Ag Nanocrystal Junctions as the Site for Surface-Enhanced Raman Scattering of Single
Rhodamine 6G Molecules
Amy M. Michaels, Jiang Jiang, and Louis Brus*
Department of Chemistry, Columbia UniVersity, New York, New York 10027
ReceiVed: July 18, 2000; In Final Form: October 3, 2000
Atomic force microscopy (AFM) measurements show that the Ag nanoparticles that yield surface-enhanced
Raman scattering (SERS) of single molecules of Rhodamine (R6G) are all compact aggregates consisting of
a minimum of two individual particles. Comparison of 514.5 and 632.8 nm excitation shows that the single
molecule R6G signal is significantly higher when the excitation wavelength is resonant with the absorption
band of R6G and suggests that the Raman excitation spectrum follows the absorption profile for R6G. We
have also observed an interesting superlinear power dependence of the SERS signal. On average, by increasing
the incident power by 2 orders of magnitude and decreasing the integration time by the same factor to maintain
constant fluence, increases of 4 to 6 times were observed in the SERS intensity. We discuss these results in
terms of model where the R6G molecule that yields single molecule SERS signals is located at the junction
of two Ag nanoparticles. We have also modeled the system using molecular resonance Raman theory to
provide further insight into the enhancement mechanism.
Introduction
Surface-enhanced Raman scattering (SERS) is a well-
established phenomenon which can enhance Raman signals of
nonresonant molecules adsorbed on noble metal particles by 5
to 6 orders of magnitude.
1-4
In certain cases, however, the
enhancement can be enormous. Several experiments have
demonstrated that the SERS enhancement can approach 10
14
to 10
15
, and, in fact, enable the detection of a single molecule.
5-10
Remarkably, the strength of the single molecule SERS signal
can be much stronger than the luminescence signal from a single
highly fluorescent molecule in the absence of metal.
We recently reported that the average single molecule SERS
cross section for the dye Rhodamine 6G (R6G) is ∼200 Å
2
for
certain “hot” SERS-active nanocrystals. This cross section
includes both sharp Raman lines and broad underlying con-
tinuum. These SERS-active nanocrystals account for less than
1% of the total particles in a heterogeneous Ag colloid. Since
the intensity of resonant Rayleigh scattering from a metal
particle is also a measure of the average local field enhancement
around the nanocrystal, we independently characterized the
SERS and resonant Rayleigh scattering spectra from individual
nanocrystals in order to probe the electromagnetic (EM)
contribution to the SERS signal.
11
These experiments demon-
strated no direct correlation between the strength or frequency
of the Rayleigh scattering resonance and the single molecule
SERS activity of the individual Ag particles. Two characteristics
common to all scattering spectra of SERS-active particles
included (1) partial resonance at the SERS excitation wave-
length, and (2) a complex scattering spectrum, consisting of at
least 2 resonances. These results suggest that some other factor,
not simply an intense plasmon resonance, is critical in distin-
guishing SERS-active particles. We interpreted these data in a
framework originally proposed by both Otto and Persson, which
considers the interaction of ballistic electrons in the metal with
a strongly chemisorbed molecule.
12,13
The number of SERS-
active particles is not limited by the number of available R6G
molecules; increasing the R6G concentration by a factor of one
hundred increases the number of SERS-active particles only by
a factor of 5. This suggests that a rare chemisorption site may
be the critical and limiting factor for single molecule SERS.
The possibility of a rare site was first discussed in prior
ensemble colloidal SERS studies, where the giant SERS signal
saturated at nanomolar concentrations of R6G that correspond
to roughly one adsorbed molecule per 55 nm Ag particle.
5
This
rare site was characterized to have a very high adsorption energy
of 65.9 kJ/mol (∼2/3 eV). This was in contrast to a second site
which had a lower adsorption energy of 35.8 kJ/mol and yielded
SERS enhancements that were weaker by more than 2 orders
of magnitude.
