particular, the deposition
current increased by a factor of approximately 4 as result of the
derivatization ͑ignoring the insertion transients͒. Using Eq. 1 with
the parameters in Table I, this corresponds to increasing the Pb
Superfill Modeling Using the CEAC Mechanism
CEAC-based feature-filling predictions were made as follows.
The first equality in Eq. 4 expressed the local current density ͑and
thus deposition rate͒ in terms of the local Au ion concentration and
Pb coverage ␪. Equation 6 defined the local evolution of the adsor-
bate coverage ␪ in terms of the local deposition rate. Required ki-
netics were from Table I, as obtained from the studies on planar
substrates. Transport of the Au ions through the boundary layer in
the agitated electrolyte and within the unfilled region of the trench
was presumed to be through diffusion, satisfying the fully time-
dependent diffusion equation and mass conservation boundary con-
ditions on the surface of the growing Au deposit for a periodic array
of trenches. Geometrical parameters defining the trenches, including
width at the trench bottom, trench pitch ͑center-to-center spacing͒,
and sidewall tilt angle were obtained from the imaged trenches ͑see
Fig. 5͒ as listed in Table II. A nearly threefold increase of measured
limiting current density in the agitated electrolyte ͑not shown͒ was
modeled as a reduced boundary layer thickness, as indicated in
Table I.
adsorbate coverage to approximately ␪ Ϸ 0.14 ͓i.e., predicted cur-
o
2
rent density going from i͑0͒ Ϸ 0.05 mA/cm to i͑0.14͒
ϳ 0.2 mA/cm , ignoring Au-ion depletion which is difficult to as-
sess accurately because of the large variations in trench width and
2
pitch across the surfaces of the patterned specimens͔.
Feature fill simulations.— Figure 8 shows simulations obtained
for deposition in derivatized trenches during Au deposition in the
Pb-free electrolyte. The trench dimensions and simulation condi-
tions correspond to those of the experimental images in Fig. 5 and 6,
using the kinetics from Table I, geometries from Table II, and initial
Pb coverage of ␪ = 0.15. The general formalism of the numerical
o
evaluation can be found in Ref. 11. The specific code used for the
36
numerical simulation is publicly available.
Estimating initial adsorbate coverage.— The experimental fill-
ing results in Fig. 5 and 6 could be modeled numerically by dividing
the deposition process into its three steps and simulating each one in
sequence. However, the offset of the fitted transients to the experi-
mental accumulation data in Fig. 1, where the accumulation of ad-
sorbate starts only after 70 s, precluded meaningful use of Eq. 6 for
assessing adsorbate coverage at the 60-s derivatization time used for
the specimens shown in Fig. 5 and 6.
Instead, the final adsorbate coverage at the end of the derivatiza-
tion step, ␪ , was extracted from the measured increase of the depo-
o
sition current from deposition step 1 to step 3 induced by the Pb
adsorption step ͑see Fig. 7͒. Filling during the third step was then
simulated with this starting coverage ␪ . Deposition during the un-
o
accelerated, first step and relatively short, still-accelerating derivati-
zation step was ignored. Neglect of the deposition associated with
the first two steps is a reasonable approximation considering the
Table II. Geometrical parameters for fill modeling as obtained
from patterned specimens.
Height, ␮m
0.2
6
0.115
0.6
Sidewall tilt, degrees
a. Width at bottom, ␮m
a. Pitch, ␮m
Figure 9. Filling in two nominally identical specimens. Filling of the same
size trenches as shown in Fig. 5 as well as one end of the wide trench
pictured in Fig. 6. Deposition for 100 s in the agitated Pb-free electrolyte,
180 s in the stagnant 10 ␮mol/L Pb derivatization electrolyte, and 400 s in
the agitated Pb-free electrolyte. All depositions were at −0.95 V SCE.
b. Width at bottom, ␮m
b. Pitch, ␮m
c. Width at bottom, ␮m
c. Pitch, ␮m
0.095
0.55
0.07
0.5
Downloaded on 2015-03-29 to IP 152.14.136.77 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, 153 ͑1͒ C11-C18 ͑2006͒
C17
ing is understood to occur because the initially high adsorbate cov-
erage precludes significant acceleration of local deposition through
the CEAC mechanism.
The corresponding experimental filling ͑specimen 2 of Fig. 9͒
has in fact lost the bottom-up filling dynamic, as predicted. How-
ever, superfill of all three trench widths has been lost rather than just
the predicted two, and voiding is observed rather than the predicted
v-notch, conformal filling. Deposition within the features is also
slower than predicted, particularly in comparison to that in the field.
