Atomically controlled substitutional boron-doping of graphene nanoribbons

Boron is a unique element in terms of electron deficiency and Lewis acidity. Incorporation of boron atoms into an aromatic carbon framework offers a wide variety of functionality. However, the intrinsic instability of organoboron compounds against moisture and oxygen has delayed the development. Here, we present boron-doped graphene nanoribbons (B-GNRs) of widths of N=7, 14 and 21 by on-surface chemical reactions with an employed organoboron precursor. The location of the boron dopant is well defined in the centre of the B-GNR, corresponding to 4.8 atom%, as programmed. The chemical reactivity of B-GNRs is probed by the adsorption of nitric oxide (NO), which is most effectively trapped by the boron sites, demonstrating the Lewis acid character. Structural properties and the chemical nature of the NO-reacted B-GNR are determined by a combination of scanning tunnelling microscopy, high-resolution atomic force microscopy with a CO tip, and density functional and classical computations.

B oron is a unique element in terms of electron deficiency and Lewis acidity, yet it has a comparable size to carbon. Incorporation of boron atoms into an aromatic carbon framework like carbon nanotubes 1 and graphene 2 offers a wide variety of functionality in, for example, chemical sensing 3,4 , nanoelectronics 5,6 , photocatalysis 7 and battery electrodes 8 . So far, the syntheses have been demonstrated by both the chemical vapour deposition method with co-or mono-depositing of small molecules, like B 2 H 6 , BCl 3 , and PhB(OH) 2 (refs 7-11), and by the reactive microwave plasma method with B(CH 3 ) 3 (ref. 12). However, besides the formation of Stone-Wales defects, 10 the coordination environment of the boron sites in the graphitic carbons, namely the doping site and density of boron, cannot be precisely controlled in these methods. This limitation can be overcome by a recently developed on-surface chemical reaction 13,14 . In fact, nitrogen atoms have been doped into the edges of graphene nanoribbons (GNRs) by employing nitrogen-doped precursor molecules 15,16 . In contrast, the intrinsic instability of organoboron compounds against moisture and oxygen has delayed the development of boron-doped nanocarbon chemistry 17,18 and has led to the unavailability of boron-doped precursor molecules. However, our recent chemical synthesis of polycyclic aromatic hydrocarbons with boron atoms proved the stability of this materials class 19 . The use of such an organoboron precursor can pave the way for boron doping into GNRs. We now present a well-structured boron-doped-GNR (B-GNR) in terms of valency, position and doping ratio and study its chemical structure and property by high-resolution atomic force microscopy (AFM) and scanning tunneling microscopy (STM).

Results
Synthesis of boron-doped graphene nanoribbons. Figure 1a shows the chemical structure of the precursor molecule, in which two pristine anthracene units and one 9,10-dibora-9,10dihydroanthracene moiety are covalently bonded while two bromine atoms are located at both termini. By annealing the molecules on a Au surface at 180°C, the bromine atoms are removed by surface catalytic dehalogenation. Consequently, the precursor molecules are linearly polymerized. At an elevated temperature of up to 400°C, the surface-assisted cyclodehydrogenation results in the formation of an aromatic graphene nanoribbon 20 . Employing this reaction, boron atoms can be doped as BC 3 Lewis acidic sites at the centre row in the armchair-edge GNR (B-GNR) with a width of seven carbon atoms (N ¼ 7). The interval between the boron sites is fixed with a doping density of 4.8 atom%.
In experiments, the precursor molecules were deposited on a clean Au(111) surface in ultra-high vacuum and were subsequently annealed at 180°C (Fig. 1b). At 4.8 K the molecules mainly stay along the valley of the herringbone structure on Au(111). We performed tip-induced lateral manipulation to check the presence of the polymer chain. After scanning the tip at a constant height along the line indicated with an arrow in Fig. 1b, several molecular units moved together, indicating successful polymerization (Fig. 1c). The successive anthracene units have opposite tilts to the Au substrate owing to the repulsion between hydrogen atoms in the adjacent anthracene moieties. After annealing at 400°C, such corrugated structures disappeared. Instead, stripe-shaped flat nanoribbons appeared (Fig. 1d).
The observed STM contrast is inverted by the polarity of the applied tip bias voltage ( À 3 and 3 V), indicating the perturbation of the electronic structure by the doped boron atoms as an electron-accepting unit (Fig. 1e,f, Supplementary Fig. 1). Pink crosses indicate the same atomic site in B-GNR. By setting close to zero bias voltage ( À 2 mV), the electronic states near the Fermi level of B-GNR are resolved (Fig. 1g). We also observed two different contrasts at the armchair type edge, where the ratio of one dark-and two bright-contrasts suggests that the darkcontrast edge relates to the boron-doped anthracene. The hydrogen atoms at the armchair edge can be desorbed by the tunnelling current at a large negative tip bias ( À 3.5 V and 100 pA, Supplementary Fig. 2) as indicated with an arrow. The simulated STM topographies (right panel in Fig. 1e-g) are in good agreement with the experimental results (left panel), so that the chemical structure can be unambiguously assigned as shown in Fig. 1h.
