Light Emission and Conductance Fluctuations in Electrically Driven and Plasmonically Enhanced Molecular Junctions

Electrically connected and plasmonically enhanced molecular junctions combine the optical functionalities of high field confinement and enhancement (cavity function), and of high radiative efficiency (antenna function) with the electrical functionalities of molecular transport. Such combined optical and electrical probes have proven useful for the fundamental understanding of metal–molecule contacts and contribute to the development of nanoscale optoelectronic devices including ultrafast electronics and nanosensors. Here, we employ a self-assembled metal–molecule–metal junction with a nanoparticle bridge to investigate correlated fluctuations in conductance and tunneling-induced light emission at room temperature. Despite the presence of hundreds of molecules in the junction, the electrical conductance and light emission are both highly sensitive to atomic-scale fluctuations—a phenomenology reminiscent of picocavities observed in Raman scattering and of luminescence blinking from photoexcited plasmonic junctions. Discrete steps in conductance associated with fluctuating emission intensities through the multiple plasmonic modes of the junction are consistent with a finite number of randomly localized, point-like sources dominating the optoelectronic response. Contrasting with these microscopic fluctuations, the overall plasmonic and electronic functionalities of our devices feature long-term survival at room temperature and under an electrical bias of a few volts, allowing for measurements over several months.


Introduction
Plasmonic nanocavities with extreme light confinement allow electromagnetic interactions with few or even single molecules to be studied and tailored. 1,2They can be electrically connected, 3 resulting in molecular junctions that provide opportunities to measure both molecular transport [4][5][6][7] and plasmonically enhanced optical signals.][10] For instance, electrical transport and spectroscopy of metal-molecule-metal junctions reveal quantum properties originating from structural and chemical rearrangements, observable even at ambient conditions. 53][24][25] Such inelastic electron tunneling (IET) emission can efficiently couple to the plasmonic modes of the host nanocavity. 268][29][30][31] Probing an invidual molecule usually requires complex systems operating at ultra-high vacuum and/or cryogenic temperatures.
(PMJ) formed around a self-assembled monolayer (SAM), 6,[32][33][34] which operate at room-temperature and feature long-term functionality.A SAM of thiol-terminated molecules is created on top of two lithographically defined electrodes bridged by a few nanoparticles.The idea of bridging nanoparticles between electrodes coated with SAM was proposed by Amlani et al. in 2002. 35But for a long time, only the electrical properties of the molecular junctions were investigated.Kern et al. used a similar approach to trap nanoparticles with native ligands between pre-designed antenna structures to create a sub-nm gap and demonstrate electrically driven plasmonic emission by IET. 26 The ligand molecules there merely played the role of an insulating spacer.Inspired by these approaches, we created electrodes with slanted sidewalls that are optimized for collection of the optical signal once a nanoparticle falls inbetween to form hybridized plasmonic resonances. 36A similar design was simulated, tested, and implemented as a dual-band nanoantenna in a previous work. 37These nanocavity junctions provide significant plasmonic enhancement that allows for fast optical measurements to probe atomic-scale fluctuations at the metal-molecule contact through its impact on light emission mediated by the tunneling of electrons across the gap.The junctions can be studied over weeks and months, allowing us to thoroughly investigate the impact of atomic fluctuations in the junction on both its photoemission and electrical transport characteristics.The devices continue to be dominated by molecular transport until they get damaged under too high voltage (several Volts), too large optical powers (mW/µm 2 ) or by electrostatic discharge (ESD).
Our main finding is that a few randomly switching metal-molecule contact sites appear to control the macroscopic behavior of the device, despite the large number of molecules acting as a spacer.This microscopic dynamics is evidenced by discrete jumps in the light emission spectrum, as well as joint fluctuations of emission intensity and conductance that are consistent with a minimal model of fluctuating atom-molecule contacts.Similar point-like emission was recently evidenced in a large-area SAM tunnel junction between gold and eutectic gallium-indium alloy (EGaIn) contacts 38 where it was attributed to conformational changes in the molecule due to the excitation of vibrational modes, and no such blinking was observed in the current through the junction. 39In Ref. 38 the resonance and polarization of the point-like plasmon source were modified by the applied bias voltage.With our junctions, we find no such shift in the plasmonic resonance with applied voltage.The plasmonic response is fixed by the nanoparticle-electrode cavity and is excited by electron tunneling through the junction, and we observe correlated blinking in both the light emission and conductance.The spectral diffusion and intensity redistribution between the different plasmonic modes observed in our junctions resembles the movement of gold atoms locally modifying the emission efficiency of the individual plasmonic modes as demonstrated in photo-induced luminescence blinking of metal-molecule junctions. 40Our results therefore suggest that a phenomenology similar to that underlying the occurrence of picocavities in SERS, 41,42 of blinking in gold nanojunction photoluminescence 40 and of flickering in electronic Raman scattering 43 can also be driven by elec-trical bias.5][46] We demonstrate that a bulk nanoparticle-electrode system can capture fluctuations of atom-molecular contacts in both the conductance and light emission and extend the optoelectronic investigation of quantum-point contacts in STM/break-junctions to a completely different regime of conductance, junction geometry, and operation conditions.