This article presents AFM measurements which demonstrate
that the SERS-active particles are compact aggregates consisting
of two or more 50 nm Ag particles. On the basis of these results,
we propose that the rare site required for single molecule SERS
lies at the junction of two aggregated particles. We also observe
an unusual power dependence in that the scattering is higher
than linear for most, but not all, SERS-active aggregates. We
have also modeled the nanocrystal/molecule complex within the
framework of resonance Raman theory in order to gain further
insight into the enhancement mechanism.
Experimental Section
Sample preparation and the microscope-based optical imaging
apparatus have been described in detail previously.
11
Briefly,
the sample was prepared by incubating a nanomolar concentra-
tion of R6G in a colloidal silver solution containing 1-10 mM
NaCl for ∼2 h. The resulting solution was diluted by a factor
of 3 with a boiled sodium citrate solution, and then spin-cast
onto a polylysine-coated quartz cover slip. For AFM studies,
the solution was similarly diluted by a factor of 30. Studies
were conducted at room temperature in air. Resonant Rayleigh
spectra were obtained using the microscope equipped with a
* Author to whom correspondence should be addressed.
11965J. Phys. Chem. B 2000, 104, 11965-11971
10.1021/jp0025476 CCC: $19.00 © 2000 American Chemical Society
Published on Web 11/23/2000
dark-field condenser and a tungsten lamp. For SERS measure-
ments, the sample was irradiated with ∼35 mW of either
continuous wave 514.5 nm (Coherent Innova 308) or 632.8 nm
(Coherent High Power HeNe) laser light in grazing incidence
geometry, yielding a power density of ∼30 W/cm
2
. In the power
dependence experiments, samples were illuminated by the laser
at 3 W/cm
2
, 30 W/cm
2
, and 300 W/cm
2
, while the integration
time was adjusted from 10 s to1sto0.1s,respectively. The
linearity of the instrumental response under the same conditions
(i.e., integration time and power) was calibrated by observation
of laser scattering from imperfections in bare coverslips. To
obtain AFM images, the SERS-active particles were first
identified optically, and the dark-field optics were then removed
from the microscope. The AFM scanner (Digital Instruments,
Bioscope) was mounted on the sample stage and the SERS-
active particles were identified and imaged in tapping mode.
Results
Figure 1 shows a correlated single molecule R6G SERS
spectrum, resonant Rayleigh scattering spectrum, and AFM
image of an individual SERS-active Ag nanoparticle. The
marked peaks in the SERS spectrum correspond to the Raman
bands for R6G. Consistent with prior results, the resonant
Rayleigh scattering spectrum shows some scattering intensity
at 514.5 nm, the laser excitation wavelength, and appears to
have at least 2 resonances. The AFM image shows that this
particular SERS-active particle is a tightly packed aggregate
consisting of ∼12 to 15 particles.
The large majority of Ag particles on the surface were
isolated, single nanocrystals, yet all of SERS-active particles
identified were compact clusters that consisted of a minimum
of 2 individual Ag particles. Tapping mode AFM images of
three additional SERS-active nanocrystals are shown in Figure
2. Each of these particles corresponds to a scattering spectrum
with multiple peaks (not shown), verifying that the presence of
multiple scattering resonances corresponds to aggregates. While
the individual particles are ∼55 nm in diameter, the SERS-
active aggregates range from ∼100 to ∼250 nm in size and
vary significantly in shape as well. These results are consistent
with two recent reports. Xu et al. observed single molecule
hemoglobin SERS only from dimers and trimers of colloidal
Ag particles.
14
Similarly, Moyer et al. observed SERS of single
carbon domains from 75% of the Ag particles in their sample,
all of which consisted of clusters of Ag crystallites.
15
It is well-known that the addition of low concentrations of
Cl
-
ions (1-10 mM) to the Ag colloid yields significant
increases in SERS signals. Since extinction spectra, TEM, and
AFM measurements have shown that such low quantities of
Cl
-
ions do not cause any significant aggregation of the sample,
Figure 1. (a) AFM image, (b) R6G SERS spectrum, and (c) resonant
Rayleigh scattering spectrum of a single SERS-active Ag particle
incubated with 10 mM NaCl and 2 nM R6G. For the SERS spectrum,
the intergration time was 60 s at ∼30 W/cm
2
. The Rayleigh scattering
spectrum was obtained with a 10 s integration time.