Electronic structure of boron-doped graphene nanoribbons. To measure the electronic structure of N ¼ 7 B-GNR, dI/dV spectra were taken with a clean Au tip above four different sites of B-GNR and, for reference, above one Au(111) site (Fig. 2a). The conductance is related to the local density of states at the tip position. At the Au(111) site, the effect of Shockley surface states locates around À 0.5 eV below the Fermi energy. The contribution, originating from the surface state, is even visible on B-GNR as modulations of the dI/dV spectrum down to À 1.0 eV, and rather stronger than on the non-doped GNR [21][22][23] . This may relate to the stronger coupling to the substrate due to the presence of boron atoms. More interestingly, in contrast to the pristine GNR 22 , the presence of these boron atoms downshifts the interface state. Nevertheless, the valence band edge locates around À 0.8 eV below the Fermi energy, with clear peaks at 1.6 eV corresponding to the conduction band edge. Consequently, the band gap of the B-GNR is 2.4 eV, which is comparable to the non-doped GNR on Au(111) (ref. 22). Previous theoretical studies indicate that the position of doped boron strongly affects the modulation of the electronic structure and so that the minimum (maximum) effect is expected with a doping at the centre (edge) 5 . In this system, boron atoms locate exactly at the centre of N ¼ 7 B-GNR and, thus, no significant electron-hole asymmetry is present (full suppression of backscattering), leaving intact the overall electronic structure (this is seen also in our calculated electronic structure, Supplementary  Fig. 3). Yet, in a detailed inspection, the presence of boron atom in the GNR can be observed. The inversions of the conductance between the doping and non-doping sites occur twice between 1.4 and 2.0 eV and between À 1.0 and À 0.4 eV both at the edge and centre. This is consistent with the scanning tunnelling spectroscopy (STS) images taken within the energy levels at constant heights (Fig. 2b), and also the location of the boron features in the calculated projected density of states.
Fused boron-doped graphene nanoribbons. After the substrate was further annealed at 510°C, B-GNRs with different widths (N ¼ 7, 14 and 21) were obtained by fusing the armchair edges ( Fig. 3a). To check the chemical structure in detail, we used an AFM with a CO-functionalized tip 24 . In this technique, when the tip-sample separation is small enough, the tip detects the short-range Pauli repulsive interaction 25 and the bond structure of molecules can be resolved via the frequency shift of the oscillating force sensor 24,[26][27][28] . Figure 3b shows the AFM image of N ¼ 7 B-GNR obtained in a constant height mode. The  Δf (Hz) -1.57 -8.24 Δf (Hz) ARTICLE periodic aromatic carbons are clearly revealed, in which every three anthracene units have a defect-like feature, with darker contrast (more negative frequency shift). To enhance the observed bond visually, a Laplace filter 27 was applied in the region around the darker contrast site (Fig. 3c). The C-C bonds in the anthracene unit are clearly observed and, even at the centre, the B-C bonds were observed.
Although the AFM imaging mechanism with a CO tip is somewhat sensitive to the total density of charge 25 , the nature of the contrast cannot fully be explained by the inherent electron-deficient nature of boron and its electron-accepting character. Thus, we consider the topographic geometry in detail. Although the van der Waals radius of the B atom (192 pm) is larger than the C atom (170 pm), the actual topographic shift of the boron atom core in B-GNR on Au obtained by our density functional theory (DFT) calculation is a 30 pm dip ( Supplementary Fig. 4), which clearly plays an important role in the more negative frequency shift observed and predicted in simulations (see also calculated the AFM image in Fig. 3d and Supplementary Fig. 5). In the absence of this topographic distortion, the difference in local charge density and resulting electrostatic interactions is not enough to produce a clear contrast change. The B-C bond (longer than the C-C bond) also deforms the honeycomb structure 19 , while the distortion of the carbon ring in the AFM image relates to the effect of tilting CO tip 27 .