Results and discussion
The plasmonic molecular junction (PMJ) is formed by gold nanoparticles (from a citrate-stabilized colloidal suspension) bridging two gold electrodes that are previously functionalized with a SAM of biphenyl-4,4'-dithiol (BPDT) molecules (Fig. 1a).One or few gold nanoparticles are trapped in the gap and establish electrical contact (see Supporting Information Sec. 1 for sample fabrication and experimental methods).Each nanoparticle linker forms in fact two junctions in series, but one of them typically dominates the series resistance and experiences most of the voltage drop, as will be discussed below.The device conductance G after fabrication falls in one of the following ranges: (1) Short-circuited contact where G ∼ 10 2 • G 0 where G 0 = 2e 2 /h is the quantum of conductance; this occurs in particular in junctions fused by ESD.(2) Open-circuited contact where G < 10 −6 • G 0 ; this occurs due to several reasons: (i) there is no nanoparticle bridging the gap; (ii) the size of the nanoparticle is smaller than the size of the gap; (iii) the nanoparticle does not establish successful contact with the electrodes on both sides.(3) Molecular contact where Sometimes an initially open-circuited junction can be brought into molecular contact by applying several volts.We also note that junctions featuring the largest conductances (G > 2 × 10 −3• G 0 ) cannot be studied for light emission, as under voltages above 1.2 V, the corresponding current exceeds the damage threshold of the device, which is typically a few µA and above which irreversible changes occur (see Supporting Information Sec.3.3).
The nanogaps formed by the SAM between the metallic parts have a width of around 1 nm that depends on the length of the molecules and their orientation with respect to the gold surfaces. 47ey support localized plasmonic resonances that offer extreme light confinement and good radiative efficiency 36 and allow to efficiently read out the plasmonically enhanced optical signals.3][24][25][26] An image of light emission from one PMJ overlaid with a dark-field scattering image is shown in Fig. 1c, with the corresponding bright-field image shown in Fig. 1d.The emission couples to the plasmonic modes of the cavities formed by the nanogaps between the nanoparticles and electrodes and can be observed in the far field.The light emission spectrum of BPDT junctions for different d.c.bias voltages and an exposure time of 20 s is shown in Fig. 1e.Note that there is a cut-off observed in the light emission spectrum at an energy close to the value of eV (V is the applied voltage), as indicated by the vertical line in Fig. 1e.This is also asserted in Fig. 1f, where the photon count rate increases nonlinearly beyond a threshold of about 1.25 V that matches the spectral response of the silicon detector having a cuton around this photon energy.While we expected the electrode-molecule-nanoparticle geometry to form two identical molecular junctions in series, the fact that the energy cut-off matches well with the applied voltage suggests that most of the voltage drops across one of the two junctions (consistent with previous observations of similar junctions 26 ) and the light emission happens through a single-electron inelastic tunneling process.Some overbiased light emission is also observed in our PMJs, with photon energies higher than the applied voltage (times the electron charge).This was previously demonstrated in STM-junctions, 27,[48][49][50][51] electromigrated junctions, 17,19,52 mechanical break junctions 53 and more recently memristive junctions. 54This emission was either attributed to hot carriers or coherent multi-electron scattering processes.In our experiment, the excess photon energy is in agreement with a moderate increase of the electron temperature in the junction 18 (see Sec. 5 in Supporting Information).
The two peaks in the spectra are attributed to plasmonic modes of the junction, also seen in the dark-field scattering spectrum (dotted lines) collected using a tungsten lamp.The nanoparticle-SAM-electrode design is adapted from a previous work involving a nanoparticle-in-groove dual resonant cavity 37 that exhibits several resonances at visible and NIR wavelengths.There is no evidence of hot electron emission in our system (see Sec. 5) and the emission is found to be shaped by the plasmonic response as discussed in Sec. 4 of Supporting Information.We refer to Ref. 37 for simulation of the visible plasmonic modes of a similar structure.The plasmonic modes are progressively populated upon increasing the voltage, as observed with the onset of emission from a plasmonic mode at higher energy for voltages above 1.7 V.
Despite using SAMs as molecular-ensemble contacts, we typically find conductance values that are rather in the range of few-molecule junctions. 7,55We, therefore, expect that the resulting light emission from IET originates from one or few spatially localized emitters (i.e., point sources) whose localization is varying over time across the electrode gap, as pictured in Fig. 1b.We now study the direct optical and electrical signatures associated with the dynamic localization of these point-like emitters as discussed in (Figs.2-4).
The current-voltage characteristics (grey curve in Fig. 1f) show the typical nonlinear electrical response of a molecular junction. 5Given the nanometer size of the molecular spacer, we propose that the main transport mechanism is tunneling.The thiol anchor group of the BPDT molecules forms strong covalent bonds with the gold atoms and the current flow should be mainly mediated by the HOMO levels. 56However, since we perform the experiment at room temperature the resonant molecular orbitals of BPDT cannot be identified from the current-voltage characteristics.Also, note that the HOMO-LUMO gap of BPDT molecules is high (3.85 eV 56 ) and hence de-excitation through the molecular orbitals is not involved in the light emission process for the voltages considered in this study.There could be additional tunneling contribution from the citrate molecules present in the gap as BPDT may not replace all the citrate molecules around the nanoparticles.However, the citrate molecules are weakly bonded to the gold atoms and form an even smaller nanogap between the electrodes. 57Pure citrate junctions are characterized in Supporting Information (Sec.3.1) which show a much lower yield of conductive junctions compared to the BPDT junctions.The presence of BPDT molecules on the surface of gold readily attracts nanoparticles in the nanogap to bridge the electrodes and establish electrical contact.
In order to further elucidate the nature of the spacers in the PMJ, a progressive addition of BPDT to a native PMJ with a pure citrate spacer is carried out.A PMJ is first formed by depositing citrate-capped nanoparticles in the electrode gap.A solution of BPDT molecules is dropped on the become more pronounced when the solution of BPDT is present.After the droplet evaporates, some BPDT molecules are expected to replace the citrate spacer in the nanogap.We find that the D.C. conductance (Fig. 2a) and the current-voltage characteristics (Fig. 2b) of the PMJ with BPDT fluctuates more than the initial PMJ with citrate spacer (see also Sec. 3.2 of Supporting Information).
