Synthesis and Local Probe Gating of a Monolayer Metal‐Organic Framework

Achieving large‐area uniform 2D metal‐organic frameworks (MOFs) and controlling their electronic properties on inert surfaces is a big step toward future applications in electronic devices. Here a 2D monolayer Cu‐dicyanoanthracene MOF with long‐range order is successfully fabricated on an epitaxial graphene surface. Its structural and electronic properties are studied by low‐temperature scanning tunneling microscopy and spectroscopy complemented by density‐functional theory calculations. Access to multiple molecular charge states in the 2D MOF is demonstrated using tip‐induced local electric fields. It is expected that a similar strategy could be applied to fabricate and characterize 2D MOFs with exotic, engineered electronic states.


Introduction
Metal-organic frameworks (MOFs) are an important class of materials that have been intensively studied in the last two decades. Despite the vast number of reports on 3D, bulk MOFs, synthesis and characterization of 2D, single layer MOFs are much more limited. [1][2][3] Intrinsic 2D MOFs are expected to attract increasing attention since they are anticipated to possess exotic electronic properties, such as high electrical conductivity, [4][5][6][7][8] superconductivity, [9,10] topologically non-trivial band structure, [11][12][13][14][15][16][17] halfmetallic ferromagnetism, [18][19][20][21] and quantum spin liquids. [22] To isolate their intrinsic electronic properties from the substrate, synthesis of 2D MOFs on inert surfaces, such as graphene, other van der Waals layered materials, and bulk insulators, is highly desired. Furthermore, understanding the performance of 2D (details given in the Experimental Section). By tuning the DCA:Cu ratio, both the DCA 3 Cu single complex (Figures S1 and S11, Supporting Information) and the DCA 3 Cu 2 honeycomb network can be fabricated. The initial sample quality can be improved by annealing the sample at 50 °C. This helps to grow larger DCA 3 Cu 2 networks up to a full monolayer, as shown in Figures 1 and S2a, Supporting Information. The proper annealing temperature is vital for the formation of a uniform monolayer 2D MOF, since the coordination bonds are relatively weak and reversible. [32,33] The networks are structurally robust, which is evidenced by the fact that they grow seamlessly across step edges of the underlying Ir(111) substrate (see Figure 1b). Annealing at a higher temperature (>70°C) yields a new phase of close-packed DCA molecules and large Cu islands (see Figure S2b, Supporting Information). Figure 1c shows a high-resolution image of the backbone of DCA 3 Cu 2 network, with the unit cell shown as a white parallelogram with a lattice constant of a = 1.98 ± 0.03 nm, which is consistent with the DFT value of 1.98 nm and in the range of previous reports of the network on a Cu(111) surface. [35,36] While the Cu atoms arrange in a honeycomb lattice in the network, the DCA molecules form a kagome pattern, as shown in the model of Figure 1e. Figure 1d shows the same area at a bias voltage corresponding to the low energy electronic band of the Cu-DCA network (see below for more detailed spectroscopy of the Cu-DCA network electronic structure). The image is acquired with a molecule-modified tip apex, which enhances the spatial resolution of the local density of states (LDOS). [37] The experimental image is nicely reproduced by the STM image simulations based on DFT calculations of the Cu-DCA structure shown in Figure 1f (see below for details). The modest contrast difference between the DCA molecules in Figure 1d is caused by the moiré pattern of graphene on Ir(111) (confer Figure S3, Supporting Information). Figure 2 shows the dI/dV spectra recorded on different high symmetry sites of the network. All the spectra in Figure 2b exhibit a broad peak in the energy range between 0 to 0.5 V. The contrast in constant height dI/dV maps in Figure 2c is not strongly bias dependent in the range from 0.1 to 0.4 V and reproduced well by the DFT simulations ( Figure S9, Supporting Information). We attribute these features to the band structure formed in the 2D network, which has been well studied in similar DCA 3 Co 2 network [28] and will be explored in more detail in Figure 3. Interestingly, the STS measured on the center of the DCA molecule (black curve in Figure 2b) shows two sharp dips around −0.6 and −1.2 V; the STS on top of the Cu atom (green curve in Figure 2b) shows a small dip around −0.6 V and a sharp peak around −1.2 V; the end of the long axis of the DCA molecule (blue curve in Figure 2b) shows a sharp peak around −0.6 V and a tiny dip around −1.2 V. The peaks/dips at these two bias values are attributed to the typical charging features, where the charge state of the molecule under the tip changes due to the tip-induced local electric field. [38][39][40][41][42][43][44][45][46][47][48][49][50] We will discuss the details of these charging features in Figure 4. The spectra of the MOF on the step edge of the underlying Ir(111) substrate ( Figure S12, Supporting Information) is consistent with the one on the flat area ( Figure 2), indicating that the electronic properties of the MOF are effectively decoupled from the metal substrate by the graphene layer. Besides, all the spectra show a small and sharp peak at or very close to the Fermi level (0-10 mV), which we attribute as a charging peak as well (confer Figure S4, Supporting Information).
