Optical probes of molecules as nano-mechanical switches

Molecular electronics promises a new generation of ultralow-energy information technologies, based around functional molecular junctions. Here, we report optical probing that exploits a gold nanoparticle in a plasmonic nanocavity geometry used as one terminal of a well-defined molecular junction, deposited as a self-assembled molecular monolayer on flat gold. A conductive transparent cantilever electrically contacts individual nanoparticles while maintaining optical access to the molecular junction. Optical readout of molecular structure in the junction reveals ultralow-energy switching of ∼50 zJ, from a nano-electromechanical torsion spring at the single molecule level. Real-time Raman measurements show these electronic device characteristics are directly affected by this molecular torsion, which can be explained using a simple circuit model based on junction capacitances, confirmed by density functional theory calculations. This nanomechanical degree of freedom is normally invisible and ignored in electrical transport measurements but is vital to the design and exploitation of molecules as quantum-coherent electronic nanodevices.


Supplementary Note 2. In-situ darkfield spectroscopy
The transparent cantilever affects darkfield spectroscopy of AuNPs because of the increased refractive index this introduces on top of the NPoM structure. We consistently observe a ~40nm blueshift in the resonance wavelength of the NPoM coupled mode, and a slight decrease in intensity of the single particle mode (Supplementary Figure 2).

Supplementary Figure 2 | Darkfield through cantilever.
Darkfield spectroscopy through the cantilever shows a blueshift in the NPoM coupled-mode peak.
The NPoM coupled mode remains clearly distinguishable in our in-situ measurements, and its evolution is tracked in real time with a stable signal (Supplementary Figure 3a). We observe no change in the amplitude or position of this peak in the tunnelling regime for |V|<1.5V, and voltage can be cycled repeatedly in this range for tens of minutes without detectable spectral changes. For |V|>1.5V we generally observe a rapid increase in conductance (Supplementary Figure 3b) and a redshift of the coupled mode (Supplementary Figure 3a), possibly due to disruption of the molecular layer and formation of metallic conductive links across the junction that increase conductance and short the coupled plasmonic mode. This redshift is only rarely reversible and suggests a morphological change in the junction structure.

Supplementary Note 3. In-situ Raman spectroscopy
Raman spectroscopy is performed in real time on molecules inside a single AuNP junction. The plasmonic enhancement of optical field in the junction gap amplifies light-matter interactions enabling surface enhanced Raman spectroscopy (SERS) with a typical signal of >10 5 counts/mW/s for the molecular vibrational signatures, allowing real time spectroscopy with <1s integration times. Electrical measurements are performed at a faster rate of 50 datapoints/s given by the Keithley SMU, so tens of I-V data points are collected for each spectrum acquisition.
Reduction in the Raman intensity for voltages above 0.5V are consistently observed for BPDT in more than 70% of the single AuNP junctions contacted with the cantilever (Supplementary Figure 4a,b). The remaining cases are accounted for by variations in AuNP size or local imperfections in the parylene insulating layer leading to poor contact with the AuNP crown. In-situ spectroscopic and electrical measurements on a particular junction in the |V|<1.5V regime can typically be cycled for tens of minutes before mechanical vibrations of the cantilever cause local displacement of the AuNP leading to permanent modifications.

Supplementary Figure 4 | In-situ Raman spectroscopy of BPDT junctions. a,b
In-situ Raman spectroscopy of a single AuNP junction with BPDT molecular layer. c, Corresponding voltage (black) and SERS intensity ratio (blue) calculated as SERS( )/SERS( =0) for the 1590cm -1 ring-ring CC stretching mode. Each data point is the average of consecutive measurements taken at the same voltage. d, Ratio of amplitudes for peaks at 1590cm -1 /1140cm -1 , showing that modulation is stronger for the 1590cm -1 peak corresponding to the inter-ring CC bond.
We observe a decrease of up to 95% in the SERS intensity for the Raman signal of BPDT for |V|>1V (Fig.S4c). The effect is first visible around |V|=0.5V, saturating for |V|>1V, and is symmetric for positive/negative voltages in most experimental realisations. The SERS peak near 1590cm -1 , corresponding to the stretching of the central CC inter-ring bond, shows a stronger modulation than the other peaks in the 1100-1250cm -1 region (Supplementary Figure 4d). This highlights that this bond is more affected by the applied voltage due to the twist between the rings, and is in good agreement with the same trend observed in DFT calculations. We also observe a decrease of the SERS background by up to a factor 2 for |V|>1V, suggesting a 2 1/4 -1=16% decrease in the electromagnetic field intensity in the gap (since SERS intensity scales as 4 in the NPoM gap 2 ). This could be due to a reduction in gap refractive index at optical frequencies, however this would produce blueshifts in DF while redshifts are seen experimentally (but at higher voltages where current flows).
DFT calculations (Figure 3 main text) predict a small redshift of a few cm -1 as the twist angle between the BPDT rings is increased from 0 to 90°. This shift is not experimentally observed in our measurements nor in previous reports 3,4 , suggesting this is due to inaccuracies in the DFT model at this order. DFT calculations also involve a single molecule attached to individual Au atoms, whereas a real device includes extended Au electrodes and close-packed molecules.
In-situ SERS of 2-naphtalenethiol in our AuNP junctions shows no changes of peak intensity correlated with voltage for any of the molecular vibration peaks (Supplementary Figure 5). Variations in the signal are compatible with minor intensity fluctuations generally observed in all measurements. The 2naphtalenethiol molecule still has two rings, but these are locked in the same plane since they share two carbon atoms, so they are unable to twist like the biphenyl molecules and thus do not produce the same modulation in SERS intensity. Different functional groups at the ends of the biphenyl ring structure affect the conductance of the AuNP junction near the molecule-electrode interfaces, producing different effects in the capacitivemolecular energy balance discussed in the main text. For BPDT we observe the maximum level of modulation, with a symmetric decrease in SERS with voltage as high as 95% (Supplementary Figure  6a). BMMBP shows a modulation of up to 75%, but stronger for negative voltages than for positive ones (Supplementary Figure 6b). Even though the BMMBP molecule itself is also symmetric, the additional bonds at its ends compared to BPDT can introduce variations in the SAM packing during assembly on the substrate or AuNP, causing the observed some asymmetry in SERS modulation. Finally, for BPT and CN-BPT we observe a symmetric SERS change of up to 30%, with the modulation occurring only at |V|>2V for CN-BPT.

