Synthesis, Characterization, and Tunable Optical Properties of Hollow Gold Nanospheres†
Adam M. Schwartzberg,‡,§Tammy Y. Olson,‡,§Chad E. Talley,§and Jin Z. Zhang*,‡
Department of Chemistry and Biochemistry, UniVersity of California, Santa Cruz, California 95064,
and Department of Chemistry and Materials Science, Lawrence LiVermore National Laboratory,
LiVermore, California 94550
ReceiVed: April 5, 2006; In Final Form: May 5, 2006
Nearly monodisperse hollow gold nanospheres (HGNs) with tunable interior and exterior diameters have
been synthesized by sacrificial galvanic replacement of cobalt nanoparticles. It is possible to tune the peak of
the surface plasmon band absorption between 550 and 820 nm by carefully controlling particle size and wall
thickness. Cobalt particle size is tunable by simultaneously changing the concentration of sodium borohydride
and sodium citrate, the reducing and capping agent, respectively. The thickness of the gold shell can be
varied by carefully controlling the addition of gold salt. With successful demonstration of ensemble as well
as single HGN surface-enhanced Raman scattering, these HGNs have shown great potential for chemical and
biological sensing applications, especially those requiring nanostructures with near-IR absorption.
Nanostructured materials of coinage metals such as gold
and silver have unique optical properties due to strong
surface plasmon absorption in the visible region of light and
provide excellent substrates for surface plasmon resonance
(SPR) spectroscopy1-9and surface-enhanced Raman scattering
(SERS).10-19The surface plasmon is a collective oscillation of
conduction-band electrons within the nanoparticle induced by
the oscillating dipole of a resonant wavelength of light.20The
electron oscillation induces a surface electromagnetic (EM) field,
which is largely responsible for the SERS effect.21The
wavelength at which a given nanoparticle is resonant depends
on the size, shape, chemical nature of the metal, and the
In solid spherical particles, there is a single resonance at
approximately 520 nm for gold and 400 nm for silver, varying
slightly depending on size and embedding media. However,
when one axis is extended, for example, a nanorod, the
resonance will break into two absorption bands, one corre-
sponding to the short axis, or transverse mode, and another to
the long axis, or longitudinal mode.25-28The longitudinal mode
has lower energy or redder absorption than the transverse mode.
This is also true for aggregated systems in which there are
multiple resonances within each given cluster of particles.29-35
Therefore, controlling the size and shape of these metal
nanostructures allows control of their optical properties that have
potential applications in nanophotonics and sensing.
Control of the structure and thereby optical properties of gold
and silver nanomaterials is especially important for SERS
applications because the SERS enhancement factor depends on
the optical absorption of the substrate.11,36-38Only nanostruc-
tures with absorption on-resonance with the incident light will
contribute to the SERS signal, whereas those with absorption
off-resonance with the incident light wavelength will contribute
none or little to SERS.39-42In ensemble average experiments
involving a large number of nanoparicles, optical and structural
variations from particle to particle average out to yield consistent
results from measurement to measurement. However, when
examined at the single-nanoparticle or single-molecule level,
the SERS spectrum and enhancement can vary significantly from
one nanostructure to another because of their different structure
and optical absorption. This is often complicated further by the
necessity of aggregation of nanoparticles to achieve enhance-
ments large enough for single-molecule observations.39,40,42
Because aggregation is generally a random process, structural
homogeneity is nearly impossible. It is essential to have high
homogeneity or uniformity in the structure and optical absorp-
tion of the different nanostructures to achieve good consistency
on a single nanostructure level. The solution to this is to design
homogeneous SERS substrates that can provide significant
single-particle enhancement without aggregation.
The first works to realize this goal were the so-called core/
shell systems.43,44There are two major advantages of the core/
shell system over standard solid gold or silver particles. First,
because the synthesis of the core silica particle is well
characterized, size tunable from 100 nm to more than a
micrometer, and monodisperse, so is the resultant core/shell
structure.45This leads to a nanostructure that is optically tunable
from the visible to the IR and highly consistent on a particle-
by-particle basis.44This is ideal for in situ biological studies
because tissue has an absorption minimum in the IR. The second
major advantage is their SERS response, even at the single-
particle level.46Using a shell of gold or silver versus a solid
particle allows the electromagnetic field to extend further from
the surface, inducing greater enhancements than single, spherical
particles.43There is also, most likely, some portion of the
increased enhancement coming from surface roughness of the
shells that is not present in the single particles. In these systems,
the shell is most likely not single-crystalline and involves
aggregates of nanoparticles.
†Part of the special issue “Charles B. Harris Festschrift”.
* Corresponding author. E-mail: email@example.com. Phone:
(831) 459-3776. Fax: (831) 459-2935.
‡University of California.
§Lawrence Livermore National Laboratory.
J. Phys. Chem. B 2006, 110, 19935-19944
10.1021/jp062136a CCC: $33.50© 2006 American Chemical Society
Published on Web 06/29/2006
As an effort to engineer so-called “hot spots” of large
enhancement in single particles, Lee et al. produced nano-
crescent structures by depositing silver over latex beads on a
surface, then dissolving away the bead.47These hollow spheres
are open-ended with a sharp edge, which greatly enhances the
EM field. This engineered hot-spot approach yields improved
SERS enhancements over core/shell systems and is of a similar
homogeneity because of the highly consistent latex beads
available. For applications requiring an extremely small probe
size, however, both nanocrescents and core/shell systems are
A system of particular interest where probe size is of utmost
importance is intracellular studies.48It has been found that
although particles larger than 100 nm can enter a cell they do
not do so readily and may interrupt some cellular functions.
Similarly, particles that are too small, less than 20 nm, will
diffuse out of the cell, rendering them useless. The ideal is a
structure that can be tuned in size between 20 and 100 nm
depending on the application.
By utilizing sacrificial galvanic replacement of silver with
gold, Xia et al. have produced nanostructures with tunable sizes
and excellent optical properties.49-52This synthetic technique
utilizes the redox potential between metallic silver and gold salt
in solution. When the Au3+ions come in contact with the silver
atoms, there is an electroless plating that reduces the Au3+ions
to gold atoms and oxidizes the silver to Ag1+ions. For every
three silver atoms oxidized, a single gold atom is reduced,
leading to structures with1/3the metal, leaving a hollow core.
