Vibrationally Resolved Fluorescence Excited with Submolecular Precision

See allHide authors and affiliations

Science  24 Jan 2003:
Vol. 299, Issue 5606, pp. 542-546
DOI: 10.1126/science.1078675


Tunneling electrons from a scanning tunneling microscope (STM) were used to excite photon emission from individual porphyrin molecules adsorbed on an ultrathin alumina film grown on a NiAl(110) surface. Vibrational features were observed in the light-emission spectra that depended sensitively on the different molecular conformations and corresponding electronic states obtained by scanning tunneling spectroscopy. The high spatial resolution of the STM enabled the demonstration of variations in light-emission spectra from different parts of the molecule. These experiments realize the feasibility of fluorescence spectroscopy with the STM and enable the integration of optical spectroscopy with a nanoprobe for the investigation of single molecules.

Single-molecule spectroscopy has potential applications in ultrasensitive chemical identification (1, 2) and as a probe of the local molecular environment and single-biomolecule dynamics in vivo (3, 4). The influence of the conformational structure on the properties of a molecule is one of the key questions in molecular spectroscopy. Conformationally driven fluorescence intensity fluctuations and photobleaching have been observed in organic dyes and biomolecules (5). However, despite the unprecedented sensitivity achieved by these optical techniques, determination of the corresponding molecular internal structure is not straightforward.

One possible way to improve the spatial resolution of single-molecule optical spectroscopy would be to use the highly localized tunneling current of the scanning tunneling microscope (STM) for excitation of light emission (6) that would be sensitive to the properties of the sample on the atomic scale. This approach has been reported for clean metal (7, 8) and semiconductor (9) surfaces, as well as for atomic and molecular adsorbates (10–13) on metal substrates. However, the reported photon emission spectra did not show identifiable molecule-related features (12). On a metal surface, the electronic levels of a molecule are considerably broadened whereas light emission is strongly quenched (14), making it difficult to detect and identify any molecule-specific emission.

Here, we report the detection of single- molecule fluorescence with the STM. We found that fluorescence could be observed when the molecule was supported on a thin aluminum oxide (Al2O3) film grown on a NiAl(110) surface. This oxide film has been used as a substrate to study STM-excited photon emission from Ag clusters (15). The oxide spacer reduces the interaction between the molecule and the metal. In contrast, for molecules deposited directly on the NiAl(110) surface, the light-emission spectra were similar in shape to the spectra from the bare NiAl(110) surface.

The experiments were conducted with a home-built ultrahigh-vacuum (UHV) STM (16). For photon collection, a lens was mounted inside the UHV chamber. The collected light was transferred to a spectrometer (Acton Research 300) and detected with a liquid N2–cooled charge-coupled device (CCD) camera (Princeton Instruments, Spec-10:100B).

Ag and W tips were used in our experiments (17). We found that photon emission efficiency could be improved by at least one order of magnitude with a Ag STM tip as compared to a W tip (18). The clean NiAl(110) surface (19) was exposed to 100 Langmuirs (1 Langmuir = 1 s of exposure at 1 × 10−6 Torr) of O2 at 750 K and then annealed to 1300 K. Nearly 50% of the NiAl(110) surface was covered by a thin Al2O3 film. The oxide film on a fully oxidized NiAl(110) surface has a thickness of approximately 0.5 nm and a structure similar to that of γ-alumina (20). Zn(II)-etioporphyrin I (ZnEtioI, Porphyrin Corp.) was thermally sublimed onto the partially oxidized surface at 13 K (Fig. 1).

Figure 1

(A) STM topograph of a partially oxidized NiAl(110) surface with ZnEtioI molecules adsorbed at 13 K. The image size is 350 by 350 Å. The blue-colored (low) region in the image is the NiAl surface. The yellow-colored (high) region in the image is a 5 Å–thick alumina island. Porphyrin molecules appear as protrusions on both the oxide and metal surfaces. The apparent height of the island is greater than 5 Å because of the underlying multiterrace step of NiAl. (B) Molecular structure of ZnEtioI superimposed over a schematic representation of the four-lobe STM image. In STM images of adsorbed ZnEtioI molecules, the four lobes often appear nonuniform.