The increased fluctuation of current density in BPDT junction is attributed to the formation of goldthiol 'staples'.The gold-thiol bond is capable of lifting adatoms on metal surfaces 41,42,58 and were found to increase the frequency of these events.Apart from an increase in the average conductance value post-functionalization (that may further evolve over long measurement times, see Supporting Information Sec.6), we observe clear telegraph noise that is consistent with a randomly switching contact at the single-molecule level (Fig. 2a).The biphenyl molecule's conductance depends on its conformation 55 and coordination geometry of the terminal thiols to the gold atoms. 59However, the changes in conductance due to these intramolecular phenomena are averaged out at room temperature where the aromatic systems are oscillating around an average position.The large discrete conductance jumps, which can reach an order of magnitude in some cases, are mainly attributed to the binding and unbinding of metal-molecule contacts, 60 the time-scale of which has been found to be on the order of a second at room temperature. 422][63][64] The conductance variation due to conformational change is much lower in magnitude. 55,65Breaking of a single molecular contact mostly occurs at the Au-Au bond which has a bond strength of 1 eV as compared to the Au-S bond strength of 1.6 eV. 66us, measuring the conductance of the PMJ provides information about atomic fluctuations at the metal-molecule interface.The point-like IET emitters resemble randomly occurring adatom protrusions at metallic nanogaps that can locally affect the plasmonic response.These atomic-scale events are usually probed by laser spectroscopy, mainly by SERS 41 and TERS 30 and more recently manipulated by STM in the regime of quantum point contact. 20,45,46Here, we show a novel probe of atomic-scale fluctuations driven by electrical current in a robust plasmonic molecular junction: our data are consistent with the picture of randomly occurring adatoms that create preferential sites for conductance and light emission.
Our PMJs remain functional over weeks and months (see Sec. 6 in Supporting Information), allowing us to perform in-depth characterizations of their dynamic behavior.In particular, we can correlate the electrical properties with optical signals associated with the fluctuating nature of metal-molecule contacts, as illustrated in Fig. 3.We investigate these fluctuations under a constant bias using a single photon counting module (SPCM) for faster acquisition synchronized with the conductance measurement.Our general finding is that light emission intensity and conductance value are strongly correlated despite the relatively large area of the junction and the presence of hundreds of molecules.It is compatible with the existence of a few molecular sites that dominate both transport and light emission, and whose number fluctuates over time.For the same junction, the conductance-emission correlation can switch between positive and negative trends over extended measurement times, as illustrated in Fig. 3e-g (additional data sets are provided in Supporting Information Sec. 9).The light emission from the IET process is, however, expected to vary linearly with current 67 and to display abrupt changes only when switching from tunneling to quantum point contact. 24,27The deviation from linearity and the periods of negative correlation observed in our PMJs will now be explained by the interplay between two nanojunctions in series.As mentioned earlier most of the voltage drop happens across one of the two junctions in series and the light emission is expected to originate from this dominating one as illustrated in Fig. 3d.
Hence, the conductance G 1 of the non-emitting junction is assumed to be much greater than G 2 , that of the light emitting junction.Consequently, small fluctuations in G 1 have insignificant impacts on overall conductance and emission intensity.The instantaneous light emission intensity depends on three major components: (i) the current through the emitting junction i(t); 27 (ii) the voltage across the emitting junction V mid − V R (through the energy-dependent responsivity of the detector); and (iii) the efficiency for converting electrical energy into photons, which varies with the emitter location 40 (see Fig. 4a,b).We will show that contributions (i) and (ii) can together explain both negative and positive correlations of photon counts vs. conductance observed in the same junction at different times, while (iii) results in a variation of the overall efficiency of the same PMJ accounted for by the prefactor A below.The intensity of light emission is therefore modeled by the equation where f (V (t)) is an experimentally determined function that accounts for the time-averaged voltagedependent emission and detection efficiency.The linear dependence on i(t) is to be interpreted as a first-order Taylor expansion valid for a limited range of conductance fluctuations.From the conductance data collected at constant bias, we fit the data from Fig. 3 to obtain the values of the parameters G 1 and G 2 (t).These parameters are used to derive the values of V mid − V R for each value of the conductance.
The empirical relation from equation 1 is then used to obtain the fit shown as solid lines in Fig. 3e,g.When the change in V mid − V R is negligible, the monotonous dependence of photoemission on the current dominates and leads to positive correlations between photon count rate and conductance.This can happen when G 1 >> G 2 such that V L − V mid approaches 0. This is the case for the fit parameters G 1 = 10 −2 G 0 and ⟨G 2 ⟩ = 10 −5 G 0 in Fig. 3e (the notation ⟨G i ⟩ is used to represent the mean value of the conductance of junction i=1,2).Conversely, negative correlations are observed when V mid − V R changes significantly so that the voltage dependence of emission embodied in f (V (t)) dominates over the dependence on current.This is the case for fit parameters 3g.Depending on the values of G 1 and G 2 (t) the two opposite regimes can be realized.In all fits, G 1 > G 2 as predicted, consistent with our initial assumption that one of the two junctions has a higher conductance compared to the other one.
More details on the data fitting are provided in Supporting Information Sec. 8 and more examples of measured and fitted correlations are presented in Sec. 9.In Fig. 4 we report the fluctuations in light emission spectrum occuring together with conductance fluctuations in the same PMJ.The light emission spectrum is dominated by two peaks at photon energies of 1.56 eV and 1.72 eV (as estimated by the Lorentzian fits in Fig. 4b), but their respective intensities can randomly change over time.Intermittent blinking in single-molecule junctions can happen from conformational or structural changes in the molecular backbone.But such effects take place in timescales of 10 −11 to 10 −9 s and are averaged out in large area molecular junction. 60,68,69A gold-SAM-EGaIn junction with a small contact area was able to capture such intermittent blinking in light emission but found no correlation with current fluctuations. 38,39The BPDT molecules can display conductance switching by a change in the tilt of the molecule in the junction, its binding conformation with the gold atoms, and the rotation between the two benzene molecules. 55,59These effects could explain blinking in the intensity of the light emission due to the change in conductance 70 but do not explain the reshaping of the plasmonic modes.
Instead, we argue that dynamically occuring localized atom-molecular contacts naturally explain the observed fluctuations in the light emission spectrum.We hypothesize that the changing ratio between emission intensities at the different plasmonic peaks is caused by the random appearance of point-like emitters at different positions inside the host nanocavity.Depending on its localization, a point-like emitter couples more efficiently to a particular mode of the plasmonic response.The far-field spectra of these point-like emitters depend on the overlap between the emitter position and the near-field distribution of the different gap modes.Hence, the spatial wandering of point-like emitters in the near-field causes an apparent spectral wandering in the far-field.This has been demonstrated in photo-excited luminescence of gold clusters in a plasmonic cavity by Chen et al.
in Ref., 40 where simulation of this effect in a gap nanocavity similar to the ones studied here was performed.
The relative conductance fluctuations in our PMJs are found to be significantly enhanced under increasing electric bias (see Supporting Information Sec.7) while there is no clear temperature rise seen in the overbias emission, within the fitting uncertainty (Fig. 16).These observations suggest that a non-thermal mechanism may contribute to the creation of new atomic protrusions and could be connected to recent findings on optically-induced picocavities, 71 where external electric fields (at the optical frequency) were argued to lower the energy barrier for the creation and relaxation of a gold adatom.However, future dedicated experiments are needed to confirm the non-thermal mechanism that may be at play in the junction.