The DFT simulated band structure of the gas-phase DCA 3 Cu 2 network ( Figure 3a) shows a kagome band structure around Fermi level which consists of a Dirac band with an  [51][52][53] The band structure of the DCA 3 Cu 2 network on graphene (the MOF states represented in purple in Figure 3b) shows very similar features with an additional avoided crossing between the MOF kagome and the graphene Dirac bands due to weak hybridization between them. While the Dirac points of the isolated DCA 3 Cu 2 network and graphene are both located at the Fermi level in the gas phase, the Dirac point in the DCA 3 Cu 2 Figure 2. a,b) STS recorded (b) on DCA 3 Cu 2 network at the positions shown in (a), the spectra were vertically offset for clarity, the spectrum in black has been magnified by a factor of 3 to highlight the weak features. Imaging parameters: 1.5 V and 10 pA. Scale bar: 1 nm. c-f) Experimentally recorded constant-height dI/dV maps at the energies indicated in the panels in the same area of (a). kagome band on graphene is 0.3 eV above the Dirac point of the graphene substrate, indicating a charge transfer between the DCA 3 Cu 2 network and the graphene substrate (confer Figure S7, Supporting Information). This is consistent with the experimental finding that the energy levels of the DCA 3 Cu 2 kagome band are mostly above the Fermi level. Note that in a clean G/Ir(111) sample, the Dirac point of graphene is 100 meV above the Fermi level as well. [54] However, the experimental dI/dV spectra did not resolve the bandgap of the MOF at the Dirac point. This is mainly due to a certain lifetime broadening and the satellite vibronic peaks coming from the intermolecular electronic coupling which broadens the spectra.
The projected density of states (PDOS) in Figure 3 show that the kagome band originates mainly from the DCA molecule (p z orbitals of C and N atoms), with very minor contributions from the Cu d orbitals (d xz + d yz ), indicating that the kagome geometry of the DCA molecular array and the π-d extended conjugation in the metal-organic framework is realized in this band structure. [4,6,8,55,56] The DFT simulated LDOS maps show uniform features at different energies within the kagome band ( Figures S5a-d). To probe the electronic structure of the DCA 3 Cu 2 network in more detail, we used a molecule-modified p-wave tip to get the high-resolution LDOS maps [57] shown in Figures S5e-h. These show homogeneous appearances which are similar at different biases as in the simulated results. The representative experimental and DFT simulated LDOS maps are shown in Figures 3c and 3d, respectively. The experimental maps match the representative DFT simulated data very well. The minor contrast difference comes from the moiré pattern and slight inhomogeneities of the graphene substrate, which is also reflected in the charging rings shown in Figure 4.
Charging behavior has been studied in detail in the case of single molecules or self-assembled molecular monolayers on coinage metal surfaces such as Au(111) [42] and Ag(111) [47,48] and more frequently found when the molecule is decoupled from the substrate by an ultra-thin film such as Al 2 O 3 , [38][39][40][41] graphene [46,49] and hexagonal boron nitride. [43][44][45]50] It can be explained by considering the tip-molecule-substrate system as a doublebarrier tunnel junction (DBTJ) as illustrated in Figure 4a,b. When a bias voltage is applied across the DBTJ, there is a potential distribution with a drop at both the tip-molecule and the molecule-substrate junctions. Consequently, applying a bias voltage V b causes the molecular levels to shift in energy by αV b , where α is the fraction of the potential drop between the molecule and the substrate with respect to the overall bias. If there are molecular energy levels close to the Fermi level, these can shift across it at some value of the bias voltage and the charge state of the molecule changes by ±e. In our particular case, the charging occurs once the band bottom of the kagome band (KB) shifts down below the Fermi level: The charging features can be distinguished from the usual molecular resonances by checking how the charging peak/dip shifts as a function of the tip-molecule distance. Bringing the tip closer to the molecule increases α and, consequently, the charging features shift toward the Fermi level monotonously with decreasing tip-molecule distance (shown in Figure 4c), which is consistent with previous studies. [38,48,49] The charging features shift by about 8.8 mV Å −1 upon reducing the tip-sample distance. On the other hand, the kagome band position at the positive bias barely changes with different tipmolecule distances, which can also be seen from Figure 4c.