The photon emission spectra and intensities varied for different tips, especially for Ag tips. The spectra typically exhibited a number of broad peaks with relative intensities that depended on the voltage applied to the tunnel junction. This phenomenon has been attributed to the existence of local polarization oscillation modes (commonly referred to as plasmons) in the tunnel junction (21). Plasmons mediate the photon emission process and thereby have a direct influence on the emission spectra. In order to improve the emission intensity and make the plasmon spectra smoother in the desired spectral region, a series of high-voltage pulses was applied across the STM junction. Tips prepared in this way were suitable for probing molecular photon emission.

Light emission was measured in the constant current mode of the STM. Typically, sample bias voltages,V bias, ranging from 2 to 2.4 V (positive Vbias corresponds to electrons tunneling from the tip to the sample) and a tunneling current I of 0.5 nA were used to excite photon emission from individual molecules on alumina. Higher voltages and currents often irreversibly changed the molecules. The spectrometer slit was kept relatively wide in our experiment to achieve statistically sufficient photon counts on a time scale of minutes. This limited the spectral resolution to 8 nm with a 300-g/mm spectrometer grating.

Alumina film grown on NiAl(110) has a large unit cell (1.79 by 1.06 nm) and a relatively high density of defects (20). Because of this surface inhomogeneity, ZnEtioI molecules experience nonuniform local conditions on the oxide surface. The porphyrin macrocycle has a relatively flexible structure (22) and is susceptible to the local molecular environment. Deformations of the porphyrin macrocycle are known to cause substantial variations in the optical, vibrational, and electrochemical properties of the molecules (23). STM imaging of ZnEtioI revealed a number of different molecular conformations, some of which are shown in Fig. 2. Each conformation has a distinct energy level structure, as determined from the differential tunneling conductance (dI/dV) spectra (Fig. 2), which give a measure of the local density of states. Moreover, a strong correlation between the conformation, dI/dVspectrum, and photon emission was observed. In particular, emitting molecules always exhibited a dI/dV feature around 0.5 V (Fig. 2, A and B). In contrast, a large percentage (60 to 80%) of molecules (Fig. 2, C to F) did not have any low-energydI/dV features and showed very little or no emission at all. The conformation of ZnEtioI could be changed either accidentally during the course of light-emission measurements or deliberately with the STM. Reversible conformational changes were observed for some of the molecules, accompanied by the corresponding changes in their emission and dI/dV spectra.

Figure 2

(A to F, left and right panels) STM images of ZnEtioI molecules in different conformations on Al2O3/NiAl(110), with (center panel) the corresponding dI/dVspectra. All images sizes are 32 by 32 Å; V bias= 2.35 V and I = 0.1 nA. Etioporphyrin molecules are typically imaged with STM as four-lobe structures (Fig. 1B); lobes are numbered in panels (A) and (B). Depending on the molecular adsorption site and conformation, the apparent relative height of these lobes can be different. STM images can therefore be used to classify the molecules into different conformational groups. The saddle conformation in (B) is the most common one (approximately 30% of the molecules). We found that dI/dV spectra taken at different positions on the molecules were similar, so only one representative curve for each conformation is shown. ThedI/dV spectrum of the oxide surface is presented for comparison. The dI/dV spectra were measured in the constant height mode by setting the gap at V bias= 2.35 V, I = 0.25 nA. Molecules A and B luminesced, whereas no emission was detected for molecules C to F in the spectral range from 500 nm to 1000 nm. The presence of a low-energy peak in the dI/dV spectrum has been found to be necessary for a molecule to be optically active in the detection range.

Light emission from NiAl and oxide film (Fig. 3, A and B) arises through the well-known inelastic electron tunneling (IET) mechanism (24–26), which involves inelastic tunneling from the tip electronic states into the lower-lying states of the sample with a simultaneous release of the excess energy in the form of a plasmon. The excited plasmon then decays into a far-field photon. The spectrum of this emission is quite broad and has a characteristic energy cutoff determined by the sample bias. NiAl spectra reflect the wavelength-dependent emission yield of the tips. Photon emission spectra from the oxide are close in shape to those from NiAl, but their intensity is extremely weak (27). The same low-intensity emission was obtained from different locations on the oxide surface, including domain boundaries, step edges, and positions right next to molecules.

