Doping-Induced Electron Transfer at Organic/Oxide Interfaces: Direct Evidence from Infrared Spectroscopy

Charge transfer at organic/inorganic interfaces critically influences the properties of molecular adlayers. Although for metals such charge transfers are well documented by experimental and theoretical results, in the case of semiconductors, clear and direct evidence for a transfer of electrons or holes from oxides with their typically high ionization energy is missing. Here, we present data from infrared reflection–absorption spectroscopy demonstrating that despite a high ionization energy, electrons are transferred from ZnO into a prototype strong molecular electron acceptor, hexafluoro-tetracyano-naphthoquinodimethane (F6-TCNNQ). Because there are no previous studies of this type, the interpretation of the pronounced vibrational red shifts observed in the experiment was aided by a thorough theoretical analysis using density functional theory. The calculations reveal that two mechanisms govern the pronounced vibrational band shifts of the adsorbed molecules: electron transfer into unoccupied molecular levels of the organic acceptor and also the bonding between the surface Zn atoms and the peripheral cyano groups. These combined experimental data and the theoretical analysis provide the so-far missing evidence of interfacial electron transfer from high ionization energy inorganic semiconductors to molecular acceptors and indicates that n-doping of ZnO plays a crucial role.


■ INTRODUCTION
Organic/inorganic interfaces attract substantial interest because of the prospect to exploit electronic and optical effects occurring at such heterojunctions for applications in optics, electronics, energy storage and conversion, catalysis, and sensing. 1 The often unusual behavior at such interfaces results from combining two materials with rather different properties, for example, electronic structure, dielectric constants, and polarizabilities. In general, depending on the relative energies of the rather localized molecular orbital levels and the more dispersive electronic bands in the inorganic semiconductor, interfacial charge transfer can take place, leading to the formation of electric fields at these hybrid interfaces. 2 A unique prospect for engineering the properties of devices exploiting these effects arises from the possibility to tune these charge density rearrangements by doping, in particular, that of the inorganic part.
However, even for an easy-to-realize class of organic/ inorganic interfacesorganic adlayers on inorganic sub-stratesthe nature of the charge transfer remains elusive. An interesting wide band gap semiconductor for such heterointerfaces is the metal oxide ZnO, and a number of studies have been reported recently in this context. 3−6 Here, a large work function increase was observed for the adsorption of strong electron acceptors, such as tetrafluoro-tetracyano-quinodimethane (F 4 -TCNQ). 3 A straightforward explanation for this observation would be a substantial charge transfer from the ZnO substrate to the adsorbed electron acceptor, similar to the situation on coinage metals. However, evidence for such a large transfer, for example, in the form of a filled lowest unoccupied molecular orbital (LUMO) appearing in the ZnO band gap (as indicated in Figure 1), is missing from ultraviolet photoelectron spectroscopy (UPS) experiments. 3 In theory, for a moderately doped semiconductor, band-bending effects could reduce the amount of transferred charge up to the point where the molecules would be essentially uncharged. 3,7  In this contribution, we challenge the interpretation of basically neutral molecular adsorbates. Our data from infrared reflection−absorption spectroscopy (IRRAS) directly reveal that hexafluoro-tetracyano-naphthoquinodimethane (F 6 -TCNNQ), a common p-dopant for organic semiconductors, 8−10 becomes strongly negatively charged upon adsorption.
For this purpose, F 6 -TCNNQ was deposited on a structurally well-defined ZnO(101̅ 0) surface. IRRAS allows detecting charge transfer across the interface by measuring adsorptioninduced changes of molecular vibrational frequencies. For metal substrates, IRRAS has already been applied successfully to evidence interfacial charge transfer by detecting frequency shifts of vibrations in molecular adsorbates. 11−14 For PTCDA, another common, planar aromatic compound, adsorbed on Ag(111), the changes in frequency were found to amount to 10−30 cm −1 . 11 However, similar studies for molecular adsorbates on wide band gap inorganic semiconductors have not yet been reported. This is mainly due to severe challenges in measuring infrared (IR) spectra in reflection for nonmetallic substrates. Only more recently, these problems have been overcome by a number of groups (for a review, see ref 15).

