Electrodeposition of gold from a thiosulfate-sulfite bath for microelectronic applications

Article abstract of DOI:10.1149/1.1541006

The feasibility of using a thiosulfate-sulfite bath to electrodeposit soft gold for microelectronic applications has been investigated. This bath is stable at near-neutral pH. shows good compatibility with positive photoresists, and does not contain any additives. It was found that the bath produced gold bumps with straight sidewalls, flat top surfaces, and a good reproduction of the resist pattern was achieved. The thickness uniformity, roughness, stress, hardness, adhesion, and shape of the plated structures were found to be compatible with the requirements for a wide range of microelectronic applications, including wafer bumping.

Full text of DOI:10.1149/1.1541006

C104
Journal of The Electrochemical Society, 150 ͑3͒ C104-C110 ͑2003͒
0013-4651/2003/150͑3͒/C104/7/$7.00 © The Electrochemical Society, Inc.
Electrodeposition of Gold from a Thiosulfate-Sulfite Bath for
Microelectronic Applications
,z
*
T. A. Green, M.-J. Liew, and S. Roy
Department of Chemical and Process Engineering, Merz Court, University of Newcastle,
Newcastle upon Tyne, NE1 7RU, UK
The feasibility of using a thiosulfate-sulfite bath to electrodeposit soft gold for microelectronic applications has been investigated.
This bath is stable at near-neutral pH, shows good compatibility with positive photoresists, and does not contain any additives. It
was found that the bath produced gold bumps with straight sidewalls, flat top surfaces, and a good reproduction of the resist pattern
was achieved. The thickness uniformity, roughness, stress, hardness, adhesion, and shape of the plated structures were found to be
compatible with the requirements for a wide range of microelectronic applications, including wafer bumping.
© 2003 The Electrochemical Society. ͓DOI: 10.1149/1.1541006͔ All rights reserved.
Manuscript submitted May 2, 2002; revised manuscript received August 21, 2002. Available electronically January 28, 2003.
Electrodeposition of soft gold has numerous applications in mi-
nately, gold sulfite baths are unstable when the pH is reduced below
8.5. This instability arises due to the disproportionation of uncom-
plexed Auϩ ions to form colloidal gold5,10-12
croelectronics and microsystem technology.1-8 In the electronic
packaging industry, for example, interconnects between integrated
circuits ICs and external devices are performed using tape auto-
mated bonding ͑TAB͒, chip-on-glass ͑COG͒, and chip-on-flex
͑COF͒ techniques.1-3 A key process in all of these technologies is
gold wafer bumping. In addition to connecting driver ICs to flat
panel displays, gold bumping is used in many other packaging ap-
plications - particularly where a high density of input/output ͑I/O͒
connections is required.4 Electrodeposition of gold is also used in
the fabrication of transmission lines and air bridges on gallium ars-
enide ICs.5 The electrodeposition of soft gold has also been used to
fabricate X-ray masks for the deep X-ray lithography electroform-
ing, and plastic molding ͑LIGA͒ process.6-8
3Auϩ ⇔ 2 Au ϩ Au3ϩ
͓1͔
A number of approaches have been presented to stabilize sulfite
baths at low pH. For example, a gold sulfite bath operable at pH 8
has been reported10,11 using 2,2 -bipyridine as a stabilizer. Presum-
Ј
ably, this ligand complexes with free Auϩ ions thereby suppressing
the disproportionation reaction. Morrissey13,14 has also developed a
gold sulfite bath which is stable to a pH of 4.5 using polyamines and
aromatic nitro compounds as additives. Ammonium sulfite baths
have also been developed which are operable in the pH range
6-8.15,16 These electrolytes appear to be stable in the absence of
additives, but have not yet found wide use.
All of the above technologies utilize the through-mask plating
technique.9 Typically this involves depositing a conductive seed
layer on a wafer substrate, followed by patterning with a photoresist
material. The wafer is then gold plated, followed by resist removal
and seed layer etching. Although the requirement for each applica-
tion varies slightly, usually, gold deposits are required to have high
purity, low porosity, low stress, and good adhesion to the
substrate.1-8 It is also important that the fabricated features exhibit
straight sidewalls, a relatively planar top surface and are uniform in
height across the wafer.1-3 For reliable thermocompression bonding
͑e.g., in TAB͒ low deposit hardness is also required.2,3 Similarly, in
transmission-line applications,5 a high electrical conductivity and
low surface roughness is desirable.
More recently, Osaka and co-workers17,18 have proposed the use
of a gold plating bath containing sulfite and thiosulfate as ligands.