Because the structure is no longer a solid sphere, the plasmon
resonance is shifted from its normal position, similar to the core/
shell systems. Recently, this work has been extended to hollow
nanocubes, which have been shown to be SERS active, however,
not at the single-particle level.53
We have shown that by using a similar technique developed
by Liang et al.,54in which cobalt is used as the sacrificial
template to create hollow gold nanospheres (HGNs), it is
possible to observe SERS at the single-particle level with 30
nm HGNs.55Along with this single-particle sensitivity, we found
significantly increased homogeneity over solid silver nanopar-
ticle systems for the application of single-particle pH sensing
using surface functionalization with a pH-sensitive probe
molecule. However, only a single particle size with limited
plasmon tunability was used in that study. The ideal substrate
would be highly tunable in both size and resonance while
To achieve broad size tunability of this HGN system, size
control of the cobalt sacrificial template is required. Although
there have been significant advances in the production of small
sized (∼5-10 nm) cobalt nanoparticles with incredible homo-
geneity,56there has been little work in the production of larger
particles. The main instance of this work is by Kobayashi et al.
who showed significant size control in the production of silica
capped cobalt particles by varying the CoCl2/citric acid ratio.57
Citric acid is the capping agent that stabilizes the particles, and
its concentration strongly affects the number of nucleation sites
that initiate particle growth. The key to controlling particle size
is in the concentration of nucleation sites. In general, at a given
metal salt concentration, the more nuclei that are formed, the
smaller the average particle size will be. This is a simple mass
distribution argument; however, controlling the production of
nuclei is not necessarily a simple matter.58,59
Since the early work by Turkevich et al. and later Frens et
al.,60,61it has been understood that in a standard colloidal gold
synthesis using the hot citrate reduction of chloroauric acid, the
particle size may be controlled by the concentration of citrate.
In general, citrate stabilizes the initially formed nuclei and the
more citrate present the more nuclei will be stabilized. However,
when trying to apply this logic to the aqueous synthesis of cobalt
nanoparticles, it is a significantly more challenging task.
Because of the stability of the cobalt salt, the reduction cannot
be done by citrate alone and a stronger reducing agent is
required. In this case, sodium borohydride is used to reduce
the salt and citrate is present only as a capping agent. In this
work, we present the synthetic route necessary to control the
particle size of the cobalt nanoparticles, which is reflected in
the resultant HGN diameter. The inner diameter, or wall
thickness, can be controlled by the amount of gold salt used,
leading to complete control of the optical properties of particles
ranging from 20 to 70 nm. This makes it possible to tune the
peak of the surface plasmon band absorption between 550 and
820 nm. For a particular diameter and wall thickness, the
absorption band is relatively narrow because of the near-
monodisperse distribution, as determined by single-nanosphere
scattering spectrum. These HGNs have been further demon-
strated to be active SERS substrates with excellent consistency
based on a single HGN SERS spectra.
Materials and Methods
Materials. Cobalt chloride hexahydrate (99.99%), chloroauric
acid trihydrate (ACS reagent grade), trisodium citrate dihydrate
(>99%), citric acid (99%), 4-mercaptobenzoic acid, and sodium
borohydride (99%) were obtained from fisher scientific. All
water used in the syntheses was 18 MΩ milli-Q filtered.
Cobalt Nanoparticle Synthesis. Cobalt nanoparticles were
synthesized with the utmost attention paid to cleanliness and
exclusion of air. All glassware was cleaned with alconox
glassware detergent and then with aquaregia to ensure the
removal of all adsorbates; it was then washed repeatedly with
ultrapure water. To ensure completely air-free solutions, all
solutions were vacuumed on a Schlenk line until gas evolution
ceased and then bubbled with ultrapure argon for 10 minutes.
This process was repeated twice to remove as much oxygen as
possible from the reaction vessel.
Fast Addition of Cobalt Chloride. Water (100 mL) was placed
into a three-neck flask with 100-800 µL of a 0.1 M solution
of sodium citrate or citric acid and deoxygenated. To this, 100-
800 µL of a fresly made 1 M sodium borohydride solution was
added. With rapid magnetic stirring, 100 µL of a 0.4-0.6 M
cobalt chloride solution was added. Hydrogen evolved im-
mediately, and the solution changed from pale pink to brown/
gray indicating the reduction of Co (II) into cobalt nanoparticles.
This solution was allowed to react for between 15 and 60 min,
(under constant argon flow) depending on sodium borohydride
concentration, until hydrogen stopped evolving, indicating
complete hydrolysis of the reductant. The addition of sodium
borohydride and cobalt chloride was also performed in reverse
Slow Addition of Cobalt Chloride. Water (75 mL) was placed
in a 500 mL three-neck flask with 400 µL of a 0.1 M solution
of sodium citrate. Water (25 mL) with 100 µL of 0.4 M cobalt
chloride was placed in a 250 mL three-neck flask. These two
solutions were deoxygenated. To the 500 mL three-neck flask,
300-400 µL of a freshly prepared 1 M sodium borohydride
solution was added. Using a cannula and argon gas to pressurize
the 250 mL flask, the cobalt chloride solution was added drop-
wise at approximately 10 mL/min. During this addition, the
solution slowly changed from colorless to brown/gray signifying
cobalt particle formation. This solution was allowed to react
for 25 min to completely hydrolyze the sodium borohydride.
19936 J. Phys. Chem. B, Vol. 110, No. 40, 2006
Schwartzberg et al.
Gold Shell Growth. Because of the ease with which sodium
borohydride is able to reduce the gold salt, it is imperative that
it be completely hydrolyzed before introducing gold. The
presence of sodium borohydride is checked by halting stirring
and inspecting the solution for bubbles, indicating the continuing
hydrolysis of the reductant. It is only when bubbling has ceased
completely that gold may be added.
High-Concentration Addition. Upon ensuring complete hy-
drolysis of the sodium borohydride, the flow of argon was
increased and a 0.1 M solution of chloroauric acid was added
at 50 µL/addition to a total volume between 150 and 450 µL.
