Stable anchoring chemistry for room temperature charge transport through graphite-molecule contacts.

An open challenge for single-molecule electronics is to find stable contacts at room temperature with a well-defined conductance. Common coinage metal electrodes pose fabrication and operational problems due to the high mobility of the surface atoms. We demonstrate how molecules covalently grafted onto mechanically robust graphite/graphene substrates overcome these limitations. To this aim, we explore the effect of the anchoring group chemistry on the charge transport properties of graphite-molecule contacts by means of the scanning tunneling microscopy break-junction technique and ab initio simulations. Molecules adsorbed on graphite only via van der Waals interactions have a conductance that decreases exponentially upon stretching the junctions, whereas the molecules bonded covalently to graphite have a single well-defined conductance and yield contacts of unprecedented stability at room temperature. Our results demonstrate a strong bias dependence of the single-molecule conductance, which varies over more than one order of magnitude even at low bias voltages, and show an opposite rectification behavior for covalent and noncovalent contacts. We demonstrate that this bias-dependent conductance and opposite rectification behavior is due to a novel effect caused by the nonconstant, highly dispersive density of states of graphite around the Fermi energy and that the direction of rectification is governed by the detailed nature of the molecule/graphite contact. Combined with the prospect of new functionalities due to a strongly bias-dependent conductance, these covalent contacts are ideal candidates for next-generation molecular electronic devices.

INTRODUCTION

Moleculn class="Chemical">ar electronics is a research field that aims at exploiting individual molecules as building blocks for electronic devices (, ). Since the first seminal studies, important experimental advances have helped to shed light on the structure-property relationships that determine the charge transport in typical metal/molecule/metal junctions (). In particular, it is now understood that the nature of the electrode materials and of the anchoring groups used for the electrode chemical functionalization plays a key role. Most studies to date have considered standard coinage metal electrodes (–), such as Au, Pt, Cu, Pd, and Ag, and a large number of anchoring groups have been proposed and thoroughly characterized (–).

The two fundamental challenges in moleculn class="Chemical">ar electronics that have limited the technological breakthroughs are the device reproducibility and stability at room temperature. Devices with coinage metal electrodes suffer from several problems: (i) The conductance is very sensitive to the atomistic details of the contacts, which show a large device-to-device variation especially at room temperature due to the low mechanical stability and high mobility of the electrode surface atoms (, ); (ii) except for gold, all other coinage metals readily oxidize under ambient conditions (, ); and (iii) the rather featureless density of states (DOS) of coinage metals around the Fermi energy limits the possible range of device functionality.

Graphite and n class="Chemical">graphene offer an attractive alternative to coinage metals because of their robust mechanical stability and the highly dispersive DOS around the Fermi energy (EF) (). On the basis of these ideas, McCreery et al. (, ) studied large-area molecular devices, where an organic monolayer was attached covalently to a graphitic substrate and to a Cu (or Au) top electrode. Recently, the charge transfer properties of molecular wires were also studied upon the integration of these molecules into large-area () or single-molecule (, –) graphene (graphite)–metal junctions via noncovalent attachment. At the same time, several authors (, ) have proposed novel single-molecule transistors, where the electrodes are created by electroburning a nanosized gap across either a graphene or a few-layers graphene flake. The contact with the molecules, which present planar anchoring groups, such as anthracene and pyrene, is typically achieved by π-π stacking (, , ). Unfortunately, all aforementioned devices present a number of limitations. On the one hand, large-area devices do not allow for a precise characterization of the molecular contact and of its importance for the transport properties. They can only measure averaged quantities and cannot discern whether the results are representative of a well-defined molecule/substrate contact or whether they are just an average of widely varying individual contact properties. On the other hand, the fabrication of graphene (graphite) single-molecule transistors involves steps that are difficult to control so that the performances cannot be easily characterized with statistically significant methods, and the transport properties are greatly affected by the largely unknown chemical and electronic structure of the gap edges (). Furthermore, the main fundamental problem for all the noncovalent attachments that have been considered so far in single-molecule devices is that they lack device reproducibility and stability in the same way as found for coinage metal electrodes.

Instead of using the noncovalent attachment, molecules can be grafted onto n class="Chemical">carbon-based electrodes via direct C–C bonding. This covalent anchoring to the graphite/graphene surface produces an sp3 node in the perfect sp2 honeycomb lattice. The effect of the formation of an sp3 node on the charge transport properties of single molecular junctions and its potential stability for room temperature applications, in particular when compared to the noncovalently bonded counterparts, is unknown. Here, we use the scanning tunneling microscopy break-junction (STM-BJ) technique combined with electronic transport simulations based on density functional theory (DFT) and nonequilibrium Green’s functions (NEGFs) to address the charge transport properties of hybrid junctions, where molecules are trapped between a highly oriented pyrolytic graphite (HOPG) substrate and a gold STM tip. We systematically compare for the first time the results for the single-molecule junctions formed with covalent (sp3 node formation; Fig. 1, B and C) and noncovalent (no sp3 node; Fig. 1D) attachment of molecules to the graphite surface, whereas the anchoring to the gold tip is achieved via the amine (NH2) group in both cases for a well-defined comparison. Both covalent and noncovalent junctions exhibit a bias-dependent conductance (G), and the rectification occurs in opposite direction for the two cases, showing that new functionalities can be tailored by specifically designing the graphite-molecule interface. Covalent functionalization of the HOPG surface results in molecular junctions with a distinct and narrow distribution of conductance values at room temperature. This is a radical improvement over noncovalently bonded molecules, which show scattered conductance values exponentially dependent on the angle between molecule and substrate. The results therefore demonstrate that the C–C direct bond ensures a room temperature–stable and precisely defined contact between the molecules and the top graphene layer. Hence, our study represents an important step toward room temperature reproducibility and stability in molecular electronics, a key requirement for technological applications.

