Current–voltage characteristics of single-molecule diarylethene junctions measured with adjustable gold electrodes in solution

We report on an experimental analysis of the charge transport through sulfur-free photochromic molecular junctions. The conductance of individual molecules contacted with gold electrodes and the current–voltage characteristics of these junctions are measured in a mechanically controlled break-junction system at room temperature and in liquid environment. We compare the transport properties of a series of molecules, labeled TSC, MN, and 4Py, with the same switching core but varying side-arms and end-groups designed for providing the mechanical and electrical contact to the gold electrodes. We perform a detailed analysis of the transport properties of TSC in its open and closed states. We find rather broad distributions of conductance values in both states. The analysis, based on the assumption that the current is carried by a single dominating molecular orbital, reveals distinct differences between both states. We discuss the appearance of diode-like behavior for the particular species 4Py that features end-groups, which preferentially couple to the metal electrode by physisorption. We show that the energetic position of the molecular orbital varies as a function of the transmission. Finally, we show for the species MN that the use of two cyano end-groups on each side considerably enhances the coupling strength compared to the typical behavior of a single cyano group.

The reaction mixture was stirred for 24 h at rt and was poured into water (30 mL) and

Sample calibration and statistics
The data presented in the manuscript have been recorded from a total of approximately 50 break-junction samples. Out of these samples roughly 20 were used for distance calibration in air and the various solvents. We recorded stretching and relaxing curves and determined the distance scale by fitting to the functional dependence for vacuum tunnelling along the lines of Grüter et al. [S1]. Our break-junction setup is designed to achieve a maximum life-time of single-molecule junctions. A drawback of this highstability design is that it enables only rather slow stretching and relaxing speeds on the S7 order of 0.1 nanometres per second, at most. This enhances measuring time to a few minutes per stretching and relaxing cycle. Additionally, the rather soft polyimide insulation layer between the metal substrate and metal electrodes undergoes a creeping motion that limits the total measuring time to a few hours. Because of the creeping, the bending angle at which a particular electrode distance is obtained drifts to higher values, such that the elastic deformation limit of the substrate is reached and no further bending is possible. These two factors restrict the total number of deformation cycles per sample to a few hundred.
For measuring the molecular contacts we used fresh samples with nominally the same geometry assuming the same mechanical behaviour as in vacuum. This sets the error bar for the distance scale to roughly 30%. Finally, 22 samples were used to investigate the molecular contacts with varying lifetimes and therefore varying distribution over the individual species, (Table S1). We started the measurements by recording stretching and relaxing curves for training the electrodes and forming narrow metal tips, indicated by steps on the order of the conductance quantum when the conductance is on the order of a few quanta. This training is performed at air or directly with the molecular solution, because we observed an even stronger drift of the breaking point of the electrodes in the solvent. Simultaneously the shape of the current-voltage characteristics recorded in the solvent developed pronounced nonlinearities (see below). Both observations indicate the formation of disordered gold contacts and chains stabilized by incorporating carbon atoms resulting from the solvent [S2]. After this training a few hundred of these stretching/relaxing cycles were performed before we started recording the current-voltage characteristics (I-V's). The described ageing effect limits the maximum amount of data that can be obtained from one sample.
Because of this effect, it was not possible to establish single-molecule contacts that showed the typical s-shaped I-V's for all samples, hence reducing the number of S8 successful samples as indicated in Table S2. We stopped measurements of a given sample when no plateaus at 1 G 0 could be observed any more.    For all species we recorded continuous stretching/relaxing curves, i.e., without interruption of the motion. We applied a dc voltage of 100 mV. The initial value of the gold junctions before breaking is on the order of 300 to 500 G 0 . The junctions are stretched by bending the substrate until the conductance drops below 10 −8 G 0 . The motion direction of the breaking mechanism is then reversed and the junctions are relaxed until the junction conductance arrives at a conductance of at least 100 G 0 . This procedure is repeated up to a few hundred times. The MCBJ setup is optimized for maintaining single-molecule contacts constant over several minutes. Figure S3 shows the results for TSC in IPA, Figure S4 for 4Py and Mn. Junctions with conductance values covering the whole range from 10 −7 G 0 to 10 G 0 are observed. We do not observe pronounced differences between the histograms for stretching and relaxing, S12 except for the fact that more data is recorded for lower conductance values G < 10 −7 G 0 when relaxing the junction. This is, however, an artefact of our measurement protocol as described above. There are no clear maxima in the histograms in the open state but an enhancement of probability for conductance values above 10 −2 G 0. In the stretching histogram of the closed form we observe a pronounced minimum around 10 −2 G 0 . The relaxing histogram of the closed form features a higher probability for rather low conductance ~10 −5 G 0 .

Testing the fit procedure
To determine  L ,  R and E 0 , we fitted the measured I-V curves with the expression for the current obtained from the Landauer formula as described in Equation 1 to Equation 3 of the manuscript. The integral was numerically evaluated with the program "Octave" applying a Levenberg-Marquardt algorithm. We evaluated the Fermi functions for k B T = 25 meV, corresponding to a temperature of 290 K.
Furthermore, we tested the quality of the fit regarding its functional dependence. As discussed in [S3], for small voltages, Equation 1 can be expanded into a sum of a linear and a cubic term of the voltage as: Both prefactors A 1 and A 3 depend on  L ,  R and on E 0 in a unique way, implying a correlation between them [S3]. If the I-V does not correspond to the functional shape described by the model (Equation 1 to Equation 3), the resulting best-fit results do depend on the voltage range [S4,S5]. A helpful method to test the validity of the model is thus given by varying the fit range. We performed this procedure for our data by varying the fit range from [−0.5 V, +0.5 V] to [−1 V, 1 V]. We discarded all I-Vs in which S13 the best-fit parameters varied by more than 20%. Finally we compared the transmission T(E,0) calculated from the best-fit results with the one deduced from the linear conductance measured in a range of [−100 mV, 100 mV]. We labeled such I-V's as "fittable" for which T(E,0) calc and T lin = G/G 0 agree within 50% error. For MN the I-V's were measured in the voltage range of [−0.5 V, +0.5 V] only. The "fittable" I-V's were selected on the basis of the latter criterion only.

Current-voltage characteristics of the solvents Tol and THF/Tol
We Examples are given in Figure S5. The fitting parameters of symmetric I-V's are shown in Figure S6 together with those for MN and 4Py. While the coupling constants are in a similar range as for the molecular junctions, the corresponding E 0 are not, enabling us to distinguish molecular contacts from "solvent" contacts. We believe that these contacts correspond to disordered metal contacts in which residues from the solvent are incorporated into the metal, as described in [S2]. In the solvent ethanol no fittable I-V's could be recorded, presumably due to the hygroscopic and polar properties of the solvent, enhancing the probability of electrochemical reactions during sweeping.
Reducing the voltage range to 0.5 V or less resulted in unreliable fitting, i.e., the fit results were strongly dependent on the fit range. S14 Figure S5:  Table S3 below. S15

Current-voltage characteristics of TSC in THF/Tol
In order to test whether the properties of the junctions that were formed depend on the solvent, we investigated the species TSC in a mixture of 50% tetrahydrofuran and 50% toluene. Within our statistics we did not find characteristic differences in the data recorded for TSC in ethanol. Figure Table S3.