Covalent organic frameworks (COFs), as a large class of porous organic materials, have attracted intense research in the past few decades due to the great potential for applications in the fields of, for example, catalysis1,2, optoelectronics3,4 and gas storage and separation5,6. Since the seminal work on COF synthesis by Yaghi and co-workers in 2005 (ref. 7), developments in both solvothermal and on-surface syntheses have increased the diversity of COF structures, which are mainly composed of light elements (B, C, N, O, H). COF syntheses employed Schiff base reactions and self-condensation of boronic acid, as well as coupling between boronic acid and catechol1,7,8,9,10. More recent efforts introducing new chemical reactions connecting aromatic multi-substituted monomers have succeeded in the synthesis of conjugated two-dimensional COFs (refs.11,12). If the heavy elements are substituted in the precursor molecules, the elemental variety is drastically enhanced13.

Silicon is an element of group 14 in the periodic table. Its four valence electrons at the outermost shell give it similar properties to carbon, yet the longer bond lengths and possible higher bonding states lead to a higher Lewis acidity and unique chemical reactivities14,15. Silicon-incorporated organic functional molecules and novel nanostructures have attracted much attention in recent decades16,17. For instance, silicates (containing Si–O bonds) are some of the most studied compounds18,19 and the successful synthesis of the silicate-COF was recently demonstrated20. The silyl group, commonly used as a protecting group in solution21, was found to induce coupling reactions on surfaces22,23,24,25. By contrast silabenzenes, having unique heterocyclic rings with C–Si bonds, represent the heavier congeners of cyclic aromatic compounds in fundamental organic chemistry26,27. These have been studied as an elusive target product for organic synthesis due to their high reactivity at ambient temperatures and their difficult isolation28,29,30. As yet, only a few silabenzene compounds have been reported in solution28,29,30,31,32,33,34,35. Silabenzenes are very promising candidates in on-surface synthesis to create COFs and other silicon-incorporated nanostructures. In this Article, we present the on-surface synthesis of disilabenzene-bridged COFs by deposition of silicon atoms on a Au(111) surface and annealing followed by reaction with 2,3,6,7,10,11-hexabromotriphenylene (HBTP). Their structures and chemical properties are analysed using a combination of bond-resolved scanning tunnelling microscopy (STM), scanning tunnelling spectroscopy (STS), photoelectron spectroscopy and density functional theory (DFT) calculations. The synthesis of the conjugated COFs by on-surface coupling of Si atoms and polyaromatic hydrocarbons may pave the way for fabrication of novel low-dimensional nanostructures.

Results and discussion

Synthesis of 1,4-disilabenzene-linked conjugated COFs

We used on-surface synthesis to realize a silabenzene-bridged COF under ultra-high vacuum conditions36,37. This bottom-up method has proved successful in the synthesis of graphene nanoribbons (GNRs) with different edges, including different atomic species38,39. Once GNRs are merged with each other at their edges, wider GNRs or even two-dimensional COFs can be synthesized40,41. Alternatively, COFs have also been synthesized with predefined and small precursors42,43. However, in some cases, such as for the unstable silabenzene, no direct precursor is available.