Since the stability constant of the Au(S2O3)32Ϫ complex is much
larger than for Au(SO3)32Ϫ (␤ ϭ 1026 vs. 1010) the former species
predominates; in effect, a gold thiosulfate bath is obtained.12 Note
that the existence of
a mixed sulfite-thiosulfate complex,
Au(S2O3)(SO3)3Ϫ, has also been proposed19 and under some con-
ditions this species may also be present in solution.12,17 Gold thio-
sulfate baths ͑without sulfite͒ have been developed earlier for the
electrodeposition of soft gold.20 However, they were found to be
unstable at neutral or acidic pH due to the disproportionation of free
thiosulfate ions to form colloidal sulfur21
For most applications requiring soft gold deposition through
a
photoresist
mask,
sulfite-based
plating
baths
are
employed.1,2,5,7,8,10,11 Cyanide-based electrolytes can also be used to
deposit soft gold,3 but due to their toxicity and poor compatibility
with many photoresists, they are rarely used. For example, the use of
gold cyanide baths often leads to the delamination of the resist from
the seed layer.3,5 This leads to gold deposition underneath the resist,
which is highly undesirable in most applications. In contrast, sulfite-
based baths exhibit better resist compatibility, and underplating is
not usually a problem.5,11 However, as gold sulfite baths are typi-
cally operated at pH values between 9 and 10,1,2,5,7,8 problems can
still arise when alkaline soluble positive resists are used. While the
dissolution of these resists is relatively slow under mildly alkaline
conditions, significant organic contamination of the plating bath can
still occur as the plating process continues. This can adversely in-
fluence deposit properties ͑e.g., hardness, stress, morphology͒ as
well as make process control difficult.
S2O23Ϫ ⇔ S ϩ SO23Ϫ
͓2͔
The addition of sulfite shifts the disproportionation equilibria to
the left, and the mixed ligand bath is stable at pH values as low as
6.17,18,21 Osaka found that these electrolytes produced deposits of
good quality whose properties matched the requirements for wafer
bumping applications. Deposit hardness was found to be high ini-
tially, but could reduced by annealing, increasing the total ligand
concentration, or by using a thallium additive.12,17,18
The primary aim of this study was to establish the suitability of a
neutral pH sulfite-thiosulfate gold plating bath for microelectronic
applications. Although Osaka and co-workers17,18 have already de-
termined the purity, morphology and hardness of gold films depos-
ited from a sulfite-thiosulfate bath, thickness uniformity, roughness,
stress, adhesion and bondability measurements are also necessary to
fully verify the performance of these electrolytes. A secondary aim
was to compare the deposit properties with those obtained from a
conventional alkaline sulfite bath, as it is important to demonstrate
that the sulfite-thiosulfate bath has equivalent or superior
performance.
The interaction between the resist and plating bath can be mini-
mized by making the solution neutral or slightly acidic. Unfortu-
* Electrochemical Society Active Member.
z E-mail: s.roy@ncl.ac.uk
Downloaded on 2014-06-26 to IP 134.117.10.200 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, 150 ͑3͒ C104-C110 ͑2003͒
C105
Table I. Composition of ECF60mod plating bath and operating
conditions.
Composition
͑M͒
0.05
0.42
0.42
HAuCl4
Na2SO3
Na2S2O3.5H2O
Current density
pH
Flow rate
Temperature
3.5, 5.5, and 7.0 mA cmϪ2
7.4 Ϯ 0.1
220 and 330 mL sϪ1
55°C
Experimental
Plating baths.—The plating baths were prepared by Engelhard-
CLAL, and their compositions are given in Tables I and II. The
ECF60 bath is a commercially available alkaline gold sulfite elec-
trolyte. The mixed sulfite-thiosulfate bath ͑designated ECF60mod͒
was of a similar formulation to that used by Osaka et al.17 The
starting material is an Au(III)ClϪ4 salt which subsequently under-
goes reduction and complexation by thiosulfate to form
Au(S2O3)23Ϫ . Although it is common1-3,7,8,13,17 to use additives such
as arsenic and thallium to improve deposit properties ͑e.g., rough-
ness, stress, and hardness͒, in this study, both electrolytes were
additive-free. The pH of the baths could be adjusted using dilute
H2SO4 or NaOH/KOH solutions, and were found to be very stable.
The standard operating conditions are also given in Table I and II.
These are based on earlier optimization experiments which involved
gold deposition on to a rotating cylinder electrode as a function of
solution composition, temperature, and current density.
