Hervé
Dekkiche
a,
Andrea
Gemma
b,
Fatemeh
Tabatabaei
c,
Andrei S.
Batsanov
a,
Thomas
Niehaus
*c,
Bernd
Gotsmann
*b and
Martin R.
Bryce
*a
aDepartment of Chemistry, Durham University, Durham, DH1 3LE, UK. E-mail: m.r.bryce@durham.ac.uk
bIBM Research−Zurich, Rueschlikon 8803, Switzerland. E-mail: bgo@zurich.ibm.com
cUniversité Claude Bernard Lyon 1, CNRS, Institut Lumière Matière, F-69622, Villeurbanne, France. E-mail: thomas.niehaus@univ-lyon1.fr
First published on 9th September 2020
We report the synthesis and the single-molecule transport properties of three new oligo(phenyleneethynylene) (OPE3) derivatives possessing terminal dihydrobenzo[b]thiophene (DHBT) anchoring groups and various core substituents (phenylene, 2,5-dimethoxyphenylene and 9,10-anthracenyl). Their electronic conductance and their Seebeck coefficient have been determined using scanning tunneling microscopy-based break junction (STM-BJ) experiments between gold electrodes. The transport properties of the molecular junctions have been modelled using DFT-based computational methods which reveal a specific binding of the sulfur atom of the DHBT anchor to the electrodes. The experimentally determined Seebeck coefficient varies between −7.9 and −11.4 μV K−1 in the series and the negative sign is consistent with charge transport through the LUMO levels of the molecules.
The energy conversion efficiency of a specific thermoelectric material is determined by the figure of merit ZT given by the ratio S2σelectric/σthermal, where S is the Seebeck coefficient (or thermopower), σelelectric is the electrical conductance, and σthermal is the thermal conductance. For energy harvesting, a value of ZT ≥ 1 is required to be technologically interesting. For cooling applications, a large power factor, S2σelectric, is needed. To date the best intrinsic efficiency has been obtained for bulk materials, typically metallic (nano)composites.11,12 However the drawback of these materials is that they are usually quite expensive and/or necessitate complicated manufacturing processes and therefore cannot easily be employed in small-scale junctions. Recently, the idea of using single molecules in thermoelectric junctions, and more specifically organic molecular wires, has emerged.13 These molecules possess an appropriately conductive delocalised π-electronic system, they can be produced on a large scale and their properties and dimensions can be finely tuned by chemical synthesis. Importantly, high Seebeck coefficient values have been predicted for some of them, such as −56 μV K−1 for a fullerene pair14 or the impressive range of −280 μV K−1 to 230 μV K−1 for selected metalloporphyrins.15 Therefore, organic molecules are perfect candidates for applications in thermoelectric devices.16
The electrical conductance of single molecules is now well understood and can be measured by using various techniques such as Scanning Tunneling Microscopy (STM) based or mechanically controlled (MC) break-junction (BJ) experiments.17 On the other hand, the experimental18 and theoretical characterisation19 of thermal conductivity in those systems is still in its infancy and only recently thermal conductance of single molecules has been measured experimentally.6,7 The measurement of the Seebeck coefficient involves performing BJ experiments at variable temperature bias between the two metal electrodes.4,18
The main issue of thermoelectric materials for the use in thermoelectric converters is the low energy conversion efficiency. The requirements for high efficiency, i.e. a high value of ZT, need careful tuning of the material to optimize the mutually related properties: electrical, S and thermal. In addition, there is a need for stable molecular films with high integrity. A prerequisite is the availability of a stable and versatile molecular platform. Such a platform would have favourable anchor groups and the ability to host side-groups. Ideally, these groups would not interfere with the electrical transport and therefore allow reduction of phonon transport.20 As a first step in this direction, there is need to demonstrate that thermoelectric transport through a promising molecular backbone can be maintained despite chemical modification through anchor points for side-group chemistry.