Conclusion
In summary, using a simple and scalable self-assembled geometry, we demonstrated how the struc- the exact location of these dominant conduction channels governs the relative coupling of IET to distinct plasmonic modes.While the switching of atom-molecule contacts is not a deterministic process, we could repeatedly drive our system in this regime by applying a sufficient voltage across the junction.
With controlled capillary assembly, 72 atomic force microscope, 26 transfer printing 73 or dielectrophoresis, 74 it is possible to make on-demand single nanoparticle junctions (some preliminary results are shown in Sec.11 of Supporting Information).This would enable simultaneous SERS or luminescence and electrical measurements on the PMJ, which are currently hindered by the presence of multiple nanoparticles nearby.The resulting single-nanoparticle PMJ offers a unique opportunity to connect the microscopic origins of various phenomena such as picocavity in SERS, 41 flares in electronic Raman scattering, 43 blinking of gold photoluminescence, 40 and flucutations in IET and conductance studied here.The PMJs should be particularly useful in understanding their formation mechanisms, including their dependence on the electric field 71 and their non-thermal origin.
1 Experimental methods

Sample fabrication
The sample fabrication procedure is illustrated in Fig. 5.A 4-inch diameter silicon wafer with a thickness of 380 µm and double-sided 300 nm SiO2 layers (NOVA wafers) is used as the substrate for the fabrication process.The substrate underwent a spin coating process, with a 700 nm layer of LOR 5A followed by a 1600 nm layer of AZ1512 photoresist.A first photo-lithography is employed to define the contacts and gold electrodes by metal deposition and lift-off.Utilizing the deep-UV photolithography technique, the electrode pattern is formed on the substrate.Following the photoresist development step, a 3-nm-thick Cr adhesion layer and a 150-nm-thick gold layer are thermally evaporated onto the substrate at a rate of 0.5 nm/s.The subsequent metal lift-off process is performed using Remover 1165 -NMP.
Subsequently, electron-beam lithography and collimated ion-beam etching are employed to carve the ∼ 150 nm gaps separating the two electrodes, creating the space for the bridging gold nanoparticles to fit.We used 100 keV electron-beam lithography and PMMA 950 K as the e-beam resist.
Following the development of the resist, the nanogaps are etched with a collimated ion-beam at an angle of -10 • , resulting in a 150 nm wide and 4 µm long V-shape trenches, oriented perpendicular to each electrode.The actual length of the gold electrode is 2 µm; the e-beam step creates a longer gap that extends to the substrate to safely mitigate alignment errors.To remove PMMA, the sample is immersed in warm Remover 1165 -NMP for a duration of 6 hours.
After the fabrication of electrodes, the wafer is diced into individual chips, with each chip size (7 mm x 10 mm) containing 25 pairs of electrodes.Following the dicing process, the chips underwent the formation of molecular spacer and nanoparticle bridges, as described in section 1.2.
Chips are then forwarded for wire bonding to connect the electrodes to a printed circuit board (PCB).This wire-bonding step enabled the electrical connection between the fabricated structures and the external circuitry, facilitating further characterization with the setup described in section 1.3.

