Quantum Tunneling Induced Optical Rectification and Plasmon-Enhanced Photocurrent in Nanocavity Molecular Junctions

Molecular junctions offer the opportunity for downscaling optoelectronic devices. Separating two electrodes with a single layer of molecules accesses the quantum-tunneling regime at low voltages (<1 V), where tunneling currents become highly sensitive to local nanometer-scale geometric features of the electrodes. These features generate asymmetries in the electrical response of the junction which combine with the incident oscillating optical fields to produce optical rectification and photocurrents. Maximizing photocurrents requires accurate control of the overall junction geometry and a large confined optical field in the optimal location. Plasmonic nanostructures such as metallic nanoparticles are prime candidates for this application, because their size and shape dictate a consistent junction geometry while strongly enhancing the optical field from incident light. Here we demonstrate a robust lithography-free molecular optoelectronic device geometry, where a metallic nanoparticle on a self-assembled molecular monolayer is sandwiched between planar bottom and semitransparent top electrodes, to create molecular junctions with reproducible morphology and electrical response. The well-defined geometry enables predictable and intense plasmonic localization, which we show creates optical-frequency voltages ∼ 30 mV in the molecular junction from 100 μW incident light, generating photocurrent by optical rectification (>10 μA/W) from only a few hundred molecules. Quantitative agreement is thus obtained between DCand optical-frequency quantum-tunneling currents, predicted by a simple analytic equation. By measuring the degree of junction asymmetry for different molecular monolayers, we find that molecules with a large DC rectification ratio also boost zero-bias electrical asymmetry, making them good candidates for sensing and energy harvesting applications in combination with plasmonic nanomaterials.

M olecular electronics holds the promise of ultimate circuit miniaturization, where single or few molecules perform electronic functions at the nanometer and sub-nanometer scale, with adaptable functionality and potential gains in energy efficiency, circuit density, and speed. 1 Fundamental aspects of molecular junctions have been extensively explored over the past few decades with a range of benchmark techniques including scanning probe microscopy, mechanical break junctions, and liquid alloy contacts. Supported by increasingly refined theoretical and computational models, these methods have unveiled the fundamental properties of transport through various types of molecules and identified the role of molecular orbitals, chemical functional groups, and molecule−electrode interfaces. 2−4 At the same time, considerable effort has been put into the design and fabrication of molecular electronic devices, generally built from fixed electrodes bridged by an ensemble of molecules. 5 Producing devices on the basis of molecular junctions however has proved challenging, as seen in the large number of approaches including growing electrodes through nanopores, 6 deposition of 2D materials, 7 coating with conductive polymers, 8 trapping within nanoparticle networks, 9 and immersion in electrochemical environments 5,10 among others. Despite these efforts many challenges still remain, such as forming junctions with a reproducible number of molecules and preserving molecular integrity during device processing, all while using simple and scalable fabrication methods and geometries that allow for exploiting the properties of small molecular ensembles.
One class of devices that exploits molecular junctions are metallic nanogap antennas, where incident light triggers electrical charge transport across the junction gap amplified by plasmon-enhanced electromagnetic fields. Nanogap antennas have attracted strong interest in recent years for optical detection and energy harvesting 11,12 and light generation, 13 but production of photocurrent from nanogaps has been particularly hard to control consistently due to poor definition of the optical confinement, preventing quantification of the local optical fields. 14,15 Here we demonstrate an advantageous molecular junction geometry for molecular optoelectronics, where a monolayer of molecules is trapped between a single Au nanoparticle (AuNP) and a flat Au surface integrated within a layered electrode structure. This electrically contacted junction is optically accessible and the AuNP in this nanoparticle-on-mirror (NPoM) geometry 16 serves the double purpose of defining the junction area and plasmonically enhancing the optical field in the gap by several hundred-fold. Upon illumination, local nanoscale asymmetries within the junction produce optical rectification, detected as a photocurrent signal that is boosted by plasmon oscillations in the NPoM structure. Molecules enable the functionality of these devices by defining the nanometer-sized tunneling gap and are used here to evaluate the contribution of various molecular parameters to junction asymmetry.

