On‐Surface Synthesis of Edge‐Extended Zigzag Graphene Nanoribbons

Graphene nanoribbons (GNRs) have gained significant attention in nanoelectronics due to their potential for precise tuning of electronic properties through variations in edge structure and ribbon width. However, the synthesis of GNRs with highly sought‐after zigzag edges (ZGNRs), critical for spintronics and quantum information technologies, remains challenging. In this study, a design motif for synthesizing a novel class of GNRs termed edge‐extended ZGNRs is presented. This motif enables the controlled incorporation of edge extensions along the zigzag edges at regular intervals. The synthesis of a specific GNR instance—a 3‐zigzag‐rows‐wide ZGNR—with bisanthene units fused to the zigzag edges on alternating sides of the ribbon axis is successfully demonstrated. The resulting edge‐extended 3‐ZGNR is comprehensively characterized for its chemical structure and electronic properties using scanning probe techniques, complemented by density functional theory calculations. The design motif showcased here opens up new possibilities for synthesizing a diverse range of edge‐extended ZGNRs, expanding the structural landscape of GNRs and facilitating the exploration of their structure‐dependent electronic properties.


Introduction
Graphene nanoribbons (GNRs) are an emerging class of nanomaterials with electronic properties that can be widely tuned by their width and edge structure. [1]The sensitive dependence of DOI: 10.1002/adma.202306311 the electronic properties on the chemical structure of the GNRs, however, demands an atomically precise synthesis.This can be attained using surface-catalyzed chemical reactions that lead to the formation of covalent bonds between and within surface-adsorbed molecular building blocks.This so-called on-surface synthesis approach [2,3] achieves atomic precision if only specific reactions are allowed, thus necessitating a careful design of molecular precursors used as building blocks.Therefore, the structural diversity of the GNRs that can be synthesizedand thus the structure-dependent properties that they can exhibit-relies critically on the available design motifs. [1,4]omocoupling by forming a single C─C bond between molecular precursors is the simplest means toward the synthesis of extended one-dimensional structures.However, this necessarily leads to coupling along the armchair direction.Consequently, most of the GNRs that have been synthesized to date are armchair GNRs (AGNRs).22] Nevertheless, the synthesis of ZGNRs and their structural diversification have remained a major challenge in the field, due to the geometric restriction imposed by precursor homo-coupling via single C─C bond formation, and consequently, only two similar design motifs have been reported to afford ZGNRs. [23,24]In particular, the on-surface synthesis of the 6-atom-wide ZGNR (6-ZGNR), the first ZGNR synthesized with atomic precision, was based on 2,12-dibromo-14-R 1 -7-R 2 -benzo[m]tetraphene as the precursor, cleverly subverting the geometric restriction on polymerization (Figure 1a) [23] : Upon debromination, polymerization occurred along the zigzag direction to form a snake-like polymer, leading to 6-ZGNR with a dimethylbiphenylyl group as R 1 , through the surface-catalyzed oxidative cyclodehydrogena-tion.The choice of various R 1 and R 2 substituents on the polymer backbone made available a platform for the conception of various GNRs with this motif.However, the synthesis of the corresponding precursor molecules is highly challenging due to the structural complexity and their thorough purification, which is essential for the on-surface polymerization, is not trivial because of their poor solubility.Consequently, only a few minor modifications of the precursor of 6-ZGNR have thus far been achieved, [23,25,26] prohibiting the exploration of their structure dependent electronic properties, e.g., by incorporation of periodic edge-extension on to the zigzag edges.
Here, we introduce a design motif to synthesize ZGNRs with diverse periodic edge-extensions, a class of novel GNRs that we name edge-extended ZGNRs.This molecular motif relies on a simple 2,7-linked-9-R 1 -10-R 2 -anthracene polymer backbone achieved by the surface-catalyzed debromination and subsequent polymerization of 2,7-dibromo-9-R 1 -10-R 2 -anthracene (Figure 1b).The inter-molecular C─C bond is formed at an angle of 30°to the growth axis.This design motif entails the incorporation of edge extensions fused onto the zigzag edge at alternating sides of the backbone, ruling out the synthesis of pristine ZGNRs.However, it provides a platform for synthesizing various types of edge-extended ZGNRs due to its versatility.[29] To establish the novel design motif based on 2,7-linked anthracenes, we synthesize a precursor with bianthryl substitution onto the R 1 position (Figure 1b).Annealing the molecular precursors on Au(111) single crystal surfaces under ultrahigh vacuum conditions yields an edge-extended 3-ZGNR with edge-fused bisanthene extensions, which we name the 3-ZGNR-E(bisanthene, 7) or 3-ZGNR-E(BA, 7) for brevity (see Section S1, Supporting Information for definition of the nomenclature).The chemical structure of the 3-ZGNR-E(BA, 7) is confirmed by bond-resolved non-contact atomic force microscopy (nc-AFM, Figure 1c).STM, STS, and Density functional theory (DFT) calculations provide further insights into its electronic properties.