Formation of molecular spacer and nanoparticle bridge
Biphenyl-4,4'-dithiol (BPDT) molecules in solid form from Sigma Aldrich are dissolved in ethanol to form a 3 mM solution.Each chip is incubated in 3 ml of this solution for 2 hours. 47The sample with the self-assembled monolayer (SAM) of BPDT is then cleaned several times with ethanol to remove unbound molecules.To form the nanoparticle bridge, 5 µl of 1:100 diluted Nanopartz 150 nm OD100 gold nanoparticle solution is drop cast on the sample and evaporated until the edge of the droplet passes across the slit.Due to the capillary forces, the edge has a higher concentration of nanoparticles and leaves some nanoparticles in the slit. 72In the final device, the electrodes with molecular SAM are bridged by a few nanoparticles as shown in the SEM image in Fig. 6.
We found that instances with a single nanoparticle have a low probability of establishing good contact with the electrodes.Having more nanoparticles increases the probability of making a successful contact.Note that all the discussion in the main text does not rely on having a single particle bridging the electrodes; having more than one simply increases the total number of molecules potentially participating in conductance and light emission by inelastic electron tunneling.2][63][64] The conductance change due to conformational change of the molecules is much lower in magnitude. 55,65e appearance of three peaks in the histogram (Fig. 9d) that are equally spaced also indicates that it does not originate from conformational changes of the molecules.

Experimental setup
3 Native ligands vs. BPDT The gold nanoparticles are non-covalently capped with citrate molecules to avoid aggregation of the suspension.When these nanoparticles are deposited in the gap between the electrodes, the BPDT molecules on the surface of the electrodes are expected to replace the citrate layer to form the molecular junction, but some ligand molecules could still be present in the nanogap.To better understand how citrates impact the behavior of the junction, we compare here BPDT junctions with pure citrate-spaced junctions (no incubation in BPDT).

Yield of the devices
Comparing BPDT and pure citrate spacers (2 × 3 chips, i.e. 75 devices each), the yield of BPDT in establishing measurable electrical contact is 43% while that of citrate is 23% (Fig. 10).The dashed line marks the measurement limit of our probe station, and the devices with conductance below   A qualitative observation is that PMJs with BPDT spacers are found to feature more pronounced fluctuations in conductance than citrate spacers.However, to clearly distinguish the two kinds of PMJs through their I-V curves, we believe that cryogenic, molecular-specific characterization like inelastic tunneling spectroscopy 75 is required.

Inelastic electron tunneling (IET) light emission
We observe light emission by IET from both kinds of PMJ (with citrate and BPDT spacers).Our PMJs get damaged when the current through them exceeds a few µA and hence a current compliance of 1 µA is maintained.As a consequence, PMJs with higher conductance (above 2 × 10 −3 G 0 at 5 mV) cannot be used for IET experiments as their currents under the voltages where IET becomes measurable are higher than the current compliance.
The voltage threshold at which IET is detected is shown for different PMJs with citrate and BPDT spacer (Fig. 12).For most devices, the voltage threshold is close to the cut-on energy of the detector efficiency.For other devices with lower conductance, the photon yield is low and they require much higher voltages for IET to be detectable.

Plasmonically enhanced emission
The design of the nanoparticle-SAM-electrode is adapted from previous work 37 where the numercial simulation of the plasmonic modes was performed.(In Ref. 37 the 2 µm groove acted as an antenna at mid-IR wavelengths, which is not needed here).The simulations predict at least two distinct radiative modes at visible wavelengths that arise from the hybridization of the individual gap modes formed by each nanoparticle-on-mirror (NPoM) cavity between the nanoparticle and each side of the electrode. 36Due to the added complexity in the electrically-contacted device studied here that often involved multiple nanoparticles in different positions within the nanogroove, we did not repeat the simulation of the plasmonic modes for the individual devices and refer the reader to Ref. 37 for a general overview of the plasmonic response.
The electrical excitation of localized plasmons is confirmed from the dark-field scattering and the polarization response.We first obtain the dark-field scattering spectrum from the junction for P and S polarization of the incoming white light.The dark-field scattering spectroscopy is obtained using illumination from a tungsten lamp at about 15 • incidence angle with respect to the sample plane as illustrated in Fig. 7.The spectrum is collected using a multimode fiber and measured with a QE pro-Ocean Insight spectrometer.The S-polarized excitation results in two dominant scattering peaks that are suppressedd under P polarization.The same peaks are excited by the IET process with an external bias voltage, indicating that they are of plasmonic origin.
The dark-field scattering spectrum and IET emission of a PMJ at 1.9 V is shown in Fig. 13 for mixed polarization.Some additional features in the dark-field spectrum are not found in the IET spectra because the dark-field scattering signal comes from the entire junction in the collection area including the nanocavities formed by multiple nanoparticles in the electrode gap.IET on the other To evidence the role of plasmonic resonances in the out-coupling of light emission, a linear polarizer is placed in front of the spectrometer, and the light emission spectrum is collected along different electric field orientations (Fig. 14).The variation of the spectrum with the polarization angle is consistent with the enhancement and reshaping of emission by the plasmonic cavity.