The present work explores the single-molecule conductance and thermoelectric properties of new oligo(phenyleneethynylene) (OPE3) derivatives (3 denotes the number of phenylene rings in the backbone) using STM-BJ experiments. Extensive computational calculations using DFT and DFTB were then performed to support the results obtained. OPE3 derivatives were chosen as they are synthetically-versatile, highly-conjugated, robust molecules and they have been widely studied in molecular electronics.21,22 OPE3 systems, in particular, can be conveniently assembled and measured in metal–molecule–metal junctions.23–25 The wide agreement between conductance data obtained for OPE3 junctions in different research groups, makes this an excellent candidate for a systematic study on the influence of side and end-groups. We chose the very stable dihydrobenzo[b]thiophene (DHBT) anchor as this group is known to bind efficiently to metal electrodes, with high junction formation probability,26–28 and without the problems associated with thiol anchors, such as oxidation and formation of metal-thiolate clusters.29 We demonstrate that these molecules possess the highly versatile structural features that are required for thermal conductance studies.
To combine electrical and thermoelectrical measurements the tip of the STM was equipped with two platinum resistors, one acting as heating element and the other acting as a thermometer. The temperature of the system was maintained at the desired value using a temperature controller (Meerstetter TEC 10–91) with an accuracy and stability better than 0.01 °C. All the measurements were performed in vacuum (≃10−7 mbar) and at room temperature (22 °C) with a custom-built STM, located in a low-noise laboratory.33
Electronic transport properties of the OPE3 derivatives were computed using Non-Equilibrium Green's Function Theory based on the Density Functional Based Tight-Binding Method (DFTB)38–40 as implemented in DFTB+ version 19.1.41 DFTB is an approximate DFT method with an appealing cost/accuracy ratio and has been successfully used in a variety of applications in the field of molecular electronics.42–44 Here we employed the auorg-0-1 Slater–Koster set45,46 with orbital dependent Hubbard parameters and used a periodic setup, where the device is replicated perpendicular to the transport direction along the surface. This entails a solution of the Poisson equation under periodic boundary conditions to obtain the charge density in the device region. The actual device model was created in several steps. We first optimised the target molecule attached to two Au20 clusters in the gas phase using the atomic orbital NWCHEM code.47 The calculations were performed with the PBE functional and a 6-31G* basis set for C, H, S and a lanl2dz_ecp basis set and corresponding core potentials for Au. The defaults for energy and force convergence were kept. With the help of the simulation environment ASE,48 the Au20–molecule–Au20 complex was then embedded between two identical Au(111) surfaces, each containing three atomic layers with 30 atoms per layer, to form a full device as shown later in the manuscript in the theory section. Note that in this process the Au–S–C binding angles (found in the gas-phase relaxations) were conserved, such that the molecular tilting with respect to the surface was correctly captured. This device was then coupled to gold electrodes, consisting of three additional gold layers that were periodically replicated in the transport direction.
Single crystals suitable for X-ray diffraction analysis were obtained for OPE3-Ph and OPE3-An by slow exchange of methanol and hexane vapours respectively into a CH2Cl2 solution of the compounds. OPE3-Ph lies at a crystallographic inversion centre. Thus, arene ring A and its axis X are parallel to their equivalents (A′ and X′), while inclined to the central ring (Ph) by 18.7° and 10.7° respectively (Fig. 2a). OPE3-An has no crystallographic symmetry. Rings A1 and A2 are inclined to the planar anthracene moiety (An) by 26.9° and 38.4°, respectively, their axes X1 and X2 by 3.3° and 19.5°. The A1/A2 dihedral angle is 21.7°, the X1/X2 angle 22.9° (Fig. 2b). In both molecules, the heterocycles adopt envelope conformations, with the CH2 group adjacent to the S atom, tilting out of this (and the fused arene) ring plane. In OPE3-An, both heterocycles are disordered between two envelope conformations with opposite tilt, in 4:1 and 10:1 ratios. Overall, the OPE3 backbone is essentially planar in each case, suggesting a good π-electron delocalisation in these molecules. Moreover, a highly organised extended network was observed in the case of OPE3-Ph. The molecules stack together and arrange in a 2D fashion, resembling what could be expected for a self-assembled monolayer (SAM) on a surface. In contrast, a dimeric arrangement, rather than continuous stacks, was observed for OPE3-An (see ESI† for figures).