■ EXPERIMENTAL SECTION
The experiments were conducted in a state-of-the-art ultrahigh vacuum (UHV) multichamber system (Prevac). 15 The apparatus allows performing IR spectroscopy in combination with other surface sensitive techniques such as X-ray-photoelectron spectroscopy (XPS) and low-energy electron diffraction (LEED). The IR spectra were obtained in the reflection− absorption mode (IRRAS) using a Fourier-transform (FT)-IR spectrometer (Bruker VERTEX 80v) directly coupled to the UHV chamber via differentially pumped KBr-windows. The base pressure of the entire chamber is in the 10 −10 mbar range.
As substrates, one nominally undoped ZnO(101̅ 0) single crystal (CrysTec) and two Ga-doped ZnO(101̅ 0) thin films grown on the bulk ZnO(101̅ 0) single crystal by molecular beam epitaxy were used. In the latter case, the charge carrier density was determined by analyzing far-IR transmission spectra in the context of the Drude model, 16 yielding n-doping levels of 7.8 × 10 19 and 2.9 × 10 20 cm −3 , respectively.
The substrates were cleaned by repeated cycles of Ar + sputtering (1 × 10 −6 mbar, 10 min) and annealing for 10 min at 800 K in a case of the undoped sample and at 700 K for Gadoped ZnO. As a final step, the samples were annealed in an oxygen atmosphere (1 × 10 −6 mbar) at 840 and 700 K for undoped and Ga-doped samples, respectively. The lower annealing temperature was applied for Ga-doped ZnO to keep the doping level unchanged. The cleanliness and structural quality of the substrate surface were monitored by XPS and LEED.
Deposition of F 6 -TCNNQ (Novaled) on ZnO(101̅ 0) was carried out by directing the flux of an effusion cell toward the substrate. Deposition rates were determined using a quartz crystal microbalance. Based on the determined deposition rate, we estimate that the coverage of the monolayers used to record the data shown in Figure 2 amounted to 0.7 ± 0.3 monolayer, that is slightly less than a full monolayer (one monolayer corresponds to a surface fully covered by flat-lying molecules). For larger coverage (i.e., longer deposition times), bilayer formation was observed in the IR data. According to the calculated adsorption geometry (see Figure SI1, Supporting Information), monolayer coverage corresponds to ∼4 × 10 13 molecule per cm 2 . This value is in good agreement with the estimates made by Katayama 17 for F 4 -TCNQ considering the larger size of the F 6 -TCNNQ molecule. For the F 6 -TCNNQ multilayer, the deposition time was 10 times longer; therefore, we estimate a film thickness of ∼70 Å.