Thermally assisted emission
The overbias contribution to the light emission spectra is attributed to the emission from a globally hot system at equilibrium.While it is possible for the electron temperature to be significantly higher than the lattice temperature, the observed overbias tails do not require such an assumption to be explained in our experiment.

Electrically induced fluctuations
The random fluctuations in the conductance are found to intensify with an increase in the applied bias voltages, as illustrated in Fig. 19.The standard deviation of the conductance from each measurement is monotonously increasing with the increase in the voltage both in the forward and reverse sweep of voltages (Fig. 19c).The device is prone to significant changes in its conductance state at high voltages as observed in the transition of conductance at 1.5 V.This indicates that fluctuations are strongly current-or voltage-driven, possibly involving a non-thermal mechanism given the very good heat dissipation provided by the large gold electrodes.
Yet, we remark that once the standard deviation of the conductance is further normalized by its mean value (Fig. 19d), the hysteresis from Fig. 19c disappears.Two lines of explanation are proposed.First, it could be that more fluctuations correlate with a higher number of conducting

Details of the fit
To determine the photon emission as a function of conductance for a fluctuating PMJ, we model the intensity of light emission by the equation where A is a scaling factor to account for the overall efficiency of IET and photon detection, i(t) is the current, and f (V (t)) is an experimentally determined function that accounts for the voltage-

PMJ with BPDT spacer
Another example of a PMJ showing the two regimes of conductance correlations is shown in Fig. 24 for a BPDT spacer.Noteworthy is the stability of conductance and emission intensity values observed between the end of a measurement and the start of the next one.This observation further suggests that the fluctuations are current-or voltage-driven, with little contribution from ambient thermal energy.

Picocavity events in SERS and conductance
To show evidence of picocavity formation during intermittent blinking, we performed a combined conductance and SERS measurement on our PMJ.The picocavities in the nanogap create strong optical field gradients that modify the Raman selection rules and create additional vibrational peaks in the SERS spectra. 41In Fig. 28, we show a few picocavity events from PMJs that are correlated with the conductance jumps.Such correlated events are very rare in SAM molecular junctions because we need to capture the conductance and SERS signal from the same molecule to observe the correlation.SERS in general is a collective signal obtained from all the molecules present in the nanogap, whereas conductance is obtained from a very few molecules connecting the electrodes.Thus the Raman signal is dominated by several other molecules that are not involved in the transport.Even in the presence of picocavities, it needs not occur from the same molecule that transmits the electrons.Such events are much less probable in PMJ as opposed to break junctions with atomic-sized tips. 76PMJs with single nanoparticle devices could be useful to perform such combined SERS and conductance measurements, where there will be better chances to observe direct evidence of picocavity events in transport.

Outlook: Formation of single nanoparticle junction
To trap a single nanoparticle in the nanogap, the dielectrophoresis (DEP) technique with feedback from optical imaging is used.DEP involves controlling a polarizable object with a non-uniform electric field. 77When a particle is exposed to a non-uniform electric field, it is polarized and builds its own induced dipole moment.The DEP forces steer the particle towards the region of higher field intensity.The DEP force experienced by a spherical nanoparticle depends on several factors including the field intensity variation, the surrounding medium, and the size of the particle.Hence, DEP can be used to trap nanoparticles of varied concentration by applying an oscillating voltage of varied magnitude and frequency.The DEP forces can be expressed by the equation, where ϵ m is the permittivity of the medium, R is the radius of the nanoparticle, f CM (ω) is the Clausius-Mosotti factor, and E rms is the rms value of the electric field.The Clausius-Mosotti factor describes the complex polarizability of the particle and is given by where ϵ p and σ p are the permittivity and the conductivity of the particle, while ϵ m and σ p correspond to that of the medium, ω = 2πf is the angular frequency.
For the trapping to occur, the DEP forces have to be larger than the thermal motion described by, where k B is the Boltzmann constant and T is the temperature.
We ideally want to trap single nanoparticles in the gap.[80][81] An optical microscope was modified to include probes that are attached to the sample holder (Fig. 29a).The probes can contact the chip directly for the DEP experiment.A water immersion objective is used to observe the dark-field image throughout the trapping process.The AC voltage is stopped once the particles are trapped, and the excess solution is blow-dried.
Sometimes the particles are trapped with AC voltage but get released when the voltage is turned off.In such cases, a small DC voltage of 100 mV helps to stick the particles in the gap.With this approach, we could obtain devices with almost 100% yield.Sometimes, if the image was not clear enough due to excess scattering from the electrodes, we might end up with more than one nanoparticle.Otherwise, a single nanoparticle junction could be achieved reproducibly.