Fig. 1

Single-molecule conductance measurements.

Left panels illustrate the schematics of Au/HOPG junctions in (A) blank experiment without adding molecules and (B to D) with trapped single molecules: (B) molecular junction with a probable dendrimeric structure, which can be obtained upon AB grafting; (C) molecular junction with grafted DMAB; and (D) noncovalently anchored molecular junction with PPD. The corresponding 1D and 2D conductance histograms, which are constructed from more than 1000 individual withdrawing traces, are shown in the middle and right panels, respectively. The STM-BJ results in (A) to (C) are obtained in argon atmosphere without solvent. The measurements shown in (D) are carried out with PPD dissolved in 1,2,4-trichlorobenzene up to a concentration of 0.1 mM. In all cases, a positive bias voltage of 100 mV is applied. In 2D histograms, Δz represented the relative vertical displacement of the STM tip with respect to the substrate.

Left panels illustrate the schematics of Au/HOPG junctions in (A) blank experiment without adding molen class="Chemical">cules and (B to D) with trapped single molecules: (B) molecular junction with a probable dendrimeric structure, which can be obtained upon AB grafting; (C) molecular junction with grafted DMAB; and (D) noncovalently anchored molecular junction with PPD. The corresponding 1D and 2D conductance histograms, which are constructed from more than 1000 individual withdrawing traces, are shown in the middle and right panels, respectively. The STM-BJ results in (A) to (C) are obtained in argon atmosphere without solvent. The measurements shown in (D) are carried out with PPD dissolved in 1,2,4-trichlorobenzene up to a concentration of 0.1 mM. In all cases, a positive bias voltage of 100 mV is applied. In 2D histograms, Δz represented the relative vertical displacement of the STM tip with respect to the substrate.

RESULTS AND DISCUSSION

HOPG/molecule structures and single-molecule conductance measurements

For the investigation of the covalent HOPG/molecule contacts, we use n class="Chemical">4-nitrobenzenediazonium and 3,5-dimethyl-4-nitrobenzenediazonium salts with tetrafluoroborate anion; the syntheses and characterization details are provided in section S1 (scheme S1 and figs. S1 to S5). The diazonium salts are electrochemically grafted onto an HOPG substrate, and subsequently, the electrochemical reduction of the nitro group to amine is performed (see Materials and Methods and section S2 for the assembly details and adlayer characterization). This protocol gives rise to the final organic submonolayers of 4-aminobenzene (AB) and 3,5-dimethyl-4-aminobenzene (DMAB). In both cases, Raman spectra show a clear indication of the formation of sp nodes on the HOPG surface (fig. S4A), and topographic STM measurements reveal that the molecules tend to assemble in small clusters (fig. S4, B1 and B2). For noncovalent bonding, we use the p-phenylenediamine (PPD) molecule.

The STM-BJ experiments with and without molecules n class="Chemical">are performed with an Au tip and give a measurable conductance range between 10−7.0 and 10−2.0 G0 for 100 mV of applied bias voltage (Fig. 1; G0 is the conductance quantum, equal to 77.5 μS). The one-dimensional (1D) conductance histograms (Fig. 1, middle panels) are constructed from all measured individual conductance traces without any data selection. Reference blank measurements without molecules for the bare HOPG substrate under argon atmosphere (Fig. 1A and fig. S6) display a peak in the 1D histogram at around 10−7 G0, which corresponds to the noise level of our setup, and a feature at around 10−2.7 G0. This is due to the formation of a soft contact between the Au tip and the HOPG substrate. No feature is observed in the range from 10−2.7 to 10−7 G0. A similar featureless 1D histogram is obtained in a blank experiment in a nonconductive solvent, such as 1,2,4-trichlorbenzene, saturated by Ar (fig. S7). For the three different molecular junctions comprising AB, DMAB, and PPD, additional peaks in the 1D histogram between 10−6.0 G0 and 10−3.0 G0 are found and assigned to the most probable conductance of the single-molecule junctions. In addition to the 1D conductance histograms, we also plot the 2D conductance-distance histograms (Fig. 1, right panels), which provide further valuable information concerning the conductance evolution during the stretching process. The distance axis is aligned with respect to 10−3 G0, which corresponds approximately to the conductance of the soft Au-HOPG direct contact, and individual conductance-distance traces obtained in this manner are binned in the 2D graph and overlaid together as color plot. In the absence of molecules, the 2D graph conductance-displacement curves show an exponential decay with distance, whereas in the presence of molecules for each peak in the 1D histogram, plateaus appear in the 2D traces (, ). Typical individual conductance-distance traces are given in fig. S8.