All IRRAS experiments were carried out at room temperature. An internal polarizer module allowed to record IR spectra with sand p-polarized light. All IR spectra shown here are difference spectra obtained by subtracting a spectrum recorded for the clean ZnO substrate before F 6 -TCNNQ deposition from that measured after molecule deposition. Note that in contrast to metal substrates, the bands in the IRRAS spectra can have different signs, positive or negative. For a detailed discussion, see the previous literature. 15,18 The angle of incidence of the IR light was set to 80°with respect to the surface normal because according to previous spectrum simulations for adsorbates on ZnO, 18 the p-polarized signal adopts a maximum at 80−85°, while for s-polarized light, the dependence on Θ is only weak. The Journal of Physical Chemistry C pubs.acs.org/JPCC Article

■ COMPUTATIONAL DETAILS
As mentioned in the introduction, obtaining a reliable prediction of the charge state of the adsorbed molecule via semi-local density functional theory (DFT) is challenging. This is because the LUMO is spuriously below the valence band of ZnO, which yields superficial charge transfer, independent of the substrate's actual Fermi level. In principle, this problem can be solved using more sophisticated methods, such as DFT + U 19 or hybrid DFT. 20 However, in either case, the amount of charge transfer depends sensitively on the parameter used to account for the Coulomb on-site energy (the Hubbard U or the amount of exact exchange) and on the precise doping concentration (and its nature), 19,20 which still prevents a stand-alone computational prediction. Therefore, we pursue a different approach here: rather than calculating the charge state directly, the aim of the calculations is to predict the shift of vibrational frequencies as a consequence of (a) charging of the F 6 -TCNNQ molecule and (b) binding of the molecule to the surface. We do this separately by calculating the shift due to charging for a free molecule in the gas phase and by calculating the vibrational frequencies for the molecule on the surface (where, in our calculations, we assume a charge transfer of approximately one electron). Assuming that these two effects are independent and additive, we can combine them to estimate the vibrational shift for arbitrarily charged molecules on the surface, as explained in more detail in the Supporting Information. Comparing the frequencies observed in the IRRAS experiment to the computed values then allows us to extract the factual charge state of the molecule on the surface. All computed vibrational modes presented here were obtained by DFT calculations using the FHI-aims program package. 21 The Perdew−Burke−Ernzerhof (PBE) 22 density functional was used and van der Waals forces were included with the vdW-TS scheme, 23 using the appropriate parameters for the ZnO substrate. 24 The vibrational modes were computed numerically via the finite displacement method using a displacement amplitude of 0.005 Å. Initial geometries were optimized until residual forces were smaller than 10 −4 eV/Å with "tight" basis sets and numerical settings.
Calculations for the free molecule were performed with open boundary conditions. The electronic structure of the F 6 -TCNNQ/ZnO interface was determined using a periodic slab approach. The ZnO substrate was modeled using a five doublelayer slab cell. For geometry optimization, the bottom three double-layers were fixed at their bulk positions and the top two layers were allowed to relax. For the vibrational analysis, we restrict the discussion to vibrations of the molecule within the harmonic approximation. By keeping the substrate fixed, we omit changes in pure surface and pure bulk modes. To avoid artifacts arising from artificial surface states located at the bottom side of the slab, the bottom layer was passivated with hydrogen atoms. Calculations for the primitive surface unit cell are performed with a 16 × 24 × 1 k-grid and scaled accordingly for larger cells. The 4 × 4 supercell including the molecule was chosen such that there is no significant intermolecular interaction between the adjacent molecular adsorbates when employing periodic boundary conditions.
The adsorption of the F 6 -TCNNQ molecules on the ZnO(101̅ 0) surface is governed by weak binding of the protruding cyano groups to the surface Zn atoms, which act as docking sites, leading to a planar molecular adsorption geometry. Two different azimuthal orientations were considered: oriented along the rills of the mixed-terminated ZnO surface (i.e., along the ZnO[12̅ 10] direction) and oriented perpendicular to the rills (the ZnO[0001] direction). The orientation perpendicular to the rills is energetically more favorable by 33 meV and was therefore used for the vibrational analysis in this work.
Vibrational frequencies usually strongly depend not only on the physics of the system but also on the density functional and the basis set used in the DFT calculations. Because of the systematic nature of the deviations, it is customary to apply uniform multiplicative scale factors to correct the computed vibrational frequencies. 25−27 The scale factor used here was determined from experimental and theoretical data for F 4 -TCNQ. 28

■ RESULTS AND DISCUSSION
To determine the charge state of the molecules in direct contact with the substrate, it is useful to first characterize a molecular multilayer, where the molecules are clearly charge neutral. This provides a baseline for comparing the vibrational signatures of the experiment and theory, allowing us to analyze shifts due to charging and binding to the surface afterward. In the following, we focus on the results for nominally undoped ZnO(101̅ 0), which is actually unintentionally n-type doped; 31,32 and the results are qualitatively equivalent to those obtained for highly The inspection of the IRRAS data from multilayer F 6 -TCNNQ (Figure 2, top curves) reveals several well-defined, sharp vibrational bands. In the frequency regime below 1700 cm −1 (CC/C−C region), four strong bands are seen in the experimental multilayer data, centered at 1639, 1552, 1460, and 1396 cm −1 . In addition, a small peak is seen at 1424 cm −1 . For their assignment, we rely on the comparison to the corresponding data for a structurally related molecule, F 4 -TCNQ, 28−30 and to the results of DFT calculations for F 6 -TCNNQ in the gas phase (see Table 1).