Between each addition, 30-60 s are allowed to pass to ensure
complete mixing. Upon completion of gold addition, the argon
flow was stopped and the vessel was opened to ambient
conditions under rapid stirring to oxidize any remaining cobalt
metal left in solution.
Low-Concentration Addition (Retaining Co at Core). Using
a cannula, 30 mL of the sodium borohydride-free cobalt
nanoparticle solution was transferred to an argon-purged gradu-
ated cylinder. This was then rapidly added to a vortexing 10
mL solution of chloroauric acid. The gold solution contained
between 20 and 60 µL of cloroauric acid diluted to 10 mL.
This solution may be kept under argon flow to retain the cobalt
core; however, by exposing the solution to air the cobalt was
completely oxidized, leaving only water and dissolved salts at
the core of the HGN. Samples with remaining cobalt cores retain
a brown color, whereas oxidized samples change to between
purple- and green-colored depending on the amount of gold
added and size of the particle.
Experimental Measurements. Low-resolution TEM mea-
surements were performed on a JEOL model JEM-1200EX
microscope, and high-resolution TEM was performed on a
Philips CM300-FEG at the national center for electron micros-
copy at Lawrence Berkeley National Laboratory. Absorption
measurements were taken on an HP 89532A spectrometer. All
spectra were fit with Igor Pro 5.0 using a Lorentzian function
with chi square values less than 0.1. Particles were sized with
imageJ image processing software.62
SERS and Rayleigh scattering measurements were performed
on a home-built confocal microscope system described previ-
ously42with the addition of transmitted light, dark field
illuminator (NA 1-1.4). For SERS experiments, a Zeiss
Apochromat 60 X, 1 NA air objective was used. Typically, the
sample was integrated for 30 s with a total power of 100 µW
from a helium-neon laser (632.8 nm, Melles Griot). Rayleigh
scattering experiments were performed with a Zeiss Apochromat
100 X, 0.7 NA oil-emersion objective.
Samples for Rayleigh scattering were prepared by immobiliz-
ing the particles on glass coverslips with trimethoxy-[3-
(methylamino)propyl]silane (APS) (Aldrich). Coverslips were
cleaned prior to the silanization step by sonication in a 2%
solution of Hellmanex, followed by 18 MΩ water. They were
then submerged in a 5 mM aqueous solution of APS to deposit
the tethering molecules. After 1-2 min, the coverslips were
rinsed with water, dried under nitrogen, and 40 µL of the as-
prepared particle solution was placed on one surface. After
several seconds exposed to the solution, it was rinsed with water
and then blown dry with nitrogen.
Results and Discussion
Effect of Cobalt Chloride, Sodium Borohydride, and
Sodium Citrate Concentration on Particle Size. The goal of
this study was to gain control of the cobalt particle size by
aqueous solution chemical methods. Previous work on this
system by Liang et al. focused more on the thickness of the
shell to control its optical properties.54Although their work
produced excellent results, further tunability is necessary to
make the system as useful as possible. Initial attempts to
reproduce the work of Liang did not yield satisfactory results.
The particles obtained were inhomogeneous and significantly
smaller than the 60 nm reported. In fact, using as close to the
original synthesis as possible, ∼25 nm cobalt particles were
obtained; however, with their method of gold addition only
inhomogeneous, gray solutions were observed. Upon determin-
ing an improved method of gold addition, this yielded excellent
results for single-particle SERS probes. However, there are many
applications that may benefit from larger particle size and further
red-shifted absorption, including SERS.
The other guiding hand in this work was provided by
Kobayashi et al, who first reported this cobalt particle synthesis,
but proceeded to cap the particles with silica shells to protect
them from oxygen.57They found that as citrate concentration
was reduced, particle size increased. This is consistent with
colloidal gold and silver syntheses and is not an unreasonable
claim. For this application, however, their trend did not hold
true. A significant difference between this work and that of
Kobayashi is the time at which the reaction could be halted. In
their work, for large cobalt particles, they were forced to add
the silica growth reagents almost immediately upon reduction
of the cobalt salt. Any delay at low citrate concentration and
the solutions would become unstable and flocculate. In this
work, however, if the gold solution is added too quickly, it is
immediately reduced by the remaining sodium borohydride
instead of the cobalt particles. This leads to an unfortunate mess
of nanoparticles. To achieve optimal particle growth, a signifi-
cant amount of time must pass in order to allow the sodium
borohydride to completely hydrolyze before the gold can be
This being said, it is also important to note that even at
relatively high concentrations of citrate where the particles are
still stable after some time there is little change in particle size
by merely altering the citrate concentration. There may be a
relatively simple explanation for this observation. Because the
particle stability is directly related to the concentration of citrate,
there may have been an aggregation affect responsible for the
size increase observed previously. As citrate concentration is
reduced, we have observed that the rate of aggregation increased.
Therefore, when capping the particles immediately after reduc-
tion, they are likely halting the aggregation at different stages
depending on citrate concentration. When concentration is low,
a larger aggregate will be formed before the silica can stabilize
it; at high concentration, a smaller aggregate will be present.
This may be responsible for the lack of crystalline structure in
the as-synthesized particles. By sintering them at high temper-
ature, they are likely fused into one crystalline particle.
Why then, does citrate not affect particle size as strongly as
previously thought? In the case of colloidal gold, the reduction
is done by the relatively weak reductant, citrate. This reaction
is slow, which allows for thermodynamic processes to control
the formation of clusters. Only as many seed particles will be
formed in the reaction as can be stabilized by the capping agent/
reductant. This means that the capping agent concentration will
have a strong effect on the number of seed particles and, hence,
particle size. In the formation of cobalt particles, however, a
much stronger reducing agent is required. Because sodium
borohydride is a significantly stronger reductant than is techni-
cally required to reduce the cobalt salt to cobalt metal, the
reduction is extremely fast, taking place in less than 1 min as
Hollow Gold Nanospheres
J. Phys. Chem. B, Vol. 110, No. 40, 2006 19937
opposed to 5-10 min for the reduction of gold salt by citrate.
Because of this, kinetic processes dominate the formation of
seed particles. The number of seeds, and therefore the size of
the resulting particle, will be more dependent on the rate of the
The rate of reduction can be controlled in several ways.