In this Article, we overcome this limitation by combining conventional surface science techniques and on-surface chemistry. We used HBTP as a building block to fabricate a two-dimensional Si-incorporated COF (Fig. 1a). Firstly, a submonolayer AuSix film was formed on Au(111) by depositing Si atoms at room temperature with a post-anneal to 420 K (Supplementary Fig. 1). HBTP molecules were deposited on the substrate held at 420 K, causing debromination. Consequently, the sample was further annealed at a higher temperature of 580 K. We found the formation of hexagonal nanoporous structures in the STM topography (Fig. 1b). The pores are surrounded by six bright spots, with equivalent sites separated by 1.75 ± 0.02 nm (Fig. 1c). The contrast of the nodal site and surrounding six bright spots changes with respect to the bias voltage (Extended Data Fig. 1). Since following a similar growth procedure method without pre-deposited Si atoms on a clean Au(111) surface resulted in formation of disordered films (Extended Data Fig. 2), we conclude that Si atoms play a decisive role in the synthesis of the structure. Before the final step of annealing at 580 K, we also found the formation of SiBrx (x = 1, 2, 3) compounds on the Au terrace as well as on the nanoporous structure (Supplementary Fig. 2)—these can be desorbed as SiBr4 molecules from the surface by annealing at 450 K (ref. 44). To resolve the inner structures of the framework, the tip apex was terminated by a CO molecule45,46. The bond-resolved STM image taken at a constant height mode indicates that the structure was composed of six triphenylene backbones at the nodal site (Fig. 1d and Extended Data Fig. 3), which are inter-connected by two different types of bond. The length of the longer one (3.10 ± 0.20 Å) is far more than the typical length of a covalent bond, indicating that it is composed of more than two atoms. We tentatively assigned this longer line as C–Si–X (X = Br or H) bonds, where Br atoms are from HBTP molecules, while H atoms are possibly from the chamber environment. In our previous study, Si and Br atoms can easily form a covalent bond on Au(111) (Fig. 1a)44. The central bond between two neighbouring triphenylene backbones has a shorter length (2.80 ± 0.20 Å), which we suggest is due to deflection of the CO tip during scanning and is not a physical bond47. In order to investigate the structure behind these images, we undertook an extensive DFT analysis of possible molecular assemblies, including a wide variety of configurations and atoms in the network. The best agreement is shown in Fig. 1e, suggesting the Si atoms in the C4Si2 ring are passivated by Br atoms (a Si atom at the edge of the Si-COF can also bond to two Br atoms (Supplementary Fig. 3)). We see a very good agreement with experiment in the STM-simulated topography (Fig. 1f), with pores also separated by 1.75 nm. We also see good agreement in the comparison between the high-resolution CO-tip STM images—the simulated image in Fig. 1g reproduces the sharp contrast over the central triphenylene backbone and shows a thick line between neighbouring triphenylenes (2.93 Å) and also a long line over C–Si–Br (3.33 Å). Note that the Si–Br bond in the COF can be cleaved by applying a bias voltage (Extended Data Fig. 4). The analysis of induced charge-density differences shows the interaction between Si and Au stabilizing the COF on the substrate (Extended Data Fig. 5). However, we observe minimal changes in bond lengths and charges for equivalent C, Si and Br sites when comparing the COF on Au with the relaxed, fully planar, system in isolation, suggesting that the properties of the planar COF in isolation and the COF on Au are similar. This is further confirmed by a comparison of the nucleus-independent chemical shift (NICS) between characteristic molecules, the isolated COF and the COF on Au (Supplementary Figs. 4 and 5). Overall, this confirms that the triphenylene blocks are connected via the planar C4Si2 ring, resulting in 1,4-disilabenzene-linked COFs.

Fig. 1: Synthesis of Si-incorporated COF.
figure 1

a, Scheme of on-surface reaction for the aryl–Si coupling reaction on Au(111). b, STM topography of the sample after annealing at 580 K. Sample bias voltage V = 100 mV and tunnelling current I = 20 pA. c, Close-up STM image of the Si-COF. V = 200 mV and I = 5 pA. d, High-resolution constant height dI/dV map of the Si-COF taken with a CO tip. a.u., arbitrary unit. e, Predicted top-view structure from DFT simulations with key bond lengths (Å) indicated. Inset, side view. Atom colour code: C, cyan; H, white; Au, gold; Si, brown; Br, purple. f,g, Associated simulated STM images showing STM topography at 200 mV and constant charge density of 1 × 10−8 e Å−3 (f) and high-resolution image at a bias of 0.5 V and a height of 0.3 nm (g).