RESULTS AND DISCUSSION
The nanogaps are formed by depositing AuNPs on a bottom Au electrode patterned on a SiO 2 or glass substrate which has been coated with a self-assembled molecular monolayer (SAM) (Figure 1a). The AuNPs are partially embedded in an insulating PMMA layer, leaving the top half of the NPs exposed, and finally this exposed region is coated with a semitransparent Au film (50% transmission in the visible) to create a top electrode. The intermediate PMMA layer prevents electric contact between the electrodes away from the NPs. The bottom and top Au electrodes are patterned by evaporation through a pair of custom-designed shadow masks (therefore without lithography), so the SAM is never exposed to photoresist, solvents, or UV light. The shadow mask geometries create arrays of approximately 300 cross-bar junction devices per sample (Figure 1b,c, of area 600−2500 μm 2 depending on the device, each containing a few AuNP junctions), which are electrically addressed individually by contacting the corresponding pair of bottom/top electrodes via two external probes. The probes are in turn connected to a function generator, source-measure unit (SMU) or lock-in amplifier depending on the measurement. AuNPs are deposited by drop-casting a colloidal NP solution, with average NP number density within the junctions controlled by NP solution concentration and deposition time. These parameters are tuned to obtain on average 1−10 NPs within the device area, with AuNPs of diameter 100 nm used for all samples here unless otherwise noted. AuNPs are nominally spherical but always partially faceted, 17 and thus trap a few thousand molecules in the planar junction underneath 18 while maintaining tight SAM packing. SAM formation and uniformity in NPoM geometry has been verified extensively in previous work via large data sets of dark-field and Raman spectroscopy. 16,18 When fabrication is completed, AuNPs within the junctions can be clearly identified with a microscope in dark-field configuration and optically addressed individually for spectroscopy or laser illumination.
To confirm that our device geometry creates working molecular junctions, we verify that device conductance decays exponentially with junction gap size at low DC bias, as expected for junctions in the direct tunneling regime. We systematically fabricate samples using linear alkanedithiol SAMs with chain lengths of 4, 6, 8, and 10 carbon atoms and measure their I−V curves in the ±100 mV range (linear response region) to extract the conductance G = I/V from linear fits to the data (Supporting Information (SI) Figure S1). For each molecule we collect conductance values from ≈100 individual devices into a logarithmic conductance histogram (Figure 1d for 6C chain). In all histograms we observe one prominent conductance peak, sometimes accompanied by other minor peaks or shoulders with sparser density distributions. We fit each set of conductance data with a Gaussian mixture model and assign the weighted mean of the conductance peaks as the characteristic conductance of devices fabricated with that molecule (SI Figure S1). The typical device yield is 40−60% and as high as 85% in some samples (yield measured as percent of junctions per sample that are not shorted and with nonzero conductance). This compares favorably with other small-area ensemble molecular junction geometries, 19 where yields reach 20−70% but without optical access. Device conductance is largely independent of the number of NPs each junction contains, suggesting that only one NP per cross-bar determines the electrical response (SI Figure S2) as verified below.