Temperature plays a strong role in the rate of reaction; however,
little change in particle size was observed between particles
synthesized at 0 °C and room temperature. A second way to
alter rate is by changing the solution pH. The reductive potential
of sodium borohydride is pH-dependent. It is important to note
at this point that, contrary to previous reports of this synthesis,
we use sodium citrate instead of citric acid. This is because the
reaction was found to be slower at the higher pH, and particle
homogeneity was superior in the neutral solution. Higher and
lower pH was also attempted by adjusting with HCl and NaOH.
These solutions, however, were unstable and immediately
crashed out. This is most likely due to the presence of excess
ions, especially Cl-, which has a strong disrupting effect on
aqueous colloidal capping. Finally, the concentration of reduc-
tant was used to change the reaction rate. This was found to be
the best method of controlling particle size without drastically
decreasing particle homogeneity.
By decreasing the amount of sodium borohydride present,
the reaction time is increased substantially. This produces larger
particles that remain stable in solution. Table 1 shows the result
of varying sodium borohydride concentration by one-quarter.
The particle size is increased by approximately 40%; however,
this is the practical limit of size tunability by this method. Lower
concentrations produce incredibly inhomogeneous results, which
are often unstable. To form larger particles, we must also alter
the sodium citrate concentration.
Although the sodium borohydride reduction of metal salts is
largely kinetics-driven, there are still some thermodynamic-type
processes controlling particle size. This is especially true as the
concentration of reductant is decreased and the reaction is
slowed. The reaction is now substantially more thermodynami-
cally controlled, making the variation in capping-agent con-
centration more effective in controlling particle size. By
decreasing both NaBH4and citrate concentration, we observed
a drastic increase in particle size, this is shown in the 3D plot
in Figure 1 and Table 2. The trend appears to be linear, at least
within the concentrations shown here. At lower concentrations,
the particle sizes could be substantially larger; however, because
they crash out of solution almost immediately this is not
something we could test. We present this as a general method
of tuning the size of cobalt nanoparticles. Using this plot, it is
possible to predict roughly what the final particle size will be
at a given sodium borohydride and sodium citrate concentration.
Influence of the Rate of Addition and Concentration of
CoCl2on Particle Homogeneity. Sample homogeneity is an
important parameter to control, especially when focusing on
optical properties. Inhomogeneous samples will have signifi-
cantly broadened ensemble-averaged plasmon bands, making
the sample less desirable for optical applications such as SERS.
To increase particle homogeneity and size, a slow addition of
low-concentration cobalt salt was attempted. It was thought that
this would artificially slow the rate of reaction. This, however,
was not the case, as is clear in Table 3. Although slightly larger
particles were achieved, the coefficient of variation increases
from 7% to 18%. This is clearly not an advantageous method
of controlling particle size. The reason for this great increase
TABLE 1: Dependence of Particle Size on Sodium
31 ( 2
44 ( 5
aAll reactions were performed in 100 mL of water. All particle sizes
are determined by examining the resulting gold particles. Reported sizes
Figure 1. Particle size as a function of citrate and sodium borohydride
concentration. All particle sizes are determined by examining TEM
images of the resulting gold structures and represent the measurement
of at least 200 particles. Reported sizes are diameters.
TABLE 2: Particle Size as a Function of Citrate and
Sodium Borohydride Concentration Shown Illustratively in
aAll reactions were performed in 100 mL of water. All particle sizes
are determined by examining the resulting gold particles. Reported sizes
TABLE 3: Influence of Rate of Addition and Concentration
of Cobalt Salt on Particle Sizea
28 ( 2
31 ( 6
50 ( 5
aAll particle sizes are diameters. The cobalt chloride solution used
for the slow addition is diluted to 25 mL with water.
19938 J. Phys. Chem. B, Vol. 110, No. 40, 2006
Schwartzberg et al.
in variation is due to the continual formation of seed particles
as the cobalt is added. When examining the particles, it is
obvious that some seeds are formed initially and result in very
large particles, whereas others are formed throughout the
addition and lead to small particles. This is clear in Figure 2,
which shows histograms of particle size from slow and fast
addition of cobalt. Not only does this exemplify the inhomo-
geneity of the slow addition sample but it also shows the
asymmetric formation of particles. While the fast addition yields
a nice, even sample, the slow addition yields a curve broadened
and asymmetrically shifted by the presence of large particles
formed early in the cobalt addition. This is clearly not the way
to increase particle size. By increasing the concentration of
cobalt while maintaining volume, however, we have found that
particle size changes drastically without excessively broadening
particle distribution; this is also clear in Table 3. Although higher
concentrations of cobalt seem to induce flocculation, it may be
possible to better control this with careful changes in citrate
Formation of Gold Shells. Along with the tunability of cobalt
particle sizes, we have been able to produce a wide variety of
sizes of the HGNs as shown in Figure 3. These are representative
TEM images of the HGNs at different sizes. Figure 3A is a
high-resolution TEM of a 30 nm particle; the lattice fringes of
gold are clearly defined and show that these particles are
polycrystalline with large single-crystalline areas. Figure 3B-F
shows the tunability of the samples, from 70 to 28 nm. The
largest particle sample in 3B clearly demonstrates the inhomo-
geneity that seems to be inherent at larger sizes. Also clear from
these images, there is some very small (2-5 nm) gold particle
formation in some samples. The origin of these particles is not
entirely clear; however, they are more prevalent at high gold
concentrations. They are likely a byproduct of the shell growth,
small particles that do not grow into the shell, but break off
early in the process.
Forming the gold shell seems to be an extremely simple
matter at first glance; however, under closer inspection it
becomes clear that there are many parameters that must be
carefully controlled in order to form high-quality samples. As
mentioned above, attempting to recreate the previous works did
not result in good samples. Another method was needed to make
homogeneous samples of high optical and structural quality like
those shown in Figure 3.
High-Concentration Gold Addition. The general consensus
on homogeneous nanoparicle formation is that a low concentra-
tion of reagents yields the best results. It is important to
remember, however, that in the addition of gold here we are
not forming a normal colloidal nanoparticle system. All that
determines particle size and shape is the sacrificial template.
For this reason, the high-concentration addition of gold should
not necessarily produce poor results. After many failed attempts,
it was found that by adding high-concentration (0.1 M) gold
salt in small volumes yielded excellent results. Adding the gold
all at once gave poor results, as did adding the solution dropwise.
