Intermolecular coupling enhanced thermopower in single-molecule diketopyrrolopyrrole junctions

: Sorting out organic molecules with high thermopower is essential for understanding molecular thermoelectrics. The intermolecular coupling offers a unique chance to enhance the thermopower by tuning the bandgap structure of molecular devices, but the investigation of intermolecular coupling in bulk materials remains challenging. Herein, we investigated the thermopower of diketopyrrolopyrrole (DPP) cored single-molecule junctions with different coupling strengths by varying the packing density of the self-assembled monolayers (SAM) using a customized scanning tunneling microscope break junction (STM-BJ) technique. We found that the thermopower of DPP molecules could be enhanced up to one order of magnitude with increasing packing density, suggesting that the thermopower increases with larger neighboring intermolecular interactions. The combined density functional theory (DFT) calculations revealed that the closely-packed configuration brings stronger inter-molecular coupling and then reduces the highest occupied molecular orbital (HOMO)-lowest unoccupied molecular orbital (LUMO) gap, leading to an enhanced thermopower. Our findings offer a new strategy for developing organic thermoelectric devices with high thermopower.


INTRODUCTION
Thermoelectric materials offer the opportunity for direct conversion of heat into electric energy via Seebeck effects [1][2][3][4], and the efficiency of thermoelectric devices can be evaluated through the dimensionless figure of merit ZT=GS 2 T/κ, where G is the electrical conductance, S is the Seebeck coefficient, T is the temperature, and κ=κ el +κ ph is the thermal conductance due to electrons (κ el ) and phonons (κ ph ) [5].Compared with inorganic thermoelectric materials, molecular thermoelectric materials exhibit lower thermal conductivity (κ ph ) and higher flexibility with tunable electronic structures [6][7][8][9], some pioneering theoretical works even suggested that molecular thermoelectric devices can reach the highest ZT value up to 5.9 [10].However, the experimental determined Seebeck coefficients of the molecule-scale thermoelectric materials are still much lower than theoretical predictions [11][12][13].Strategies to enhance the Seebeck coefficient are crucial for developing high thermoelectric performance devices at the single-molecule scale.
To increase the ZT of the molecular devices [14], various strategies have been investigated to improve the thermopower, including varying molecular length of organic building blocks [11,15,16], changing connectivity of molecular cores [17][18][19][20], tuning molecule-electrode coupling [21], exploring different anchor groups [22] and substituent groups [23].More importantly, theoretical advances revealed that the intermolecular interactions, such as π-π stacking, will enhance ZT by suppressing the phonon contribution to lowering the thermal conductance [12], suggesting that the intermolecular interactions are essential for the design of molecular devices with high ZT [24,25].However, the role of intermolecular interactions in the thermopower of molecular devices has not yet been experimentally investigated, which is mainly due to the challenges in the control of intermolecular interactions at the single-molecule level.Previous studies have demonstrated that the π-π interaction between the porphyrins in the mixed self-assembled monolayers (SAMs) decreases with reducing the concentration of porphyrin during the assembling process [26], which offers the strategy to tune intermolecular couplings by varying packing density in the SAMs and explore how the presence of intermolecular coupling affects the thermoelectric properties from molecular level [27][28][29].
In this work, we investigated the Seebeck coefficient of diketopyrrolopyrrole (DPP) molecular junctions by varying the packing density of their SAMs.The three target DPPs of difuranyl-DPP (F-DPP), dithienyl-DPP (T-DPP), dithiazolyl-DPP (Thia-DPP) with p-methylthiobenzenes at both ends are shown in Figure 1A (see Figure S1 for synthesis information).To explore the effect of packing density on thermoelectric properties of the single-molecule junctions, we assembled the molecule on a gold surface through immersion in different solutions with different concentrations [30].We found that the Seebeck coefficients of all DPPs increase with growing molecular packing density.Besides, density functional theory (DFT) based calculations revealed that stronger intermolecular coupling associated with higher packing densities generally reduces the energy gap and leads to an enhanced thermopower on DPP molecules.