Figure 1. Schematic of wafer showing the positions of the four electrode
contacts. The central grid area corresponds to the active bump area. The
cross-hatched regions correspond to resist-free areas. The wafer diam is
10 cm.
Wafer plating.—Wafers were plated in vertical flow channel,
which has been fully described in an earlier publication.22 In brief,
the wafer is mounted in one wall of the channel and a platinized
titanium mesh anode of equal diameter ͑i.e., 10 cm͒ is mounted in
the opposite wall. The gap between the anode and wafer was 1 cm
and the width of the channel was 12 cm. The wafer sits flush with
the channel wall, and electrical contact was made via four insulated
probes which contacted the front side of the wafer near the periph-
ery ͑Fig. 1͒. Note that the placement and contacting in this cell is
different from the original design22 where the wafer was mounted in
a small ͑0.3 cm͒ recess and electrical contact was made from the
back side of the wafer. This difference in arrangement is due to
practical problems regarding contacting and sealing and may lead to
some difference in mass transfer and current distribution as com-
pared to that described in Ref. 22.
Wafer preparation.—All electrodeposition experiments were per-
formed on 10 cm diam glass-like wafers. An exception was the
stress measurements, which were performed on 10 cm diam Si͗100͘
wafers. Both types of wafers were polished on one side and had a
surface roughness of less than 1 nm. The wafers were metallized
with a 20 nm Ti adhesion layer followed by a 30 nm Au seed layer.
They were then coated with an 18 ␮m thick positive photoresist
͑Shipley, SJR5740͒. After exposure and development of the photo-
resist, the wafers were cleaned in oxygen plasma to remove any
remaining resist residues.
The mask design used in the plating trials is shown in Fig. 1 and
2. It consisted of an array of 25 ϫ 26 squares with an overall di-
mension of 7.5 ϫ 7.8 cm. The positioning of this area with respect
to the wafer and electrical contacts is shown in Fig. 1. Each indi-
vidual square contained a 9 ϫ 9 array of bump features, as shown
in Fig. 2. The bump dimensions were 100 ϫ 100 ␮m and the dis-
tance between each bump was 300 ␮m. This regular pattern design
is expected to minimize non-uniformities in thickness arising from
variations in the resist density.9 The two cross-hatched regions in
Fig. 1 were completely resist-free to allow electrical connection be-
tween the electrode contacts and seed layer. All other areas of the
wafer were patterned with photoresist.
The plating system consisted of the following components: a
vertically mounted Perspex ͑polymethyl methacrylate͒ flow chan-
nel, a polypropylene plating reservoir equipped with a poly͑tet-
rafluoroethylene͒ encased heater, a magnetically coupled pump and
a filter chamber. The various components were connected using
Table II. Composition of ECF60 plating bath and operating con-
ditions. Note that these baths contained no proprietory additives,
buffers or conductive salts.
Composition
͑M͒
0.05
0.24
Na3Au(SO3)2
K2SO3
Current density
3.5, 5.5, and 7.0 mA cmϪ2
9.5 Ϯ 0.1
pH
220 and 330 mL sϪ1
45°C
Figure 2. SEM micrographs of plated gold bumps in a 0.3 ϫ 0.3 cm square
cell located on the wafer bump area.
Flow rate
Temperature
Downloaded on 2014-06-26 to IP 134.117.10.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C106
Journal of The Electrochemical Society, 150 ͑3͒ C104-C110 ͑2003͒
polypropylene piping and the flow was controlled via a network of
pneumatically controlled valves. Flow rates of 220 and 330 mL sϪ1
were used, and this corresponded to a turbulent flow regime.22 The
solution was continuously filtered through a pair of polypropylene
filters ͑1 ␮m retention͒ to remove particulates.
The electrodeposition experiments were performed under con-
stant current conditions using a computer-controlled direct current
power supply ͑Hewlett Packard, HP6613C͒. The flow rate was mea-
sured using an in-line paddle rotameter, and the temperature in the
flow channel was measured via a calibrated thermocouple. Overall
control of the plating system and the collection of all relevant pro-
cess parameters ͑i.e., cell voltage and current, flow rate and tempera-
ture͒ was performed using a data acquisition card and software writ-
ten in LABVIEW.