For AB molen class="Chemical">cules, the peaks in the 1D conductance histogram are smeared out (Fig. 1B), and the conductance-distance traces show a large spread in the conductance plateau positions. A small conductance peak is found at 10−5.5 G0, and a shoulder is found at about 10−4.0 G0. These ill-defined conductance values are consistent with the formation of dendrimeric oligomer structures during the fabrication process (schematics in Fig. 1B and fig. S5), which is discussed in section S2. We assign the low peak at 10−5.5 G0 to the conductance through long dendrimeric structures, such as those depicted in Fig. 1B and fig. S5, whereas the shoulder in the 1D histogram at ~10−4.0 G0 (marked by a dashed line) is attributed to the case when the Au tip is attached to the amine group of the benzene ring directly linked to graphite.

We overcome the problem of ill-defined conductance due to dendrimer formation by grafting DMAB molen class="Chemical">cules, which grow as a submonolayer of monomers only and do not form dendrimers (section S2 and fig. S4). The conductance-displacement trace for DMAB displays a nearly flat and well-defined plateau (Fig. 1C), and the 1D histogram shows a corresponding very sharp peak at 10−4.0 G0. The observation of these features allows us to conclude that there is a rigid molecular orientation and a very stable and well-defined contact with respect to the top graphene layer. Additionally, this stable covalent contact shows a reproducible conductance that persists even when the junction is stretched until breaking of the Au-molecule contact. We note that the conductance of 10−4.0 G0 is consistent with the shoulder that is found at the same value for AB, which confirms the attribution of that shoulder to the conductance through a single benzene ring.

For comparison, we also present the result for the noncovalently attached n class="Chemical">PPD molecules (Fig. 1D). In agreement with reported results (), the 1D conductance histogram shows a very broad structure, with a small and smeared peak at about 10−3.9 G0. Accordingly, a flat molecular plateau in the 2D conductance-distance histogram is absent. There is only a minor deviation from the perfect exponential decay found for the junction without molecule, which indicates that even when a molecular contact is formed, the increase in the molecule-HOPG angle upon stretching leads to an approximately exponential decay of conductance. This result is very different from the one obtained for covalently attached DMAB molecules, which gives a well-defined conductance plateau. No unique conductance value can therefore be assigned to the graphite-molecule noncovalent contact, whereas we show that this problem is overcome for the covalently bonded DMAB molecule.

Conductance calculations

Electron transport results are computed using the Smeagol code (–), which combines DFT as implemented in the Siesta code () with the n class="Chemical">NEGF method for electron transport (DFT + NEGF). Within the NEGF framework, the total current through the system is obtained from the energy (E)– and voltage (V)–dependent transmission coefficient, T(E,V), aswhere e is the magnitude of the electronic charge, G0 = 2e2/h is the quantum of conductance (h is the Planck constant), μB = EF + eV/2 (μT = EF − eV/2) is the chemical potential of the bottom (top) electrode, and f(E) is the Fermi-Dirac distribution. In our simulation setup, the HOPG is the bottom electrode, and Au forms the top electrode. The definition of T(E,V) itself is given in Materials and Methods. Note that the integrand is significantly different from zero only in an energy range of approximately EF − eV/2 to EF + eV/2, which is called the bias window. The molecular conductance, given by I/V, depends critically on the contact geometry between molecule and electrodes. For amine groups attached to Au electrodes, it is well established that the N atom binds preferentially to undercoordinated Au atoms on the surface (). We use a three-atom tip on a flat Au(111) surface as the top contact, which shows a smooth DOS for energies around EF (). On the other hand, the local structure of the covalent bond between DMAB and the top HOPG layer needs to be determined. The main mechanism for chemisorption of the considered molecules on HOPG consists in the transition from an sp2 to an sp3 bond at the absorption site. This transition releases two electrons: the first is involved in the chemical bond with the molecule, whereas the second is an unpaired electron located in a singly occupied orbital, which is extended across the graphite C atoms around this bond. The formation of this singly occupied dangling bond state is also found in the case of the adsorption of a single H atom on graphene. However, this is a high-energy state, and even for low H coverage densities, it has been shown that H atoms bind to graphite in close pairs (), where the electrons in the two dangling bonds pair up to form a low-energy state. This pairing is promoted by the fact that once the first H atom is adsorbed, there is no energy barrier for the adsorption of the second H (). For covalently bonded DMAB, an analogous passivation of the dangling bond is expected, either in the form of two molecules binding in pairs or by adsorption of H atoms from the solvent present when grafting the molecules on HOPG. This is confirmed by the experimental evidence for clustering of molecules discussed above and in the Supplementary Materials. To evaluate the most favored distance between two adsorbed molecules, we calculate the energy as a function of their separation. To this aim, we perform DFT total energy calculations for the two molecules at increasing distances, where at each distance we relax the structure to its lowest energy state. Details of the computational method are given in Materials and Methods, and the exact used binding sites are described in section S4 and fig. S9. A representative structure is shown in Fig. 2A, and all resulting total energies are shown in Fig. 2E. It can be seen that the nearest-neighbor bonding arrangement has significantly lower energy than the other configurations. An analogous result is found when passivating a dangling bond with a single H. We therefore conclude that for most of the adsorbed molecules, the dangling bond around the bonding site is passivated by a second molecule or an H atom at the nearest-neighbor position, and use this structure for the calculation of the conductance.