Electronic properties of Si-COFs

The electronic properties of the Si-COF were measured in more detail using STS (Fig. 2a). Due to the interaction with the Au substrate, the occupied state was detected only at the centre of the triphenylene block (light blue curve) and was significantly broadened as indicated by the grey areas. In fact, such a transition of the contrast can be clearly seen in a series of constant height dI/dV maps (Fig. 2b,c and Supplementary Figs. 6 and 7). The interaction between the Si-COF and Au(111) surface also leads to the broadening of the unoccupied state peaks, which are localized around the Br sites. Nevertheless, we assigned the unoccupied and occupied states as 1.6 and −0.7 V, resulting in a band gap of approximately 2.3 eV. On the other hand, the distinct electronic states of a Si atom incorporated in the Si-COF and a single Si atom adsorbed on a clean Au(111) surface were also measured by STS, indicating a different chemical nature (Supplementary Fig. 8). The calculated density of states (DOS) (Fig. 2d) shows similar features to the STS, with matching rapidly rising densities of states around −1.5 and 2.0 V. This rapid rise is associated with the states coming from the molecular backbone of the COF, as the contribution from the C atoms significantly increases, while the contributions from the Si and Br atoms remain more or less constant in the interval of energy shown in Fig. 2d. The direct comparison of the simulated occupied states image at −0.3 V (Fig. 2e) reproduces the experimental dominance of the Br atoms, with nothing seen on the molecular backbone. Supplementary Fig. 9 shows that the Br dominance is actually observed in several occupied states within the window of −1.0 to 0.0 V, and in the empty states as well, up to 0.8 V. However, for higher biases, contributions from both Br atoms and the molecular backbone are observed uniformly, until the general feature previously observed is inverted. Figure 2f, for example, shows this inversion at 2.2 V, where the strong dominance of the molecular backbone seen in the experiment is finally reproduced.

Fig. 2: Electronic properties of Si-COF.
figure 2

a, dI/dV curves were recorded above the Si-COF (inset image) and the Au(111) surface for contrast. The corresponding sites are indicated by coloured crosses. a.u., arbitrary unit. b,c, Constant height dI/dV maps were measured with a CO-terminated tip at −0.7 V (b) and +1.6 V (c). d, Calculated density of states (DOS) for C, H, Si and Au as well as for the total of the atoms. e,f, Simulated constant height dI/dV maps of the Si-COF at biases of −0.3 V (e) and +2.2 V (f).

Photoemission spectroscopy in each reaction step

To investigate the chemical properties of the Si-COF, we carried out synchrotron photoemission spectroscopy measurements after each reaction step. The clean surface of the Au(111) substrate was first ensured by the presence of well-defined spin–orbit doublet peaks of Au 4f7/2 (83.8 eV) and Au 4f5/2 separated by 3.69 eV. These energies are in excellent agreement with established numbers (Fig. 3a)48,49,50. After depositing Si atoms on a clean Au(111) surface, Au 4f doublet peaks (Fig. 3a) were significantly broadened by other components with a higher binding energy (BE) of 0.51 eV, which corresponds to the AuSix alloy (Supplementary Fig. 10)51,52,53. After formation of the Si-COF, the Au 4f spectra became comparable to those of clean Au 4f doublet peaks (Fig. 3a). As observed in the STM measurements, the dissociated Br atoms from HBTP molecules tend to react with Si atoms and consequently remove the AuSix layer from the surface by forming highly volatile SiBr4 molecules34. This phenomenon was also identified in the Br 3d spectra (Supplementary Fig. 11) where we observed a significant reduction in the signals of Br 3d after further annealing at a higher temperature.

Fig. 3: Photoemission spectroscopy measurement.
figure 3

a, Au 4f curves at each reaction step (top to bottom: clean Au(111) substrate, AuSix/Au(111) and Si-COF/Au(111) surfaces). Au 4f spectra of clean Au(111) substrate show doublet peaks of Au 4f7/2 (83.8 eV) and Au 4f5/2 separated by 3.69 eV. Au 4f doublet peaks of AuSix/Au(111) were broadened by other components with a higher binding energy (BE) of 0.51 eV. Au 4f spectra of Si-COF/Au(111) became comparable to those of clean Au(111). b, Si 2p core level spectra at each reaction step. Si 2p spectra of the AuSix/Au(111) surface show the characteristic Si 2p doublet peaks with a small separation from spin–orbit coupling of p orbitals (0.61 eV) as the Si 2p3/2 peak is located at 99.75 eV. Si 2p spectra of the Si-COF/Au(111) surface have four sets of doublet peak components after fitting, showing three different charge states of the Si atom (those components are located 0.71, 2.0 and 3.18 eV higher in BE than elemental Si, respectively). Open circles in the plots denote the Shirley-background-subtracted XPS data.