RESULTS
A customized scanning tunneling microscope break junction (STM-BJ) instrument [31,32] was used to simultaneously measure the conductance and thermopower of the single-molecule junctions at room temperature (Figure 1B).A Peltier device mounted under the substrate was used as a heater to establish a stable temperature difference (ΔT=T substrate -T tip ) between the tip (at room temperature, ~298 K) and the substrate (heated) [11,33].The thermopower measurement was performed for four temperature differences (ΔT=0, 5, 10 and 15 K).To obtain a statistical distribution of ΔV of an Au-Molecule-Au junction, we have collected more than 1000 consecutive data at each ΔT and selected those that sustained a molecular junction through the entire "hold" period (see Figure S2 for details of thermoelectric measurement process).
The typical individual thermovoltage traces and histograms are shown in Figure 1C for a single F-DPP molecule (1 mmol/L immersion concentration).It is found that a larger temperature difference brings significantly higher ΔV, which could be further demonstrated in the distribution histograms of the individual traces (right panel of Figure 1C).To quantitatively determine the Seebeck coefficient, we constructed the distribution histograms of thermovoltage at different temperature differences (Figure 1D) from more than 1000 individual traces, and the most probable thermovoltage could be determined from the Gaussian fitting of the distribution.Because the fluctuation of the molecular junction configurations inevitably occurs during the measurement process, the thermoelectric voltage from different trapped molecular junctions might exhibit different distributions [11].By fitting the slope of the most-probable thermovoltage value versus the temperature differences, the Seebeck coefficient could be determined to be −18.05±1.64μV/K.To investigate the role of packing density in SAMs, the Seebeck coefficient was measured for F-DPP with increasing packing densities through immersing the gold electrode into corresponding 0.01, 0.1, and 1.0 mmol/L molecule solutions.The previous work had revealed that molecular interaction in the SAMs could be controlled by modulating the concentration of solution during the assembling process [26].The Xray photoelectron spectroscopy (XPS) semiquantitative analysis was conducted to confirm chemical composition of molecular monolayer and its atomic ratio (Table S1) [34].The results suggest that more molecules existed on the gold sample under the immersion of the molecule solutions with increased concentration, based on the atomic concentration of the S2p peak signal (Figure S3), and the XPS analysis indeed had demonstrated the formation of Au-S bonds in the SAMs.
Besides, analytical electrochemistry was also used to quantitatively determine how much gold surface was covered with F-DPP molecules at different immersion concentrations, according to a method modifed from a previously published work [35].The gold electrode with molecules assembled on was used as a working electrode, and a cyclic voltammetry (CV) was performed in 2.5 mmol/L K 3 Fe(CN) 6 /K 4 Fe(CN) 6 solution containing 0.1 mol/L KNO 3 as the supporting electrolyte with a sweep rate of 100 mV/s, and the potential was controlled between −0.2 and 0.6 V to avoid other redox reactions from assembled F-DPP.A typical CV at different immersion concentrations was obtained (Figure 2A), there are apparent redox peaks from K 3 Fe (CN) 6 /K 4 Fe(CN) 6 in all cases.The current at the region with potential over redox peak was mainly controlled by reactants diffusion, and the diffusion areas could be obtained by applying the Cottrell equation [36] (see Figure S4 for the apllication of Cottrell equation in this work).The diffusion areas of the bare gold electrode (without molecules assembled) were determined to be 0.0051 cm 2 , while it decreased to 0.0036 cm 2 for molecule-assembled gold electrode with 1 mmol/L immersion concentration (Table 1).The diffusion areas could reflect the assembled-molecule-occupied electrode areas, because SAMs with higher packing density left fewer areas for redox reaction and diffusion.Finally, the SAM surface coverage fraction of assembled molecules with different immersion concentrations was obtained according to the diffusion area compared to bare gold (Table 1 and inset of Figure 2B), varying from 7.84% (0.01 mmol/L immersed F-DPP) to 29.41% (1 mmol/L immersed F-DPP).In addition, a micro thermal gravimetric analyzer (μ-TGA) [37] was also applied to determine their packing density differences (Figure S5).All these characterizations indeed proved that the packing density could be controlled by modulating the immersion concentration during the assembling process.
It is found that the Seebeck coefficients of a single F-DPP molecule at increasing packing densities are  2).These showed that the Seebeck coefficients of F-DPP maintain the same sign (negative) and increase dramatically with the increase of molecular packing density.More importantly, the Seebeck coefficient of the single-molecule junction could be enhanced near one order of magnitude, which is much larger than the mainstream tuning method called destructive quantum interference (DQI) with only two times enhancement up to now [20].Furthermore, we observed a similar trend for T-DPP and Thia-DPP.[38].This trend is the same as that observed in oxidized oligothiophenes derivatives, in which the dominant charge carriers changed from holes to electrons with increasing molecular length [16].
To elucidate the origin of the experimentally observed trends, we investigated the transport properties of F-DPP, T-DPP and Thia-DPP junctions connected to the gold source and drain electrodes via -SMe groups as shown in Figure 3A-3C (see more binding configurations for F-DPP, T-DPP and Thia-DPP in Figures S9  and S10) using DFT combined quantum transport theory.The material-specific mean-field Hamiltonian of each geometry obtained from SIESTA [39] was combined with quantum transport code Gollum [40] to obtain the electronic transmission coefficient, which controls electrical and thermoelectric properties.
To investigate the dependence of conductance and Seebeck coefficients on the packing density, we calculated the transport properties for junctions containing either a single-molecule (i.e., a monomer junction) or in the presence of two nearby molecules (i.e., optimized geometry of trimer junction is shown in Figure 3A-  3C).We found that the trimer junctions have smaller highest occupied molecular orbital (HOMO)-LUMO gaps than monomers due to the splitting of HOMO and LUMO peaks caused by the interaction with nearby molecules [41].The modelling demonstrates that the reduced HOMO-LUMO gap of the trimer generally increases the slope of transmission coefficients.Consequently, for a wide range of Fermi energies within the HOMO-LUMO gaps, the Seebeck coefficients of trimers are greater than or equal to monomers, and trimers have similar or higher conductance than the monomers (see Figures S11 and S12 for more details).These features are in qualitative agreement with the observed experimental trends (see Figures S5 and S6 for conductance measurements).DFT has difficulties in predicting the correct HOMO-LUMO gaps [42], charge transfer and Coulomb interaction for charged systems and particularly for these strong acceptor DPP-cores [ 43,44].On the other hand, the Fermi energy is located somewhere inside the HOMO-LUMO gap.Therefore, to compare results for monomer and trimer molecular junctions, we aligned their mid-gap of the conductance curve (Figure 3D S2, the Thia-DPP molecule gains more electrons than F-DPP and T-DPP, which could induce more vital Coulomb interaction for the negatively charged molecule, moving the energy levels upwards and pushing the HOMO closer to the Fermi energy.Indeed, if the Fermi energy is slightly below the mid-gap position for Thia-DPP, then the sign of the Seebeck coefficient would be positive, as shown to the left of the crossing point (S=0 and E F =−0.3 eV) in the Seebeck plot of Figure 3I, which also reveals a higher Seebeck coefficient for the trimer junctions.Conversely, for F-DPP and T-DPP, the right sides of the crossing point of S=0 have a negative Seebeck coefficient and greater magnitudes for the trimers (Figure 3G and 3H). Figure 3G-3I clearly shows that the Seebeck coefficient for the trimer junctions (solid lines) typically has a higher magnitude than the monomers (dotted lines), in agreement with the experimental trends.