Figure 3

(A and B) Light-emission spectra acquired on the different lobes of molecules A and B from Fig. 2, together with spectra acquired on bare NiAl and Al2O3/NiAl(110) surfaces. The spectra in (A) and (B) are indexed according to the lobe-numbering patterns ofFig. 2, A and B, respectively, and are offset vertically for clarity. Series A and B were taken with two different Ag tips, as evident from the differences in corresponding NiAl spectra. The quantum cutoffs for NiAl and oxide spectra are outside the displayed energy ranges. In (A), the spectra were acquired at V bias = 2.35 V, I = 0.5 nA, with an exposure time of 100 s; the NiAl and oxide spectra have been multiplied by factors of 4 and 15, respectively. In (B), V bias = 2.2 V,I = 0.5 nA, and exposure time = 300 s; the oxide spectrum has been multiplied by a factor of 3. (C) Variation of curve 1 in (A) as a function ofV bias [the same tip was used as in (A)]. The inset shows the dependence of the 800-nm peak intensity on current (V bias = 2.35 V). Linear dependence was found for all wavelengths in the measured spectral region. (D) Variation of curve 1 in (B) as a function ofV bias [the same tip was used as in (B)]. The photon spectra were not corrected for the wavelength-dependent sensitivity of the detection system. The signal is given in photon counts as detected by our CCD camera (the total number of points in each spectrum is 1340).

A characteristic photon emission with sharp features can be seen when the STM tip is positioned directly above a molecule (Fig. 3, A and B). Furthermore, the light-emission spectrum is very sensitive to the tip position inside the molecule. The emission spectra acquired over the four unequal lobes of molecule A in Fig. 2 exhibit substantial differences in the relative intensities of the spectral features (Fig. 3A, curves 1 to 4). However, as might be expected from the apparent molecular symmetry, emission spectra from the like pairs of molecular lobes of molecule B in Fig. 2 are similar in shape and intensity (Fig. 3B, curves 1 and 3 and curves 2 and 4). Lobes 1 and 3 give highly structured emissions, whereas the spectra of lobes 2 and 4 show a “step” at ∼740 nm. The observed spectral features are very sharp, in contrast to spectra obtained from bare NiAl and oxide surfaces.

In principle, the light-emission variations measured on different molecular lobes could be caused by variations in the apparent heights of the lobes, because the plasmon spectrum depends on the height of the tip above the sample (12). Plasmon modes have lateral dimensions of around several nanometers (28),thus they are expected to be insensitive to the lateral position of the tip above the molecule, because the distance between the lobes is ∼1 nm. We did not observe any appreciable changes in emission yield (photons per electron) for all measured wavelengths, when the height of the tip above the molecules was varied by changing the tunneling current at a given sample bias. The linear dependence of emission intensity with tunneling current (Fig. 3C, inset) indicates constant emission yield. Therefore, the lobe-dependent variations in emission cannot be explained by changes in the plasmon mode properties.

The series of emission spectra for different sample biases (Fig. 3, C and D) further clarifies the nature of the observed spectral features. In Fig. 3C, the spectral peaks do not shift whenV bias is varied. This result indicates that these peaks did not originate from transitions between the electronic states of the tip and those of the substrate. The sharp spectral features could only be observed when the STM tip was positioned on a molecule. We therefore attributed them to transitions inside the molecule. The existence of a cutoff voltage (approximately 2 V in Fig. 3C) for excitation of the sharp features is expected if an excited electronic level participates in emission. The difference between this cutoff voltage and the photon energy of the shortest-wavelength feature in the spectra of Fig. 3C (∼1.57 eV), lies in the range of the low-energy dI/dV peak for this molecule (Fig. 2A) (29). This relationship between the cutoff voltage, photon energy, and the dI/dV peak was found to be characteristic for molecules that showed sharp spectral features in their emission spectra.