The electrical conductance and Seebeck coefficient for the three synthesised molecules have been measured via STM Break Junction techniques. The experimental setup is represented schematically in Fig. 3a. The chips were coated with the material following the procedure described in the Experimental section. Applying the two cleaning methods (flame annealing or oxygen plasma + ion milling) resulted in comparable, high-quality histograms without obvious signs of contamination, such as seen sometimes at intermediate conductance.
Fig. 3 (a) STM break junction setup for the electric and thermoelectric characterization of single molecules; (b) typical example of opening trace: I(t) and Vbias(t) for a single-molecule junction. |
The experiment is based on a modified version of the STM – Break Junction technique, where the current between the two electrodes is recorded as a function of the separation distance, at different temperature difference between the electrodes. A typical experiment runs as follows. First, the temperature bias between the two electrodes is set using the resistive elements present in the tip holder. The substrate is maintained at room temperature, while the overall temperature difference, ΔT, ranges from room temperature up to 60 K. The temperatures in the setup are then allowed to settle to a steady state for more than 1 h at each temperature, before starting a new measurement. After the thermal equilibrium has been reached, the tip is moved closer to the gold substrate where molecules have been deposited (closing trace) with a speed of 5 nm s−1. Then, after a good electrical contact has been established (>5 G0, where G0 denotes the electrical conductance quantum), the two electrodes are slowly moved apart (opening trace), at speeds of 3 nm s−1. For the sake of simplicity, we will consider only the closing traces, since statistically they show a larger number of molecules trapped in the junction. During the trace the current is measured at constant bias of ΔV = 9 mV. This is a value over the junction plus a series resistance, that is added in front of the current-to-voltage converter to prevent it from saturation. In fact, the resistance of the junction usually varies from few ohms ∼Ω, when the two electrodes are in contact, to ∼GΩ, when the electrodes are separated by a single molecule.
A typical trace is shown in Fig. 3b. Three different regions can be distinguished: (I) At low resistance (1 to 10 G0), the conductance is dominated through a metal junction between the gold electrodes (Au–Au). During the thinning down of the Au–Au junction (I), the electrical conductance and consequently the current decreases in a step-like fashion, because of the quantized availability of electron channels in 1D systems. In this framework, the last few channels correspond to the last atoms bridging the contact, down to a single atom, immediately before the rupture.49 (II) After breaking the metal junction, some opening traces also show a second regime (II), where the electrical current exhibits a plateau with a constant value of current (and therefore a constant value of conductance), implying the formation of a molecular junction (Au–molecule–Au). Since for the OPE3 derivatives, the molecular plateau has been theoretically predicted to have a conductance around 10−4G0, as confirmed by the conductance histograms, we define a molecular region as the interval of conductance between 10−3G0 and 10−5G0. In this region the speed of the piezo-scanner is lowered to 1 nm s−1. (III) After the molecular junction is broken, a much smaller tunnelling current is measured (Au–tunnelling–Au) and the current signal falls into the noise level of the setup.
To measure the Seebeck coefficient, we proceeded as follows. When a molecular plateau is found within the molecular region, the following steps were performed while continuing to retract the tip.
(i) The conductance Gbefore of the first 50 points of the plateau is analysed. If the conductance value stays between set thresholds (10−3G0–10−5G0), then it is assumed a molecular junction is still present;
(ii) Then, the bias is turned off for a window of 80 datapoints, and the thermoelectrical current ISeeb is measured;
(iii) Subsequently, the bias is restored and the conductance Gafter of the following 50 datapoints is compared to the value before the switching. If both values of conductance, before and after the bias switching, are found to be within the bounds of the molecular region and Gafter = Gbefore ± 20%, the switching is repeated until the conditions of Gbefore, Gafter are not fulfilled anymore.
This technique enables the determination of the electrical conductance and thermoelectric current within one single measurement, and therefore for the simultaneous determination of Gel and S for each molecule.