Figure 2. AFM images of 3 different SERS-active Ag aggregates
prepared by incubating a Ag colloid with 10 mM NaCl and 2 nM R6G.
The heights of the aggregates ranged from ∼50 to ∼200 nm.
11966 J. Phys. Chem. B, Vol. 104, No. 50, 2000 Michaels et al.
it has often been proposed that the anions serve as “activating
agents”. This activation may occur either by generating unique
“SERS-active sites” on the nanocrystal or by stabilizing the
chemisorbed molecule.
5,16
To test the importance of a potential
“anion effect”, we screened for single molecule R6G SERS in
a sample initially incubated with polylysine and ascorbic acid
in place of NaCl. Polylysine, in conjunction with ascorbic acid,
is a well-known aggregating agent.
17
For this sample, the same
observations were noted as for NaCl: (1) only a small
percentage of Ag particles exhibited SERS-activity, and (2) the
scattering spectra of all such SERS-active particles were
characterized by multiple resonances. This suggests that the
addition of NaCl serves to induce aggregation in the colloid
and provides further evidence that aggregates are indeed
responsible for the enormous SERS enhancements.
These results strongly suggest that the special site for single
molecule SERS, whose rarity leads to signal saturation at very
low concentration in bulk colloids, lies at the junction of two
Ag particles in a compact cluster. Classical electromagnetic
calculations show that this junction, for Angstrom-scale separa-
tion of the two particles, experiences a significantly higher
electromagnetic field and ballistic carrier surface flux than any
other location on the surface of a single metal particle. We will
return to this point in the theory section.
Each SERS-active aggregate gives a somewhat different
Raman signature. As previously reported, some single molecule
SERS signals blink or show intensity fluctuations on a second
time scale under cw laser excitation, while others do not.
9,11
Additionally, the absolute intensity and relative ratio of the sharp
R6G Raman lines to the underlying continuum vary from one
aggregate to the next. Figure 3 shows three consecutive R6G
SERS single molecule spectra taken with 60 s integration times
for which the relative intensities of the Raman bands show
fluctuations of approximately 20%. The spectral widths of these
Raman lines range from ∼12 to 15 cm
-1
. Spectra obtained with
shorter integration times have shown significant frequency
variations in the Raman signals as a function of time, therefore
the fairly large width of these lines compared to classical Raman
scattering may be partially accounted for by these spectral
variations.
9
Decreasing the integration time from 60 to 10 s in
an attempt to eliminate the time-averaging of these spectral
variations yields no noticeable change in the spectral width.
In addition to the varied behavior described above, the
individual SERS-active aggregates also show different laser
power dependencies. It is difficult to study the power depen-
dence systematically in the presence of intensity fluctuations
at constant laser power. Therefore, for a single SERS signal,
we measure the integrated Stokes-shifted SERS intensity
(including underlying continuum) for four consecutive periods
at constant power, to obtain a measure of the signal reproduc-
ibility before changing laser power. Most remarkably, about 80%
of the SERS-active aggregates showed greater scattering ef-
ficiency at higher powers, as shown in Figure 4. By increasing
the laser power by a factor of 100 and decreasing the integration
time by the same factor, we observed reproducible increases in
SERS signals ranging from a factor of 4 to 6. A small number
of SERS signals exhibited a linear power dependence, and
several showed decreases in efficiency with increasing power.
Figure 4b shows a SERS signal from a single particle where
there is some fluctuation superimposed upon a varying power
dependence over the course of the experiment. There is no
Figure 3. Three consecutive R6G SERS spectra obtained at ∼30
W/cm
2
with 60 s integration time. The fwhm of the Raman lines ranges
from ∼12 to 15 cm
-1
. The magnitude of the change in relative
intensities of the Raman bands is ∼20%.