Deposition of 6 onto a clean Au(111) surface results in large islands of self-assembled intact molecules (Figure 2b).Upon annealing to 170 °C, the molecules undergo debromination followed by homocoupling to yield polymers of 7 (Figure 2c).The STM image shows linear polymers arranged in self-assembled islands.The formation of linear polymers confirms that the bianthryl substituents do not sterically hinder access to the radical formed by debromination in 6, demonstrating the viability of the proposed 2,7-dibromo-9-R 1 -10-R 2 -anthracene design motif for edge-extended ZGNR synthesis.Annealing to 310 °C triggers cyclodehydrogenation, resulting in planar GNRs with fretpatterned edges characterizing the 3-ZGNR-E(BA, 7) (Figure 2d).The GNRs are found to grow reasonably long, with an average length of 9 nm (see Section S3, Supporting Information for the statistics) The atomically precise synthesis of 3-ZGNR-E(BA, 7) is confirmed with bond-resolved nc-AFM images on a 16-unit long segment (Figure 1c).It clearly resolves the three zigzag rows constituting the 3-ZGNR backbone, the three additional short zigzag rows within the edge-fused bisanthene extensions and their terminating armchair edge segments.
The electronic state localized on the zigzag edge of graphene is topological in origin, [31] and therefore it must be localized on the boundary of the system.In this picture, incorporating the bisanthene extensions onto the zigzag edge pushes that boundary locally further out.The edge state on the original zigzag "edge" of the "3-ZGNR" is thus locally quenched.Since the bisanthene extensions also possess zigzag edges, the edge state is now expected to be localized on this new boundary.These heuristic arguments are fully supported by spin-unrestricted DFT band structure calculations for the 3-ZGNR-E(BA, 7) in the gas phase (Figure 3a), that show the spin-polarized state hosted on the edges of this GNR (Figure 3b).DFT calculations neglecting spins (i.e., spinrestricted) reveal an ultralow bandgap of 0.01 eV.Incorporating the spin degree of freedom increases this bandgap to 0.24 eV.The states that are almost at the Fermi level in the spin-restricted case, the conduction band (CB) and the valence band (VB) at the Γ point, are both localized on the edges of the GNR (Figure 3c).When electron-electron interactions are included, it is more favorable for the electrons to spin-polarize, forming the so-called Mott-Hubbard gap.Indeed, based on the DFT calculations, the spin-polarized GNR is more stable by 35 meV and thus the open-shell electronic configuration better represents the ground state of the free standing system.Additionally, we perform tightbinding calculations with a mean-field Hubbard Hamiltonian to investigate the development of the magnetization as a function of the length and the strength of the on-site Coulomb repulsion (U).We find GNRs have spin polarized ege-states irrespective of their length at reasonable values of U (see Section S4, Supporting Information).
The characterization of ZGNRs with spin-polarized frontier states is challenging to achieve on metallic surfaces. [23]Due to the spin polarization, the ZGNRs have an open-shell ground state, and consequently, these reactive GNRs can strongly couple to the metal surface.It required a decoupling layer of NaCl to unravel the intrinsic electronic properties of 6-ZGNR. [23]Somewhat unexpectedly, the 3-ZGNR-E(BA, 7) exhibits spectroscopic signatures in scanning tunneling spectroscopy (STS) measurements even on the Au(111) surface (Figure 4a,b).The spectroscopic features are most pronounced near the ends of the GNR (see Section S5, Supporting Information for a discussion).Moreover, a constant-height current map at −10 mV bias reveals striking molecular orbital features, evidencing the presence of electronic states near the Fermi level, as opposed to a bandgap of ≈0.24 eV expected for the open-shell ground state of the GNR in gas-phase (Figure 3a).