In a typical experiment, wafers were first weighed and then
mounted in the flow cell. The solution was then allowed to circulate
for 3 min prior to plating to allow sufficient time for thermal equili-
bration and to dislodge any air bubbles that may be trapped in the
resist pattern. After this time, a current corresponding to the desired
deposition rate was applied. Current densities of 3.5, 5.5, or
7.0 mA cmϪ2 were used based upon an active ͑i.e., resist-free͒ plat-
ing area of 16.8 cm2. From earlier measurements,23 the limiting cur-
rent densities of both plating baths at 220 and 330 mL sϪ1 are esti-
Figure 3. SEM micrographs of gold test structures plated in the ECF60mod
bath at 5.5 mA cmϪ2. The nominal widths of the gold lines are 6, 9, and 12
␮m.
mated to be 12.5 and 14.5 mA cmϪ2
, respectively. All gold
deposition experiments, hence, were performed where mass transfer
limitations should not exist. The plating duration was selected to
achieve a final Au thickness of approximately 15 ␮m. After plating,
the wafers were rinsed, dried, and then weighed in order to deter-
mine the current efficiency. The photoresist was then removed with
a commercial resist stripper and the deposit properties were mea-
sured. Each deposition experiment was repeated three times to en-
sure that results were reproducible.
ECF60mod baths were 98.2 and 99.1%, respectively. These results
are in good agreement with earlier studies that indicate the gold can
be deposited from sulfite and thiosulfate baths at nearly 100%
efficiency.11,13,16 There was no notable influence of current density
on the plating efficiency.
Deposit structure.—After the resist was removed, the plated
structures were examined by optical and scanning electron micros-
copy. For both electrolytes, a good reproduction of the photoresist
mold was obtained. The bumps typically had straight sidewalls, a
planar top surface, and are largely defect free. The gold films exhib-
ited a columnar microstructure, which was similar to that obtained
from the ECF60 bath.23 The overall quality of the plated structures
can be seen in Fig. 3. This figure also shows that it is possible to
replicate features with linewidths as low as 6 ␮m using the
ECF60mod bath. This level of resolution is sufficient for all soft
gold applications.
Deposit and bath characterization.—The deposit thickness and
surface roughness was measured using an optical profilometer
͑Veeco, NT2000͒. Note that the roughness values are reported as the
arithmetic roughness, Ra . Stress measurement was performed by
measuring the wafer curvature before and after electrodeposition
with a stylus profilometer ͑Ambios Technology, XP2͒. Hardness
measurements were performed with a nanoindenter ͑Hysitron Tri-
boindenter͒ using a variable force of 100-1000 ␮N. Deposit mor-
phology was characterized using a scanning electron microscope
͑SEM, JEOL, JSM-5600͒. Wire bondability tests were carried out by
thermosonically bonding 125 ␮m gold ribbon to each bump and
then performing a bond pull test with a tension gauge according to
MIL-STD-883E ͑method 2011.7͒.
Deposit uniformity.—The deposit thickness for each individual
square was characterized by measuring the height of thecentral
bump in each square cell. For example, in Fig. 2, the bump adjacent
to the letter L is measured. Typically, the thickness at each square
corresponding to every third row or column in Fig. 1 ͑i.e., 45 bumps
per wafer͒ was measured in order to map the thickness uniformity
across the pattern.
As the plating trials proceeded it was noticed that the average
thickness in patterned area was higher than that calculated from the
weight gain, the active plating area, and the density of gold ͑using
Faraday’s law͒. For example, when the current densities and plating
times employed corresponded to an average deposit thickness of 12
␮m, an average bump height of between 14-16 ␮m was typically
obtained. It was found that the gold thickness of the two crescent-
shaped electrode contact pads ͑cross-hatched regions in Fig. 1͒
counteracted this increased thickness. These resist-free areas, which
constitute approximately 50% of the total plating area, were found
to be typically 10 ␮m thick. It is well known that current tends to
concentrate in areas where the resist density is highest9,24 and this
may lead to a higher deposit thickness in the patterned area and a
lower thickness at the resist-free regions.
Results
Bath stability and resist compatibility.—For the evaluation of the
ECF60mod plating bath, a total of 30 wafers were plated over two
weeks. During this time there was no apparent degradation of the
plating solution, or changes in the overall plating behavior. In con-
trast, conventional sulfite baths solutions rapidly change color from
clear to purple when operated at near-neutral pH due to the forma-
tion of colloidal gold.10,11 The pH of the bath was also relatively
constant ͑i.e., Ϯ0.1) during the trial period. The bath was also stable
with regard to chemical decomposition, and extraneous gold
deposition5 on internal surfaces of the plating system was not ob-
served. After plating, each wafer was inspected for signs of dissolu-
tion, cracking, or delamination of the photoresist. For both the
ECF60 ͑pH 9.4͒ and ECF60mod ͑pH 7.3͒ baths, there were no in-
dication of any significant resist attack, nor were there any instances
of underplating. However, in the case of the ECF60 bath, some fine
bubbles or pits were visible within the resist after plating, indicating
some interaction between the solution and the resist. In contrast, the
resist exposed to the ECF60mod bath looked identical before and
after plating.
