Imaging and tuning molecular levels at the surface of a gated graphene device.

Gate-controlled tuning of the charge carrier density in graphene devices provides new opportunities to control the behavior of molecular adsorbates. We have used scanning tunneling microscopy (STM) and spectroscopy (STS) to show how the vibronic electronic levels of 1,3,5-tris(2,2-dicyanovinyl)benzene molecules adsorbed onto a graphene/BN/SiO2 device can be tuned via application of a backgate voltage. The molecules are observed to electronically decouple from the graphene layer, giving rise to well-resolved vibronic states in dI/dV spectroscopy at the single-molecule level. Density functional theory (DFT) and many-body spectral function calculations show that these states arise from molecular orbitals coupled strongly to carbon-hydrogen rocking modes. Application of a back-gate voltage allows switching between different electronic states of the molecules for fixed sample bias.

Combining organic molecules with graphene creates new opportunities for fabricating hybrid devices with tailored properties. Previous experiments have shown that electronic,[1−18] magnetic,[6,19,20] and optical[21−23] characteristics as well as chemical reactivity[22,24,25] of graphene devices can be tuned through molecular adsorption. Such measurements have been performed primarily using electrical conductivity and optical spectroscopy techniques. These measurements, however, do not directly yield local microscopic information regarding the hybrid graphene/molecule interface. Additional electronic structure information on molecule/graphene systems in nongated configurations has been provided by STM[14,15,20,26−200] and photoemission spectroscopy[7,13,14,18,29,33,34] experiments, including measurement of the energy location of molecular orbitals. Such measurements, however, have so far precluded the hybrid molecule/graphene electronic structure from being electrostatically tuned through the use of a back-gate, and molecular vibronic properties in these systems remain poorly understood.[32]

Here we describe a single-molecule-resolved STM study of a molecular monolayer adsorbed onto a back-gated graphene device (Figure a) that allows both characterization and gate-induced modification of molecular electronic properties. 1,3,5-Tris(2,2-dicyanovinyl)benzene (CVB) molecules were adsorbed onto a graphene device in ultrahigh vacuum (UHV) and studied via STM spectroscopy at cryogenic temperatures. Hybridized electronic levels of individual CVB molecules on graphene were imaged, and the electronic states were observed to exhibit unexpectedly strong vibronic satellites. The hybridized vibronic electronic structure of the CVB molecules rigidly shifted in energy as a voltage was applied to the device back-gate, thus allowing the electronic local density of states (LDOS) at fixed sample bias to be switched between different molecular orbitals. Identification of the experimentally observed molecular orbitals was facilitated via density functional theory (DFT) based spectral function simulations which accurately reproduce the orbital structure imaged by STM. These simulations also allow identification of the vibronic satellites through calculation of the CVB electron–phonon coupling. Although CVB molecules exhibit a broad spectrum of vibrational modes, only C–H rocking modes of the CVB molecules having an energy close to 200 meV are seen to contribute significantly to the molecule/graphene electron–phonon coupling. The energy of these modes is in good agreement with the energy spacing of vibronic satellites observed experimentally for CVB on graphene.

Figure 1

CVB molecules on a graphene/BN/SiO2 FET device. (a) Sketch of the back-gated graphene device used in these STM/STS measurements, as well as a model of the CVB molecule. (b–d) STM images of a monolayer of CVB molecules on graphene/BN show the hexagonal lattice of the CVB molecules at different zoom values (VS = 2.0 V, It = 10 pA, T = 4 K). Isolated vacancies are observed in (b) and (c).

Results and Discussion

Parts b–d of Figure show STM images of a monolayer-high self-assembled island of CVB molecules on graphene/BN. The Moiré pattern arising from interaction between the graphene lattice and the underlying BN lattice is clearly visible (Figure b,c) even though the graphene is covered with a layer of molecules. The molecules form a hexagonal lattice with a lattice constant of a = 1.13 ± 0.01 nm, and isolated defects having triangular shape can be observed. The close-up STM image in Figure d shows a spiral-like electronic LDOS that is associated with the molecules (the opposite chirality was also observed in different islands). The presence of single-molecule vacancies (Figures b,c) allows us to assign the location of CVB molecules in the film.