Fig. 2

Junction geometries and transmission coefficients.

(A) Representative structure used for the evaluation of the dependence of (E) the binding energy to graphene on the distance between two adsorbed DMAB molecules. Transmission coefficient, T, as a function of energy, E, for (F) DMAB and the longer dendrimer, and for (G) PPD at different angles. The used structure for the transmission calculation for DMAB is shown in (B), the one for AB with an aminobenzene unit attached to model a longer dendrimer is shown in (C), and the structure for PPD at an angle of 21° from the HOPG plane is shown in (D) (the PPD structures at different angles are shown in fig. S12). For PPD, the lowest energy is found at an angle of 3°, and the energy increases with increasing pulling angle. Note that the full scattering region used in the calculations includes six layers of graphene for the HOPG electrode and five layers of Au, whereas here only two layers are shown for each electrode for clarity. A detailed explanation of the structural relaxations performed to obtain the atomic structures shown in (B) to (D) is given in Materials and Methods.

(A) Representative structure used for the evaluation of the dependence of (E) the binding energy to graphene on the distance between two adsorbed n class="Chemical">DMAB molecules. Transmission coefficient, T, as a function of energy, E, for (F) DMAB and the longer dendrimer, and for (G) PPD at different angles. The used structure for the transmission calculation for DMAB is shown in (B), the one for AB with an aminobenzene unit attached to model a longer dendrimer is shown in (C), and the structure for PPD at an angle of 21° from the HOPG plane is shown in (D) (the PPD structures at different angles are shown in fig. S12). For PPD, the lowest energy is found at an angle of 3°, and the energy increases with increasing pulling angle. Note that the full scattering region used in the calculations includes six layers of graphene for the HOPG electrode and five layers of Au, whereas here only two layers are shown for each electrode for clarity. A detailed explanation of the structural relaxations performed to obtain the atomic structures shown in (B) to (D) is given in Materials and Methods.

The computed transmission curves for n class="Chemical">DMAB (Fig. 2, B and F) show a strong dependence on the energy of the incoming electrons. This reflects the fact that HOPG has a very small DOS around EF so that most electrons at those energies are reflected back to the Au tip. This is in contrast to the typical situation of the symmetric junction with Au electrodes on both sides of the junction, where the T around EF is usually rather weakly dependent on the energy (). The transmission value in the range of ±0.1 eV around EF is of the order of 10−4, in good agreement with the measurements. The conductance resulting from the integration of the transmission over energy for different bias voltages is compared quantitatively to the experimental data in the next subsection. In section S4, we evaluate the transmission through DMAB for different configurations of the DMAB (figs. S10 and S11). We find that T is only weakly dependent on the angle between the molecule and graphite substrate or between the molecule and the contacting Au atom. There is also no difference whether the dangling bond resulting from the attachment of DMAB is passivated by an H atom or by another molecule (see fig. S10). This confirms the experimentally found robustness of the conductance for covalently bonded DMAB. The conductance for a longer dendrimer, which is evaluated by adding a single aminobenzene unit to the AB (Fig. 2, C and F), is about one order of magnitude lower than the one for DMAB over the whole energy range around EF. This is also in good agreement with the experimental observation and confirms that the high conductance peak in the AB measurements at around 10−4 can be assigned to an Au contact with the N atom attached to the benzene ring closest to the HOPG. In contrast, the conductance peak at about 10−5.5 is assigned to the tip contacting the N atom on the second benzene ring. For longer dendrimers, the conductance is expected to be even lower, moving below the experimental noise limit.

In comparison to the covalently bonded n class="Chemical">DMAB, the transmission of the noncovalently bonded PPD molecule (Fig. 2, D and G, and fig. S12) is strongly dependent on the angle between molecule and substrate, in agreement with published results (), and which is due to the absence of a strong molecule-graphite bond. In this case, the transmission is directly proportional to the overlap between the wave function of the PPD and that of the top HOPG layer. This overlap is maximized when the PPD molecule is parallel to the HOPG surface, whereas it becomes progressively smaller as the angle between PPD and HOPG is increased. Moreover, with increasing angle, the distance between the center of the molecule and the electrodes also increases, which in turn reduces the image charge correction to the energies of the frontier molecular orbitals. The orbitals are thus shifted further away from EF, leading to a larger effective tunnel barrier. Both effects combined result in the decreasing transmission with increasing angle shown in Fig. 2G. Overall, the results of this section confirm the experimental findings. They show that the noncovalently bonded PPD molecule is not suited for device applications that require a well-defined molecular conductance. On the other hand, they show that molecules with a covalent bond such as DMAB provide an excellent device platform for graphene-based molecular electronics stable at room temperature.