Figure 4. R6G SERS signal for two different Ag aggregates as a
function of laser power under conditions of constant fluence. At each
power, four consecutive signals were recorded to measure variations
in the signal intensity as a function of time. For each particle, the power
dependence was measured three consecutive times.
Ag Nanocrystal Junctions in SERS of R6G J. Phys. Chem. B, Vol. 104, No. 50, 2000 11967
apparent correlation between the laser intensity and the fre-
quency of fluctuations. Despite this superlinearity, both the
underlying continuum and the sharp Raman lines show a nearly
identical power dependence, as shown in Figure 5 for three
different SERS aggregates.
If the Raman species were a photoproduct, then one might
expect superlinear behavior, but we see the same sharp Raman
lines at low and high power. We also measured the 514.5 nm
Rayleigh scattering for the same time periods and laser powers
for particles throughout the entire sample. Within a reproduc-
ibility of about 20%, the Rayleigh scattering is linear, as
expected. Thus the vibronic coupling between the resonant
optical state and the molecular vibration is stronger at higher
optical fields. To our knowledge, this type of superlinearity has
never been seen in a Raman experiment.
We explored the SERS excitation spectrum by comparing
SERS activity with 514.5 and 632.8 nm laser excitation. As
previously stated, at nanomolar concentrations of R6G, ap-
proximately 1 out of every 200 Ag particles yields single
molecule SERS, when the sample is incubated with 10 mM
NaCl and excited at 514.5 nm. In one sample, using 514.5 nm
excitation, 32 such SERS-active particles were identified, all
of whose spectra consisted of sharp R6G Raman lines super-
imposed upon a broad continuum. All strong SERS signals
contain both continuum and sharp Raman lines; occasionally
signal at 10× weaker is observed consisting only of continuum.
When the excitation wavelength was changed to 632.8 nm, 10
of these particles showed significant Stokes shifted intensity.
However, 9 of these 10 particles showed only continuum
emission without any evidence of R6G Raman lines under 632.8
nm exctitation, while the remaining one particle showed both
R6G Raman and continuum emission. With 632.8 nm excitation,
no new particles exhibited SERS activity that were not part of
the 514.5 nm active group. When the laser wavelength was
returned to 514.5 nm, the original SERS activity remained in
these 32 particles.
This result indicates that a Ag aggregate that yields both
Raman and Stokes-shifted continuum emission under 514.5 nm
excitation has a 1/3 probability of emitting a broad continuum
alone (no Raman lines) when excited at 632.8 nm. The excitation
spectrum of the underlying SERS continuum therefore appears
to differ, in this limited sample, from that of the R6G Raman
scattering, by extending further red in wavelength. The con-
tinuum excitation spectra may, in fact, more closely follow the
resonant Rayleigh scattering spectrum. While Rayleigh scat-
tering spectra were not acquired in this experiment, many of
the 32 SERS-active particles likely had strong resonances at
632.8 nm based upon results of our prior studies, which showed
that the scattering spectra of SERS-active particles often extend
over the entire visible spectrum. Since excitation at 632.8 nm
lies below the absorption band of aqueous R6G (not chemi-
sorbed) which peaks at 530 nm, but still lies well within the
plasmon resonance of many of the Ag particles present in the
sample, this result suggests that HOMO-LUMO resonance of
the molecule plays a crucial role in yielding sharp R6G Raman
lines. This observation is also consistent with colloidal SERS
study by Hildebrandt and Stockburger, which reported the
excitation spectrum of the giant SERS signal to follow the
absorption spectrum of R6G.
5
Yet, while the Raman and
underlying continuum emission appear to originate from two
different spectroscopic processes, both show the same super-
linear dependence and are clearly linked. If the signal blinks,
both the continuum and Raman lines blink on and off together.