dI/dV spectra were measured while holding the STM tip above the CVB monolayer. The spectra were essentially featureless over the range −0.5 V < VS < +1.6 V, but for VS > 1.6 V clear resonance features were observed (Figure a). For VS < −0.5 V the CVB molecules tended to jump to the STM tip, and so reproducible STM spectra at sample voltages lower than −0.5 V could not be obtained. The Dirac point of the underlying graphene substrate could be seen when the tip height was lowered by 4 Å relative to the typical dI/dV measurement tip-height (Figure a inset), but this usually led to CVB molecules jumping to the STM tip. The empty state spectrum for larger tip heights, however, was quite stable up to nearly 3 V, as seen in the spectrum of Figure a (this spectrum was reproduced with numerous different tips and samples). Four pronounced peaks can be seen in the spectrum, labeled 1–4. A statistical analysis of our spectra (using Gaussian fits to the peaks) yields the following energy locations for the four peaks (where E = |e|VS): E1 = 1.86 ± 0.02 eV, E2 = 2.06 ± 0.02 eV, E3 = 2.28 ± 0.02 eV, and E4 = 2.68 ± 0.03 eV. The energy differences between peaks 1–2 and peaks 2–3 are quite similar, whereas the energy difference between peaks 3–4 is twice as big: E2 – E1 = 0.20 ± 0.03 eV, E3 – E2 = 0.22 ± 0.03 eV, and E4 – E3 = 0.40 ± 0.04 eV. dI/dV maps taken at sample biases covering the range of the first three peaks (1.85 V < VS < 2.4 V) show no significant differences in the spatial distribution of the electronic LDOS (Figure b, first two panels). However, the dI/dV map obtained at the energy of the fourth peak (VS = 2.65 V) shows a significantly different spatial distribution of the electronic LDOS (Figure b, third panel).

Figure 2

STM spectroscopy of CVB/graphene/BN reveals vibronic response. (a) dI/dV spectrum measured with STM tip held above a monolayer of CVB molecules on a graphene/BN FET device (VG = 0). Spectrum is featureless over the range −0.5 V < VS < 1.6 V but shows four clear molecule-induced resonances (marked 1–4) in the range 1.6 V < VS < 2.8 V (junction set-point parameters VS = 2.7 V, It = 160 pA; the spectrum is normalized by its value at 2.6 V). Inset shows a section of the dI/dV spectrum over the range −0.6 V < VS < 0.6 V where the tip has been lowered by 4 Å relative to other spectra (junction set-point parameters: VS = 0.6 V, It = 40 pA; tip is closer because VS now lies in the HOMO–LUMO gap). Here the Dirac point can be observed at VS ≈ 0 V (VG = 0 V). Inset also shows onset of peak 1 (0.6 V < VS < 1.9 V) for typical junction set-point parameters: VS = 1.9 V, It = 40 pA. Peaks 1 and 4 are interpreted as LUMO and LUMO+1, respectively, while peaks 2 and 3 are interpreted as vibronic satellites of the LUMO (see text). (b) Experimental dI/dV maps obtained at voltages VS = 1.85, 2.4, and 2.65 V (VG = 0 V). dI/dV maps taken in the range 1.85 V < VS < 2.4 V probe the local density of states (LDOS) of peaks 1–3 and look very similar. The dI/dV map taken at VS = 2.65 V probes peak 4 and yields a LDOS pattern that is different from the pattern observed for peaks 1–3. (c) Calculated density of states (DOS) of vibrational modes of CVB molecules on graphene (gray lines), as well as the electron–phonon coupling strength between the CVB vibrational modes and the CVB LUMO state (vertical blue lines). The blue curve shows the calculated electron–phonon coupling broadened with a Gaussian function of width 16 meV.

One of the unique aspects of this study is that we were able to perform STM spectroscopy and imaging while modifying the hybridized molecule/graphene electronic doping using an electrostatic back-gate. Figure a shows dI/dV spectra taken on a CVB monolayer island at two different back-gate voltages (VG). The black trace shows the spectrum acquired at VG = 0 V while the red trace shows the spectrum taken at a gate voltage of VG = 60 V. The red trace is rigidly shifted by ∼0.2 V toward lower sample bias but does not exhibit any other significant changes in its features. We see a similar gate–dependent shift in the Dirac point energy via STM spectroscopy for these graphene devices without molecular layers,[35] corresponding to a change in the carrier concentration from n = 4 × 1010 cm–2 (VG = 0 V) to n = 4 × 1012 cm–2 (VG = 60 V). We next acquired dI/dV maps at a fixed sample bias of VS = 2.4 V, but for different gate voltages VG. The dI/dV map acquired at VG = 0 V (Figure b) shows the same features as observed at biases corresponding to peaks 1–3 (Figure b, first two panels), but the dI/dV map acquired at VG = 60 V (Figure c) exhibits a significantly different LDOS that is similar to what was observed previously for peak 4 (Figure b, third panel). The LDOS at this fixed energy with respect to EF can thus be toggled between two different molecular orbitals via application of a positive gate voltage (this orbital switching is reversible and has no hysteresis).