It is interesting to note that, despite the very different nature of the junction for covalently attached DMAB molen class="Chemical">cules and noncovalently attached PPD molecules, the transmission values for small angles (for example, 3°) of the PPD molecule and for DMAB are rather close and low. Covalently attached molecules break sp2 hybridization of the topmost HOPG layer, leading to sp3 hybridization. These sp3 nodes have weaker electronic coupling with the surrounding carbon atoms and with underlying graphite layer, which results in a lower junction conductance. Thus, even weak van der Waals interactions of the amine group of PPD with graphite, provided that the angle is small, can lead to a similar conductance as that for covalently attached molecules.

Bias dependence

The pronounced energy dependence of the transmission is expected to lead to significant changes of the conductance as a function of the bias voltage, and this is demonstrated by the experimental results in Fig. 3A. The plotted experimental conductance value at different voltages corresponds to the position of the peak in the 1D conductance histogram measured at that voltage. We note that the voltage is defined in such a way that for a positive bias, the electrons flow from the graphite substrate to the n class="Chemical">Au tip. The very low bias points, below 100 mV for covalent bonding and below 50 mV for noncovalent bonding, are not reliably accessible in the experiment because of the rather noisy current response and instabilities of molecular junctions. The experimental data show a pronounced rectification ratio R, which is defined as the ratio between the conductance for a given positive bias V and the conductance measured by reversing this bias, that is, R = G(V)/G(−V). On the one hand, the PPD junction shows a rectification ratio R ≈ 0.33 at 100 mV so that the conductance is larger for the negative bias. This is in qualitative agreement with previous results obtained for terphenyldiamine (the longer oligomer of PPD) at a higher voltage (), where R was found to be in the range between 0.26 and 0.6. In contrast, covalently bonded DMAB presents a rectification ratio R ≈ 3.7 at 100 mV. It means that for covalently bonded DMAB, the current is larger for the positive than for the negative bias. Additionally, the covalently bonded DMAB molecule shows an increased rectification ratio of 16 at 170 mV. Therefore, the results demonstrate that the nature of the graphite-molecule contact determines the transport characteristics (Fig. 3B).

Fig. 3

Bias dependence.

(A) Bias-dependent conductance of the HOPG/molecule/Au junctions for DMAB and PPD molecules obtained from both experiment (symbols) and theory. The calculations are performed at an electronic temperature of 300 K (solid lines) as well as 30 K (dashed lines). For DMAB, we use the structure shown in Fig. 2B, and for the PPD molecule, we used the configuration at 3°, where the molecule is nearly flat between Au and the graphite substrate, which exhibits the highest transmission. Because of the temperature-induced broadening of the Fermi distribution, the drop in conductance at very low bias is washed out at 300 K, whereas it is visible at 30 K. (B) Illustration of rectification behavior in opposite direction for covalent and noncovalent molecule-graphite attachment.

(A) Bias-dependent conductance of the HOPG/molecule/n class="Chemical">Au junctions for DMAB and PPD molecules obtained from both experiment (symbols) and theory. The calculations are performed at an electronic temperature of 300 K (solid lines) as well as 30 K (dashed lines). For DMAB, we use the structure shown in Fig. 2B, and for the PPD molecule, we used the configuration at 3°, where the molecule is nearly flat between Au and the graphite substrate, which exhibits the highest transmission. Because of the temperature-induced broadening of the Fermi distribution, the drop in conductance at very low bias is washed out at 300 K, whereas it is visible at 30 K. (B) Illustration of rectification behavior in opposite direction for covalent and noncovalent molecule-graphite attachment.

The opposite rectification for PPD and n class="Chemical">DMAB is well captured by our DFT + NEGF calculations (Fig. 3A). For conventional systems, the rectification is often explained in terms of tunnel barrier heights and the asymmetry in the potential drop across the contacts in the junction caused by an asymmetry in the electronic coupling (). Because DMAB and PPD are both very short molecules, the effective tunnel barrier corresponds to the distance in energy of the closest molecular orbital from EF, which, in this case, is the highest occupied molecular orbital (HOMO). For DMAB, we find the HOMO to be about 1.8 eV below EF, and for PPD, it reduces to values between 1.4 and 0.6 eV because of the larger image charge correction caused by the smaller effective molecule/electrode separation (see Materials and Methods and fig. S11 for details). The effective tunnel barrier is proportional to the energy separation of the HOMO from EF and generally changes when a bias voltage is applied, depending on the asymmetry of the potential profile across the molecule. To evaluate the effects of changes on rectification for this system in the effective tunnel barrier, we compare the conductance for a linear drop of the potential (ΔVH,1) to the one for a drop very close to the HOPG (ΔVH,2) or Au (ΔVH,3) surfaces, respectively (Fig. 4, A to C). The effective tunnel barrier is approximately constant as a function of bias for the linear drop, whereas it shifts to higher (lower) values for positive bias when the potential drop occurs entirely close to the Au (HOPG) surface. As shown in Fig. 4C, changes in the asymmetry of the potential drop only lead to minor variations in the bias-dependent conductance for the considered bias voltage for DMAB. We find the analogous behavior also for PPD. These results show that for these systems, the large rectification cannot be explained in the conventional way by the asymmetric potential drop across contacts. We note that whereas these effects can therefore be neglected at the small voltages considered here, for larger voltages, we expect the changes in effective tunnel barriers induced by asymmetric potential drops to be of increasing significance.