Results are shown in Fig. 4 for the ECF60mod bath at three
different current densities. It can be seen that the thickness unifor-
mity is relatively flat along the flow direction but there is a slight
asymmetry perpendicular to the flow, particularly at the higher cur-
rent densities. The variation in thickness did not exhibit any clear
trends, such as consistently higher thickness at the right for
The current efficiency for the gold deposition was determined
gravimetrically. The mean current efficiencies for the ECF60 and
Downloaded on 2014-06-26 to IP 134.117.10.200 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, 150 ͑3͒ C104-C110 ͑2003͒
C107
Figure 6. The dependence of the surface roughness, Ra , on the applied
current density and flow rate. ͑ᮀ͒ ECF60 at 220 mL sϪ1
and ͑᭡͒
ECF60mod at 220 mL sϪ1, and ͑᭹͒ ECF60mod at 330 mL sϪ1
,
.
current density. The data in Fig. 5 also indicates that the thickness
variation was lowered when the flow rate was increased from 220 to
330 mL sϪ1, indicating that the flow conditions have some influence
on the deposit uniformity.
Figure 4. The thickness distribution in the active bump area for wafers
plated in the ECF60mod bath at three different current densities.
Deposit roughness.—The deposit roughness at various current
densities and flow conditions for the ECF60 and ECF60mod baths is
illustrated in Fig. 6. Note that the roughness values reported are an
average of three individual bump measurements per wafer on three
different wafers. As per the figure, ECF60mod clearly produces
rougher deposits than ECF60. This is also visible in the SEM images
͑Fig. 7͒ and in the bump cross sections shown in Fig. 8. Figure 6
also shows that the roughness is nearly independent of either current
density or flow rate. This result is not unexpected since all experi-
ments were carried out well below the mass transfer limiting current
͑the highest applied current was only 55% of the limiting current
density͒. An interesting result with the ECF60mod bath was that the
minimum roughness occurred at an intermediate current density
͑i.e., at 5.5 mA cmϪ2). At higher or lower current densities, the de-
posits were rougher and more nodular.
5.5 mA cmϪ2 and at the left for 7.0 mA cmϪ2. The nonuniformity,
therefore, is attributed to small variations in the position of the con-
tacts in each deposition experiment, which is likely to change both
the current distribution and fluid flow. Nevertheless, the overall
thickness uniformity is good.
Figure 5 shows a comparison of the thickness uniformity ob-
tained for the two plating baths as a function of the applied current
density and flow rate. Note that the thickness uniformity is ex-
pressed as a percentage of the ratio of the standard deviation to the
mean, and the value at each current density is the mean variation of
three separate wafers. In all cases, the uniformity of the ECF60mod
is significantly better than that of ECF60, particularly at the lowest
Deposit stress.—Figure 9 summarizes the stress of gold films
electrodeposited on silicon wafers from sulfite and sulfite-thiosulfate
electrolytes. In addition to the ECF60 and ECF60mod, the stress for
an ECF64 electrolyte is also included. The latter bath operates at a
neutral pH and is formulated from ammonium gold sulfite.15 For all
of these experiments, a blanket 15 ␮m thick Au film was deposited
on the wafers. Stress values ranged from 6 to 70 MPa ͑tensile͒ with
the ECF60mod bath producing the most stressed deposits at a cur-
rent of 3 mA cmϪ2. Since the stress measurements were performed
at ambient temperatures, i.e., 25-35°C lower than the deposition
temperature, the measured values include a contribution from the
stress arising from the differing thermal expansion coefficients of Si
and Au. For gold electroplated on to silicon this corresponds to a
correction of around 1 MPa °CϪ1 25 When corrected for the thermal
.
expansion effect, the sulfitic baths show compressive stresses of the
order of 20 MPa. This correction also significantly lowers the de-
posit stress from the sulfite-thiosulfate electrolyte to between 5 and
40 MPa, although the stress is still tensile.
Adhesion and wire bondability.—The adhesion of the plated gold
bumps to the wafer surface was determined using a simple tape test.
All bumped wafers passed, but this test is qualitative and only gives
an approximate assessment of adhesion. Nevertheless, it does indi-
cate reasonably good adhesion of the bumps to the substrate. The
suitability of the plated gold for wire bonding5,6 was also determined
Figure 5. The dependence of the intrawafer thickness uniformity on the
applied current density and solution flow rate. The uniformity is expressed as
a percentage of the ratio ͑␴/mean͒. ͑ᮀ͒ ECF60 at 220 mL sϪ1
, ͑᭡͒
ECF60mod at 220 mL sϪ1, and ͑᭹͒ ECF60mod at 330 mL sϪ1
.