Figure 3

Gate-induced shift of the electronic levels of CVB molecules on a graphene/BN FET device: (a) dI/dV spectra of CVB/graphene/BN at two different gate voltages: VG = 0 V (black trace) and VG = 60 V (red trace). Increasing the gate voltage to VG = 60 V causes a rigid downward shift of the molecular electronic resonances by 0.2 eV, consistent with the gate-induced shift seen in the Dirac point for graphene/BN devices without adsorbed molecules (spectra are normalized by their respective values at VS = 2.6 V). (b) Experimental dI/dV map obtained with VS = 2.4 V and VG = 0 V. (c) Same as (b) except that VG = 60 V. (d) Theoretical local density of states map of the CVB/graphene LUMO state calculated using DFT. (e) Same as (d) except for LUMO+1 state. These maps show that changing the device gate voltage allows the STM to access different molecular orbitals for a fixed sample bias.

We are able to understand our gate-dependent local electronic structure measurements of the hybrid CVB/graphene system through the use of first-principles simulations. The main questions we wish to answer here concern the origin and behavior of spectroscopic peaks 1–4 (Figures and 3). To do this, we performed density functional theory (DFT) calculations of the combined CVB/graphene system using an exchange-correlation functional which combines the PBE functional together with a semi-empirical dispersion correction to take into account van der Waals interactions between the molecules and graphene.[36,37] After correcting for electron–electron[38−40] and electron–phonon interactions[41,42] in the electron self-energy (see the Supporting Information), we obtained a HOMO–LUMO energy gap of 6.3 eV and a (LUMO)–(LUMO+1) energy gap of 0.8 eV. Although our experimental energy range does not allow us to experimentally verify the predicted HOMO–LUMO gap, we note that the theoretical (LUMO)–(LUMO+1) energy almost perfectly matches the experimental energy difference between peaks 1 and 4 (ΔE14 = 0.82 ± 0.04 eV). This suggests that peak 1 is the CVB/graphene LUMO and that peak 4 is the LUMO+1 state. To further test this hypothesis we calculated the theoretical LDOS of the LUMO and LUMO+1 states and compared it to the experimental dI/dV maps measured at the energies of peaks 1 and 4. The theoretical LDOS of these different states (Figure d,e) is in good agreement with the experimental LDOS maps (first panel and third panel in Figure b), thus providing additional evidence that peaks 1 and 4 correspond to the system’s LUMO and LUMO+1 states (this procedure also allowed us to confirm the molecular orientation shown in Figures and 3).

An important remaining question is the origin of peaks 2 and 3, which do not appear in the calculated DOS obtained via DFT. These peaks clearly have some relation to the LUMO state, since their experimental dI/dV maps are essentially identical to the dI/dV map for peak 1 (Figure b), and they strongly resemble the theoretical LDOS map that was calculated for the LUMO state (Figure d). We believe that these additional peaks (2 and 3) arise due to the existence of vibronic modes of the CVB molecule on graphene. Such modes reflect coupling between the electronic and vibrational states of a molecule and result in new features in the electron spectral function. Vibronic modes have been seen previously in STM spectroscopy of molecules that are decoupled from a metallic substrate[43−45] (single-particle vibronic modes are quenched by direct molecular coupling to a metallic electrode[43,45]). The spacing between vibronic satellites corresponds to the quantum of vibrational energy (ℏω) for the molecular vibrational state that is entangled with the electronic orbital. Vibronic states do not appear in DFT calculations of electronic levels because the Born–Oppenheimer assumption prevents mixing of electronic states with vibrational modes.