Overall, the agreement between the positions of the bands in the experimental multilayer IR spectra and the DFT results for the free molecule is very good. Note that the small deviation between the experiment and theory is similar to that reported in previous work for F 4 -TCNQ multilayers. 29 On the basis of this comparison, in Table 1 (column "theory, gas phase, neutral"), we assign the four strong bands to the C−C vibrations b u ν 51 , b u ν 52 , b u ν 53 , and b u ν 55 and the weak feature to b u ν 54 . In the C N stretch regime, we observe four features at 2224, 2215, 2205, and 2194 cm −1 (Figure 2b, top curve). We assign the band at 2224 cm −1 to the degenerate b u ν 49 /a g ν 1 vibrations and the band at 2215 cm −1 to the degenerate b u ν 50 /a g ν 2 vibrations (see Table  1). This assignment is in analogy with that reported for F 4 -TCNQ multilayers. 29 Two features at lower wavenumbers are ascribed to signals still originating from the monolayer, see below. We would like to note that for F 4 -TCNQ/(co)polymer blends, an additional strong band at 2169 cm −1 has been found. 29 We were unable to observe such vibration band in our data, and all features seen in Figure 2b in the range 2180−2100 cm −1 are below the noise level.
Unfortunately, the bulk crystal structure of F 6 -TCNNQ is unknown. Also, in the case of thin films, Duva et al. 33 were unable to deduce structural parameters because out-of-plane diffraction peaks for F 6 -TCNNQ films could not be detected. However, we note that all vibrational modes discussed above have their transition dipole moments oriented within the molecular plane of F 6 -TCNNQ. Note that the experimental ppolarized multilayer data (Figure 2) do not reveal any vibrational bands with a negative signal (for a discussion of the interplay between orientation of transition dipole moments, the polarization of the incident IR-light, and the sign of vibrational bands in IRRAS at dielectric surfaces, see the Supporting Information). This observation implies that all F 6 -TCNNQ molecules must be orientated with their molecular plane parallel to the ZnO substrate, see Figure 3. IRRAS data recorded with s-polarization (see Figure SI11 in the Supporting Information) are fully consistent with parallel stacking in multilayers of this aromatic compound.
For the F 6 -TCNNQ monolayer on ZnO(101̅ 0), the IRRAS data reveal pronounced differences compared to the multilayer spectra (see Figure 2a, bottom curve). For the monolayer, the most pronounced peak is that at 1507 cm −1 . Lower intensity bands are observed at 1630, 1450, 1410, and 1389 cm −1 , and a small feature is observed at 1545 cm −1 .
In case of a neutral F 6 -TCNNQ adsorbate, the DFT results predict a small blue shift (15 cm −1 ) of the strong b u ν 52 band at 1552 cm −1 relative to the molecule in the gas phase because of the bonding of the molecule to the surface. 34 Because experimentally, the peak actually shifts in the opposite direction, we conclude that the presence of neutral F 6 -TCNNQ molecules at the interface is unlikely. Conversely, for a negatively charged molecular species, the calculations predict a red shift of 30 cm −1 for the b u ν 52 band (see Figure SI5 in the Supporting Information), in agreement with the experimentally observed shift of 45 cm −1 . Such a substantial lowering of frequencies is typical for electron transfer to this organic acceptor. 30 In line with this hypothesis, the second sharp band for the monolayer observed at 1450 cm −1 can be assigned to the b u ν 53 band (1460 cm −1 in the multilayer), that is, a red shift of 10 cm −1 is observed for the monolayer. Also, in this case, the DFT results for the neutral species predict a small blue shift, whereas for the charged species, a small red shift is expected (5 cm −1 ). Inspection of Table 1 confirms that the other vibrations also agree well with the theoretical results for a F 6 -TCNNQ anion at the interface with ZnO.