By using approximately 50 µL per addition over five to eight
additions, spectrally narrow, optically dense samples were
Figure 2. Histograms showing the size dispersion of cobalt nanopar-
ticles produced by slow and fast addition of cobalt chloride. Solid lines
are best fits demonstrating particle dispersion. Particles sizes are
determined by measuring low-resolution TEM images.
Figure 3. Transmission electron micrographs of the HGNs. (A) High-
resolution TEM of a single 30 nm HGN. The wall thickness is
approximately 4 nm and large areas of crystallinity are clearly visible.
(B-E) Low-resolution TEM images of particles of 71 ( 17 nm (B),
50 ( 5 nm (C), 40 ( 3.5 nm (D), and 28 ( 2.3 nm.
Hollow Gold Nanospheres
J. Phys. Chem. B, Vol. 110, No. 40, 2006 19939
The explanation for this is a fairly simple one; it is a matter
of mixing. The reaction of gold salt with the cobalt particle is
very fast, happening almost instantaneously upon the addition
of the gold. There is also a secondary shell-mediated growth,
which takes place on a slightly longer time scale, where free
citrate in solution will reduce excess gold salt onto the formed
shells. This can result in significantly thicker shells when too
much gold is added. When a small amount of gold is introduced
to the stirred solution, all particles at the site of the addition
will immediately be oxidized completely in the presence of such
a high-concentration gold. If there is excess gold at this site,
then it will diffuse through the solution being reduced onto the
cobalt particles until there is no more gold. If the volume of
gold solution is too low, that is, dropwise, then the immediate
impact will be relatively small, but due to the small size of the
droplet it will dilute quickly. As the gold dilutes into the water,
less and less will be reduced onto the cobalt, resulting in a
gradient of shell thicknesses: thickest at the site of addition
and thinner shells moving away from the concentration center.
This leads to an incongruous sample in which some shells are
badly underformed and some are overgrown by seed-mediated
growth. An excellent example of this overgrowth is in Figure
3C. The second particle from the top has some slight over-
growth, which looks like small particle stuck to the surface.
When the concentration is excessive, this becomes a much more
pronounced feature of the particle.
At the other end of the addition rate scale is the all-at-once
addition of the gold. This suffers similar problems to the
dropwise addition; however, there is significantly more over-
growth, and less underformed particles. We were able to
overcome this problem by using a middle of the road approach.
By using 50 µL per addition, the resulting particles were uniform
and we did not observe excessive overgrowth. The choice of
this volume was not obvious and was discovered only by
experimental trials. This method does, however, have one major
flaw. Because such high concentrations are used, we were not
able to readily control the shell thicknesses. In theory, if the
gold is added correctly, the shell thickness should be a function
of the amount of gold added. This was achieved by using
relatively large volumes of low-concentration gold.
Low-Concentration Gold Addition. It was determined early
on in this study that using low concentrations of gold would
not produce satisfactory results; however, this assessment was
not entirely correct. Several factors are required for the low-
concentration addition of gold to work properly. The first is
that the solution mix very well as quickly as possible. If the
cobalt is added to the gold solution too slowly, then most of
the gold will be utilized by a small number of particles, which
will lead to poor sample homogeneity. Second, the volume of
the gold salt to which the cobalt is added must be large enough
that mixing can happen very quickly. With low volumes of gold
at higher concentrations there is still a pronounced mixing
problem, leading to poor samples. This is the problem we
observed in reproducing the work of Liang et al. Although the
larger volumes of gold produced reasonable results, using 5 or
8 mL of gold salt gave widely varying results and consistency
was a major issue. Because mixing is the biggest issue in
producing consistent results, it was hypothesized that by holding
the volumes of gold and cobalt solutions constant, a more
consistent result could be obtained.
By diluting varying volumes of gold salt to 10 mL with water
and adding the cobalt as quickly as possible under rapid stirring
we were able to produce homogeneous HGN with tunable wall
thicknesses, similar to the work of Liang et al. Shell thickness
varies linearly with gold concentration, indicating that homo-
geneous mixing is taking place, as shown in Table 4. These are
representative values from a single sample and are consistent
with all other data.
Effect of Particle Size and Wall Thickness on Optical
Properties. One of the major intents of all this size tuning is
the control of the optical properties of the HGN. We have found
that by varying wall thickness and particle size it is possible to
tune the plasmon absorption across much of the visible spectrum
as in Figure 4a, UV-vis data, and 4b, an image of HGN
solutions to illustrate the possible color range. These spectra
are representative of many experiments and show the full range
of tunability of this system. Although the full width at half-
maximum (FWHM) of the spectra remains relatively unchanged
from 500 to 750 nm and between 50 and 100 nm, the last two
spectra are fairly broadened to over 200 nm. This is likely due
to the formation of gold shells and their aggregates. At large
particle size, there is less capping agent to stabilize the colloids.
This likely brings about aggregation, which will significantly
shift the absorption of the shells. The weak shoulder at 700 nm
may be due to the presence of single shells, whereas the peak
is due to the aggregates. At this time, however, it is not possible
to determine the exact effect of the presence of the aggregates.
By increasing particle size at a constant wall thickness, the
absorption band will red-shift as the plasmon oscillation
decreases in energy. However, increasing wall thickness at
constant particle size will blue-shift the absorption band. The
band shifts to higher energy because as the inner diameter of
the HGN decreases it takes on more solid-particle-like proper-
ties. Because solid gold particles at these sizes have plasmon
bands at approximately 520 nm, the absorption will always shift
in this direction as wall thickness increases. This is predicted
in the work of Schatz et al. is shown experimentally here in
Figure 5a and b.23These plots show the affect of the aspect
ratio of particle size and wall thickness on plasmon absorption
(a) as well as the affect of particle size and wall thickness on
plasmon absorption. Representing 13 independent experiments,
the trend is clearly shown here. Because the work of Schatz et
al. is for particles of different sizes than those made here, we
are not able to directly correlate their results to our data.
However, we are currently working on similar calculations that
should determine if these results match well to the theory.