To verify this picture, we calculated the theoretical vibrational modes for CVB molecules on graphene as well as the electron–phonon coupling that exists between the molecular LUMO state and each vibrational mode. While the molecular vibrations span a wide energy range, we seek to understand whether strong electron–phonon coupling exists for any modes having energy similar to the energy difference between peaks 1 and 2 (0.2 eV), as well as between peaks 2 and 3 (0.22 eV). The gray lines in Figure c show the theoretical vibrational modes of CVB on graphene, calculated within the framework of DFT perturbation theory. As expected, they span a wide energy range, up to 400 meV. The calculated electron–phonon coupling strength between these modes and the CVB LUMO state is shown by the bold blue trace in Figure c (see eq SI-6 in the Supporting Information for a detailed definition of this quantity). The electron–phonon coupling shows a strong peak at the modes near 200 meV, in good agreement with the experimental value of ΔE12 and ΔE23 which have an average value of 210 meV. From this calculation we are able to determine that the modes with high electron–phonon coupling at 200 meV mainly involve C–H rocking vibrations (which have representations A1′, E1′, and E2′ of the C3 point group) suggesting that this particular type of vibration is the origin of the vibronic satellites observed experimentally as peaks 2 and 3 (a detailed calculation of electronic LDOS within the first order cumulant approximation for the electron–phonon coupling is shown in the Supporting Information).

This vibronic interpretation of the dI/dV spectrum helps to explain the gate-dependent orbital switching observed in dI/dV maps obtained at a fixed tip–sample bias of VS = 2.4 V. Here a gate voltage of VG = 0 V yields a dI/dV map reflecting the LUMO density (Figures b,d) while a gate voltage of VG = 60 V yields a dI/dV map reflecting the LUMO+1 density (Figures c,e). Within a vibronic picture the fixed tip–sample bias at VG = 0 V probes the peak 3 resonance which is a vibronic LUMO satellite, whereas gating at VG = 60 V shifts the tunnel current to the peak 4 resonance which reflects the LUMO+1 state. Electrostatic gating thus allows a fixed tip–sample bias to switch between imaging the LUMO and LUMO+1 orbitals.

Conclusions

In conclusion, we have shown that CVB molecules adsorbed onto a graphene/BN device self-assemble into a hexagonal lattice and develop vibronic peaks that correspond to coupling of electrons to the A1′, E1′, and E2′ rocking modes of the CVB carbon–hydrogen bonds. The fact that vibronic peaks can be so readily resolved in the molecule/graphene spectra shown here suggests that substrate-induced lifetime broadening is weak on graphene due to electronic decoupling of adsorbed molecules.[43,45] Gating the hybrid molecule/graphene device allows electronic switching between two different molecular states (LUMO and LUMO+1) for a fixed tip–sample bias voltage. Extensions of this approach can be envisioned that might allow exploration of gate-controlled changes in molecular functionality for hybrid graphene devices at the single-molecule level.

Methods

We used back-gated graphene/BN/SiO2 devices[46−48] similar to the one schematically depicted in Figure a. The graphene sample was grown by the CVD method described in ref (48). Hexagonal boron nitride flakes were exfoliated onto heavily doped silicon wafers coated by a 285 nm thermal oxide. The graphene was transferred on top of the BN/SiO2[48] using a polydimethylsiloxane (PDMS) stamp, and electrical contact was made by depositing Ti (10 nm thick)/Au (30 nm-thick) electrodes using a stencil mask technique.

1,3,5-Tris(2,2-dicyanovinyl)benzene (CVB) was synthesized through a Knoevenagel condensation of benzene-1,3,5-tricarboxaldehyde and malononitrile.[51,52] The molecules were evaporated from a Knudsen cell onto the graphene device in UHV while the device was held at T = 5 K. The device was then briefly annealed at room temperature before being cooled back to 4 K. CVB was chosen for this experiment due to its extended π-system and the high electron affinity of its dicyanovinyl groups.

STM/STS was performed using an Omicron LT STM at T = 4 K. STM differential conductance (dI/dV) was measured in constant-height mode (both for point spectroscopy and for spatial maps) by lock-in detection of the a.c. tunnel current generated by a 6 mV rms 316 Hz signal added to the sample bias.

DFT calculations were performed using an exchange-correlation functional that combines the PBE functional together with a semi-empirical dispersion correction to take into account van der Waals interaction between the molecules and graphene.[36,37] We used a supercell containing one CVB molecule and 42 graphene atoms, with a lattice constant of 1.13 nm. The combined CVB and graphene system was allowed to relax in these simulations to its most stable configuration. Frequencies of vibrational modes and electron–phonon matrix elements were calculated in the framework of density functional perturbation theory (DFPT)[41] using the QUANTUM ESPRESSO package.[42] We performed GW[38,39] corrections employing the BerkeleyGW package[40] to account for electron–electron correlations not captured in DFT. The molecular orbital energies were further corrected for electron–phonon interactions using the first-order cumulant approximation (see the Supporting Information).