The broad spectroscopic features deriving from these frontier states indicate significant hybridization with the underlying metal surface.To identify the origin of the observed spectral features, we acquired differential conductance (dI/dV) maps at various energies on a 9-unit long GNR (Figure 4e).These maps are in qualitative agreement with the gas-phase orbitals shown in Figure 3c, permitting us to assign these features to their origin in the spin-restricted band structure of the GNR.The corresponding points are indicated below the acquired images.On the Au(111) surface, the minimum of the second-highest valence band (VB-1 Γ) is at −0.8 eV, and the maximum of the second conduction band (CB+1 Γ) is at 1.15 eV.This corresponds to a "bandwidth" of ≈2.0 eV for the four bands (CB+1, CB, VB, and VB −1), which is somewhat (and unsurprisingly) underestimated by the gas-phase DFT calculations (≈1.4 eV).The assignment above compared a finite GNR segment adsorbed onto a substrate with an infinite gas-phase GNR.The objective of such a comparison was to link the orbitals of a finite system to the bands of an infinite GNR.However, to capture the interaction of the GNRs with the metal substrate, we obtain the DFT-optimized geometry of a 9-unit segment of 3-ZGNR-E(BA, 7) adsorbed on a 4-layer Auslab.Spin-restricted STM (Figure 4c) and STS (Figure 4d) simulations were performed on the optimized geometry.The simulated STS results exhibit broad spectroscopic features.This shows that the DFT calculations capture the hybridization between GNR and Au states, therefore provide a better comparison with the experimental observations than the gas phase calculations.Simulated dI/dV maps shown in Figure 4f are in qualitative agreement with the corresponding experimental results.A series of the simulated dI/dV maps from −1.0 to 1.0 eV in steps of 0.1 eV are shown in Section S6 (Supporting Information) and additional experimental results are shown in Section S7 (Supporting Information).
Experimentally, the broad spectroscopic features are largely confined to the ends of the GNR segment.Naively, one may be tempted to assign the peak near the Fermi level seen at the ends of the GNR (Figure 4b) to a "topological end-state."However, this is clearly not the case.Tight-binding simulations of a finite length GNR segment exhibit no mid-gap end states, showing that the system is topologically trivial (see Section S8, Supporting Information).The origin of the high intensity of the STS signal at the ends of the GNR lies in the symmetry of the wavefunctions in the VB and CB, as has previously been observed for the armchair edges of the 7-AGNR. [32,33]A detailed discussion of this effect is given in Section S5 (Supporting Information).
The absence of a substantial bandgap (>0.24 eV) shows that the GNR is closed-shell when adsorbed on Au(111), despite having an open-shell ground-state in gas-phase.To show that the interaction with the substrate leads to the loss of spin-polarization, we first obtain the gas phase orbitals of the adsorbed 9-unit GNR segment.We then calculate the projected density of states (PDOS) of all the states of the optimized GNR-Au substrate system onto these gas-phase orbitals.This is obtained by taking a dot product of the orbital wavefunctions with every state of the combined GNR-Au slab system.The corresponding plots (see Section S9, Supporting Information) reveal that the PDOS for each orbital is significantly broadened due to hybridization of the GNR with the Au-slab.Moreover, the gas phase HOMO, HOMO-1, and HOMO-2 of the 9-unit segment are substantially depopulated when the GNR is adsorbed onto the Au surface (Figure S10, Supporting Information).Similar depopulation has been observed in the case of the end-states of the 7-atom wide AGNR (7-AGNR). [34]Depopulation of the states near VB Γ results in the 3-ZGNR-E(BA, 7) segment being closed-shell on Au(111).
However, the 7-AGNR end-states were re-populated when the 7-AGNR segments were transferred onto thin insulating NaCl decoupling layers. [34,35]A similar strategy may enable us to uncover the open-shell ground state of 3-ZGNR-E(BA, 7).
Adsorption of 3-ZGNR-E(BA, 7) on the Au surface thus quenches its intrinsic magnetic properties, and electronic decoupling from the metal substrate is thus essential for exploring the expected intriguing magnetic properties. [26]However, strong interaction of a GNR with the underlying metal surface makes the STM tip-assisted manipulation of a GNR segment on top of a NaCl layer highly challenging.The recent success in growing 7-AGNRs on a TiO 2 surface points to a possible alternative to electronically decouple GNRs, [36] but is beyond the scope of the present work.
Recent results on a nitrogen-substituted 6-ZGNR suggest the possibility to spectroscopically access the edge states of the substituted 6-ZGNR even on the Au(111) surface, without the need for a decoupling layer. [26]Similar nitrogen substitutions might also be achieved with the design motif presented here.Finally, we note that the work done here focused entirely on R 1 substituents.Future work could incorporate additional R 2 substituents on the polymer backbone to obtain wider ZGNRs, such as to decrease the hybridization between opposite zigzag edges.Successful incorporation of R 2 and R 1 substituents along with heteroatom substitutions provide a versatile route to the synthesis of various novel GNR structures utilizing this design motif.Their expected magnetic and optical properties make these systems technologically highly appealing.