On the basis of these observations, we state that the on-surface chemical reaction was accomplished as demonstrated in Fig. 1a. Some defects, presumably relating to the on-surface dissociation of B-C bonds in the first annealing process, are also observed ( Supplementary Figs 6 and 7). We further analysed B-GNRs with wider widths (Supplementary Fig. 8). For N ¼ 14 B-GNRs, two different fused configurations with the out-of-phase (Fig. 3e) and the shifted phase (Fig. 3f) were observed. Since the configuration is determined by the relative position in the fusing process, the shifted phase N ¼ 14 B-GNR has a mirror configuration, thus, in total three configurations. In such a process, the C-C link seems to be caused by a zipping up process through dissociating hydrogen atoms at the armchair edge ( Supplementary Fig. 9). For N ¼ 21 B-GNRs, nine different textures are theoretically expected. Among them, the perfectly ordered out-of-phase structure is shown in Fig. 3g. The boron atoms can stay at the designated atomic site throughout the annealing process at the higher temperature of 510°C. Such high stability may reflect the robustness of the BC 3 site 29 . Therefore, once a boron atom is perfectly incorporated in graphene, the thermal stability is remarkably high (Supplementary Fig. 10).
Lewis acidity of the boron site. Since the structural and electronic characters of the BC 3 sites were demonstrated, we next evaluate the Lewis acidity of the boron sites by exposing the B-GNRs to nitric oxide (NO) gas. Figure 4a shows a large-scale STM image, where small dots correspond to dosed NO molecules. It is well-known that the NO molecules adsorb at the elbow site of the Au(111) herringbone structure (blue arrow) 30 . More preferentially, NO molecules attach to the edge of the B-GNR (red arrow). However, it is more important that the NO molecules were most efficiently trapped at the centre of the B-GNR (yellow arrow), corresponding to the boron site. By setting the tip closer to the sample surface, this NO molecule can be easily manipulated by the scanning tip.  Fig. 4c corresponds to the manipulated NO by the scanning local probe. Since the NO molecule showed a hopping behaviour on specific boron sites, the BC 3 moieties exhibit a Lewis acidic role for the stable gas adsorption. The simulated STM image (Fig. 4e) with optimized adsorption model (Fig. 4f,g) shows that the NO molecule is bound to the B site via the N atom. The calculated binding energy of 1.5 eV is ideal for a chemical sensor, since a continuous adsorption at room temperature is expected to show a linear response to the NO density. The NO chemical adsorption significantly changes the electronic structure of the B-GNR, introducing new states into the band gap ( Supplementary Fig. 3), which suggests a strong electric response in a device. To refresh the device and avoid saturation, a small increase in thermal energy will cause desorption.

Discussion
We present a programmed synthesis of boron-doped graphene nanoribbons with uniform texture of Lewis acidic site. Ullmann type on-surface polymerization and the following thermal cyclodehydrogenation reactions were employed for a triarylborane precursor with an elaborate design. The uniform BC 3 sites are doped at designated positions with a fixed doping ratio according to the precursor. Namely, two boron atoms exist at the centre row in every three anthracene moieties and the doping density is 4. only static control. The tunable response presented here offers great potential in many technologies, particularly where dynamic control would be a disruptive development. Besides this application, the fabrication of boron-doped nanoribbons could allow for the design of mobility gaps and novel types of graphene transistors 32,33 . In the N ¼ 7 B-GNR, the boron locates exactly at the centre of the ribbon, so that the backscattering is fully suppressed. Yet, in the fused B-GNRs, the symmetry can vary. Further, the asymmetry can be increased by co-depositing boron-doped and non-doped precursor molecules. To fabricate such devices, transferring on-surface synthesized B-GNRs is a critical issue, yet we believe that the recently developed thin film etching technique can solve this 16,34 .

Methods
Experimental measurement. All experiments were performed with an Omicron STM/AFM with a qPlus configuration 35 , operating at 4.8 K in ultrahigh vacuum. A clean Au(111) surface was in-situ prepared by repeated cycles of standard sputtering and annealing. The tungsten tip of a tuning fork sensor was ex-situ sharpened by a focused ion beam milling technique and was then in-situ covered with Au atoms by contacting to the sample surface. The resonance frequency of the self-oscillating qPlus sensor was detected by a digital PLL circuit (Nanonis: OC4 and Zurich Instruments: HF2LI and PLL). 9,10-bis(10-bromoanthracen-9-yl)-9,10-dihydro-9,10-diboraanthracene was deposited on the substrate from a crucible of Knudsen cell, resistively heated at 240°C. Subsequently, the sample was annealed at 180 and 400°C to synthesize the N ¼ 7 B-GNR. The fused B-GNRs with N ¼ 14 and 21 were obtained by further annealing at 510°C. Further, NO gas was introduced into the ultrahigh vacuum chamber (100 Langmuir) while the sample stayed in the cold microscope. In STM mode, the tip was biased while the sample was electronically grounded. The topographic images were taken in a constant current mode. In AFM mode, the tip apex was terminated by a CO-molecule and all images were taken at a constant height mode.