Because wall thickness plays such an important role in the
position of the plasmon absorption, it is important to understand
how this corresponds to the amount of gold added to the
solution. Figure 6 shows the nonnormalized absorption spectra
of three samples made from a single batch of 35 nm cobalt
nanoparticles. The highest concentration sample, at 60 µL of
0.1 M gold salt added absorbs most strongly at 638 nm, is the
most blue-shifted of the three as would be expected and has a
wall thickness of 7 ( 0.8 nm. The lower concentration samples
at 35 and 25 µL are red shifted to 685 nm (wall thickness 5.6
( 0.6 nm) and 702 nm (wall thickness 3.7 ( 0.6 nm),
respectively. Interestingly, as the band shifts the fwhm changes
only slightly from 80 nm for the 60 µL sample, to 91 nm for
TABLE 4: Wall Thickness as a Function of the Volume of
Gold Salt Added
0.1 M HAuCl4
diluted to 10 mL
40 ( 6
40 ( 6
40 ( 6
6.2 ( 0.6
6.9 ( 0.8
8 ( 0.7
19940 J. Phys. Chem. B, Vol. 110, No. 40, 2006
Schwartzberg et al.
the 35 µL sample, and to 82 nm for the 25 µL sample. This is
not the trend one might expect given the propensity of solid
gold nanoparticles to broaden significantly in spectrum with
increasing size. This broadening is due to the introduction of
new multipole modes, which are nonradiative and broader in
energy than the normal dipole plasmon mode.63,64In fact, upon
close examination of Figure 4a, it is clear that with the exception
of the last two spectra, the FWHM changes little regardless of
particle size or shell thickness. The explanation for this is tied
to the electron mean free path in gold. Because the wall
thickness is much less than this length (∼50 nm), longer axes
will dominate the plasmon oscillations and the multipole modes
that appear in large particles will be minimized. Interestingly,
this also explains why only one absorption band is observed
for this system, whereas nanorods, which also have multiple
axes of oscillation, will show two.
It may be noted that as the concentration of gold added
decreases there is a decrease in optical density as well. This is
not a matter of particle concentration, because 10 mL of gold
is added to each sample, and the total number of HGNs is fixed
to the number of cobalt particles present in the original solution.
This is a function of absorption cross section of the HGNs due
Figure 4. (a) UV-visible absorption spectra of nine HGN samples with varying diameters and wall thicknesses. (b) Image showing the color
range of HGN solutions. The vial on the far left contains solid gold nanoparticles, the rest are HNGs with varying diameters and wall thicknesses.
Hollow Gold Nanospheres
J. Phys. Chem. B, Vol. 110, No. 40, 2006 19941
to the different thicknesses of gold. As the wall grows thicker,
it will have a larger absorption cross section.
Homogeneous Line Width and Inhomogeneous Broaden-
ing. To determine if, and to what extent, the absorption spectrum
is broadened by inhomogeniety in the sample, we examined
the Rayleigh scattering spectra of the HGNs. Although the
FWHM of the ensemble-averaged solution of 30 ( 2.6 nm
particles is 75 nm, the single-particle FWHM is 47 nm as shown
in Figure 7. This is a broadening of 27 nm, which shows that
the samples are slightly inhomogeneously broadened. This is
to be expected to some point but is impressively small
considering how sensitive these structures are to variance in
wall thickness and local environment.65,66The sensitivity to local
environment is clear upon examination of the spectral shift
between the ensemble-averaged and scattering spectra. This is
a shift of 14 nm and is consistent with all particles examined.
The scattering spectra were taken from particles immobilized
on glass substrates in air while the ensemble-averaged spectra
were taken in aqueous solution. The refractive index of the
imbedding medium decreases from 1.33 to 1 in going from water
to air in these two scenarios. This substantially changes the
optical properties of the HGNs. However, a decrease in
refractive index has been shown to correspond to a blue shift.
It is not entirely clear why we observe a red shift from the
ensemble-averaged sample; however, one possibility is the effect
of particle immobilization to the glass slide. Although the
ensemble-averaged solution particles are completely surrounded
by water, the single particles were on a glass slide, partially
surrounded by air, residual water, and glass. The cumulative
effect of this compound refractive index may be enough to
explain this anomalous red shift in the spectrum. Further studies
are currently underway to better understand this effect.
Surface-Enhanced Raman Scattering. SERS experiments
were performed on solutions of as-prepared HGNs with mer-
captobenzoic acid (MBA) added to a final concentration of 1
mM. Although MBA tends to oxidize to form disulfides in
aqueous solution, only fresh solutions were used to perform
these studies. This should leave the thiol group free for binding
to the gold surface or possibly cobalt. Solutions are normally
washed to remove excess Co2+; however, some may remain.
This may cause sensitivity problems and is currently under
Figure 5. (a) Plasmon absorbance maximum wavelength (λmax) as a
function of the aspect ratio of shell thickness and shell diameter. The
line is a best-fit approximation to guide the eye. (b) 3D plot of plasmon
absorbance maximum wavelength (λmax) as a function of the shell
thickness and shell diameter. Each point on both plots represents an
individual set of experiments and the average measured lengths.
Figure 6. Spectral dependence on volume of added gold salt. Gold
solutions were diluted to 10 mL with water before 30 mL of a cobalt
solution made by the fast cobalt addition method with 100 µL sodium
borohydride and 600 µL of citric acid. Average particle size is 35 ( 2
Figure 7. Comparison of ensemble-averaged absorption and single-
particle Rayleigh scattering of 30 ( 2.6 nm HGNs.
19942 J. Phys. Chem. B, Vol. 110, No. 40, 2006
Schwartzberg et al.
At this concentration of MBA, there was no spectral shift
observed, which would indicate aggregation; therefore, we can
nominally say that the resulting spectra are from nonaggregated
or at least minimally aggregated HGNs. This was confirmed in
our previous work on SERS of single HGNs, which showed
that enhancement is observable from nonaggregated HGNs.55
Here we show the ensemble-averaged SERS spectrum of MBA
in Figure 8. In terms of enhancement, when compared in SERS
intensity to aggregated Lee and Meisel silver particles, the
standard high-enhancement SERS substrate, we achieve about
10% of the signal. This is an excellent result for nominally
nonaggregated particles and significantly better than many
current single-particle systems.