Conclusion
We have developed a new bottom-up graphene nanoribbon design motif based on a family of 2,7-dibromo-9-R 1 -10-R 2anthracene precursors and deployed it for the successful onsurface synthesis of a novel edge-extended ZGNR.The essential characteristic of the 2,7-dibromo-9-R 1 -10-R 2 -anthracene design motif is its versatility, and several possible edge-extended ZGNRs can be attained with synthetically feasible modifications of the precursor.An intriguing possibility would be the incorporation of spinful edge-extensions that could interact via the edge states.Moreover, the tremendous body of knowledge generated in synthesizing dibromo-bianthryl precursors for AG-NRs is also applicable with the novel design motif to synthesize edge-extended ZGNRs.Heteroatom substitutions, incorporation of macrocycles, different edge morphologies, and GNRheterojunctions, which were hitherto largely limited to AGNRs, are expected to now become available also in the zigzag direction.

Experimental Section
Sample Preparation: Details of the synthesis and characterizations of monomer 6 and all the other new compounds are reported in Section S2 (Supporting Information).Single crystal Au(111) substrates (MaTeck GmbH) were cleaned by repeated cycles of sputtering with Ar + -ions followed by annealing to 400-450 °C.Monomer 6 was deposited onto the cleaned Au(111) surface in situ by sublimating the precursor molecules from quartz crucibles.A custom built evaporator, by Lukas Rotach (Empa) was utilized to this end.A quartz crystal microbalance was used to ensure a sublimation rate corresponding to a rate of ≈2.3 monolayer h −1 .The onsurface synthesis was triggered by annealing the Au(111) single crystal at the specified temperature for 15-60 min.Longer times were used to trigger the polymerization, while shorter times were used to trigger cyclodehydrogenation.The surface temperature was measured with an optical pyrometer (Optris).The sample preparations were carried out in a preparation chamber (base-pressure: 2 × 10 −10 mbar) and the samples then transferred to the attached STM chamber (base-pressure: 7 × 10 −11 mbar).
The STM and nc-AFM experiments were performed using a lowtemperature scanning tunneling microscope (Scienta Omicron) operated at 4.7 K with a tungsten tip placed on a qPlus tuning fork sensor. [37]The tip was functionalized with a single CO molecule. [38]The CO was deposited onto the surface by filling the STM chamber with the gas at a pressure of 3 × 10 −8 mbar, with the sample in the cold STM stage.To pick up the CO molecule from the Au(111) surface, the surface at −20 mV and 120-200 pA was scanned till a change in height was observed.To perform the nc-AFM measurements, the qPlus sensor was driven at its resonance frequency (24787 Hz, quality factor 20.6k) with a constant amplitude <100 pm.A phase-locked loop (PLL, HF2Li PLL by Zurich Instruments) was used to track the frequency shift of the tuning fork.The Δz values are measured with respect to the tip-surface distance where the feedback loop is switched off.Increasing Δz corresponds to increasing the tip-surface distance.
Computational Details: DFT computations were executed using the AiiDAlab environment, [39] based on AiiDA, [40] and the CP2K code. [41]urface-adsorbate setups were modeled within a periodic slab scheme.The simulation cell included four Au atomic planes along the [111] orientation.Hydrogen atoms passivated one face of the slab to mitigate Au(111) surface states.A 40 Å vacuum layer was included to isolate the system from its periodic images along the axis orthogonal to the slab.Electronic wavefunctions were represented via TZV2P Gaussian basis set [42] for C, H, and DZVP for Au.Plane-wave basis cutoff for the charge density was set at 600 Ry.Norm-conserving Goedecker-Teter-Hutter pseudopotentials [43] were employed.The PBE GGA functional [44] and included Grimme's D3 scheme [45] was adopted for van der Waals interactions.Au supercells varied in size depending on the adsorbate, ranging from 28.12 × 26.54 Å 2 (corresponding to 480 Au atoms) to 89.37 × 32.42 Å 2 (1728 Au atoms).Geometry optimization were performed with the bottom two atomic planes constrained while relaxing others until forces were below 0.005 eV Å −1 .For nc-AFM studies in AiiDAlab, equilibrium geometries and CP2K-calculated electrostatic potentials were used alongside Hapala's probe-particle code. [46]TM simulations were based on the Tersoff-Hamann approximation, [47] utilizing the Kohn-Sham orbitals of the slab and the adsorbate.Charge density decay was corrected by extrapolating orbitals to the vacuum.Band structure DFT for GNRs was carried out in Quantum Espresso, [48] using PBE and a 400 Ry cutoff for charge density, along with PAW pseudopotentials and a 13 × 1 × 1 k-grid.Convergence criteria for forces were set at 0.001 a.u.Molecular orbital projections were computed through scalar products with full-system orbitals, utilizing gas-phase molecular orbitals in their adsorption conformations.