The corresponding Si 2p spectra of the AuSix layer on Au(111) show the characteristic Si 2p doublet peaks with a small separation from spin–orbit coupling of p orbitals (0.61 eV) as the Si 2p3/2 peak is located at 99.75 eV (Fig. 3b and Supplementary Fig. 12). We assume that charge transfer from Si to Au is responsible for the small shift to higher energy, compared with a value measured in the bulk (99.3 eV)54. After the synthesis of the COF, the Si 2p spectra became complex. Our best fitting is consistent with four sets of doublet peak components, which can be associated to three different charge states of the Si atom (in our case, those components are located 0.71, 2.0 and 3.18 eV higher in BE than elemental Si, respectively). Although these numbers differ slightly from those of integer charge states of Si1+ (1 eV), Si2+ (1.81 eV) and Si3+ (2.63 eV) measured in bulk inorganic form55,56, the Si-COF exhibits two major species with relatively large peak area that are located at 0.71 and 2.0 eV higher BE than the elemental Si position. Since the component shifted by 2 eV is almost comparable to a Si2+ charge state (1.81 eV), it is reasonable to assign it with the proposed COF structure (Fig. 1a,d,e) where the Si atom is covalently bonded with two carbon atoms and a bromine atom. Therefore, as both carbon and bromine have higher electronegativities than Si, a slightly higher charge state than Si2+, but less than that of Si3+, can be expected for these C4Si2Br2 linker units. We attributed the other major component that is shifted by 0.71 eV in BE to the Si atoms that are incorporated on the edges of the Si-COF and are bonded with one carbon and one Br atom (Supplementary Fig. 3). The peak at 3.18 eV is from the Si at the edges, which is bonded with one carbon and two Br atoms (Supplementary Fig. 3).

Thermal transformation of 1,4-disilabenzene to silole in a linear COF analogue

To demonstrate the generality of the Si–C bond formation by co-deposition of Si atoms and bromo-substituted molecules, we used 4,5,9,10-tetrabromopyrene (TBP) molecules, in which two groups of ortho-bromine atoms are introduced at both sides of the pyrene backbone (Fig. 4a). After TBP molecules were deposited on the partially covered AuSix layer on Au(111) held at 420 K, short oligomers appeared (Fig. 4b). The close-up view (Fig. 4c) shows that bright dots are located at the edges of the longitudinal axis. A constant height dI/dV image taken with a CO-terminated tip (Fig. 4d) and the corresponding Laplace filtered image for enhancement of bond features (Fig. 4e) show the detailed structures, in which the pyrene backbones are connected via disilabenzene. The bright dots indicated by arrows in Fig. 4c correspond to Br atoms. Comparison to the simulated structure and STM images (Fig. 4f) confirms that a Si-doped cove-edge graphene nanoribbon (Si-Cove GNR) was synthesized. As seen in Fig. 4b and Supplementary Fig. 13, the contrast of the Br atoms significantly varies. Since the tip-induced manipulation allowed iterative switches (Extended Data Fig. 6), the difference in adsorption height should be responsible for the difference in contrast, indicating a certain flexibility of the non-planar C4Si2 ring. Note that the difference of the adsorption heights of Br atoms in the C4Si2 rings between the Si-COF and Si-Cove GNR also relates to the difference of the apparent Br atom sizes (Extended Data Fig. 5). The electronic properties of Si-Cove GNR were investigated by STS measurement and DFT calculations (Extended Data Fig. 7). We determined that the occupied and unoccupied states were located at around −0.5 and + 1.1 V, respectively, resulting in a band gap of 1.6 eV.