Avoiding Oxygen at All Costs. Finally, a few words about
the effects of oxygen on HGN formation. Cobalt is extremely
sensitive to oxygen, especially in aqueous solution. If the
solution is not properly deoxygenated, or if air is allowed to
enter the reaction vessel, then the results can be disastrous.
Although it is still possible to perform the reduction of gold
salt on partially oxidized cobalt particles, it produces very poor
results. The physical result of this is shown in Figure 9. Although
the oxidized cobalt will dissolve in the solution, it does not
oxidize homogeneously, which results in malformed HGNs.
Optically, this has extremely deleterious results greatly broaden-
ing the absorption band due to the random nature of the
oxidation. When solutions are badly oxidized, the percentage
of these types of particles tends to increase.
Nearly monodisperse HGNs of tunable interior and exterior
diameter have been synthesized by sacrificial galvanic replace-
ment of cobalt nanoparticles. We have been able to control the
position of the surface plasmon band between 550 and 820 nm
by carefully controlling particle size and wall thickness. Cobalt
particle size, the sacrificial template that controls the resulting
HGN size, is tunable by simultaneously changing the concentra-
tion of sodium borohydride and sodium citrate, the reductant
and capping agent, respectively. This varies from all previously
reported aqueous syntheses of cobalt particles. We also show
that by controlling the addition of gold carefully the thickness
of the gold shell can be varied. These HGNs have been further
demonstrated to be excellent SERS substrates in terms of
spectral consistency. They are promising for chemical and
biological sensing applications, particularly those requiring near-
Acknowledgment. We thank Luis Liz-Marza ´n for useful
discussions, and we appreciate financial support from the
National Science Foundation, the Petroleum Research Fund/
American Chemical Society, the University of California at
Santa Cruz, the student employee graduate research fellowship
at Lawrence Livermore National Laboratory, the Center for
Biophotonics Science and Technology at the University of
California at Davis, and the Genomics:GtL program of the
Office of Science of the U.S. Department of Energy. The Center
for Biophotonics, an NSF Science and Technology Center, is
managed by the University of California, Davis, under Coopera-
tive Agreement No. PHY 0120999. This work was performed
under the auspices of the U.S. Department of Energy by
University of California Lawrence Livermore National Labora-
tory under contract No. W-7405-Eng-48.
Note Added after ASAP Publication. This article was
released ASAP on June 29, 2006. Figure 6 and references 34
and 55 have been revised. The correct version was posted on
August 9, 2006.
References and Notes
(1) Sherry, L. J.; Chang, S. H.; Schatz, G. C.; Van Duyne, R. P.; Wiley,
B. J.; Xia, Y. N. Nano Lett. 2005, 5, 2034.
(2) Hicks, E. M.; Zhang, X. Y.; Zou, S. L.; Lyandres, O.; Spears, K.
G.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2005, 109, 22351.
(3) Whitney, A. V.; Elam, J. W.; Zou, S. L.; Zinovev, A. V.; Stair, P.
C.; Schatz, G. C.; Van Duyne, R. P. J. Phys. Chem. B 2005, 109, 20522.
(4) Haes, A. J.; Hall, W. P.; Van Duyne, R. P. Laser Focus World
2005, 41, 105.
(5) Lahav, M.; Vaskevich, A.; Rubinstein, I. Langmuir 2004, 20, 7365.
(6) He, L.; Smith, E. A.; Natan, M. J.; Keating, C. D. J. Phys. Chem.
B 2004, 108, 10973.
(7) Penn, S. G.; He, L.; Natan, M. J. Curr. Opin. Chem. Biol. 2003, 7,
(8) Nenninger, G. G.; Tobiska, P.; Homola, J.; Yee, S. S. Sens.
Actuators, B 2001, 74, 145.
(9) Yee, S. S.; Homola, J.; Gauglitz, G. Sens. Actuators, B 1999, 54,
(10) Creighton, J. A. Metal Colloids. In Surface Enhanced Raman
Scattering; Chang, R. K., Furtak, T. E., Eds.; Plenum Press: New York,
1982; p 315.
Figure 8. Ensemble-averaged surface-enhanced Raman scattering
spectrum of mercaptobenzoic acid on the HGNs.
Figure 9. High-resolution TEM of an HGN formed from a slightly
oxidized cobalt particle.
Hollow Gold Nanospheres
J. Phys. Chem. B, Vol. 110, No. 40, 2006 19943
(11) Moskovits, M. ReV. Mod. Phys. 1985, 57, 783.
(12) Yonzon, C. R.; Stuart, D. A.; Zhang, X. Y.; McFarland, A. D.;
Haynes, C. L.; Van Duyne, R. P. Talanta 2005, 67, 438.
(13) Tian, Z. Q. J. Raman Spectrosc. 2005, 36, 466.
(14) Haynes, C. L.; Yonzon, C. R.; Zhang, X. Y.; Van Duyne, R. P. J.
Raman Spectrosc. 2005, 36, 471.
(15) Kneipp, K.; Kneipp, H.; Itzkan, I.; Dasari, R. R.; Feld, M. S. Chem.
ReV. 1999, 99, 2957.
(16) Nie, S. M.; Emery, S. R. Science 1997, 275, 1102.
(17) Otto, A. J. Raman Spectrosc. 2005, 36, 497.
(18) Otto, A.; Mrozek, I.; Grabhorn, H.; Akemann, W. J. Phys.:
Condens. Matter 1992, 4, 1143.
(19) Campion, A.; Kambhampati, P. Chem. Soc. ReV. 1998, 27, 241.
(20) Quinten, M. Appl. Phys. B 2001, 73, 317.
(21) Moskovits, M. J. Raman Spectrosc. 2005, 36, 485.
(22) Creighton, J. A.; Eadon, D. G. J. Chem. Soc., Faraday Trans. 1991,
(23) Hao, E.; Li, S. Y.; Bailey, R. C.; Zou, S. L.; Schatz, G. C.; Hupp,
J. T. J. Phys. Chem. B 2004, 108, 1224.
(24) Link, S.; El-Sayed, M. A. Annu. ReV. Phys. Chem. 2003, 54, 331.
(25) Nikoobakht, B.; El-Sayed, M. A. Chem. Mater. 2003, 15, 1957.
(26) Chang, S. S.; Shih, C. W.; Chen, C. D.; Lai, W. C.; Wang, C. R.