Plotting the characteristic conductance against the size of the junction gap (Figure 1e), we observe the expected exponential decay typical of direct tunneling. From an exponential fit to the conductance vs gap size d of form e −βd we obtain a decay constant β = 4.9 ± 0.7 nm −1 or β = 0.61 ± 0.09 per CH 2 unit, slightly lower than β = 0.8−1.0 per CH 2 unit often reported in literature. 20 Taking our value of conductance for an 8C chain device and assuming a single molecule conductance 21 of 5nS and that molecules are connected in parallel, from conductance we estimate that on average ∼400 molecules contribute to electrical current in each junction. A 100 nm AuNP with a w = 50 nm facet diameter (as typically observed 17 ) accommodates ∼9000 molecules (with 0.21 nm 2 /molecule 21 ), so only f ∼ 5% of the molecules underneath one NP are effectively contacted and contribute to junction conductance. This is typical of many previous ensemble junctions (e.g., nanopores) since not all molecules are connected at both ends, attributed previously to surface roughness. 21 One advantage of our device geometry is that this number can be tuned by changing the NP diameter D. By fabricating junctions with smaller NPs (D = 60−100 nm) the device conductance can be reduced by almost an order of magnitude (Figure 1f), with a power law dependence extracted, G ∝ D 2.9 . This exactly matches the scaling of the facet area (Figure 1f dashed line; note facet diameter decreases faster than NP diameter, 17 as validated by Raman measurements). Smaller NPs than D < 60 nm become hard to observe optically (scattering strength decreases as D 6 ). Extrapolating G down to NP diameters of 2−3 nm, which are predicted to have contact facet areas comparable to the size of a molecule, we obtain a conductance value ∼1 nS. This is indeed compatible with the conductance G m of a single molecule (although Coulomb blockade effects may occur in that regime). These results therefore demonstrate an effective strategy to tune junction conductance over several orders of magnitude, in static and reproducible devices.
The thin top electrode of our devices allows optical access to molecular junctions underneath each individual AuNP. Tight confinement and large field enhancement, enabled by plasmonic coupling between the AuNP and underlying Au surface, strongly amplify the optical fields under the NP within the molecular junction gap, 16 which is the region responsible for electric conduction. In a classical picture, light in the gap is modeled as an AC field V AC (t) = V opt cos ωt oscillating at optical frequency ω with amplitude V opt . In the presence of an external DC bias V DC and within a simplified quasistatic approximation, where charges can tunnel through the junction fast enough to follow the optical field, the total current can be expanded to second order as 22 ω ω ω ω An optical field in the gap can thus generate a net DC photocurrent, I opt = I″V opt 2 /4, if the junction has a nonzero I″ nonlinearity, even at zero bias where V DC = 0 and therefore I 0 = 0. A nonzero I″ can originate from local nanoscale asymmetries within the junction area, such as roughness features or protrusions. 23,24,22 These asymmetries give rise to potential barriers of slightly different shapes in the two directions, 25 so electron tunneling rates are different for positive and negative voltages resulting in asymmetric I−V response and nonzero I″ (see below). Note that this electrical nonlinearity I″ differs from any optical nonlinearities such as multiphoton absorption (which are not seen here since all effects are linear in laser power).
Individual NPoM junctions are first illuminated with an intensity-modulated 633 nm laser, and the resulting device photocurrent detected with lock-in amplification at the modulation frequency. Devices used for photocurrent measurements are fabricated on glass to prevent spurious signals that can originate from Si substrates. Initially we work at zero DC bias and use either 1,6-hexanedithiol (6Cdt), 1,8-octanedithiol (8Cdt), or 1,10-decanedithiol (10Cdt) as molecules for the SAM. When the focused laser beam is positioned on top of a NPoM junction, photocurrent signals are observed only when the laser is on (Figure 2a), and these rapidly decay as the focal spot is moved away from the NP (Figure 2b). This confirms that the NPoM is the only active region of the device. Even when there are several NPs within the device area and each is individually illuminated under the same conditions, we detect photocurrent from one NP only and find no measurable signal in any other location. This active NP is likely the one whose local interface with the bottom electrode results in the largest conductance and junction asymmetry, therefore dominating the others and determining the overall electrical properties of the device. We assign an upper bound to the photocurrent response time of <10 ms, which is the minimum lock-in time constant below which the signal falls below noise.
The photocurrent under continuous laser illumination remains stable for minute time scales as long as the average optical power does not exceed 0.3 mW. Above 0.3 mW, for many junctions we observe fluctuations, flips in direction, and large photocurrent intensities, typically accompanied by a permanent increase in conductance and I″ or even electrical shorts (arising from light-induced welding). To extract the power dependence of the photocurrent, the responsivity is measured when changing the average laser power while maintaining constant (small) power modulation amplitude. We obtain a constant responsivity for low power, indicating a linear power dependence up to ∼0.3 mW (Figure 2c), which is disrupted at higher power when changes in junction morphology permanently alter conductance and I″.