Fig. 4: Synthesis of Si-doped GNRs.
figure 4

a, Scheme of on-surface synthesis of two types of Si-doped GNR on Au(111). b, STM topography of Si-covered Au(111) held at 420 K after depositing TBP molecules. Sample bias voltage V = 200 mV and tunnelling current I = 5 pA. c, Close-up view of Si-Cove GNR. V = 200 mV and I = 5 pA. d,e, Constant height dI/dV image of the area in c (d) and the corresponding Laplace filtered image (e). f, Simulated chemical structure of Si-Cove GNR and associated simulated STM image at 200 mV and constant charge density of 1 × 10−8 e Å−3. g, STM topography of sample after annealing at 580 K. V = 100 mV and I = 10 pA. h, Close-up view of Si-Armchair GNR. V = 100 mV and I = 20 pA. i,j, Constant height dI/dV image of the area in h (i) and the corresponding Laplace filtered image (j). k, Simulated chemical structure of Si-Armchair GNR and associated simulated STM image at 200 mV and constant charge density of 5 × 10−8 e Å−3.

After annealing at a higher temperature of 580 K, the structure of oligomers further changed as the bright dots at the edges form a zigzag arrangement along the longitudinal axis (Fig. 4g,h). The high-resolution constant-height dI/dV image (Fig. 4i) and the corresponding Laplace filtered image (Fig. 4j) show that the C4Si2 six-membered rings were transformed into the C4Si five-membered rings (siloles). The apparent length of C–Si–Br is 3.14 ± 0.20 Å (indicated in Fig. 4i). Unlike the C4Si2 rings in Si-COF, the C4Si2 six-membered rings in Si-Cove GNR are not stabilized within the network structure of COF, and thus sequential cyclization from desilicification of the disilabenzene and subsequent dehydrogenation of the pyrene backbones proceed upon thermal activation. Again, the simulated structure and STM images support our analysis (Fig. 4k). The electronic properties of Si-Armchair GNR were investigated by STS measurement and DFT calculations (Extended Data Fig. 8), in which the occupied state peak is at –0.47 V and the unoccupied state peak is at +0.95 V, resulting in a band gap of approximately 1.4 eV. Hence, the high reproducibility of the Si-incorporated COF structure is unambiguously proved.


In summary, we synthesized 1,4-disilabenzene-bridged COFs by reacting bromo-substituted HBTP molecules and Si atoms on a Au(111) surface. The linked structure of C4Si2 rings passivated by Br atoms can be determined by bond-resolved STM images combined with DFT calculations as well as XPS measurements. TBP molecules can also form the C4Si2 rings after reacting with Si atoms and these can then be transformed into C4Si rings after desilicification and dehydrogenation. These results demonstrate the high generality of the C–Si on-surface coupling by depositing Si atoms and subsequent polyaromatic hydrocarbons on Au(111). This may further extend the possibilities for syntheses of various low-dimensional nanostructures.


Synthesis of precursors

Materials: All chemicals were purchased from Sigma Aldrich or Kanto Chemicals and were used without further purification unless otherwise described. High-resolution mass spectra (HR-MS) were recorded using a Bruker mirOTOF II with an APCI II module. HBTP was purified by recrystallization from o-dichlorobenzene.

Synthesis of 4,5,9,10-tetrabromopyrene (TBP): As per a similar literature procedure57, to a 30 ml round-bottom flask, 1,2,3,6,7,8-hexahydropyrene (0.622 g, 3.0 mmol), iron (0.117 g, 2.1 mmol), dichloromethane (20 ml) and bromine (1.70 ml, 33.0 mmol) were added. The reaction mixture was stirred at 45 °C for 23 h. The precipitate was filtered off and washed with acetone and warm chloroform. Recrystallization from o-dichlorobenzene afforded white needle-shape crystals (0.391 g, 25%). HR-MS calculated for C16H6Br4: 517.7162, found: 517.7179 (Supplementary Fig. 14).

