The molecular precursor DB-DPBA (1) consists of a bianthracene backbone with two pyridyl groups attached at opposite sides and two Br atoms at opposite edges. Precursor (1) was synthesized in-solution, as shown in the top row of Fig. 1. Bistriflate (2) was initially prepared according to previously reported procedures44 and subjected to the Miyaura-Ishiyama borylation to provide diboronic ester (3) in 75% yield. Then, the Suzuki-Miyaura coupling of (3) and 2-bromopyridine afforded 2,2’-di(pyridin-2-yl)-9,9’- bianthracene (4) in 66% yield. Finally, DB-DPBA 1 was obtained by the bromination of (4) using N-bromosuccinimide (NBS) at room temperature in 90% yield and characterized by 1H and 13C NMR spectroscopy as well as high-resolution mass spectrometry (see SI 1–6 in the Supporting Information).
Subsequently, precursor 1 was sublimation-deposited on a clean Au(111) substrate in ultra-high vacuum (UHV) to perform on-surface reactions. A first annealing step at around 180°C triggers the Ullmann-type coupling between DB-DPBA precursors to form non-planar polymers. A second annealing step at 300°C induces complete planarization to form the targeted Py-7-AGNR. Due to the asymmetry in the lateral pyridyl units and their possible rotations during the annealing process, the positions of the N atoms along the ribbon are not predetermined. In particular, the pyridyl group can cyclize under formation of either a C-N or a C-C bond, leading to a graphitic-N or a pyridinic-N (highlighted in blue and green, respectively, in Scheme 1). To understand if one of the two possible N configurations is favored, we deposited a sub-monolayer coverage of DB-DPBA on an Au(111) substrate kept at room temperature. Recently, Mugarza et al. synthesized a closely related pyridyl-extended precursors with N atoms in 3 or 4 position; however, they reported that sublimation of these precursors were impossible since they already react and polymerize in the crucible45. We did not face this problem with DB-DPBA 1, indicating that the position of the substituted atoms can greatly affect the chemical reactivity of this kind of precursors.
A large-scale STM topography image of the surface after DB-DPBA deposition is shown in Fig. 2a. The molecules tend to self-assemble into one-dimensional (1-D) chains at the fcc regions of the herringbone reconstruction of Au(111). Locally, it is possible to find some large islands of close-packed units. The inset of Fig. 2a shows a zoom-in image of a self-assembled chain of molecules. In STM images, the precursors are characterized by two bright, slightly asymmetric features separated by roughly 0.75 nm and an apparent height of 1.9 nm. To elucidate the molecular structure on Au(111), we conduct DFT simulations. Figure 2b shows the relaxed geometry of DB-DPBA on Au(111). The precursor adsorbs with the pyridyl lateral groups parallel to the substrate, while the bianthracene backbone is tilted away from the surface, with a maximum height of roughly 0.67 nm. The topmost parts of the two anthracene units have a distance of 0.70 nm, comparable to our experimental values. The corresponding simulated STM image of a DB-DPBA unit (Fig. 2c) matches well with our topography images, where two bright dots are resolved. This adsorption geometry is similar to the one reported for the pristine version (i.e. without N atoms) of the precursor11, 46.
After deposition, it is not possible to distinguish the position of the N atoms by STM, since all the possible geometries look identical. This is confirmed by DFT-based STM simulation of different pyridyl configurations (see figure SI 7 and Table 1 in the Supporting Information). The only difference is in terms of adsorption energy, where the structure with the N atoms of the pyridyl pointing outside is favored (-0.11 eV). DFT simulations allow us to estimate the energy barrier needed to flip the pyridyl ring between the two configurations as 0.45 eV at 95°, due to the steric hindrance between the H atoms of the two rings (Figure SI 8 in the Supporting Info). For larger rotation angles, the energy decreases. Rotation of the pyridyl ring, hence, sees an energy barrier of 0.45 eV and will thus be accessible already at moderate annealing temperatures.
Annealing the sample to 180°C triggers the formation of polymeric units on the surface, as shown in Fig. 3a. This step results in the growth of polymers (with an average length of 40 nm) which are characterized by bright dots alternating along the main axis with a periodicity of 0.85 nm, as expected from covalently bonded and alternatingly tilted anthracene units. Upon a further annealing step at slightly higher temperature (250°C) most of the units remain non-planar. However, as shown in the inset of Fig. 3b, some segments have already reacted. This partial planarization process starts at one GNR end and then proceeds along one edge, until a defect (e.g. a bend) is encountered. The DFT-optimized model and the corresponding STM simulation of a partially closed single unit (see Figure SI 9 in the Supporting Information) matches well with our STM images, where only one brighter dot related to the unreacted edge is seen. Some short units (mainly monomers or dimers) are already completely reacted and planarized.