To confirm that the measured photocurrent signal originates from optical rectification, we verify the correlation between photocurrent and (near) DC electrical nonlinearity I″ for a set of devices. In this case I″ is measured directly by applying a small AC voltage modulation across each device junction and recording the resulting second-harmonic current with a lock-in amplifier. Comparing the conductance and I″ of each device before and after photocurrent measurements allows laser damage to be identified and these data to be discarded. We observe a positive correlation between optical responsivity and I″ ( Figure 2d) over a range of I″ spanning almost 2 orders of magnitude, confirming the origin of the photocurrent from optical rectification set by the junction asymmetry I″. From a linear fit to the data in Figure 2d we obtain V opt 2 = 0.0075 V 2 / mW, which for the average laser power of P = 0.1 mW mostly used in our experiments gives a typical optically induced voltage in the gap of V opt = 28 mV corresponding to an electric field in the gap of ∼6 × 10 7 Vm −1 mW −1/2 , comparable to previous reports for nanoscale plasmonic junctions. 26 The precise plasmonic geometry here provides a strong advantage in allowing a quantitative comparison. Prior theory 16 tion laser spot diameter s = 1.5 μm, plasmonic near-field enhancement F ∼ 200, and gap size d ∼ 1.2 nm. This gives estimated V opt ∼ 49 mV, in good agreement with the experiments. We note that the resulting optically generated fields can easily exceed the breakdown field of molecules. A thermal origin of the observed photocurrent can be ruled out because continued heating from the laser would establish a steady-state temperature gradient across the molecular gap, resulting in a DC thermal current that would be simply proportional to junction conductance, whereas we observe no correlation between photocurrent intensity and device conductance. Additionally, plasmon oscillations excited by the laser would heat up the NP considerably more than the underlying Au electrode, the former being surrounded by a relatively insulating polymer compared to the large bottom electrode that dissipates heat. A thermoelectric current of this origin would therefore always be directed from the NP toward the bottom electrode, whereas we observe photocurrents in both directions with approximately equal occurrence rates. The typical Seebeck coefficient of alkanedithiol junctions S = 5 μV/ K 27 would imply a temperature differential of ΔT = 5600 K from thermoelectric effects via V th = SΔT to match the 28 mV potential from optical rectification. This would rapidly melt the NP, damage the SAM, and disrupt the junction.
We attribute the dispersion of data points in Figure 2d to a mismatch between the location of plasmonic hotspots underneath the NP (Figure 2d inset) and the position of the nanoscale features that generate electrical asymmetry. The main NPoM plasmon-coupled mode is centered at the middle of the NP bottom facet, with a lateral intensity full width at half-maximum (fwhm) of ε Dd/ g ≈ 9 nm (ε g gap refractive index), 16 while higher order modes have intensity maxima and nodes distributed across the facet area, depending on its exact shape (as in inset Figure 2d). The overlap of local geometric asymmetries and regions with highest optical field determine the overall photocurrent, introducing some variability compared to the electrically measured value of I″, which integrates the 5% contacted molecules and asymmetries over the whole junction facet.
To further confirm the plasmonic enhancement of the photocurrent, its wavelength dependence is measured by illuminating a NP junction with a spectrally filtered supercontinuum laser source (spectral bandwidth 20 nm). The photocurrent signal matches well the dark-field scattering spectrum of the same NPoM (Figure 2e), which shows the plasmonic modes of the polymer-coated NPoM system. 16 This allows us to record both the near-field (photocurrent) and farfield (scattering) spectra on the same nanostructure, using plasmon-enhanced optical rectification.