These observations suggest substantial metal ion or adsorbate deple-
tion within the trench. However, the Au-ion depletion within the
feature was accounted for in these simulations, and other simula-
tions ͑not shown͒ indicated minimal depletion of accelerator would
be expected to occur during the Pb derivatization step.
Figure 10. The recorded deposition currents for specimens derivatized for
80 s, including the two specimens pictured in Fig. 8. The currents are nor-
malized by the steady-state currents recorded during the first deposition step
in the Pb-free electrolyte.
Nonideal aspects of superfill.— Several nonideal aspects of su-
perfill in this system might underlie the tendency of the model to
overestimate deposition rates as well as the difference between pre-
dicted and observed dependence of fill time on trench width.
1
Free cyanide concentration.— If the same kinetics apply, the CN
that is rejected from the interface during the Au deposition would be
expected to induce a gradient of opposite sign and of twice the
Comparison to experiment.— The CEAC model predictions in
Fig. 8 capture the geometry of superfill in the wide trench in Fig. 5,
though the fill time is somewhat underestimated. The predictions
also capture the experimentally observed transition from bottom-up
filling to v-notch that accompanies decrease of the trench width
magnitude of that of the arriving Au-cyanide ion ͓two CN for each
−
2
Au in the Au͑CN͒ ͔. For transport-limited deposition this could
increase the free-cyanide concentration adjacent to the surface by as
much as 40% ͑from 0.1 mol/L to 0.14 mol/L given the 0.1 mol/L
KCN and 0.02 mol/L KAu͑CN͒2 concentrations͒. Such an increase
would be expected to shift the potential for local equilibrium such
that further Au deposition would be retarded. This would be ex-
pected to be more significant in regions of higher areal density and
in narrower features and would be consistent with the observed de-
crease of deposition rate with decreasing trench width ͑and pitch͒
visible in Fig. 5 and Fig. 9 ͑specimen 1͒. Simulations including this
effect would require data on the impact of cyanide concentration on
the Pb-accelerated Au deposition rate. This task, going beyond the
scope of the envisioned CEAC-based modeling effort, has been left
for the future.
͑
Fig. 5, 400-s deposition͒. While fill time is predicted to decrease
somewhat with decreasing trench width, experimental filling of the
wider trench is clearly faster than that of the intermediate trench.
This result is quite interesting in that it contrasts with both copper
and silver superfill, which exhibit the predicted increase of fill time
12,15
with feature width.
Discussion
There was significant variability of the experimental Au fill re-
sults. Specifically, runs with nominally identical processing, i.e., de-
rivatization time and deposition time, exhibited varying degrees of
feature filling. Examples of two specimens with substantially differ-
ent filling in spite of nominally identical processing, 180-s derivati-
zation and 400-s final deposition, are shown in Fig. 9. The origin of
some of the variability in filling can be understood from the corre-
sponding current transients in Fig. 10, which shows all the transients
from specimens that were derivatized for 180 s. Specimen 2 with its
thick, voided deposits yielded the uppermost ͑decaying͒ current
transient after the derivatization ͑scaled current ϳ11, corresponding
Adsorbate consumption.— The simulations capture the impact of
area change, accumulation, and consumption on adsorbate coverage
and feature filling through Eq. 6. The consumption term, which
quantifies the failure of the adsorbate to remain on the surface, is a
deviation of the actual system from the ideal CEAC mechanism of
superfill. For the parameters in Table I, the consumption term yields
exponential decay of the preadsorbed Pb coverage by a factor of 1/e
to ␪ Ϸ 0.46, ignoring Au-ion depletion͒. The only other specimen
after deposition of 1/k Ϸ 0.3 ␮m Au. As the trenches used for
o
−
to exhibit voiding ͑not shown͒ yielded the other gradually decaying
transient, with its high scaled current ͑ϳ10͒. In contrast, specimen 1
of Fig. 9, with its bottom-up filling, yielded one of the lower clusters
experimental superfill were significantly narrower than this value,
predicted consumption of adsorbate on the sidewalls during deposi-
tion on these surfaces was fairly small. For conditions where super-
fill did occur, coverage on the upward-moving bottom surface was
predicted to be dominated by the area change term, as the adsorbate
from the eliminated sidewall area all ended up on this surface. Thus,
the consumption term was predicted to have played only a small role
in the observed filling and is not believed to underlie the observed
differences between prediction and experiment. Using values from
Table I, the Pb incorporation, equal to k−⌫␪ and maximum at
of rising curves ͑scaled current ϳ6, corresponding to ␪ Ϸ 0.23,
o
ignoring Au-ion depletion͒.