Fig. 4

Finite bias transmission and conductance.

(A) Scattering region for the DMAB molecule between HOPG and Au and (B) independent non–self-consistent potential profiles for 0.2 V considered in the calculation of the (C) conductance at finite bias for this structure. We compare the conductance for a linear drop of the potential (ΔVH,1) to the one for a drop very close to the HOPG (ΔVH,2) or Au (ΔVH,3) surfaces, respectively. Transmission at ±0.2 eV for (D) DMAB and (E) PPD, for the geometries used to calculate the bias-dependent conductance in Fig. 3A. The yellow shaded area corresponds to the bias window (EF ± eV/2) for an applied bias voltage with an absolute value of 0.2 V so that the total current at ±0.2 V of bias is approximately proportional to the area underneath the corresponding transmission curves in that energy range (see Eq. 1).

(A) Scattering region for the DMAB molen class="Chemical">cule between HOPG and Au and (B) independent non–self-consistent potential profiles for 0.2 V considered in the calculation of the (C) conductance at finite bias for this structure. We compare the conductance for a linear drop of the potential (ΔVH,1) to the one for a drop very close to the HOPG (ΔVH,2) or Au (ΔVH,3) surfaces, respectively. Transmission at ±0.2 eV for (D) DMAB and (E) PPD, for the geometries used to calculate the bias-dependent conductance in Fig. 3A. The yellow shaded area corresponds to the bias window (EF ± eV/2) for an applied bias voltage with an absolute value of 0.2 V so that the total current at ±0.2 V of bias is approximately proportional to the area underneath the corresponding transmission curves in that energy range (see Eq. 1).

The origin of the rectification in these HOPG-based junctions at low bias is due to a novel effect. To understand its origin, we analyze the bias-dependent transmission for DMAB and n class="Chemical">PPD (Fig. 4, D and E), because the total current is given by the integral of T at each bias voltage over the corresponding bias window (Eq. 1). It can be seen that the changes of T for different voltages largely correspond to a shift in energy, caused mainly by the shift of the HOPG DOS as a function of bias. The area under the transmission curve inside the bias window (±0.1 eV in this case) for DMAB (Fig. 4D) is larger for positive bias than for negative bias, whereas it is larger for negative bias than for positive bias for PPD (Fig. 4E). This causes rectification in opposite directions. These observations lead to the following conclusion: The low-bias rectification found is mainly driven by the nonconstant, highly dispersive DOS of graphite around EF, combined with the detailed nature of the molecule/graphite contact (covalent or noncovalent), which drives the slope of T when one moves away from EF. In the energy range around EF, which dominates the low-bias transport, T is larger below EF than above EF for PPD, whereas T is smaller below EF than above EF for DMAB. We note that for electrodes with flat DOS around EF, such as Au, this effect is absent (). We also remark that measurements at room temperature cannot resolve the very low transmission dip within ±0.05 eV around EF due to the thermal broadening of the Fermi distribution. In low-temperature experiments, this sharp decay of conductance is expected to be accessible, as indicated by our conductance calculations for a Fermi distribution at a temperature of 30 K (Fig. 3A).

CONCLUSION

We explore the anchoring group chemistry effects on the conductance properties of graphite-molen class="Chemical">cule contacts by probing them with gold STM tips. Our main finding is that the controlled design of covalent attachment of molecules to the highly ordered graphite surface via direct C–C binding results in a well-defined narrow distribution of conductance values and unprecedented contact stability at room temperature. The asymmetric hybrid molecular junctions formed with graphite and gold electrodes show a strong bias-dependent conductance with an opposite rectification behavior for covalent (rectification ≈ 3.7 at 100 mV and 16 at 170 mV) and noncovalent (rectification ratio ≈ 0.33 at 100 mV) cases. We demonstrate that this rectification is due to a novel effect caused by the nonconstant, highly dispersive DOS of HOPG around EF, combined with the detailed nature of the molecule/graphite contact. These results reveal a new way to tune the device functionality via interfacial anchoring chemistry. This work reports a key finding for the development of reproducible single-molecule devices operating under ambient condition and therefore overcomes the main limitations of state-of-the-art molecular electronic technology that uses coinage metal electrodes, leading the way to next-generation molecular electronic devices.