One mechanism to consider is the generation of hot carriers through decay of surface plasmon polaritons. 28−30 In NPoMs, hot carriers would come predominantly from the nanogap and thus the resulting current would have no preferential direction; it would not scale with d 2 I/dV 2 , and would be asymmetric in voltage since electrons or holes are responsible for transport depending on bias polarity and have different hot carrier energy distributions. The observed spectral dependence of photocurrent can thus be assigned to the increased optical field in the junction gap near the plasmonic resonance, rather than to generation of hot carriers.
Photocurrents under nonzero DC bias follow the voltage dependence expected from I″(V) (Figure 2f). I″(V) is measured directly through second harmonic lock-in detection at nonzero bias, and in our devices it is typically linear at very low bias V < 0.05 V while showing a saturation behavior for V ≥ 0.1 V, as previously reported for many types of nanogap junctions. 31 The expected photocurrent of an individual junction at nonzero bias is calculated from I opt (V) = I″(V) V opt 2 /4, where V opt is extracted from the photocurrent and I″ at V = 0. The measured and calculated photocurrents are in good agreement (Figure 2f), confirming the photocurrent originates from optical rectification. Operating the junction at V > 50 mV boosts the responsivity by over an order of magnitude compared to zero bias, so this regime is most promising for detection applications.
In all junctions considered so far, symmetric alkanedithiol molecules are used that bind to Au electrodes on both ends via thiol functional groups. This implies that the molecules themselves do not contribute significantly to junction asymmetry and I″ originates primarily from local roughness features at the upper/lower molecule−electrode interfaces. This is supported by I″ measurements on many 8Cdt junctions (Figure 3a), showing the distribution of asymmetry parameter at zero bias for different junctions. This reveals that for ∼50% of junctions, conduction is favored toward the substrate and ∼50% toward the NP (Figure 3b). These geometric asymmetries do not translate into DC rectification, measured as ρ m = |I(V m )/I(−V m )| at V m = 0.5 V, which is close to 1 for all junctions (Figure 3c). Fully symmetric molecules such as 8Cdt mainly define the gap size and geometry, in a way similar to the role of vacuum or air in previous reports of photocurrent detection in nanoscale metallic gaps. 32−34 When the thiol group on the NP side is replaced by an amine (8-amino-1-octanethiol, 8Cat), the rectification ratio remains close to 1 (Figure 3g), but 72% of junctions now favor conduction toward the bottom electrode (Figure 3f). Even though I″ for 8Cat is on average smaller than for 8Cdt ( Figure  3a,e), once I″ is normalized by junction conductance to obtain the electrical responsivity ER = I″/I′, the contribution to junction responsivity is slightly higher for 8Cat than for 8Cdt (Figure 3d,h). By comparison, using a ferrocene-based alkanethiol (Fc6Ct) gives DC rectification ratios > 1, as expected for this type of molecule 35,36 (Figure 3k), with 84% of junctions favoring conduction toward the bottom electrode ( Figure 3j). Significantly, even though the average I″ is smaller than for 8Cdt and 8Cat (Figure 3i), the average electrical responsivity is now almost an order of magnitude larger (Figure 3l). Maximizing electrical responsivity is important for detection applications because it is proportional to the rectified voltage and quantum efficiency of rectification. 22,37 Photocurrent measurements on Fc6Ct show that indeed the same light-induced tunneling mechanism operates (SI Figure  S3). In the simplest model ER = I″/I′ ≃ 2V m −1 (ρ m − 1)/(ρ m + 1), thus linking the rectification ratio with the plasmonenhanced optical rectification, although this typically overestimates the measured values (SI Section SA). Experimentally we find ER = 0.1−1 fluctuates between different junctions. The optical responsivity is then given by I opt (V)/P = I″(V) × 2(dF) 2 /(cε 0 πs 2 ) = ER(V) 2I′(Fd/s) 2 /(πcε 0 ). Since the field enhancement in the NPoM geometry is known and can be approximated 11 as F 2 ∼ 40n g D 2 /d 2 (for gap refractive index of n g ) while I′ ≃ G m f(w/a) 2 [1 + VG′/G] where a ∼ 1 nm is the effective separation of molecules, we can thus estimate the magnitude and scaling of the optical responsivity as Using the parameters noted above, we thus predict I opt (V)/P ∼ 20 nA/mW, which is indeed comparable to what is measured (Figure 2f). This gives insight into how to optimize the photocurrent response, which requires high molecular conductance and large electrical responsivity, and as noted above, this is best at higher voltage. However, when a DC voltage is applied, a background tunneling current is produced which has to be distinguished from the photocurrent and creates additional shot detector noise, limiting the optimal bias that can be applied. It is thus crucial to provide the largest ER(V) at the smallest voltage.