Synthesis of precursor molecule. The precursor 1, 9,10-bis(10-bromoanthracen-9-yl)-9,10-dihydro-9,10-diboraanthracene, was obtained from compounds 2 and 3 through the synthetic route shown in Supplementary Fig. 11a. Although some derivatives of the precursor 1 with extra bulky substituents have been reported 36 , the precursor 1 has not been characterized because of its low solubility in common organic solvents. We successfully characterized the isolated precursor 1 by the NMR analyses in tetrachloroethane at 100°C and by X-ray crystal structure analysis using chlorobenzene as solvent. The crystalline sample of the precursor 1 can be stored at ambient conditions as a result of the steric protection of the boron atoms by the anthracene groups. Compounds 2 (ref. 37) and 3 (ref. 38) were prepared according to literature procedures. The compound 3 was stored in a glove box due to the moisture sensitivity. All reactions were carried out under a nitrogen atmosphere. Commercially available solvents and reagents were used as received. Dry ether and n-BuLi (1.6 M in hexane) were purchased from Wako Pure Chemical Industries. Dry toluene was purchased from Kanto Chemical.
Characterization of precursor molecule. NMR spectra of the precursor 1 are shown in Supplementary Fig. 11d. 1 H and 13 C NMR spectra were recorded with a JEOL A-600 spectrometer (600 MHz for 1 H and 151 MHz for 13 C). 11 B-NMR spectrum was recorded with a JEOL AL-400 spectrometer (128 MHz for 11 B). Chemical shifts are reported in d ppm with reference to residual protons and carbons of 1,1,2,2-tetrachloroethane-d 2 (d 6.00 ppm in 1 H-NMR and d 73.78 ppm in 13 C NMR). The external standard of BF 3 Á OEt 2 was used for 11 B NMR. The mass spectrum was measured with a Bruker Daltonics micrOTOF Focus using a positive-mode Atmospheric Pressure Chemical Ionization (APCI)-time of flight method in a CHCl 3 solution. Thin layer chromatography was performed on plates coated with 0.25-mm thickness of silica gel 60 F 254 (Merck). Column chromatography was performed using silica gel PSQ 100B (Fuji Silysia Chemical).
Theoretical calculation. All first principles calculations in this work were performed using the periodic plane-wave basis Vienna ab-initio simulation package (VASP) code 39,40 implementing the spin-polarized DFT. To accurately include van der Waals interactions in this system we used the optB86P functional [41][42][43] . Projected augmented wave potentials were used to describe the core electrons 44 . A kinetic energy cutoff of 550 eV was found to converge the total energy of all the systems to within 10 meV. All STM simulations were done using the basic Tersoff-Hamann approach 45 . The properties of the bulk and surface of Au and graphite, and a NO molecule, were carefully checked within this methodology, and excellent agreement was achieved with experiments. Systematic k-point convergence was checked for all systems and a 5 Â 1 Â 1 mesh was found to be sufficient for the large GNR-Au systems. All reference calculations for adsorption energies, charge densities and projected density of states were performed with 4-anthracene GNR on gold, which had a lattice mismatch of only 1.4% (accommodated in the ribbon). When adding boron (B-GNR), this mismatch reduced to 0.3% as the ribbon expands slightly (Supplementary Fig. 4). To have a comparable anthracene separation of B-B sites as in experiments, particularly for the STM simulations, we also performed calculations on isolated 3-anthracene GNRs (for this size the mismatch with the Au surface would be over 10% unless an unfeasibly large unit cell is used). The simulated STM images were very similar, and hence were used for the image comparison in Fig. 1 for simplicity.
Calculated AFM image. Recently, several mechanical AFM models have been proposed 46,47 , where the approach of a CO-functionalized AFM tip over a substrate can be simulated To obtain an accurate 3D force map. Key to these models is that during approach they allow the CO molecule to tilt under influence of the forces it experiences. This tilting proved to be key to understand the sharp contrast observed in atomic bonds. Here we used the model developed by Hapala et al. 46 and extended it by adding electrostatic interactions on top of the van der Waals interactions. The B-GNR structure was taken from the DFT simulations of B-GNR on Au and the charges were extracted from the DFT calculations via Bader analysis 48 . The mechanical AFM model relies on empirical Lennard-Jones parameters, which were taken from the Chemistry at HARvard Molecular Mechanics (CHARMM) force field 49 , with additional boron parameters 50 . All other parameters are the same as intended by Hapala et al., and the simulated AFM scan is performed at a resolution of 0.025 Å (in all directions), with a force tolerance criterion of 4 Â 10 À 6 eV Å À 1 . The 3D force field is subsequently converted into a frequency shift image 51 , using the experimental parameters (f ¼ 24.76 kHz, A ¼ 53 pm and k ¼ 1,800 N m À 1 ). In Fig. 3d the simulated AFM image for the approximate same height as in the experiments is shown (the height in the bottom left corner of the figure denotes the distance of the carbon of the CO-molecule to the plane going through the carbons of the B-GNR).