Figure 1 :
Figure 1: (a) Schematic illustration of the PMJ.(b) Illustration of atomic fluctuations and light emission at the metal-molecule-metal contacts.(c) Grayscale dark field image overlaid with the colored light emission image under 1.6 V bias for a typical PMJ.(d) Corresponding bright field image.(e) Light emission spectrum for various DC voltages with 20 s exposure time.The vertical lines mark the onset of overbias emission.Dotted lines represent the darkfield scattering spectrum from a white light source.(f) Current vs. voltage characteristics of the junction (gray curve) and simultaneously collected photon counts (red curve) as a function of applied voltage.The shaded background represents the quantum efficiency of the detector with voltage values translated to photon energy in eV.Inset: zoomed-in view with the vertical line showing the onset of photon detection.

Figure 2 :
Figure 2: (a) Conductance of a particular PMJ with citrate (red) spacer in a dry state, during deposition of BPDT molecules (blue) and after deposition of the BPDT molecules (green).(b) Current-voltage characteristics of the PMJ before and after deposition of BPDT molecules.(c) Magnitude of conductance changes evaluated from several measurements where clear intermittent blinking was observed.The data is collected from 6 distinct devices(see Fig. 9 in Supporting Information).

Figure 3 :
Figure 3: (a-c) Conductance (blue lines) and photon counts simultaneously measured with SPCM (red lines) for a particular PMJ.Both data are summed into 500 ms time bins to improve the signal-to-noise ratio.(d) Schematic representation of the device with fluctuating conductance model.(e-g) Corresponding correlation plots of the data from (a-c) display the switching between positive and negative correlations (see transition within (f) -the data points are color-coded from black to red, representing the progression of time).The solid lines in (e) and (g) are from the model discussed in the text, with the corresponding parameters indicated on the right panel.

Figure 4 :
Figure 4: (a) Atomic fluctuations captured in the light emission spectra (top panel) and the electrical conductance G/G 0 (bottom panel) collected simultaneously from the same PMJ.A constant DC bias voltage of 1.9 V was applied, spectra were collected with 1 s exposure time and the conductance data were measured simultaneously at a 2 kHz sampling rate.(b) The grey curve represents the average of all the spectra in (a), and the black solid line illustrates the fit obtained from two Lorentzian peaks (orange and green lines).(c) A subset of electrical transport data from (a) showing discrete jumps in the conductance.The histogram of the conductance data is shown in the right panel, suggesting a quasi-continuum of conductance states.
rearrangements of atom-molecule contacts in a plasmonic molecular nanojunction are imprinted on the electrical transport as well as the tunneling-induced light emission fluctuations.The non-monotonous correlations between electrical conductance and IET intensity are explained by a simple phenomenological model taking into account that our devices consist of a double junction in series.The large intermittent fluctuations in conductance and IET emission are proposed to arise from the binding and unbinding of molecular contacts due to the movement of gold adatoms;

Figure 6 :
Figure 6: SEM image of a plasmonic molecular junction (PMJ) with multiple nanoparticles in the gap.
then connected by wires to an external circuit consisting of a voltage source and current measurement unit.The NI 9263 voltage output module can output between -10 V and +10 V. Current-voltage characteristics are measured by applying a triangular voltage sweep.The current is fed to the DLPCA 200 transimpedance amplifier which is automatically gain-switched depending on the magnitude of the measured current by a home-built LABVIEW program.The voltage output of the a NI 9040-cRIO controller and are synchronously operated via LABVIEW with an internal clock.Finally, to synchronize the optical measurements with the spectrometer, the NI 9402 pulse generator is used to send a trigger pulse to start the spectral acquisition.The NI 9402 also acts as a digital readout module for photon counting measurements with SPCM.

Figure 8 :
Figure 8: Image of a chip with electrodes wire-bonded to a PCB, which will be further connected to an external electrical circuit.

Figure 9 :
Figure 9: Intermittent blinking in conductance observed in different measurements across different devices at a D.C. voltage of 5 mV this range are considered open circuits.Considering only the PMJs that we identify as molecular contacts (10 −5 G 0 < G < 10 −1 G 0 ), the histogram peaks at around 3 × 10 −4 G 0 for the BPDT junction and around 3 × 10 −2 G 0 for the citrate junctions, which could be related to the smaller size and possibly flatter orientation of the citrate molecules.Here, G 0 = 2e 2 /h refers to the quantum of conductance.

Figure 10 :
Figure 10: (a) Conductance measured at DC bias of 5 mV for several devices with BPDT (green) and citrate (red) spacer.The dashed line marks the measurement limit of the probe station.BPDT PMJs have a yield of 43% and the conductance histogram has a peak around 3 × 10 −4 G 0 .Citrate PMJs have a yield of 23% and the conductance histogram has a peak around 3 × 10 −2 G 0 .Molecular structure of (b) BPDT and (c) citrate molecule

Figure 12 :
Figure 12: Photon counts versus voltage for several PMJs for (a) citrate spacer and (b) BPDT spacer.Vertical lines mark the voltage threshold for onset of observable IET.(c) Voltage threshold plotted in the form of histogram.

Figure 13 :
Figure 13: Light emission spectra at 1.9 V (red) and dark-field scattering spectra of a PMJ with S-(dotted black) and P-Polarized (dotted blue) excitation.

Figure 14 :
Figure 14: Light emission spectra at 1.8 V collected for various orientations of a polarizer in front of the spectrometer (0 • -black; 45 • -blue; 90 • -red represent the angle of the polarizer with respect to the electrode length), showing spectral reshaping of emission through the plasmonic cavity.Dark lines are the averages of individual exposures shown in faded lines.Inset -bright field image of the PMJ showing the orientation of the electrode.In all experiments of the main text, no polarizer is used in the detection path.