Downloaded on 2014-06-26 to IP 134.117.10.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C108
Journal of The Electrochemical Society, 150 ͑3͒ C104-C110 ͑2003͒
Figure 7. Surface morphology of gold
bumps plated in the ECF60 and
ECF60mod baths.
lowest hardness occurred at 3.5 mA cmϪ2 but tended to increase as
the current density was raised. Deposits from the ECF60 bath were
typically much harder ͑ca. 95-20 HV͒ but the hardness decreased
significantly as the current density was increased.
using a standard pull test on gold wire loops that had been ther-
mosonically bonded to the bumps. For the dimensions of the wire
used, a minimum pull strength of 12 g is required for wire bonding
applications ͑MIL-STD-883E, method 2011.7͒. For both baths, the
pull strength exceeded 20 g in all tests and was typically of the order
30-40 g. The breakage modes were, however, slightly different. For
the ECF60 bath, 100% of the breaks occurred on the ribbon wire. In
the case of the ECF60mod baths the breaks were distributed as
follows: ribbon wire ͑66%͒, ribbon/bump interface ͑21%͒, and
bump/substrate interface ͑13%͒. This latter does suggests that the
adhesion is slightly inferior to bumps plated in ECF60 but delami-
nation did require a relatively large force (Ͼ30 g) to initiate. This
weaker adhesion may relate to the tensile stress measured for the
ECF60mod bath.
Discussion
The results shown in the previous section clearly show that gold
electrodeposited from the ECF60mod bath has properties that are
compatible with most microelectronic applications. In terms of so-
lution stability, resist compatibility, wire bondability, adhesion, and
bump shape, the thiosulfate-sulfite bath has performance which is
comparable or exceeds that for a conventional alkaline sulfite bath.
Only in the case of stress and roughness does the ECF60mod bath
offer poorer performance, but the measured values are still adequate
for most uses. Probably the most critical performance criteria for
Deposit hardness.—Figure 10 shows the dependence of the de-
posit hardness on the applied current density. The nanoindenter mea-
surements indicated that the gold films deposited from the
ECF60mod bath typically had a mean hardness of 69-93 HV. The
Figure 9. Measured deposit stress for 15 ␮m gold films deposited from
various plating baths. Deposition temperatures are: 45°C ͑ECF60, ECF64͒ or
55°C ͑ECF60mod͒. All stresses are tensile.
Figure 8. Typical profiles of gold bumps deposited in the ECF60mod bath at
͑a͒ 3.5 mA cmϪ2 and ͑b͒ 5.5 mA cmϪ2
.
Downloaded on 2014-06-26 to IP 134.117.10.200 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, 150 ͑3͒ C104-C110 ͑2003͒
C109
gold bumps deposited from ECF60mod bath appeared to be ad-
equate. However, for the fabrication of gold X-ray masks a very low
͑i.e., Ͻ20 MPa) deposit stress is usually required.7,8 In its current
form the ECF60mod bath may not be suitable for the preparation of
such masks. The stress of gold films deposited from sulfite baths has
been shown7,8,11 to depend on many parameters ͑e.g., current den-
sity, temperature, and additives͒ and it may be possible to lower the
measured stress further by making appropriate adjustments to the
experimental conditions.
The hardness of the as-deposited gold observed in this study is
comparable to those obtained from other sulfite1,2,11,15 and
thiosulfate-sulfite17,18 plating solutions. For pure sulfite baths, hard-
ness values ranging from 60-120 HV have been typically reported.
Similarly, Osaka et al.17,18 obtained hardness values ranging be-
tween 80-120 HV from a thiosulfate-sulfite bath. Hardness is an
important deposit property in applications such as TAB which re-
quires simultaneous bonding of many individual leads to their cor-
responding gold bumps.2 A low hardness is desirable in this process
as it allows the gold bumps to deform when compression is applied.
In this manner, small variations in the thickness of the gold bumps
can be tolerated during the bonding process. Typically, a hardness of
60-80 HV is recommended for TAB,2,3 however, Gemmler et al.2
have shown that reliable thermocompression bonding can be
achieved with a deposit hardness as high as 100 HV. The
ECF60mod bath typically produced quite soft gold ͑ca. 69-93 HV͒
and is well suited to TAB processing. However, the gold hardness
from the ECF60 bath is on the borderline of acceptability for TAB
Figure 10. The dependence of the deposit hardness on the applied current
density. ͑ᮀ͒ ECF60 at 220 mL sϪ1; ͑᭡͒ ECF60mod at 220 mL sϪ1
.
microelectronic applications relate to thickness uniformity, rough-
ness, stress, and hardness of the deposits, and these will be discussed
in more detail in the following sections.