The final annealing step at a temperature of 300°C forms completely planar Py-7-AGNRs, as shown in Fig. 3c. These growth conditions can give rise to long (up to 80 nm) and straight GNRs following the elbow sites of the Au(111) reconstruction. Similarly long ribbons have been reported for the pristine version of the same precursor11, 47; in contrast, bipyrimidine substituted analogues (i.e. two N atoms in each ring) have been reported to grow shorter due to the stronger interaction with the substrate which hinders their diffusion on the surface45. Small scale images of the Py-7-AGNR (inset in Fig. 3c), reveal the expected periodic pattern related to the pyridine edge extensions.
In our N-substituted case, complete planarization is reached already at 300°C instead of 400°C as in the nitrogen-free 7-AGNR counterpart11, 47. This lower cyclization temperature is attributed to the presence of N atoms in the pyridyl rings, which is in line with observations in other N-substituted small nanographenes42, 43, 45.
To better understand if N has a role in the planarization/cyclization steps, we investigated the structure of the synthesized Py-7-AGNR in detail. Figure 4a shows an STM image of a ribbon segment acquired at negative bias (occupied states). Most of the edge extensions seem asymmetric with an apparent "up-bend" on the left and "down-bend" on the right edge with the exception of two units in the bottom right part (highlighted by green arrows) which appear more symmetric and rounded. Measurements taken at positive bias (unoccupied states) (Fig. 4b), instead, show alternating bright dots present along the backbone. The distance between two consecutive brighter units is 1.28 nm with an angle of 63° with respect to the ribbon axis. As for the negative bias image (Fig. 4a), almost all the edges present these features with the exception of the same two highlighted by green arrows.
To understand the reason of this difference, we performed nc-AFM measurements with a CO molecule attached to the tip48. In this way, it is possible to achieve single bond resolution, such as in Fig. 4c. Here, the periodic pattern related to the pyridine edge extensions is clearly visible. In addition, the majority of the pyridine extensions reveal bonds that are imaged darker than the ribbon backbone. We attribute these darker bonds to graphitic-N-C bonds, and the two brighter ones (highlighted again by green arrows) to pyridinic-N-C. Figure 4d shows a high-resolution nc-AFM image of a short segment with one unit (blue rectangle) with edges in graphitic-N configuration and the adjacent one (green rectangle) with pyridinic-N. Nc-AFM simulations done on the two possible structures (Fig. 4e) clearly confirm our assignment. As expected, graphitic-N edges appear dark, while pyridinic-N does not have any evident contrast difference, just a slight distortion at the N position. The darker appearance of graphitic-N in nc-AFM measurements, already reported for other structures24, 25, 49, 50, derives from its different short-range repulsion distance compared to C atoms and the lower adsorption height from the Au substrate (as proven by DFT). In addition, STM simulations of Py-7-AGNR containing graphitic-N and pyridinic-N (Figure SI 10) reproduce the experimental findings where the extension assigned to graphitic-N appears slightly tilted at negative bias while having an apparent bright dot when scanned at around 0.8 eV. Pyridinic-N, on the other hand, appears symmetric and rounded.
We made statistics on the type of N bonding using nc-AFM images (see also SI 11 in the Supporting Info). From 306 edge extensions that we analyzed, 9.2% were exhibiting a pyridinic-N configuration, while all others were graphitic-N. Considering the design of the DB-DPBA precursor and the required on-surface synthesis steps implies that cyclization proceeds via preferential C-N bond formation (blue in Scheme 1). Moreover, this ring closure takes place at a lower temperature than the analogous C-C one for the pristine version. For a deeper understanding of this preferred cyclodehydrogenation path, we applied a DFT-based constrained optimization approach to determine the energy barriers for closing the C-N and C-C bonds. As model systems, we used two relaxed geometries of monomers which, upon cyclodehydrogenation, should form a graphitic-N or pyridinic-N bond (left panels in Fig. 5b). The energy difference of the starting configurations for the C-N versus C-C path is due to the different orientation of the pyridyl extension. We used the distance between the two atoms forming a bond as collective variable (atoms connected by dotted lines in the inset of Fig. 5a and indicated with red arrows in Fig. 5b). While slowly changing the constraint value of the collective variable (at increments of 0.05Å) we relaxed within DFT the geometry (all degrees of freedom except the constraint) thus identifying a transition state (activation energy). Figure 5b shows some configurations of the molecules at each steps.
Figure 5a shows the resulting energy profiles of the two bond formation pathways. Total energies increase as the distance between atoms is reduced, until at 2.6 Å the C-N formation becomes energetically favored. Both energies then reach a maximum at 1.9 Å, but C-N bond formation sees a lower barrier than C-C bond formation. The energy difference of ≈ 0.1 eV comes from the absence of the H atom bonded to the N, which decreases the steric hindrance and the required energy to form the bond. Upon further approach of the atoms, the bonds between the bianthracene units start to form and the energy drops substantially, with the graphitic-N case being lower in total energy. The actual mechanism of C-N bond formation has been previously discussed43. Applying a Boltzmann statistics to the barrier height difference of 0.11 eV, the probability to obtain a C-C, i.e. pyridinic-N, is 11%; which is in good agreement with the experimental value of 9.2%. It is important to note that the pyridyl orientation giving rise to the C-N bond is less favorable than the one for the C-C bond after deposition on Au(111) surface. For this reason, the majority of the pyridyl rings will initially have the wrong orientation and they will need to rotate in order to reach the preferred configuration for the C-N bond formation. As shown before, this rotation is indeed easily accessible during the different annealing steps (see Figure SI 8 in the Supporting Information).