MATERIALS AND METHODS

Materials

For the covalent modification of an HOPG surface, we used electrochemical grafting of diazonium salt derivatives: n class="Chemical">4-nitrobenzenediazonium tetrafluoroborate (Sigma-Aldrich, >97%) and 3,5-dimethyl-4-nitrobenzenediazonium tetrafluoroborate, synthesized by us (see section S1 for synthesis procedure and characterization). For the noncovalent case, we used PPD molecules (Sigma-Aldrich, >99%) in 1,2,4-trichlorobenzene solvent (Sigma-Aldrich, anhydrous, >99.0%). The electrolyte solutions for grafting and for the electroreduction of nitro group to amine group were prepared from the following: (i) tetrabutylammonium hexafluorophosphate (TBA-PF6; Sigma-Aldrich, ≥99.0%) in acetonitrile (AcN; Fisher Chemical, HPLC grade) and (ii) potassium chloride (Alfa Aesar, 99.995%) in the mixture of water (Milli-Q, 18.2 megohm⋅cm) and ethanol (EtOH, absolute, p.a.). To have a clean electrode surface, the top layers of the HOPG electrode were peeled out with a Scotch tape before the experiments.

Electrochemistry, STM, and Raman measurements

For all electrochemical, STM imaging, and STM-BJ experiments, we used a liquid cell: A Kel-F part was mounted on top of a HOPG substrate via a Kalrez O-ring. Platinum and silver wires served as counter and reference electrodes, respectively, for electrochemical experiments, which were cn class="Chemical">arried out with an Autolab PGSTAT30. High-purity Ar (Carbagas, 5N) was passed above the solutions during the electrochemical experiments.

The STM measurements (imaging and BJ) were carried out with a PicoSPM system (Molen class="Chemical">cular Imaging) in a sealed, Ar-filled chamber. To remove air and water vapor, Ar was passed through the chamber for 20 min before the STM measurements. STM tips were prepared by electrochemical etching of an Au wire (0.25 mm diameter). For image processing, the WSxM software was used ().

The following protocol was applied for the STM-BJ experiments: The tip was brought to a preset tunneling position typically defined by a setpoint current ISP = 50 to 100 pA and a n class="Disease">bias voltage Vbias = 0.10 V, followed by imaging the substrate. Current-distance measurements were performed at a fixed lateral position of the Au tip, with STM feedback switched off. The vertical movement of the tip was controlled by a ramp unit; the details of the instrument are described in our previous publications (, ). The measuring cycle was performed as follows. The controlling software drove the tip toward the HOPG substrate, and the approach was stopped when a predefined upper conductance limit was reached, typically 10−2 G0 (G0 is the conductance quantum equal to 77.5 μS). The upper conductance limit of 10−2 G0 ensured a rather soft physical contact between the Au tip and HOPG. After a short delay (~100 ms), the tip was retracted by 2 to 5 nm until a low-current limit of 1 pA was reached. Depending on the experimental conditions, a stretching rate of 14.5 or 58 nm/s was used. The entire current-distance traces were recorded with a digital oscilloscope (Yokogawa DL750, 16-bit, 1-MHz sampling frequency) in blocks of 186 individual traces. We have collected typically 1000 to 2000 conductance-distance curves at different spots on the sample at each experimental condition to guarantee the statistical significance of the results (5 to 11 spots on sample; at each spot, typically a set of 186 curves were recorded). The 1D and 2D conductance histograms were constructed from withdrawing individual conductance traces without any selection. Gaussian fits to the conductance peaks in the 1D histograms give the most probable conductance value, and the SD of the peak was used as an error bar for the conductance value. The presence of a single Gaussian-type molecular peak in the resulting 1D conductance histograms and of a single molecular plateau in the 2D histograms (constructed from several hundreds of individual curves without any data selection) indicates only one type of measured events. That is, there is only one statistically dominant junction composition at the moment preceding the breaking of the electrode-molecule(s) contact. Because the simultaneous breaking of two or more molecular junctions is much less probable than the breaking of just a single-molecule junction, it can be excluded. These BJ experiments therefore measure the conductance of individual molecules [for further details, we refer to the study of Moreno-García et al. ()]. Typically, more than two samples were tested, and all samples yielded similar distribution of conductance.

Ex situ Raman measurements were performed using a LabRAM HR800 confocal microscope (HORIBA Jobin Yvon). We used a long working distance objective lens (×50 magnification, 8-mm focal length) with a numerical aperture of 0.1 to fon class="Chemical">cus a diode-pumped solid-state laser beam (excitation wavelength, 532 nm; power, 5 mW) on the sample. The Raman signal was collected in backscattering geometry. The normal Raman spectra were recorded under ambient conditions. We collected 8 to 10 spectra at different areas on bare and modified HOPG electrodes and averaged to obtain spectra with high signal-to-noise ratio. The spectra were quantitatively reproducible.

Covalent modification of graphite surface by target molecules and adlayer characterization

Covalent attachment of the molecules to the HOPG electrode surface was performed via electrochemical reduction of n class="Chemical">4-nitrobenzenediazonium or 3,5-dimethyl-4- nitrobenzenediazonium salts, followed by the electrochemical reduction of nitro group to amine (see details in section S2) (). This protocol led to the final organic submonolayers of AB and DMAB. The covalent modification of HOPG was confirmed by STM and Raman spectroscopy (see details in fig. S4). As known, the grafting through diazonium chemistry often leads to the oligomeric growth of molecules () because an active intermediate, aryl radical, can attack the meta positions (respective diazonium group) of already attached molecules. Our STM results explicitly show the formation of dendrimeric molecular structures in case of AB and only monomers for DMAB (fig. S4).