Fig. 4 shows the results for the series of three OPE3 derivatives. The first column shows the 2-dimensional histograms for the electrical conductance at room temperature for the different molecules. The colour bar gives the number of counts in a logarithmic scale. The second column compares the 1-dimensional conductance histograms, obtained during the Seebeck measurements, for four different temperature gradients applied to the junction. Here the maximum of each distribution represents the most probable value of the electrical conductance, during the bias switching, at each given ΔT. In the third column the Seebeck voltage is plotted for different value of ΔT. The thermovoltage is calculated from the measured thermocurrent using the following equation:
(1) |
All the molecules show a negative Seebeck coefficient, indicating LUMO-dominated conductance. Also, the introduction of the side-groups in OPE3-Ph(OMe)2 and OPE3-An slightly decreased the absolute value of the Seebeck coefficient compared to OPE3-Ph, while the overall electrical properties were not greatly affected. As a side-remark, for OPE3-An a higher thermal stability is observed, that allowed a doubling of the temperature difference range compared to the other molecules (Fig. 4h and i). The mechanical arrangement of a molecular junction can be thermally unstable, which is visible at higher temperatures through a reduced yield of measuring a molecular plateau in individual opening traces. Such thermal instability of the junction will leave the molecule intact and is different from thermal decomposition of the molecules, which is found to be higher than 250 °C in measurements of the molecular crystals.
To support the experimental results, first principles calculations were performed as detailed in the method section. The metal–molecule binding configuration is not directly accessible in the measurements and will also strongly depend on the electrode distance. We therefore investigated the binding of OPE3-Ph to ideal gold surfaces without adatoms and defects. This allows us to get a first impression of the relevant binding motifs that should also be of relevance for the rough surfaces in break junctions. Geometry relaxations have been performed for OPE-Ph attached to the Au(111) surface using periodic plane wave DFT calculations. The sulfur atom of the DHBT anchoring group was positioned at the top, bridge, fcc and hcp positions of the surface. Configurations that differ by rotations of the molecular backbone around the surface normal have not been considered. This task remains for a later and more complete study.
The size of the periodic unit cell was chosen such that the distance between molecular images is around 9 Å, so that a low surface coverage is effectively simulated. As shown in Fig. 5 the most stable binding configuration is the top position with an Au–S bond length of 2.55 Å. The bonding angles indicate a pyramidalisation of sulfur, consistent with a dative bond where one of the S lone pairs interacts with Au d-orbitals. We also found stable conformations in which the OPE3 backbone is oriented perpendicular to the surface (see ESI†). These are 1.1–1.4 eV higher in energy than the tilted conformers and will not be discussed in detail. The adsorption energy of Eads = −0.51 eV for the most stable conformer shows that the DHBT group provides an effective anchor. Moreno-García and co-workers26 investigated the binding energy of several aurophilic functional groups. Specifically, they looked at cyano (CN), amino (NH2), thiol (SH), 4-pyridyl (PY) and the present DHBT and found the energetical ordering NH2 < DHBT ≈ CN ≈ PY < SH, where NH2 denotes the least stable anchor group. Although the OPE3 molecular wire was coupled to an Au pyramid in the Moreno-García study, the reported binding energy of Eads = −0.41 eV is close to our result for the top position. Considering the other configurations in Fig. 5, we find that binding at hollow sites of the Au(111) surface is not likely: the hcp starting structure relaxes into a top configuration (and is therefore not shown), while the fcc structure shows a low adsorption energy of Eads = −0.16 eV and a very long Au–S distance of 3.45 Å. Optimisation of the bridge configuration leads to a structure where the DHBT sulfur is nearly at a top position.
In summary, we find very little conformational flexibility. In contrast to the more common thiol anchor groups, DHBT features a sulfur lone pair with fixed orientation with respect to the molecular plane, and this reduces the available conformational space. As also proposed in other studies,3,50 this feature might explain the rather narrow conductance histograms seen above.