Raman Scattering Theory
Large Ag particles more than 30 nm diameter have huge
electronic transition dipoles with femtosecond radiative widths
and negligible vibronic coupling to low frequency Ag normal
modes. Such particles show intense Rayleigh scattering without
any Stokes shifted emission. A very simple theoretical approach
to SERS is to assume that chemisorption vibronically couples
the R6G vibrations to the plasmon transition dipole, without
specifying the microscopic nature of this coupling. This can be
done by treating the Ag/molecule system in the framework of
conventional molecular electronic resonance Raman theory,
which is well developed for biological chromophores of high
oscillator strength.
18,19
When the laser wavelength is resonant
with an electronic absorption band, the intensities of the Raman
lines are determined primarily by the properties of the excited
electronic state. In particular, the coupling of the vibrational
modes of a molecule to its electronic state can be described as
a function of a dimensionless displacement, ∆, between the
ground and excited electronic potential energy surfaces along
a single normal coordinate. For resonant scattering, the total
cross section for a Raman transition from initial state |i〉 to final
state |f〉, in the Born-Oppenheimer and Condon approximations,
can be written as
18
where
The cross section σ is in units of area, E
L
and E
S
are the energies
of the incident and scattered photons, and R
if
is the resonance
Raman polarizability transition matrix element between states
|i〉 and |f〉 as given by the Kramers-Heisenberg-Dirac dispersion
Figure 5. R6G SERS spectra at low power (left) and high power (right)
for three SERS-active Ag aggregates.
σ
iff
(E
L
) )
8π
9η
2
c
4
(E
S
)
3
E
L
|R
iff
|
2
(1)
R
iff
) M
2
∑
v
〈f|v〉〈v|i〉
v
-
i
+ E
0
- E
L
- iΓ
(2)
11968 J. Phys. Chem. B, Vol. 104, No. 50, 2000 Michaels et al.
formula. In eq 2, M is the electronic transition dipole moment;
|v〉 is a vibrational level of the excited electronic state;
v
and
i
are the energies of the states |v〉 and |i〉, respectively; E
0
is
the energy difference between the lowest vibrational levels of
the ground and excited electronic states; and Γ is the homoge-
neous line width of the electronic transition. The terms 〈f|v〉
and 〈v|i〉 are the appropriate Franck-Condon overlap integrals.
Assuming that both the excited and ground state potential energy
surfaces are harmonic and differ only in their equilibrium
position, ∆, the overlap integrals for Rayleigh, fundamental
Raman, and overtone scattering, respectively, are given by
20
We consider a spherical particle with a diameter of 100 nm
and which is characterized by the optical dielectric constants
of bulk Ag. Mie theory calculations show that the scattering
resonance peaks at 397 nm with a fwhm of ∼90 nm.
21
As the
scattering cross section is far larger than the absorption cross
section, the large spectral width of the resonance corresponds
directly to the femtosecond radiative decay rate Γ. Relating the
peak intensity and spectral width of the resonance to the
transition dipole shows that for a particle that is 100 nm in
diameter, M approaches a remarkable 21 000 D.
For the 1575 cm
-1
C-C aromatic stretch for R6G, the
experimentally measured Rayleigh scattering and SERS cross
sections agree with the values calculated for ∆ ) 0.012 and
the values of E
0
and Γ calculated from Mie theory as described
above. At an excitation wavelength of 514.5 nm, the model
predicts a fundamental Raman scattering cross section of 1.5
× 10
-15
cm
2
and a Rayleigh scattering cross section of ∼2 ×
10
-10
cm
2
. Assuming that the sharp 1575 cm
-1
band contributes
approximately 5% of the total measured SERS cross section of
2 × 10
-14
cm
2
, the measured SERS cross section for this band
is 1 × 10
-15
cm
2
. The measured Rayleigh scattering cross
section, which is approximately 6000 times larger than the total
SERS cross section, is ∼1 × 10
-10
cm
2
. While there have been
no reports to date of overtones observed in single molecule
SERS spectra, ensemble studies have measured that the signals
from fundamental Raman scattering of R6G SERS are approxi-
mately 10 times stronger than the overtone signals.