Figure 11 shows simulations using a starting coverage of
␪o = 0.45 ͑as per the uppermost current transient in Fig. 10 and
corresponding to specimen 2 of Fig. 9͒. The superfill dynamic is
predicted to occur only for the widest trench, with deep v-notch
cusps of conformal filling predicted for the two other trenches; seam
formation is avoided only because the tilted sidewalls permit geo-
metrical leveling to fill the trench. This shift toward conformal fill-
␪ = 1, is less than 2 ϫ 10− mol/cm , or, using the 7.11 cm molar
5
3
3
volume of Cu, less than 3 ϫ 10− atomic fraction.
6
Figure 11. Simulations of trench filling
for preadsorbed Pb coverages of ␪o
=
0.45 filling in the Pb-free electrolyte.
Kinetics used for the simulations can be
found in Table I. Trench dimensions and
pitches used for the modeling are given in
Table II.
Downloaded on 2015-03-29 to IP 152.14.136.77 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C18
Journal of The Electrochemical Society, 153 ͑1͒ C11-C18 ͑2006͒
Surface diffusion.— Surface diffusion of the adsorbed accelerator,
by reducing spatial gradients of adsorbate coverage, works against
the CEAC-induced variations that underlie the superfill mechanism.
Surface diffusion was neglected in this study, as in previous CEAC
models. The presence of sharp corners on the growth contours in Cu
superfill, in full agreement with CEAC predictions, supports neglect
of surface diffusion in that system. In contrast, the rounding of the
filling contours in Fig. 5, as opposed to the sharp corners of the
corresponding CEAC predictions, suggests that surface diffusion of
the Pb adsorbate might be more significant in the system studied
here. This is not unreasonable given that the depositions times are
approximately 2 orders of magnitude longer than those typical of Cu
superfill. Nonetheless, while surface diffusion would be expected to
be most detrimental to filling of the smallest trenches, consistent
with observation, it would not be expected to retard deposition rates
within the trenches, as suggested by the pinch-off voids in Fig. 9
specimen 2, nor does it seem likely that it would vary substantially
over the small range of dimensions studied here. Modeling of the
impact of surface diffusion of the catalyst on feature superfill is left
for the future.
References
1
. P. C. Andricacos, C. Uzoh, J. O. Dukovic, J. Horkans, and H. Deligianni, IBM J.
Res. Dev., 42, 567 ͑1998͒.
2. A. V. Davydov, L. A. Bendersky, W. J. Boettinger, D. Josell, M. D. Vaudin, K. S.
Chang, and I. Takeuchi, Appl. Surf. Sci., 223, 24 ͑2004͒.
3
4
. T. J. Kim and P. H. Holloway, Crit. Rev. Solid State Mater. Sci., 22, 239 ͑1997͒.
. D. Josell, C. R. Beauchamp, D. R. Kelley, C. A. Witt, and T. P. Moffat, Electro-
chem. Solid-State Lett., 8, C54 ͑2005͒.
. Z. Hu and T. Ritzdorf, Abstract 147, The Electrochemical Society Meeting Ab-
stracts, Vol. 2004-1, San Antonio, TX, May 9–13, 2004.
5
6
. Y. Cao, P. Taephaisitphongse, R. Chalupa, and A. C. West, J. Electrochem. Soc.,
1
48, C466 ͑2001͒.
7
8
. A. C. West, S. Mayer, and J. Reid, Electrochem. Solid-State Lett., 4, C50 ͑2001͒.
. D. Josell, D. Wheeler, W. H. Huber, and T. P. Moffat, Phys. Rev. Lett., 87, 016102
͑2001͒.
9. T. P. Moffat, D. Wheeler, W. H. Huber, and D. Josell, Electrochem. Solid-State
Lett., 4, C26 ͑2001͒.
0. D. Josell, D. Wheeler, W. H. Huber, J. E. Bonevich, and T. P. Moffat, J. Electro-
chem. Soc., 148, C767 ͑2001͒.
1. D. Wheeler, D. Josell, and T. P. Moffat, J. Electrochem. Soc., 150, C302 ͑2003͒.
2. T. P. Moffat, D. Wheeler, M. Edelstein, and D. Josell, IBM J. Res. Dev., 49, 19
͑2005͒.
1
1
1
13. 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, J. Electrochem. Soc.,
Phase formation.— Lead and gold form several intermetallic
phases. While the consumption rate experiments detailed in Fig. 2
indicate the adsorbed lead generally segregates to the surface at the
coverages studied, it is possible that such phase formation could
spontaneously occur at sufficiently high fractions ␪ of the saturation
coverage ⌫ used. As the location with highest ␪ is generally the
upward-moving bottom surface, with its CEAC-enhanced coverage,
such phase formation, if associated with increased consumption,
would be expected to negatively impact superfill. In the absence of
data to the contrary, consumption of the Pb adsorbate has been mod-
eled using the kinetics obtained from modeling of the transient data
in Fig. 1 and 2.