Another fingerprint of a charging phenomenon is the charging ring feature. As shown in Figures 4d-j, the elliptical rings observed in the LDOS maps at different bias voltages represent the onset of charging as the tip is moved toward the charging site. Notably, in the previous experiments, both the charging peaks [39][40][41]46,49] and dips [42,47,48] have been found at negative bias. The presence of a peak or dip in the dI/dV spectrum depends on whether the LDOS at the tip position increases or decreases due to the charging event, which might also depend on the local adsorption registry between the MOF and the underlying graphene substrate. The charging peak/dip rings features are around two different bias voltages: one around −1.2 V (Figure 4e-g) and the other −0.6 V (Figures 4h-j). The energy difference of the two charging rings reflects the Coulomb charging energy (≈0.6 eV). The peak and dip features can either coexist around a specific molecule or one of them can be dominant. Around both biases, the charging ring perimeter decreases as the bias becomes less negative, which is also consistent with the previous works. Figure 4d highlights all the possible charging rings with dotted purple ellipses. However, the charging ring positions and perimeters in the experimental data are not entirely identical  (Figure 4e-j). Each molecule has slightly varying adsorption environment (e.g., due to the moiré pattern on graphene on Ir(111) which is known to give rise to a work function modulation of a couple of hundred meV [58][59][60][61] ), varying the exact on-set bias of the charging. [42,45,48,50] The elliptical rings are mostly around the DCA molecules, which is consistent with the DFT results that the band above the Fermi level (which is pulled below E F at negative bias) has most of its density on the DCA molecules. These local charging features demonstrate that electron-electron interactions (characterized by the Coulomb charging energy, here ≈0.6 eV) are significant and they can be expected to be of a similar magnitude compared to the overall band width of the Cu-DCA network (here several hundreds of meV). Cu-DCA has been predicted to be an intrinsic 2D topological insulator in a non-interacting model, [31] but our results indicate the need to go beyond this simple picture and consider electron-electron interactions.

Conclusion
In summary, we study the structural and electronic properties of monolayer Cu-DCA MOF on a G/Ir(111) substrate under UHV conditions using experimental (STM/STS) and theoretical (DFT) methods. We demonstrate successful synthesis of a large-scale monolayer MOF that can grow across the terrace of the graphene substrate. The 2D Cu-DCA MOF possesses a kagome band structure near the Fermi level. We expect that a similar strategy could be applied to fabricate and characterize 2D MOFs with exotic electronic states on weakly interacting substrates, for example, 2D MOFs with heavy metal atoms [62][63][64][65][66][67][68][69] possessing strong spin-orbit couplings and the possibility of realizing an organic topological insulator. [70] In addition, multiple molecular charge states are observed and modified by the tip-induced local electric fields. This highlights the role of electron-electron interactions that are likely to be of a similar order of magnitude as the overall band width. Depending on the relative magnitudes of the different energy scales, this can give rise to magnetically ordered or spin liquid ground states or-when coupled with spin-orbit interactions-result in a quantum anomalous Hall insulator or more exotic electronic states. [71][72][73][74]

Experimental Section
Sample preparation and STM experiments were carried out in an ultrahigh vacuum system with a base pressure of ≈10 −10 mbar. The Ir(111) single crystal sample was cleaned by repeated cycles of Ne+ sputtering at 2 kV and annealing in an oxygen environment at 900 °C followed by flashing to 1300 °C. Graphene was grown by adsorbing ethylene and flashing the sample to 1100-1300 °C in a TPG (temperature programmed growth) step followed by a CVD step where the Ir(111) substrate at 1100-1300 °C is exposed to ethylene gas at 5 × 10 −7 mbar pressure for 1 min. [75] This gives approximately a full monolayer coverage of graphene (G/Ir(111)).
The DCA 3 Cu single complex and DCA 3 Cu 2 network can be fabricated by the sequential deposition of 9,10-dicyanoanthracene (DCA, Sigma Aldrich) molecules and Cu atoms onto the G/Ir(111) substrate held at room temperature. Further annealing the sample at 50 °C results in DCA 3 Cu 2 network growth until monolayer coverage is reached. DCA molecules were thermally evaporated from a resistively heated aluminum oxide crucible at 100 °C. Subsequently, the samples were inserted into a low-temperature STM (Createc GmbH), and all subsequent experiments were performed at T = 5 K. STM images were taken in the constant current mode. dI/dV spectra were recorded by standard lock-in detection while sweeping the sample bias in an open feedback loop configuration, with a peak-topeak bias modulation of 15-20 mV at a frequency of 526 Hz. In the local density of states (LDOS) maps, each dI/dV spectrum was normalized by I/V spectra to minimize the height difference contribution. [76,77] The STM images were processed by Gwyddion software. [78] The DFT calculations were performed with the QUANTUM-ESPRESSO distribution. [79] The optB86b-vdW functional [80,81] was used to optimize the structure of the DCA 3 Cu 2 MOF both as an isolated layer and on graphene. To describe the interaction between electrons and ions PAW pseudopotentials were used, [82] while the electronic wave functions were expanded considering a plane-wave basis set with kinetic energy cutoffs of 90 Ry. Integrations over the Brillouin zone (BZ) were performed using a uniform grid of 4 × 4 × 1 k-points, and a twice denser grid was used to obtain band structures and PDOS. The electronic structure for the STM simulations and LDOS maps were calculated via FHI-AIMS package [83] from the previously optimized geometry as a single point calculation. For this calculations the Perdew-Burke-Ernzerhof exchangecorrelation functional [84] was used with Γ k-point only. The s and p xy wave STM simulations and LDOS maps were then computed by means of the PP-STM code with fixed tip, where the broadening parameter η was set to 0.3 eV. [85] The fixed tip simulations fingerprint only the electronic structure of the sample in the following manner: The s-tip dI/dV signal is given by: LDOS E r , [57,85] The normalized LDOS maps have been simulated via plotting isosurface of the corresponding dI/dV signal.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.