Ferrocene-based molecules are widely known to be better rectifying because their ferrocene component, whose HOMO level is just below the Au Fermi energy, introduces an intermediate energy level that allows transport by sequential hopping-tunneling in one direction while only direct tunneling is possible with opposite bias. 35 Detailed studies of the transport mechanism in these molecules have shown that large rectification ratios are generally associated with tight SAM packing, optimal van der Waals molecule−electrode coupling, and minimal presence of defects on electrodes. 36 If sequential tunneling was responsible for increased transport asymmetry in our junctions, we would therefore anticipate devices with larger rectification ratio to also show larger responsivity. However, we observe no correlation of rectification ratio with optical responsivity or I″, suggesting that the increased junction asymmetry is not caused by a change in the type of molecular transport through the junction.
Two effects are likely responsible. Sequential tunneling would be slower than direct tunneling, and thus unable to follow the field at optical frequencies. In addition, the 0.1 V modulation amplitude used to directly measure I″ is insufficient to reach the molecular HOMO and trigger sequential tunneling. To explain the symmetry breaking from Fc6Ct, we instead propose that van der Waals interactions of the top NP electrode with Fc6Ct, which is more physisorbed than the chemisorbed S−Au at the bottom electrode, aid restructuring of the molecule−electrode interface upon NP deposition and favor formation of protrusions in the top electrode (Figure 4a). The amine group on 8Cat, which also binds less strongly than thiols, has a similar effect but with reduced magnitude given the linear chain that maintains SAM packing, supporting previous reports on electrical properties of molecular junctions with asymmetric functional groups. 38 By contrast 8Cdt does not introduce asymmetries in interface restructuring and thus shows the smallest responsivity and little directionality (Figure 4b). We note the asymmetry of 8Cat and Fc6Ct and of their molecule−electrode interfaces might also create an asymmetric tunneling barrier leading to asymmetries in tunneling currents. Additional experiments on a wider range of molecules are needed to develop this model, but it suggests the optimization of optical tunneling photocurrents has to focus on maximizing molecule−electrode binding geometry asymmetries, besides enhancing DC rectification ratios.
The suggested origin of the tunneling asymmetry comes from the non-uniform potential shape and electron image charge as it tunnels away from a perturbation with radius of curvature r (Figure 4c). 25,37 As it emerges, the image charge attractive potential is amplified by the curvature, reducing the barrier for tunneling from this side, but has little effect when at the end of the tunneling trajectory from the other planar electrode.
Metal−insulator−metal (MIM) tunneling junctions have been utilized as photodetectors, 31,39−41 because their "rectenna" response is independent of wavelength, as seen from eq 2, which can now be used to estimate the efficiency of such sub-wavelength-scale tunneling detectors. Typical zero-bias optical responsivities of 20 nA/mW here are lower than those of nanoscale Schottky junctions 42 that can reach 10 6 nA/mW. However, the latter typically operate only in the UV and have microsecond to millisecond response times, whereas the speed of MIM junction detectors is fundamentally limited only by electron tunneling times (typically femtoseconds) and the rectenna RC time constant. 43 Our geometry provides a way to simply reduce the time constant by reducing junction capacitance and resistance with smaller NPs and more conductive molecules that also boost junction asymmetry. A simple estimate using 20 nm facet width, 1 nm junction gap, and 1 kΩ resistance gives RC ∼ 2 fs (using the gap capacitance), 16 promising for ultrahigh-speed detection in the visible. Conjugated oligophenylene molecules with asymmetric functional groups are likely good candidates to maximize optical responsivity while keeping device resistance low.