We conclude that, for ZnEtioI, the two major channels for light emission involve IET and molecular fluorescence (Fig. 4). A similar picture has been presented in a study of luminescence from the quantum well structures formed by Na layers on a Cu(111) surface (30). This luminescence, assigned to the transitions between the electronic levels of the quantum well structure, was observed together with the IET emission.

Figure 4

Diagram showing the two major processes contributing to STM-excited light emission from a molecule adsorbed on the oxide surface. In process A, the IET channel, an electron inelastically tunnels from the Fermi level of the STM tip into an unoccupied molecular orbital with simultaneous excitation of a plasmon. In process B, the fluorescence channel, an electron tunnels into the higher unoccupied orbital of the molecule. The charged molecule then relaxes to a lower vibrational level of the same electronic level, with subsequent radiative (excitation of a plasmon) transition to the lower electronic level. The final step involves tunneling of this extra electron into the NiAl substrate. The plasmons are detected as photons in the far field. E F, Fermi energy;hv, photon energy.

In the IET channel (process A in Fig. 4), electrons inelastically tunnel from the tip into the lower unoccupied electronic state of the molecule, as evidenced by the variation of the cutoff photon energy with V bias (Fig. 3D). This energy is given by the difference between the Fermi level of the tip and the molecular state shown in Fig. 2B. However, if IET final states were those of NiAl, the cutoff photon energy at a givenV bias would have been higher than the displayed spectral range, as in the case of the NiAl and oxide emission spectra in Fig. 3B. A similar effect has been studied in STM light-emission experiments on Au particles supported by Al2O3film grown on an Al substrate (31), where the quantum cutoff was shown to be lowered by approximately the particle-charging energy.

In the fluorescence channel of emission (process B inFig. 4), electrons tunnel elastically into the molecule to create an electronically and vibrationally excited anionic state of the molecule. The molecule can undergo vibrational relaxation in the excited state before making a radiative transition to the lower electronic state. This radiative transition involves excitation of a plasmon, which is then detected in the far field as a photon. The last step involves the tunneling of the extra electron through the alumina film into the NiAl. Because the fluorescence spectral pattern did not change whenV bias was changed (Fig. 3, C and D), it is most likely that the molecule relaxed to the ground vibrational state of the excited electronic state before exciting a plasmon. (The increase in the fluorescence peak intensity with higherV bias can be explained by the increased number of vibronic states available for excitation of the molecule during the initial tunneling event.) We therefore attribute the series of peaks in the light-emission spectra to excitation of molecular vibrations in the lower electronic state. The spectral range of the observed fluorescence is near the Zn-etioporphyrin monoanion luminescence band (32).

Light emission from different molecular lobes can display large variations, even when the dI/dVspectra of different lobes are similar. Molecular fluorescence involves the transition between two molecular states. Whereas the lower orbital is revealed by the dI/dV spectra (Fig. 2, A and B), the higher energy part (above 2V) of dI/dVspectra is dominated by the background NiAl states, so that no definite conclusion can be drawn about the properties of the higher molecular state. Thus, the variations of photon emission for different lobes can be tentatively attributed to the spatially nonuniform properties of the higher-energy molecular states.

The fluorescence pattern strongly depends on the molecular conformation. However, light emission is almost identical for molecules of the same conformation. This reproducibility is demonstrated by the photon spectra of three “saddle” molecules (Fig. 2B), measured with different STM tips (Fig. 5A). Even though the apparent spectral shapes look different, the positions of the spectral features are very reproducible for different Ag and W tips. Figure 5B shows the light-emission spectra from clean NiAl surfaces recorded with the same tips as in Fig. 5A. From Fig. 5B it is clear that the tips have different plasmon modes. Thus, it is expected that the differences in molecular spectra of Fig. 5A may be caused by the tip-specific plasmon modes. In order to eliminate the influence of plasmons from these spectra, we divided each molecular spectrum (Fig. 5A) by the corresponding NiAl spectrum (Fig. 5B) taken with the same tip (33). The resulting three curves in Fig. 5C show remarkable similarity, indicating that the saddle molecules in fact have nearly identical intrinsic light-emission properties, independent of the tip. This result gives additional support to identification of the spectral peaks in Fig. 5A as fluorescent features, refuting the possible plasmon origin of these peaks (34, 35). Indeed, the influence of plasmon modes on light emission is strongly dependent on the tip (Fig. 5B). This influence must be absent in the curves of Fig. 5C, because they are very similar for different tips.