First-principles calculations

Within the DFT + NEGF framework, the system is subdivided in a semi-infinite bottom electrode and a semi-infinite top electrode, and a so-called scattering region between the two (). The n class="Chemical">current is calculated with Eq. 1, where the energy- and voltage-dependent transmission coefficient, T(E,V), is integrated over energy. This is given by T(E, V) = Tr[ΓBG†ΓTG], where G is the retarded Green’s function of the scattering region, given by G = [(E + i δ)S − H − ΣB − ΣT]− 1, with δ a vanishingly small positive number. S and H are the overlap and Hamiltonian matrices of the scattering region, respectively, and ΣB (ΣT) is the self-energy matrix for the bottom (top) electrode and is computed with the algorithm described by Rungger et al. (). The so-called coupling matrices ΓB and ΓT are given by and . For the HOPG/molecule/Au transport calculations, a rectangular cell was used in the plane, extending over 8 (16) C–C bonds along the zigzag (armchair) HOPG edge, including 64 C atoms per layer. For the Au(111) layers, the in-plane unit cell vectors correspond to the vectors (3 aAu)[1-10] and (aAu)[11-2] in terms of face-centered cubic Au lattice vectors (aAu is the lattice constant) and contain 24 Au atoms per layer. We fixed the C–C in-plane distance to the experimental value of 1.42 Å, resulting in an area of the unit cell of 9.84 × 17.04 Å2. For the Au electrode, the use of this cell leads to an Au-Au distance of 2.84 Å, which is slightly strained by 1.5% compared to the experimental value of 2.88 Å. For the vertical stack, we used six A-B stacked layers of graphene for HOPG and five A-B-C stacked Au layers along the [111] direction. We used a three-atom tip added to the flat Au surface to model the Au contact. For the considered supercell, we found that the electronic structure was converged for a k-point mesh of 8 × 4 k-points. The transmission and current were then calculated for a given converged charge density with a higher density of k-points of 32 × 16, and an additional finer k-grid was added around the K-point in the Brillouin zone to resolve the small DOS for energies around the EF. We used a double-ζ plus polarization basis set for all atoms of the molecule, a double-ζ basis for the C atoms in HOPG, and a double-ζ plus polarization basis for the Au s orbital, whereas for the Au d orbital we used a single-ζ basis. For a number of test systems, we verified that using a double-ζ basis for the HOPG gives results in good agreement with those obtained, including additional polarization orbitals. The real-space mesh was set by an equivalent energy cutoff of 400 Ry, and we used the local density approximation (LDA) for the exchange-correlation functional. To describe appropriately the energy level alignment between molecular orbitals and the metal electrodes, one needs accurate work functions for the electrodes and correct energies for the HOMO and lowest unoccupied molecular orbital (LUMO). The LDA-calculated work functions of 5.2 eV for Au and of 4.65 eV for HOPG are in good agreement with experimental values. On the other hand, for the molecular orbitals, the LDA typically underestimates the HOMO-LUMO gap, and moreover, it does not capture the renormalization of the gap of molecules close to metal surfaces (, –). We therefore added a scissor operator correction for these states to evaluate the transmission, which had been shown to lead to good agreement with experimentally measured conductances (, , ). Whereas this correction leads to an increase of about 2 eV of the LDA gap for molecules oriented vertically with respect to the electrodes’ surfaces, the correction is negligible for the molecule flat between HOPG and Au, which agrees with results obtained for similar systems (details of the applied scissor operator are given in the Supplementary Materials) (). Because at the low bias voltages considered here the molecule conducts in the tunneling regime and does not change its charging state, we performed the out-of-equilibrium calculations for the evaluation of the current at finite bias non–self-consistently, where we added a potential ramp between the Au tip and HOPG to the equilibrium potential (, ). As shown in Fig. 4, the detailed shape of the applied potential drop had only minor effects on the resulting current.

The electron transport calculations were performed for interface atomic structures on class="Chemical">ptimized using the Fritz Haber Institute ab initio molecular simulations (FHI-aims) DFT all-electron code (). The standard numerical atom-centered orbitals basis sets “tier 1” and “tier 2” were considered for Au and H, C, and N, respectively. To include van der Waals (vdW) interactions (), we used the exchange-correlation density functional by Tkatchenko and Scheffler (DFT + vdW) (). This returned a graphite interlayer distance equal to 3.345 Å, which agreed remarkably well with the experimental value of 3.35 Å (). For the structural relaxations, we considered a supercell containing three layers of HOPG and an Au tip of four layers, and all forces were relaxed to below 0.01 eV/Å. We fixed the bottom HOPG and the top Au layers, and a k-point mesh of 8 × 4 k-points was used. In the electron transport simulations, we extended the DFT supercell to six HOPG layers and five Au layers on each side of the molecule, as described above. For total energy calculations of Fig. 2 (A and E), we used the same basis set and a k-grid of 8 × 8 k-points to ensure that energy differences were converged to below 0.1 meV. The positions of all atoms in the supercell were optimized until forces were smaller than 0.01 eV/Å, and all calculations were spin-polarized.