Figure 15 :
Figure 15: SEM image a PMJ with (a) electrode alone and (b) electrode with nanoparticle in the gap.(c) Their corresponding SERS spectra.
Fig.16shows the temperature estimated from the tail of the overbias emission with Boltzmann fit,I thermal (λ) = A.hc λ.(exp( hc λk B T ) − 1) + B (2)where λ is the wavelength, h is Planck's constant, c is the speed of light, k B is Boltzmann constant, T is temperature, A and B are constants.Even without factoring out the plasmonic response (which is difficult to do properly due its fluctuating and voltage-dependent contribution) the thermal fit reasonably estimates slightly elevated temperatures of the PMJ compared to room temperature.

Figure 16 :
Figure 16: Thermal fit of the light emission spectra in the overbias region for various DC bias voltages (exposure time -20 s).The energy values in the x-axis are computed with respect to the maximum photon energy corresponding to the bias voltage.
. The conductance state changes over time mostly during measurements as shown for two devices in Fig.17a-b.These two PMJs, for example, remained functional for more than 100 days.

Figure 17 :
Figure 17: Conductance of two repeatedly measured PMJs (a)-(b) and several other PMJs (c) over several days (DC bias -5 mV).Each color in (c) corresponds to an individual PMJ.The devices show excellent long-term survival.

Figure 18 :
Figure 18: Stability of a PMJ for 1 hour (DC bias -5 mV) in blue.Red line shows the open circuit noise from the electronic measurement unit.
channels.Second, it may indicate that more fluctuations correlate instead with the increase in the dissipated power through local Joule heating.Distinguishing the thermal and non-thermal contributions to the conductance fluctuations therefore requires further investigation of these effects with spectroscopic thermometry techniques or direct temperature scanning experiments.

Figure 19 :
Figure 19: (a) Conductance of a BPDT junction measured for 120 s consecutively at each voltage from 0.1 V to 1.5 V and back to 0.1 V in steps of 100 mV.(b) Corresponding histograms of the conductance data.(c) Standard deviation (σ) of the conductance of each measurement in (a) plotted against the applied DC voltage.(d) The standard deviation of the conductance divided by its mean value (µ) plotted against the voltage.

Figure 20 :
Figure 20: Experimental data of photon emission as a function of voltage (red dots) and the corresponding fit (black line).

Finally, from
the values of f (V (t)) and i(t), the emission from PMJ can be fit with the equation 3 with a factor A. The fit obtained for different sets of the fitting parameters is shown in Fig.21.

Figure 21 : 9 . 1
Figure 21: Multiple fits of experimental data discussed in the main text with different values for G 1 , G 2 (t), and A for (a) positively correlated and (b) inversely correlated regimes of conductance.

Figure 22 :
Figure 22: Correlatated fluctuations in conductance and light emission obtained from different PMJs with varied numbers of nanoparticles in the electrode gap along with their SEM image

Figure 23 :
Figure 23: (a) Conductance and (b) Photon counts represented with errorbar for the devices in Fig. 22, with the color of each data point corresponds to the color of the correlation scatter plot.

Figure 24 :
Figure 24: (a)-(e) Conductance (blue lines) and photon counts (red lines) simultaneously measured with SPCM for a particular PMJ with BPDT spacer.Both data sets are summed into 500 ms time bins.(f)-(j) corresponding correlation plots displaying the switching between positive and negative correlations.Solid line in each plot depicts the fit of the experimental data and the corresponding fit parameters are mentioned in the box below.

Figure 25 :
Figure 25: Correlation between photon count rate and conductance measured on different BPDT-spaced PMJs.Each plot from (a)-(f) represents an individual PMJ.The different colors indicate the corresponding applied DC bias.

Figure 26 :
Figure 26: (a)-(h) Conductance (blue lines) and photon counts (red lines) simultaneously measured with SPCM for a particular PMJ with citrate spacer.Both data are summed into 500 ms time bins.(i)-(p) corresponding correlation plots displaying the switching between positive and negative correlations.Solid line in each plot depicts the fit of the experimental data and the corresponding fit parameters are mentioned in the box below.

Figure 27 :
Figure 27: Correlation between photon count rate and conductance measured on a few other citrate-spaced PMJs.Each plot from (a)-(b) represents an individual PMJ.The different colors indicate the corresponding applied DC bias voltages.

Figure 28 :
Figure 28: Time series of SERS spectrum along with the conductance showing picocavity events.
About 50 µl of DI water is placed between the objective and the sample for imaging.8 µl of 10 OD nanoparticle solution in DI water is added to the solution.AC voltage between 1-3 V with a frequency of 500 kHz -1 MHz is used to trap the nanoparticles.Different samples (normally with varied electrode widths and nanogap) require slightly different voltages and frequencies to trap the particles.The trapping can be monitored live from dark-field imaging.The DF image before and after trapping the nanoparticles is shown in Fig. 29b,c.

Figure 29 :
Figure 29: (a) Optical microscope with probes for DEP trapping.DF image of the electrode structures before (b) and after (c) trapping the nanoparticles by DEP.For better visibility, white dotted lines are overlayed on the DF image in (b,c) to mark the boundaries of the electrodes.White arrows point to the electrode gap.The yellow circle marks the location of the trapped particle.