Figure 5

(A) Light-emission spectra for three different experimental runs with three different tips. The spectra were acquired over one of the smaller lobes of the saddle molecules, as marked in the inset image. The raw data are plotted together with the corresponding smoothed curves (colored lines) to facilitate the identification of the peaks (red bars). Spectra 1 (V bias = 2.35 V, I = 0.5 nA, exposure time = 300 s) and 2 (V bias = 2.35 V, I = 0.6 nA, exposure time = 200 s) were taken with two different Ag tips. Spectra 1 and 2 have been offset vertically for clarity. Spectrum 3 (V bias = 2.3 V, I = 1 nA, exposure time = 600 s; the original data have been multiplied by 3) was obtained with a W tip. The differences in the spectra are caused by different tip plasmon properties. In particular, because of the higher amount of dielectric losses characteristic for W, the emission rate for the W tip is ∼30 times lower. (B) NiAl light-emission spectra measured with same tips as in (A); the curve sequence is consistent with that of (A). The raw data are plotted together with the corresponding smoothed curves (colored lines). The spectra were acquired with the same voltages as in (A) and scaled, so that the photon yields of different tips could be directly compared (the differences in the levels of statistical noise are caused by the different acquisition times for the three spectra; the data of curve 3 have been multiplied by 25). (C) Smoothed molecular spectra from (A), divided by the corresponding smoothed NiAl spectra from (B) and normalized to fit the same scale (33). The inset shows the photon energy of each peak determined for the three spectra, as marked in (A).

Within the accuracy of our experiment, the vibrational features in Fig. 5 are equidistant in energy, with peak spacing of 40 ± 2 meV (Fig. 5C, inset). In addition, the intensity distribution of the fluorescence peaks in Fig. 5C is reminiscent of a Franck-Condon vibrational progression pattern. These findings suggest that the peaks in these spectra belong to the same vibrational progression, although the possibility that the progression contains more than one vibrational mode cannot be ruled out, because of the limited spectral resolution of the experiment. A number of porphyrin skeletal vibrational modes have energies close to the measured vibrational peak spacing. However, precise identification of the progression-forming modes is difficult. When the molecular symmetry is reduced, the selection rules can change and originally inactive modes may appear in the spectrum. This effect has been demonstrated in resonant Raman experiments on different conformations of nickel octaethylporphyrin (36). Furthermore, the molecules are distorted by a pronounced Jahn-Teller effect that is present in porphyrin anions (37). Theoretical calculations of the porphyrin conformational and electronic properties, as well as high-resolution spectroscopy, would enable detailed analysis of the observed vibrational features. Vibrational progressions in STM-excited photon spectra have been observed and identified for relatively thick (approximately 200 nm) polymer films (38).

Weakness of interaction between a molecule and the substrate is essential for observation of the STM-excited molecular fluorescence. In our experiment, we used an ultrathin alumina film to decouple the molecule from the metal substrate. However, this is not the only possible choice. Other thin oxide films, such as TiO2, epitaxially grown on solid surfaces are expected to have a similar effect. Self-assembled monolayers of organosulfur molecules, especially long-chained alkanethiols, might be another attractive spacer candidate for a variety of noble metal surfaces.

STM-excited fluorescence spectroscopy combined with imaging and scanning tunneling spectroscopy can be used for probing the interdependence between the conformational structure, energy levels, and optical properties of single molecules. The high spatial resolution of STM can be used to obtain spatial maps of fluorescence spectra for individual molecules, enabling the determination of the inner mechanisms involved in the light-emission process. Our approach can be useful in studying with submolecular resolution such important issues as electron dynamics in conjugated polymers and organic molecules and intermolecular fluorescence resonance energy transfer and its modification by local plasmon modes.

  • * These authors contributed equally to this work.

  • To whom correspondence should be addressed. E-mail: wilsonho{at}


View Abstract

Stay Connected to Science

Navigate This Article