The overall thickness uniformity achieved in these experiments
is comparable to those reported in the literature and is adequate for
wafer bumping.2,3,8,15 In such applications, a maximum intrawafer
thickness variation of 5-8% is usually specified.2,3,15 This specifica-
tion could be met with ECF60mod bath as long as a current density
of 3.5 mA cmϪ2 ͑or less͒ is used. Note that it may be possible to
increase the thickness uniformity even further, since the plating con-
ditions and mask design have not yet been fully optimised. More
importantly, the uniformity of the ECF60mod is clearly superior to
ECF60 under identical experimental conditions. Since the concen-
tration of the supporting electrolyte is greater in the ECF60mod, the
ohmic drop in this electrolyte is also lower and the relative impor-
tance of activation overpotential is therefore greater. Since higher
kinetic limitations for electrodeposition usually produces better cur-
rent distribution, deposits obtained by ECF60mod show better uni-
formity as compared to ECF60.
For many applications, surface roughness is not particularly im-
portant and in some cases may actually improve the reliability of
bonding.3 Of more concern is the appearance of large nodules,
‘‘rabbit-ears,’’ or sloping profiles, which can limit the mechanical
contact to the bump and weaken the bond.2,3,15 For the ECF60mod,
the deposit was slightly nodular at some current densities but the
overall roughness (Ra Ϸ 200-250 nm) is still adequate for wafer
bumping.2,3 Moreover, as shown in Fig. 8, the surfaces are relatively
flat and did not exhibit any surface irregularities which would com-
promise bonding. The levels of roughness observed may be of more
concern in transmission line applications5 where it can cause attenu-
ation of microwave signals, and also in the fabrication of X-ray
masks.7,8 Note, however, that the bath has not been fully optimized
with respect to minimizing roughness and further improvements
may be feasible by additional changes to the solution composition or
plating conditions. It is always possible to reduce roughness by the
use of grain refiners and brighteners ͑e.g., arsenic, thallium͒ and
many soft gold plating processes1-3,7,8,13,17,18 make use of such ad-
ditives. Another possibility is pulse plating which is often employed
in soft gold plating to minimize grain size and surface
roughness.1,3,7 For all of the above reasons, the relatively high sur-
face roughness produced in the prototype ECF60mod bath is not
considered a serious limitation.
and may require a low temperature anneal11,15,17,18 before bonding.
Conclusion
The feasibility of using a thiosulfite-sulfite plating bath to elec-
trodeposit soft gold for microelectronic applications has been dem-
onstrated. This bath can operate at near-neutral pH and was stable
for periods of several weeks. The bath produced good quality gold
deposits without the addition of additives, and exhibited good com-
patibility with positive photoresists. Typically, the deposits had
straight sidewalls, flat top surface and were largely defect-free, mak-
ing them suitable for most wafer bumping applications.
The thickness uniformity of the deposit from the thiosulfate-
sulfite bath was found to be adequate for all of the applications
considered, and was superior to a conventional alkaline sulfite bath
operating under identical conditions. The surface roughness was,
however, higher than for the sulfite bath, but was still adequate for
the majority of applications. The deposit stress level was higher than
for sulfite baths, but this did not appear to significantly compromise
adhesion or reliability. The lowest stress and roughness occurred at a
current density of 5.5 mA cmϪ2, although the best thickness unifor-
mity was obtained at 3.5 mA cmϪ2. The deposit hardness was also
determined; it was sufficiently low to be suitable for applications
requiring thermocompression bonding. Finally, the wire bondability
was also examined and found to be adequate. Future work will focus
on modifying plating conditions to further reduce the surface rough-
ness and stress, and also investigate the influence of additives and
pulse plating.
Acknowledgment
This work was partially supported by the EPSRC͑UK͒ under
grant no. GR/N08254/01. The authors would like to thank S.J. Bull
͑University of Newcastle͒ for performing hardness measurements
and P.J. O’Hare ͑Ambios Technology͒ for arranging stress measure-
ments.
The University of Newcastle assisted in meeting the publication costs of
this article.