The optical responsivity of our alkanedithiol devices at 0.1 V is typically in the 100−200 nA/mW range, similar to alternative plasmonic detectors based on hot electron generation demonstrated for the visible and IR. 44 Electrical responsivity in junctions with Fc6Ct is 0.1−1 V −1 at zero bias and 1−5 V −1 at 0.1 V, which is already in line with terahertz responsivities of MIM junctions rationally designed to optimize detection in the terahertz regime. 45 Our device geometry could provide an alternative to semiconductor photodetectors for high-speed data transfer in integrated photonic circuits. 46 In terms of energy harvesting applications, the typical electrical power generated by one junction for 0.1 mW optical power is around 10 −5 mW (P = V opt 2 G, with V opt = 28 mV and G = 20 μS), corresponding to a power conversion efficiency of 10 −4 . Considering however that the NP diameter is 100 nm, illuminated by a 1.5 μm diameter laser spot, the efficiency for the active device area becomes η = 1%. This does not compete with photovoltaic generators but could find applications where device miniaturization and ease of fabrication are critical.

CONCLUSIONS
In conclusion we show that combining reliable plasmonic constructs with self-assembly of molecules provides an effective strategy to create nanometer-sized gaps for sensing and energy harvesting from light. Molecules play a crucial role in defining the transport characteristics of these nanodevices and their resultant optoelectronic properties. Selection of optimal molecules is essential, but those that show large DC rectification ratios are not necessarily the best candidates for optical-frequency rectification at zero bias, due to the molecular transport mechanisms involved. In our devices, an incident optical power of 100 μW can easily generate an AC voltage of tens of millivolts in the plasmonic cavity, while much higher optical fields disrupt the molecular junction. We derive an estimate for the optical responsivity which matches experiments well. To exploit molecules for optical rectification, they should be designed to display large rectification ratios (at zero or nonzero DC bias) within the safe potential window. Microscopy, Dark-Field Spectroscopy, and Probe Setup. Imaging and dark-field spectroscopy are achieved using a modified BX51 microscope with confocal fiber-coupled spectrometer (QE65000, Ocean Optics) with 1.5 μm acquisition spot diameter. Electrodes on the sample are contacted with tungsten probes (American Probe & Technologies) mounted in a custom probe station integrated with the microscope. All measurements are conducted at room temperature and ambient conditions. Electrical Measurements. I−V curves are acquired with a source-measure unit (2635A, Keithley). I″ is measured through a function generator (33220A, Agilent) and lock-in amplifier (SR810, Stanford Research Systems) in series with the junction being measured, with lock-in set to second harmonic current detection mode using 500 Hz sine wave modulation with typical amplitude 0.1 V peak-to-peak and zero or nonzero DC offset depending on the measurement.
Photocurrent Measurements. A red laser (633 nm MatchBox series, Integrated Optics) is coupled into the microscope from free space and focused to a 1.5 μm diameter spot on the sample. Laser power is controlled with an acousto-optic modulator and calibrated with a power meter (S120C, Thorlabs) at the sample location. The laser is sine wave modulated at 1 kHz with average optical power of 0.1 mW and peak-to-peak modulation amplitude of 0.1 mW. The modulation is used as reference for the lock-in amplifier in first harmonic current detection mode to measure the photocurrent. For photocurrent spectra the red laser is replaced by a filtered supercontinuum source (Fianium fiber laser, SuperChrome filter unit) with 20 nm fwhm bandwidth calibrated to apply the same optical power on the sample at all wavelengths, modulated by a chopper (MC2000B, Thorlabs) at 788 Hz and with typical average optical power of 0.025−0.05 mW. "Asymmetry direction" in Figure  3b,f,j is obtained from the phase of the signal measured by the lock-in amplifier.
Additional conductance data for alkanedithiol and ferrocene molecules, AFM data for surface roughness, equation derivations, and simulations of thermal expansion in NPoM junctions (PDF)