The effect of N atoms on the electronic structures is investigated using DFT and compared to experimental results. DFT-based simulation (Figure SI 12 a and b in the Supporting Information) of a Py-7-AGNR segment containing graphitic-N shows the reorganization of the energy levels upon adsorption on the metal substrate. In particular, VB-1 and VB, localized at negative energies in gas phase, shift to positive energies once the Au(111) substrate is considered. This indicates a positive charging of the Py-7-AGNR on the metal substrate and the loss of electron. Figure 6 and Figure SI 13 show DFT band structures in gas phase of infinitively long Py-7-AGNR with the possible N configurations compared to the pristine one (partial density of states (PDOS) of the frontier bands are reported in SI 13). The pyridinic-N and the pristine GNRs share similar properties: the band gap in both cases is close to 0.8 eV. The effect of the pyridinic-N is the expected rigid downshift of valence band (VB) and conduction band (CB) by roughly 50meV. The most significant contribution of pyridinic-N is a flat band at -2.1 eV (highlighted in yellow) which is mainly localized at the N positions. In contrast, graphitic-N has a stronger impact on the electronic properties. First, graphitic-N was considered as charge-neutral (Figure SI 13c). Because of the extra electrons shared from the N atoms (one from each), the system starts to fill energy levels which are empty in the pristine case. In particular, the electrons populate the CB of the unsubstituted case, which now becomes the VB. This new ordering of the bands will shift Ef and it will give a small bandgap of 50meV. However, as shown at the beginning of the paragraph, the Py-7-AGNR containing graphitic-N is expected to lose the electrons upon adsorption on Au(111); for this reason graphitic-N was also modelled removing two electrons (one electron from each N). This allows to have a better comparison with experimental results. Similar charging behavior due to graphitic-N have been reported for other compounds upon adsorption of a metal substrate 49. In this way, the same ordering of the frontier bands with respect to the pristine case is retrieved: VB2+, localized on the 7-AGNR sections (Figure SI 13c), and CB2+, which partially spreads over the pyridine extensions close to the N atoms, have the same charge distribution than the VB and CB, respectively, of the undoped case. The band gap is slightly reduced to 0.7 eV. The CB + 12+, instead, is only localized close to the N atoms in the pyridine rings, and there is no analogous band for the pristine case.
Experimentally, we characterized the electronic properties of the Py-7-AGNR by means of STS. An example is reported in Fig. 7a. By analyzing the corresponding nc-AFM image, most of the edges present graphitic-N, except two on the right side (green arrow). The dI/dV spectra recorded at different locations are shown in Fig. 7b. The most prominent feature is an intense peak at 0.9V, which is predominantly seen close to the N atoms of the pyridine extension (red and green spectra). Importantly, this contribution is only present at the graphitic-N edges. This is confirmed by taking a dI/dV spectrum close to a pyridinic-N (blue spectrum in Fig. 7b) which does not have any peak in this range, but is similar to the pristine GNR47. This agrees well with the gas-phase simulation shown in Fig. 6c, where the pyridinic-N states are localized far from the Fermi level (Ef). Similar considerations can be made for the peak at 1.5V, which is more intense only at the graphitic-N edges. These differences between the two distinct arrangements of N atoms point out that the electronic properties are affected by the specific position of the introduced heteroatom. There are two additional peaks close to Ef at -0.1 V and 0.05 V which we assign to the onset of the VB and CB and a related band gap of 0.15 eV for the Py-7-AGNR adsorbed on Au(111). To confirm the band assignment, we took dI/dV maps at selected bias voltages in order to map the distribution of these states in the ribbon, as shown in Fig. 7d, and compared them to simulated LDOS of a Py-7-AGNR with two electrons per unit cell removed (Fig. 7e). The corresponding Py-7-AGNR is reported in panel c, where the nc-AFM image shows that the edges are all graphitic-N. The dI/dV map at -0.1 V reveals intensity mainly on the 7-AGNR backbone, while the one at 0.1 V also has some contribution on at the pyridine edge extensions. There is good agreement with the simulated LDOS maps of VB2+ and CB2+ (Fig. 7e). This confirms the low band gap value of 0.15 eV, lower than the pristine GNR counterpart of 1.0 eV47. The dI/dV map recorded at 0.9 V shows bright dots localized at the pyridine extensions closer to the N atoms. This behavior is well reproduced by the CB + 12+, which is able to reproduce the inhomogeneous distribution at the edges. Finally, the state at 1.4 V is mainly localized at the pyridine extensions with a more symmetric distribution, in line with the simulation of CB + 22+.