Next, we investigated the transport properties of the various OPE3 derivatives by means of Non-Equilibrium Green's Function theory. Molecular Green's function, molecule-surface coupling and surface Green's function are all evaluated at the DFTB level, which is an approximate and numerically efficient DFT method. Similar to previous studies,26 the electrodes are modelled as multiple layers of the Au(111) surface connected to Au20 clusters that are finally attached to the anchor groups (Fig. 6). As detailed in the ESI,† binding to the surface leaves the OPE3 backbone largely unchanged with respect to the experimentally determined X-ray crystal structure. The calculations provide the energy dependent transmission function t(E), which is directly related to the electrical conductance G: G = t(EF) G0, where EF denotes the Fermi energy of the junction and G0 is the quantum of conductance.2 As Fig. 7 shows, the side groups at the central OPE3 ring have an important effect on the HOMO energy. The electron-donating OMe groups in OPE3-Ph(OMe)2 lead to an upward shift of approximately 0.15 eV, while the extension of the π system in OPE3-An has an even stronger effect with a shift of 0.75 eV with respect to OPE3-Ph. The LUMO energies exhibit a much weaker variation. Since conduction at low bias occurs through the tails of the frontier orbital that is situated closer to EF, the small differences in conductance found for both experiment and theory (see Table 1) become understandable. While the LUMO transmission for OPE3-An reaches the maximal value of 1.0 (indicating perfect transmission), both OPE3-Ph(OMe)2 and OPE3-Ph show reduced values below 0.04. This could be related to an asymmetric coupling to the left and right contacts, but since our models are symmetric, we attribute this fact to partial destructive interference of transmission paths through energetically close lying molecular orbitals. Similarly, the dip in transmission just below EF could be of similar nature.
Fig. 6 Device geometry for DFTB transport simulations. Shown is OPE3-Ph(OMe)2 connected to semi-infinite gold leads. |
G exp [10−4G0] | G theo [10−4G0] | S exp [μV K−1] | S theo [μV K−1] | |
---|---|---|---|---|
OPE3-Ph | 0.92 | 0.34 | −11.4 ± 0.5 | −57.4 |
OPE3-Ph(OMe)2 | 2.03 | 0.93 | −7.9 ± 1.3 | −66.2 |
OPE3-An | 1.82 | 0.79 | −8.7 ± 1.6 | −61.4 |
The computed transmission can also be used to estimate the temperature dependent Seebeck coefficients S(T) for the OPE3 derivatives,2
(2) |
The theoretical values obtained are reported in Table 1 and compared to the measured values. As mentioned above, the experimentally measured negative values for S indicate a LUMO based transport mechanism, which is in agreement with the theoretical findings. Additional exploratory calculations for different plausible metal–molecule binding scenarios indicate that the Fermi level positioning is influenced by sulfur pyramidalisation and the bonding to undercoordinated Au atoms (see ESI†), both factors favouring a negative Seebeck coefficient.
In general, rather small variations for the different OPE3 derivatives are observed. Experimentally, OPE3-Ph has the largest conductance and the largest (in absolute value) Seebeck coefficient. Since the transmission derivative is usually dominating over the transmission in the denominator of eqn (2), this is consistent with a LUMO position of OPE3-Ph that is slightly closer to EF than for OPE3-Ph(OMe)2 or OPE3-An. The theoretical simulations do not reproduce these subtle differences for the investigated junction models and generally overestimate the Seebeck coefficients. To a certain extent this might be related to known difficulties of DFT and DFTB to accurately predict the transport gap and level alignment.51–53 In addition, it would be desirable to sample a larger space of metal–molecule geometries to approach the experimental conditions, but already at this stage, the theoretical simulations confirm the main transport mechanism and small variability in the thermopower. In this context, we mention a recent review that highlights challenges in theory and experiment as well as the currently achievable agreement between the two.54
Footnote |
† Electronic supplementary information (ESI) available: Synthesis and characterisation of the OPE3 derivatives; X-ray crystallographic data. CCDC 1996615 and 1996616. For ESI and crystallographic data in CIF or other electronic format see DOI: 10.1039/d0nr04413j |
This journal is © The Royal Society of Chemistry 2020 |