Assuming a charge transfer of 1 electron/molecule for the full monolayer, ∼10 14 electrons for the transfer from the substrate to F 6 -TCNNQ are needed. This value is absolutely reasonable for the n-doping levels present in the substrate.
In the CN-stretch region, the experimental monolayer data reveal only one rather broad feature (2212−2188 cm −1 ) centered at 2205 cm −1 (Figure 2b, bottom curve). The center of this band is red-shifted by 10−20 cm −1 with respect to the strongest multilayer peaks at 2215 cm −1 and 2224 cm −1 . Although the lower intensity of the CN bands in the experimental data prohibits a more detailed analysis, the red shift in the measured monolayer data is consistent with the theoretical results for the negatively charged species, where for all CN stretch vibrations, red-shifts between 40 and 60 cm −1 are found.
Analysis of the experimental data confirms the plane of the F 6 -TCNNQ monolayer to be orientated parallel to the substrate, as also indicated by the theoretical results.
Overall, our findings for the nominally undoped ZnO substrate clearly show that the adsorbate is negatively charged, in variance to the naive expectation depicted in Figure 1b. Importantly, this is in agreement with the observation that the measured band bending for semiconductor/organic interfaces is significantly smaller than that required to obtain charge-neutral adsorbates. 3,35 Notably, although theory suggests that the amount of charge transfer should be dependent on the doping concentration, 3,20 experiments for more strongly doped samples showed no difference to the nominally undoped sample. The Journal of Physical Chemistry C pubs.acs.org/JPCC Article As a final remark, we note that our experimental and theoretical findings cannot reconcile the apparent absence of filled molecular states in previous ultraviolet photoelectron spectroscopy experiments. Possible origins include strong charge localization in the molecular layer, which could shift the filled states far below E F , 36,37 or a pronounced hybridization of the surface states with the molecular orbitals, which could also lead to masking of ebbing molecular states by the ZnO valence band. However, a conclusive answer would require significant improvements over state-of-the-art theory, which is outside the scope of the present contribution.

■ CONCLUSIONS
The experimental IRRAS data obtained for F 6 -TCNNQ molecules deposited on a ZnO(101̅ 0) surface reveal the formation of multilayers with the molecular plane orientated strictly parallel to the substrate. In the CN region for the F 6 -TCNNQ monolayer, two sharp vibrational bands are observed, with pronounced red shifts relative to the multilayer. As confirmed by DFT results, these shifts are inconsistent with a neutral adsorbate. Instead, from the good agreement with theoretical results obtained for a molecular anion, we conclude that F 6 -TCNNQ molecules in the monolayer are negatively charged. This information provides the so-far missing evidence of interfacial electron transfer from high ionization energy inorganic semiconductors to molecular acceptors. In this case, the transfer is made possible by n-doping. Consequently, the comprehensive understanding of charge rearrangements at inorganic/organic interfaces is substantially ameliorated and can now be exploited to further tailor such interfaces for advanced applications.
Top and side view of the theoretically found lowest energy local adsorption site of F 6 -TCNNQ on the ZnO surface; MODOS and orbital population analysis of F 6 -TCNNQ on ZnO; most relevant molecular orbitals for charge transfer upon adsorption of F 6 -TCNNQ on ZnO, two most prominent features that appear in gas-phase calculations; depiction of CC/C−C and CN frequency shifts of flat-laying F 6 -TCNNQ; theoretical results of C−C/CC stretching vibrations; theoretical results of CN stretching vibrations; chemical structure and double bond locations within the molecule; bond lengths of all bonds within the F 6 -TCNNQ molecule; and experimental IRRAS spectra of F 6