DISCUSSION
In conclusion, we experimentally investigated the Seebeck coefficients of a series of single DPP derivatives varying the packing density of SAMs on the electrode surface using a modified STM-BJ technique.We discovered that the conductance channel of molecules could be changed from LUMO-dominated to HOMOdominated by altering the adjacent aromatic rings of the DPP core.More importantly, the thermopower of molecular junctions could be enhanced by up to one order of magnitude via the increase of the packing density in SAMs.Combined DFT calculation revealed that the higher packing density leads to more substantial intermolecular coupling effects, which reduces the HOMO-LUMO gap and increases the Seebeck coefficient.Our results revealed that intermolecular coupling is of fundamental importance for designing highly efficient molecular thermoelectric devices and materials in the future.

Materials
The target molecules were synthesized according to the previous reports [45][46][47].For more details, see Figure S1.To obtain SAMs of different packing densities, the gold substrates, which are prepared by coating 200 nm Au film on silicon wafers, were immersed into 0.01, 0.1, 1 mmol/L molecule solution using the solvent of 1,2,4-trichlorobenzene (TCB, 99.9%, Sigma Aldrich) for 4 h.After that, the surface of the gold substrates with the assembled monolayer was rinsed by TCB and dried with N 2 gas.

Single-molecule conductance and Seebeck coefficient measurement
The single-molecule conductance and Seebeck coefficient measurements were performed using the homebuilt STM-BJ technique as described in previous reports [31].The temperature was modulated by proportion integral differential (PID) control.Single-molecule junctions were fabricated following the electrical conductance measurement [48].Once the conductance plateau was determined, the tip would be hovered and the tip/substrate distance fixed, followed by cutting off the bias voltage and the current amplifier.Instead, the voltage amplifier was connected to record the thermovoltage directly.After a period of time interval, the voltage amplifier would be cut off while the current amplifier is switched back to measure the conductance again.During the experiment, the tip withdrew from the sample until the tunneling current decreased to achieve the given threshold value.For further details, see Figure S2 in the Supplementary Information.