5
The
resonance Raman model predicts that the ratio of fundamental
Raman scattering to overtone scattering for this system should
approach 6 orders of magnitude and is consistent with the fact
that overtones have not been observed at the single molecule
level.
The excited state of the particle is very weakly coupled to
the molecule, yet the Raman cross section is so large because
the Ag transition dipole is huge. The Raman scattering is similar
to resonance Raman scattering from a dissipative continuum in
a molecular system, because the excited state dephases so
quickly. Despite this insight, the model is too simple because
the Raman excitation spectrum (eq 1) is predicted to be the
same as the resonant Rayleigh scattering spectrum for the Ag
particle. Our limited data examining the nature of the excitation
profiles of the particles indicate the true situation is more
complex.
The plasmon excited state is a correlated, many-electron
excitation, which Kawabata and Kubo originally described as
a coherent sum of all possible electron-hole excitations, for
which the energy separation of each pair corresponds to the
laser frequency.
22
The molecular Raman model described above
essentially treats vibrational coupling to all components of this
wave function equivalently. If some components couple more
strongly to the molecule than others, then the resonant state is
different than the zero order plasmon state and a more complex
model is required. In Otto’s and Persson’s models, individual
ballistic electrons or holes generated via excitation of the surface
plasmon couple to the chemisorbed molecule LUMO or HOMO
as shown in Figure 6. In this description, the relative energy of
the HOMO and LUMO with respect to the Ag Fermi level is
critical. If the LUMO is too high in energy with respect to the
Fermi level, or the HOMO similarly lies too low in energy,
neither the excited electrons or holes in the metal can efficiently
couple to the molecule. The interaction between the metal and
molecule will be maximized when the HOMO-LUMO separa-
tion is resonant with the excitation wavelength and symmetric
with respect to the Ag Fermi level.
This coupling between the metal electrons and chemisorbed
molecules clearly occurs. In Ag particles, the dephasing rate of
the surface plasmon increases in the presence of chemisorbed
molecules which have low-lying LUMO levels that can interact
with ballistic electrons in the metal.
23
The faster decay rate
occurs due to scattering of metal electrons off the chemisorbed
molecules via transient negative ion formation.
24
Similarly, DC
resistivity measurements have shown that only those molecules
that yield a change in resistivity upon adsorption to the metal
contribute to the SERS signal. In other words, only those
molecules with low-lying LUMOs that can interact with the
metallic electrons yield an observable SERS signals.
1
This type
of resonance may be the origin of reported single molecule
SERS in the near-IR.
A second interaction mechanism between the metal and
molecule involves exchange mixing of Ag neutral electron-
hole pairs with the HOMO-LUMO neutral excited state of the
chemisorbed molecule. This mechanism would yield a neutral
excited R6G type resonant state, rather than a charged resonant
molecular state. As the photon energy becomes resonant with
the HOMO-LUMO gap, mixing should increase. If all pairs
equally mix with the neutral excited molecular state at a given
photon energy, independent of the position of the Fermi level,
then this idea is mathematically equivalent to ∆ being wave-
length dependent. A wavelength-dependent ∆ is not a useful
physical concept, yet such behavior seems to appear in the single
molecule SERS experiment, as the SERS excitation spectrum
is similar to the aqueous R6G absorption spectrum. Possible
Figure 6. Schematic of relative energies of excited electron-hole pairs
generated via excitation of the surface plasmon in the metal with respect
to the HOMO and LUMO of the chemisorbed molecule.
〈0|v〉〈v|0〉 )
∆
2v
2
v
v!
exp(-∆
2
/2)
〈0|v〉〈v|1〉 ) 〈0|v〉〈v|0〉
(
∆ -
2v
∆
)
/
x
2
〈0|v〉〈v|2〉 ) 〈0|v〉〈v|0〉(∆
2
- 4v + 4v(v - 1)/∆
2
)/2
x
2
Ag Nanocrystal Junctions in SERS of R6G J. Phys. Chem. B, Vol. 104, No. 50, 2000 11969