The tight-binding (TB) calculations were performed by solving the standard MFH-Hamiltonian: where c † i, is the creation operator for an electron with spin  at site i and c j, is the annihilation operator for an electron with spin  at site j. can be either spin-up (↑) or spin-down (↑), with σ indicating the opposite spin component compared to . 〈i, j〉 are 1st nearest-neighbor pairs of the lat-tice and 〈l, k〉 are the 3rd nearest neighbor pairs.The hopping for nearest neighbor pairs, t 1 = 3 eV and the hopping over the 3rd nearest neighbor pairs, t 3 = 0.3 eV, unless otherwise stated.Due to particle-hole symmetry the next nearest neighbor hopping (t 2 ) is set to zero.The on-site Coulomb repulsion term is also set to zero to simulate the non-magnetic system.To compute the magnetic properties of the gas-phase system, shown in Section S4 (Supporting Information), U is non-zero, where a typical value for U would be 3 eV, i.e., 1 × t 1. n i, denotes the spin-dependent number operator and 〈n i, 〉 is the mean occupation number at site i.
|E n 〉 are the eigenvectors of ĤTB : |E n ⟩ = ∑ i  i,n |i⟩ with |i⟩ as the vectorspace basis of lattice sites.The wavefunctions in real-space are calculated using Slater 2p z orbitals ( 2p z ) at each site i.
To simulate the dI/dV maps, LDOS(⃗ r), shown in Section S5 (Supporting Information), the orbital electron densities were summed over the relevant energy window [E n 1 , E n 2 ]:

Figure 3 .
Figure 3. Gas phase band structure and frontier orbitals of the 3-ZGNR-E(BA, 7).a) Spin-restricted (in red) and spin-unrestricted (in blue) DFT bandstructure calculations show a bandgap of 0.01 and 0.24 eV, respectively.The dashed grey line indicates the Fermi level for a charge neutral GNR, the first and second conduction and valence bands are labeled.The reciprocal space is shown from the Γ-point to the Χ-point.b) Spin density plot of the GNR.c) The spin-restricted gas phase orbitals of the GNR at the Γand Χ-points are shown for the first two conduction and valence bands.The images are labeled with the points in the band structure that they represent.

Figure 4 .
Figure 4. Electronic characterization of the 3-ZGNR-E(BA, 7) on Au(111).a,b) The STS measurements in (b) are acquired on the 9-unit GNR segment shown in (a), and the spectra are color-coded to their STS positions in (a).Broad spectroscopic features are seen around the Fermi level and near the ends of the GNR.c,d) STM and STS simulations on a DFT-optimized geometry of a 9-unit GNR segment on an Au(111) slab.The spectra shown in (d) are the local density of states (LDOS) as a function of energy, color-coded to their acquisition positions on the simulated STM image in (c).e) dI/dV maps acquired over the GNR shown in (a).Acquisition bias is shown above each map, and the best-fitting gas-phase orbital of an infinite GNR is indicated below each maps.f) Simulated dI/dV maps for the system shown in (c) that best fit the experimental results are shown with the corresponding energy indicated above the individual maps.(Tunneling parameters in (a): V = −20 mV, I = 50 pA.STS settings in (b): Feedback switched off over the indicated positions at −1 V, 250 pA.Simulation settings in (c): Simulated STM image with V = −0.1 V. d simulated STS spectra by plotting the LDOS versus energy at 4 Å.Imaging conditions for the maps in (e): map at −0.8 V was acquired at a constant-height above the GNR, feedback loop switched off at −0.8 V, 200 pA on the GNR.The map at −0.01 V is the current-image acquired at a constant-height over the GNR, feedback loop switched off at −0.01 V, 100 pA.dI/dV maps at other biases are acquired with constant-current of 200 pA.dI/dV maps and STS in (b) were recorded with a peak-to-peak excitation voltage of 20 mV.(f): dI/dV maps are simulated at a constant-height of 4 Å above the GNR).