References
For gold deposition from sulfite baths, film stresses of between
10 and 100 MPa have been reported.7,8,11,13 These are comparable to
the results obtained in this study. Large film stresses can cause
delamination of the film from the substrate, but the adhesion of the
1. H. Watanabe, S. Hayashi, and H. Honma, J. Electrochem. Soc., 146, 574 ͑1999͒.
2. A. Gemmler, W. Keller, H. Richter, and K. Ruess, Plat. Surf. Finish., 81, 52 ͑1994͒.
3. J. Traut, J. Wright, and J. Williams, Plat. Surf. Finish., 77, 49 ͑1990͒.
4. J. Jasper and D. Shiels, European Semiconductor, 22, 86 ͑2000͒.
Downloaded on 2014-06-26 to IP 134.117.10.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

C110
Journal of The Electrochemical Society, 150 ͑3͒ C104-C110 ͑2003͒
5. K. Kosaki, M. Matsuoka, Y. Seiwa, S. Orisaka, K. Nishitani, and M. Otsubo,
Electrochemical Microfabrication, M. Datta, K. Sheppard and D. Snyder, Editors,
PV 92-3, p. 317, The Electrochemical Society Proceedings Series, Pennington, NJ
͑1992͒.
6. W. Chu, H. I. Smith, S. A. Rishton, D. P. Kern, and M. L. Schattenburg, J. Vac. Sci.
Technol. B, 10, 118 ͑1992͒.
7. W. J. Daukler, D. J. Resnick, W. A. Johnson, and A. W. Yanof, Microelectron. Eng.,
23, 235 ͑1994͒.
8. W. Chu, M. L. Schattenburg, and H. I. Smith, Microelectron. Eng., 17, 223 ͑1992͒.
9. L. T. Romankiw and E. J. M. O’Sullivan, in Handbook of Microlithography, Mi-
cromachining and Microfabrication, Vol. 2, P. Rai-Choudhury, Editor, p. 197, SPIE
Press, Bellingham, WA ͑1997͒.
10. H. Honma and Y. Kagaya, J. Electrochem. Soc., 140, L135 ͑1993͒.
11. H. Honma and K. Hagiwara, J. Electrochem. Soc., 142, 81 ͑1995͒.
12. Y. Okinaka, Gold Bull. (Geneva), 33, 3 ͑2000͒.
13. R. J. Morrissey, Plat. Surf. Finish., 80, 75 ͑1993͒.
14. R. J. Morrissey. U.S. Pat. 5,277,790 ͑1994͒.
16. P. Laude, E. Marka, and Z. Morrens, U.S. Pat. 4,192,723 ͑1980͒.
17. T. Osaka, A. Kodera, T. Misato, T. Honma, and Y. Okinaka, J. Electrochem. Soc.,
144, 3462 ͑1997͒.
18. T. Osaka, M. Kato, J. Sato, K. Yoshizawa, T. Honma, Y. Okinaka, and O. Yoshioka,
J. Electrochem. Soc., 148, C659 ͑2001͒.
19. T. Inoue, S. Ando, H. Okudaira, J. Ushio, A. Tomizawa, H. Takehara, T. Shimazaki,
H. Yamamoto, and H. Yokono, Proceedings of the 45th IEEE Electronic Compo-
nents Technology Conference, p. 1059 ͑1995͒.
20. A. M. Sullivan and P. A. Kohl, J. Electrochem. Soc., 144, 1686 ͑1997͒.
21. Y. Okinaka and M. Hoshino, Gold Bull. (Geneva), 31, 3 ͑1998͒.
22. S. Roy, Y. Gupta, and T. A. Green, Chem. Eng. Sci., 56, 5025 ͑2001͒.
23. S. Roy and S. Caprodossi, in Electrochemical Technology Applications in Electron-
ics III, C. Madore, T. Osaka, L. T. Romankiw, and Y. Yamazaki, Editors, PV 99-34,
p. 145, The Electrochemical Society Proceedings, Pennington, NJ ͑1999͒.
24. S. Mehdizadeh, J. Dukovic, P. C. Andricacos, L. T. Romankiw, and H. Y. Cheh, J.
Electrochem. Soc., 140, 3497 ͑1993͒.
15. J. Simon, W. Zilske, and F. Simon, in Proceedings of the 1995 International Flip
Chip, Ball Grid Array, TAB and Advanced Packaging Symposium, Semiconductor
Technology Centre, Inc., p. 275 ͑1995͒.
25. T. C. Hodge, S. A. Bidstrup-Allen, and P. A. Kohl, IEEE Trans. Compon., Packag.
Manuf. Technol., Part A, 20, 241 ͑1997͒.
Downloaded on 2014-06-26 to IP 134.117.10.200 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).