Scanning tunnelling microscopy measurement

All the experiments were conducted in a low-temperature scanning tunnelling microscopy (STM) system (home made) at 4.3 K under a high-vacuum environment (<1 × 10−10 mbar). The bias voltage was applied to the sample while the tip was electrically grounded. Au(111) surfaces were cleaned through cyclic sputtering (Ar+, 10 min) and annealing (720 K, 15 min). Si atoms were deposited on a clean Au(111) surface with an electron beam evaporator (SPECS GmbH). HBTP and TBP molecules were deposited from Knudsen cells (Kentax GmbH). The STM tip was made from chemically etched tungsten. For bond-resolved imaging, the tip apex was terminated by a small CO molecule picked up from the surface58. The bias voltage was set close to zero. The modulation amplitude was 7 mVrms and the frequency was 510 Hz.

Photoemission spectroscopy measurement

Photoemission spectroscopy measurements were conducted on the BL2B beamline at the UVSOR-III Synchrotron, which features a monochromatic light source with a photon energy ranging from 23 to 205 eV. The high-resolution Au 4f, Si 2p and Br 3d spectra were taken with a photon energy of 130 eV, measured in normal emission mode with an overall energy resolution of 0.33 eV. All spectra were processed with a Shirley-background subtraction as the binding energies are respective to the Fermi edge of the Au(111) substrate. The core level spectra fitted by GL(m) and SGL(m) functions represent the product and sum of Gaussian and Lorentzian functions, respectively. The parameter m indicates a ratio between the two functions, as m = 0 is a pure Gaussian and m = 100 is pure Lorentzian.

Theoretical calculations

All first-principles calculations on the gold substrate in this work were performed using the periodic plane-wave basis VASP code59,60 implementing spin-polarized DFT. To accurately include van der Waals interactions in this system, we used the DFT-D3 method with Becke–Johnson damping61,62— various other van der Waals functionals were tested and no significant differences were observed. Projected augmented wave potentials were used to describe the core electrons63 with a kinetic energy cutoff of 500 eV (with PREC = accurate). Systematic k-point convergence was checked for all systems with sampling chosen according to the system size. This approach converged the total energy of all the systems to the order of 1 meV. The properties of the bulk and surface of Au were carefully checked within this methodology and excellent agreement was achieved with experiments. For calculations of the assemblies on the surface, a vacuum gap of at least 1.5 nm was used. A 3 × 3 × 1 k-point grid was used and the upper three layers of Au (five layers in total) and all atoms in the assemblies were allowed to relax to a force of less than 0.01 eV Å−1. Atomic structure visualizations were made with the VMD package64. Standard simulated STM images were calculated using the CRITIC2 package65,66 based on the Tersoff−Hamann approximation67. For the high-resolution CO-tip STM images, we have made use of the FHI-AIMS code68 with the previous optimized geometry used in a single point calculation. For these calculations the Perdew–Burke–Ernzerhof exchange–correlation functional was used69 with a Γ k-point only and the standard ‘light’ basis set. The high-resolution CO-tip STM images were then computed by means of the PP-STM code with a fixed tip, where the broadening parameter η was set to 0.2 eV (ref. 70). The CO tip was approximated by 13% of the signal coming from the s orbital and 87% originating from the pxy orbitals on the probe particle. This gave a good agreement with close approach CO-STM and CO-dI/dV images71. Since for longer range CO-STM images also a 50/50 s/pxy ratio was earlier reported72, we show an additional comparison of s/pxy ratios for all the calculated voltage dI/dV images in Supplementary Fig. 9. DOS analysis was made using the VASPKIT package73.

Calculations of the nucleus-independent chemical shift were realized with the ORCA code74 at the PBE level. The def2-SZP75 basis set was used in all calculations for the lighter atoms. For Au atoms, a zeroth-order regular approximation (ZORA) scalar relativistic Hamiltonian combined with the def2-SVP basis set76 and a segmented all-electron relativistically contracted (SARC) basis set combined with the ZORA-TZVP basis set was used. An auxiliary basis set of the type def2/J was introduced to calculate the NMR chemical shifts77 in the geometric centre of each ring taken into consideration.