1
49, C423 ͑2002͒.
4. D. Josell, B. Baker, C. Witt, D. Wheeler, and T. P. Moffat, J. Electrochem. Soc.,
49, C637 ͑2002͒.
1
1
15. B. C. Baker, C. Witt, D. Wheeler, D. Josell, and T. P. Moffat, Electrochem. Solid-
State Lett., 6, C67 ͑2003͒.
6. B. C. Baker, M. Freeman, B. Melnick, D. Wheeler, D. Josell, and T. P. Moffat, J.
Electrochem. Soc., 150, C61 ͑2003͒.
1
1
7. D. Josell, S. Kim, D. Wheeler, T. P. Moffat, and S. G. Pyo, J. Electrochem. Soc.,
1
50, C368 ͑2003͒.
18. R. Yu. Bek, Russ. J. Electrochem., 38, 1237 ͑2002͒.
1
2
2
2
2
2
9. R. Yu. Bek and L. I. Shuraeva, Russ. J. Electrochem., 39, 229 ͑2003͒.
0. R. Yu. Bek and L. I. Shuraeva, Russ. J. Electrochem., 39, 872 ͑2003͒.
1. B. Bozzini and A. Fanigliulo, J. Cryst. Growth, 243, 190 ͑2002͒.
2. J. D. E. McIntyre and W. F. Peck, Jr., J. Electrochem. Soc., 123, 1800 ͑1976͒.
3. Dj. Davidović and R. R. Adzić, Electrochim. Acta, 33, 103 ͑1988͒.
4. P. Bindra, D. Light, P. Freudenthal, and D. Smith, J. Electrochem. Soc., 136, 3616
Conclusions
͑
1989͒.
5. D. Poškus, G. Agafonavas, and I. Jurgaitiene, J. Electroanal. Chem., 425, 107
1997͒.
6. E. T. Eisenmann, J. Electrochem. Soc., 125, 717 ͑1978͒.
Bottom-up superfill of gold in trenches with a preadsorbed sub-
monolayer coverage of Pb was explored quantitatively. The impact
of Pb adsorption from the electrolyte on the rate of Au deposition
and the kinetics of the Pb consumption process were explored
through studies on planar substrates. The obtained kinetics were
then used in the CEAC mechanism of superfill to quantitatively
predict Au superfill in trenches with submicrometer dimensions.
Agreement of prediction with experimental filling geometry was
good for the trenches studied, including bottom-up fill geometry and
transition from superfill to v-notch filling with decreasing trench
width. In contrast to prediction, however, filling rates decreased with
trench width and voiding rather than seam formation occurred at the
highest adsorbate coverages. Thus, while this work presented the
first quantitative prediction of Au superfill, it also signals the need
for further study.
2
2
͑
27. S. Nakahara and Y. Okinaka, J. Electrochem. Soc., 128, 284 ͑1981͒.
28. T. Sawaguchi, T. Yamada, Y. Okinaka, and K. Itaya, J. Phys. Chem., 99, 14149
͑
1995͒.
9. F. Huerta, C. Mele, B. Bozzini, and E. Morallon, J. Electroanal. Chem., 569, 53
2004͒, And references therein.
0. Y. Okinaka and M. Hoshino, Gold Bull. 31, 3 ͑1998͒.
2
͑
3
31. D. Davidović and R. R. Adžić, J. Serb. Chem. Soc., 53, 499 ͑1988͒.
32. R. R. Adzic, in Advances in Electrochemistry and Electrochemical Engineering,
Vol. 13, H. Gerischer and C. W. Tobias, Editors, J. Wiley and Sons, New York
͑1994͒.
3
3. D. Roha and U. Landau, J. Electrochem. Soc., 137, 824 ͑1990͒.
4. J. S. Newman, Electrochemical Systems, 2nd ed., p. 363, Prentice-Hall, Englewood
Cliffs, NJ ͑1991͒.
35. T. P. Moffat, D. Wheeler, C. Witt, and D. Josell, Electrochem. Solid-State Lett., 5,
C110 ͑2002͒.
3
The National Institute of Standards and Technology assisted in meeting
the publication costs of this article.
36. D. Wheeler, J. Guyer, and J. A. Warren, “FiPy: A Finite Volume PDE Solver Using
Python” at http://www.ctcms.nist.gov/fipy
Downloaded on 2015-03-29 to IP 152.14.136.77 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).