Computational methods
The optimized geometry and ground state Hamiltonian and overlap matrix elements of each structure were self-consistently obtained using the SIESTA implementation of DFT [39].SIESTA employs norm-conserving pseudo-potentials to account for the core electrons and linear combinations of atomic orbitals to construct the valence states.The generalized gradient approximation (GGA) of the exchange and correlation functional is used with the Perdew-Burke-Ernzerh of parameterization (PBE), a double-ζ polarized (DZP) basis set, a real-space grid defined with an equivalent energy cut-off of 200 Ry [49].The geometry optimization for each structure is performed to the forces smaller than 10 meV/Ang.
The mean-field Hamiltonian obtained from the converged DFT calculation was combined with our homemade implementation of the non-equilibrium Green's function method, Gollum [40], to calculate the phasecoherent, elastic scattering properties of each system consisting of left gold (source) and right gold (drain) leads and the scattering region (molecule F-DPP, T-DPP and Thia-DPP).The transmission coefficient T(E) for electrons of energy E (passing from the source to the drain) is calculated via the relation: In this expression, In the linear response, the quantity of Lorenz number L , ) n F is given

Figure 1
Figure 1 Schematic diagram of experimental setup and Seebeck coefficient measurement.(A) Molecular structures of DPP derivatives studied in this work.(B) Schematic of the experimental setup.Closely packed molecules enhance the Seebeck coefficient due to stronger intermolecular coupling.(C) Typical measured individual thermovoltage traces and histograms for single F-DPP molecule with 1 mmol/L immersion concentration at a series of ΔT (0, 5, 10, and 15 K).(D) Histograms of single F-DPP molecule thermoelectric voltage measurements with different immersion concentrations (0.01, 0.1, 1 mmol/L).Gaussian fits were plotted in a black dash curve.The horizontal black dash line indicates the baseline of thermoelectric voltage at ΔV th =0.The Seebeck coefficients were obtained from the thermovoltage as a function of ΔT.Solid lines are linear fitting.

Figure 2
Figure 2 CV characterization on SAM surface coverage and Seebeck coefficient of three types of DPP.(A) CV measurements with F-DPP assembled gold as the working electrode at different immersion concentrations in 2.5 mmol/L K 3 Fe(CN) 6 /K 4 Fe(CN) 6 containing 0.1 mol/L KNO 3 as the supporting electrolyte, standard calomel and Pt working as reference and counter electrode respectively.(Inset: surface coverage fraction of self-assembled F-DPP as a function of immersion concentration.)(B) Experimentally measured Seebeck coefficient value for F-DPP (blue squares), T-DPP (brown cycles) and Thia-DPP (magenta triangles).Error bars are the standard deviation in Gaussian fitting of thermoelectric voltages.The solid line indicates that the absolute Seebeck coefficient of the three DPPs increases with solution concentration.The green dashed line indicates that the Seebeck coefficient S=0.

Figure 3
Figure 3 DFT calculated electrical and thermoelectric properties of the three DPP derivatives attached to gold electrodes via -SMe anchor groups, the distance between the central backbone and adjacent molecules is around 3.3 to 3.6 Å for F-DPP, T-DPP and Thia-DPP trimers.(A)-(C) Models for F-DPP, T-DPP and Thia-DPP, respectively.The solid line corresponds to three molecules in the junction (i.e., a trimer), and the dotted line corresponds to a single molecule in the junction (i.e., a monomer).(D)-(F) The calculated room-temperature electrical conductance of F-DPP, T-DPP and Thia-DPP as a function of the Fermi energy (E F ) relative to the mid-gap E g .The solid lines represent trimers and dashed lines represent monomers.(G)-(I) The calculated room-temperature Seebeck coefficients of F-DPP, T-DPP and Thia-DPP as a function of Fermi energy (E F ) with the same shifting corresponding to conductance.The solid lines represent trimers and dashed lines represent monomers.
-3F).The electrical conductance for F-DPP trimer (solid curve) and F-DPP monomer (dotted curve) remain in the same value (close to 10 −4 G 0 ), and those for T-DPP, Thia-DPP trimers and T-DPP, Thia-DPP monomers possess the conductance around 10 −4 -10 −4.5 G 0 , which demonstrated a considerable consistency with the experimental data.The corresponding Seebeck coefficients are displayed in Figure 3G-3I.In terms of the net Voronoi charge distribution shown in Table describes the level broadening due to the coupling between left (L) and right (R) electrodes and the central scattering region, retarded Green's function, where H is the Hamiltonian and S is the overlap matrix.Using the obtained transmission coefficient T(E), the electrical conductance G(T) and the Seebeck coefficient S(T) can be calculated through the following formula: quantum, e is the charge of a proton; h is the Planck's constant; E F is the Fermi energy; Fermi-Dirac distribution function, T is the temperature, and k B =8.6×10 −5 eV/K is Boltzmann's constant.

Table 1
Immersion concentration-dependent surface coverage of F-DPP SAMs

Table 2
Single-molecule Seebeck coefficient and conductance measurements