Supplementary Note 4. Electrical characteristics
NPoM junctions are characterised using a source-measure unit (Keithley SMU, see Methods main text) in 4-wire configuration. To ensure that the cantilever contacting method applied to NPoM junctions does not affect the measured values of molecular conductance, we separately verify the electrical performance of the cantilever-NPoM system.
The conductance of the cantilever itself is measured by electrically contacting a set of Au wires of varying width patterned directly on SiO2 with e-beam lithography (Supplementary Figure 7a). Each wire is contacted from the top in a direction parallel to the cantilever. For the smallest Au wire of width 50nm we measure a linear ohmic response with conductance G=30G0 (Supplementary Figure  7b), while for 100nm wide wires we measure 80G0. For wider wires up to 50µm we always measure a conductance of 80G0, indicating that this is the conductance limit of the cantilever contacting method, set by the conductive coating on the cantilever. Figure 7 | Cantilever conductance. a, Cantilever contacting of Au wires defined with e-beam lithography; scale bar is 50µm. b, Current-voltage characteristic of cantilever-wire contact is linear with G=30G0 for 50nm wide wires, and G=80G0 for ≥100nm wires.

Supplementary
The conductance of the NPoM system contacted with the cantilever is measured by fabricating a sample with the same geometry described in the main text but no molecular SAM in the NPoM gap, with the AuNP thus directly contacting the bottom Au mirror. In these single particle devices we consistently measure a linear ohmic I-V response with conductance G=13G0=10 -3 S (Supplementary Figure 8). Conduction occurs through the AuNP, but a simple conductance calculation from the nominal geometry of a NP (20nm wide facet, 100nm diameter, Au resistivity 2.2×10 -8 Ωm) predicts a conductance of ≈0.1S. This suggests that conduction actually happens through a smaller contact area, likely defined by the local geometry of the NP facet in the gap and roughness of Au mirror. From previous investigations of conductance in nm-sized Au constrictions 5 and given that the bottom Au mirror and Au coating on the cantilever are deposited in the same conditions, we infer that conduction in our system is limited by two equal contact points with diameter 4nm, respectively at the cantilever-AuNP and AuNP-mirror interfaces. In presence of a SAM, we thus estimate that ≈50-60 molecules within this contact region determine the electrical properties of our devices.  Figure 9b). This is >5 orders of magnitude smaller than the conductance across a single AuNP without SAM, and we thus conclude that the cantilever contacting technique with AuNPs has a negligible effect on the measurements of molecular conductance. We attribute the observed variability in the conductance across different junctions to variations in the number of contacted molecules due to surface roughness and NP facet size, but note that most junctions show conductance in the 10 -6 G0<G<10 -5 G0 range (Supplementary Figure 9c).

Supplementary Note 5. Single molecule twist switching
As noted in the main text, the 20nm wide facets of 100nm AuNP nanogaps accommodate ~100 molecules in a single junction However an effect explored in the last few years [see main text ref 17 ] allows single molecules SERS to be tracked in real time in the same nanocavities. Strong enough optical fields in the NPoM gap can transiently pull out atoms from the facet surface, which further enhances light confinement and produces fleeting SERS lines in addition to the normal Raman. We detect these 'picocavity' events ( Supplementary Figure 10a), revealing again the same SERS intensity modulation on the lines which are shifted by molecular-interactions with the Au adatom (Supplementary Figure  10b). In particular the voltage-induced suppression of single-molecule SERS matches that of the main molecular SERS peaks (dashed box). This shows that the same voltage-induced twist model applies to single-molecule junctions (Supplementary Figure 10b). Bias is not found to modulate these Au adatoms in different ways, nor does the modified Au-molecule interaction modify the twist forces dramatically. Generally all picocavities observed show the same twist switching, matching that of the main SERS peaks. Similarly compressing the S-S distance from 10.70 to 9.70Å bends the molecule requiring large energies.