C. Langmuir 1999, 15, 701.
(27) Brioude, A.; Jiang, X. C.; Pileni, M. P. J. Phys. Chem. B 2005,
(28) Yu, Y. Y.; Chang, S. S.; Lee, C. L.; Wang, C. R. C. J. Phys. Chem.
B 1997, 101, 6661.
(29) Grant, C. D.; Schwartzberg, A. M.; Norman, T. J.; Zhang, J. Z. J.
Am. Chem. Soc. 2003, 125, 549.
(30) Quinten, M. J. Cluster Sci. 1999, 10, 319.
(31) Quinten, M. Appl. Phys. B. 2000, 70, 579.
(32) Quinten, M.; Kreibig, U. Appl. Opt. 1993, 32, 6173.
(33) Schwartzberg, A. M.; Grant, C.; Wolcott, A.; Bogomolni, R.; Zhang,
J. Z. SPIE Proc. 2003, 5221-21, 100.
(34) Norman, T. J.; Grant, C. D.; Schwartzberg, A. M.; Zhang, J. Z.
Opt. Mat. 2005, 27, 1197.
(35) Kreibig, U. Optical Properties of Metal Clusters; Springer: Berlin;
New York, 1995; Vol. 25.
(36) Kneipp, K.; Kneipp, H.; Itzkan, I.; Dasari, R. R.; Feld, M. S. J.
Phys.: Condens. Matter 2002, 14, R597.
(37) McFarland, A. D.; Young, M. A.; Dieringer, J. A.; Van Duyne, R.
P. J. Phys. Chem. B 2005, 109, 11279.
(38) Schatz, G. C.; Van Duyne, R. P. Electromagnetic Mechanism of
Surface-Enhanced Spectroscopy; John Wiley and Sons Ltd.: Chichester,
(39) Brus, L. Abstr. Papers Am. Chem. Soc. 2001, 221, 112.
(40) Michaels, A. M.; Jiang, J.; Brus, L. J. Phys. Chem. B 2000, 104,
(41) Jiang, J.; Bosnick, K.; Maillard, M.; Brus, L. J. Phys. Chem. B
2003, 107, 9964.
(42) Schwartzberg, A. M.; Grant, C. D.; Wolcott, A.; Talley, C. E.;
Huser, T. R.; Bogomolni, R.; Zhang, J. Z. J. Phys. Chem. B 2004, 108,
(43) Jackson, J. B.; Westcott, S. L.; Hirsch, L. R.; West, J. L.; Halas,
N. J. Appl. Phys. Lett. 2003, 82, 257.
(44) Oldenburg, S. J.; Westcott, S. L.; Averitt, R. D.; Halas, N. J. J.
Chem. Phys. 1999, 111, 4729.
(45) Halas, N. Gold Bull. 2004, 37, 137.
(46) Talley, C. E.; Jackson, J. B.; Oubre, C.; Grady, N. K.; Hollars, C.
W.; Lane, S. M.; Huser, T. R.; Nordlander, P.; Halas, N. J. Nano Lett.
2005, 5, 1569.
(47) Lu, Y.; Liu, G. L.; Kim, J.; Mejia, Y. X.; Lee, L. P. Nano Lett.
2005, 5, 119.
(48) Chithrani, B. D.; Ghazani, A. A.; Chan, W. C. W. Nano Lett. 2006,
(49) Chen, J. Y.; Wiley, B.; McLellan, J.; Xiong, Y. J.; Li, Z. Y.; Xia,
Y. N. Nano Lett. 2005, 5, 2058.
(50) Wiley, B.; Sun, Y. G.; Chen, J. Y.; Cang, H.; Li, Z. Y.; Li, X. D.;
Xia, Y. N. MRS Bull. 2005, 30, 356.
(51) Wiley, B.; Sun, Y. G.; Mayers, B.; Xia, Y. N. Chem.sEur. J. 2005,
(52) Sun, Y. G.; Mayers, B.; Xia, Y. N. AdV. Mater. 2003, 15, 641.
(53) McLellan, J. M.; Xiong, Y. J.; Hu, M.; Xia, Y. N. Chem. Phys.
Lett. 2006, 417, 230.
(54) Liang, H. P.; Wan, L. J.; Bai, C. L.; Jiang, L. J. Phys. Chem. B
2005, 109, 7795.
(55) Schwartzberg, A. M.; Oshiro, T. Y.; Huser, T. R.; Zhang, J. Z.;
Talley, C. E. Anal. Chem. 2006, 78, 4732.
(56) Lisiecki, I.; Pileni, M. P. Langmuir 2003, 19, 9486.
(57) Kobayashi, Y.; Horie, M.; Konno, M.; Rodriguez-Gonzalez, B.;
Liz-Marzan, L. M. J. Phys. Chem. B 2003, 107, 7420.
(58) Chow, M. K.; Zukoski, C. F. J. Colloid Interface Sci. 1994, 165,
(59) Biggs, S.; Chow, M. K.; Zukoski, C. F.; Grieser, F. J. Colloid
Interface Sci. 1993, 160, 511.
(60) Turkevich, J.; Stevenson, P. C.; Hiller, J. Discuss. Faraday Soc.
1951, 11, 55.
(61) Frens, G. Nat. Phys. Sci. 1973, 241, 20.
(62) Abramoff, M. D.; Magelhaes, P. J.; Ram, S. J. Biophotonics Int.
2004, 11, 36.
(63) Payne, E. K.; Shuford, K. L.; Park, S.; Schatz, G. C.; Mirkin, C.
A. J. Phys. Chem. B 2006, 110, 2150.
(64) Millstone, J. E.; Park, S.; Shuford, K. L.; Qin, L. D.; Schatz, G.
C.; Mirkin, C. A. J. Am. Chem. Soc. 2005, 127, 5312.
(65) Nehl, C. L.; Grady, N. K.; Goodrich, G. P.; Tam, F.; Halas, N. J.;
Hafner, J. H. Nano Lett. 2004, 4, 2355.
(66) Sun, Y. G.; Xia, Y. N. Anal. Chem. 2002, 74, 5297.
19944 J. Phys. Chem. B, Vol. 110, No. 40, 2006
Schwartzberg et al.