Supplementary Note 9. Parameters for electrical model of BPDT in nano-junction
We use the model in the main text, with two molecule-metal junctions (capacitance 1,3 and conductance 1,3 ) for the back-to-back Schottky diodes, and a junction between the two conducting aryl rings in the centre of the molecule with capacitance 2 and conductance 2 . Because a set of similar molecules with different twist angles have already been measured experimentally, these are not free parameters.
Conductive AFM and STM break-junction experiments in liquid indicate that the conductance 2 through biphenyl molecules is controlled by the inter-ring twist, 2 = (1 + cos 2 ) where ~76 μS when the rings are exactly orthogonal so there is no − overlap (see refs [23][24][25][26] in main text). The value of is then set from the total conductance when the rings are planar (note that what is measured in STM is the total conductance and not just the molecular conductance), using ( = 0)~1.5 mS (ref [26] in main text) which implies 1~ 2.4 mS, and that ~50. This is indeed what prior experiments quantitatively show, that the ring-ring conduction increases by a factor of 50 when the ring twist angle approaches zero.
For the capacitance of these junctions, we calculate = / using the charge distribution and potential profile obtained from DFT modelling of the junction under bias (see Supplementary Note 11), and obtain 2 =0.015aF and 1 =0.04aF. These values are reasonably close to simple parallel-plate estimates: 1Å. This suggests that the DFT values for 2D Au contacts are compatible with a real 3D junction given that it has a much larger contact area. We find that an adjusted value of 1 =1.5aF provides the best fit of the model to experimental data, and gives the fit described in the main text and Fig.4b,c. Indeed the junction capacitance from molecule -metal contact increases, because of the larger charge screening possible for the 3D Au contact geometry. The controlling parameter is the ratio √ 1 / 2~1 5, which can be intuitively thought as representing how much wider the contact size for the metal-molecule contact is than the C-C link between the rings inside the molecule. For the BMMBP, we can estimate the reduction in conductivity along the chain from its well characterised effect on tunnelling. 15,16 A universal scaling is found when inserting an additional carbon in the chain, with ~0.9. Using this in the model shows that the change in SERS at bias of 1V reduces as this molecular conductivity reduces (Supplementary Figure 14a).
For both BPT and BPTCN, the top contact conductance and capacitance are expected to decrease. Again using this in the model, including both 1 , 1 and 3 , 3 given the molecular asymmetry, shows that the data would be explained for capacitances more than 30x times smaller than the BPDT contact capacitance where the thiol binds directly to the Au (Supplementary Figure  14b).

Supplementary Note 11. DFT modelling of BPDT junction under non-equilibrium condition
A single molecule BPDT junction is first modelled with a 3D lattice of Au atoms in the leads (Supplementary Figure 17a), using double-zeta polarized basis set to verify consistency with the simplified 2D lattice, and with single-zeta basis set adopted later. This is to avoid computationally extremely expensive calculations using the 3D lattice of Au atoms in the leads. The energy profile shows qualitative agreement between the two models (Supplementary Figure 17). The simpler model with a 2D lattice described in the main text provides results consistent with the 3D lattice system and allows calculation of potentials and transport properties for various bias voltages including at higher bias voltages. Our calculations using this model show that the minimum energy configuration at zero bias is ~20°, which shifts to higher angles as the bias in increased (Supplementary Figure 18). Furthermore, it is found that the induced potential and charge fluctuations swap sides when the applied bias is reversed (Supplementary Figure 19). This confirms that one contact acts as rectifying while the other is conducting, as expected for back-to-back Schottky diodes (discussed in the main text). Initially upon SAM formation, electron transfer from Au to S creates a Schottky barrier at each electrode (blue arrows), with quasi-Fermi level shown dashed. a, At =90° the central CC bond presents a high barrier, which dominates transport. Hence for voltage applied the potential drops mostly across the centre of the molecule. b, At =15° there is no barrier in the centre of the molecule, so the only barriers are those set by the contact dipoles. Since the electron flux is proportional to | | 2 (with the flux velocity) and is conserved, regions with higher barrier give smaller ( ∝ ∝ 1/√2 ) and thus a larger wavefunction amplitude. This means that applying a positive bias to the right hand electrode (decreasing electron potential there) builds up electrons on the link-Au on the left side. This reduces the lhs surface dipole, decreasing the tunnel barrier which allows current to flow. The dipole on the rhs of the molecule is unchanged because there is no charge build-up (Fig. 4d