Atomic-Scale Structural Fluctuations of a Plasmonic Cavity

Optical spectromicroscopies, which can reach atomic resolution due to plasmonic enhancement, are perturbed by spontaneous intensity modifications. Here, we study such fluctuations in plasmonic electroluminescence at the single-atom limit profiting from the precision of a low-temperature scanning tunneling microscope. First, we investigate the influence of a controlled single-atom transfer from the tip to the sample on the plasmonic properties of the junction. Next, we form a well-defined atomic contact of several quanta of conductance. In contact, we observe changes of the electroluminescence intensity that can be assigned to spontaneous modifications of electronic conductance, plasmonic excitation, and optical antenna properties all originating from minute atomic rearrangements at or near the contact. Our observations are relevant for the understanding of processes leading to spontaneous intensity variations in plasmon-enhanced atomic-scale spectroscopies such as intensity blinking in picocavities.


Single-atom deposition
We prepare atomic-scale contacts used in our study by depositing individual atoms on the Au (111) surface. It can be controllably achieved by approaching the surface with the tip until contact 1 (Fig. S1).
First, the surface is imaged to ensure it is clean, as presented in Fig. S1(a). Next, at a set-point of U = 1 V, I = 100 pA, the tip is moved towards the surface by 650 pm which results in a jump-to-contact event leading to a conductance close to 1 G0 ( Fig. S1(b)); the tip is subsequently retracted. Afterwards, the surface is imaged again to confirm successful atomic deposition (Fig. S1(c)). We find this procedure to be highly reproducible, which on occasion may also deposit a small cluster of a few atoms, which can be readily identified by a higher maximum conductance during the approach-retract curve and a higher topographic appearance than the one presented in Fig. S1(d). Since both the tip and the sample are made of Au, also the deposited structures consist of Au atoms.   S2 presents optical spectra recorded, before, during and after measurements presented in Fig. 2 of the main manuscript. We find that the shape of the spectrum did not change significantly after the measurement. The total integrated intensity, however, increased by 14 % (Fig. S2(a)). In Fig. S2(b) we compare these reference spectra in tunnel conditions (100 pA) with the spectrum recorded in contact (77 A) and overbias emission condition. The main mode is slightly red-shifted with respect to the spectra recorded in tunneling. The overall higher intensity in the low energy regime (<1.8 eV) is a result of the overbias emission, which is less efficient at higher energies.   Fig. S3(a). To evaluate the permanent modifications of the junction after the experiment, the surface was rescanned ( Fig. S3(b)). We find that the investigated structure moved by 1 nm and likely an atom has been transferred to the surface, as indicated by the increased apparent height shown in Fig.   S3(c). In this particular experiment, we contacted a larger adsorbed structure consisting of more than one atom because such junctions are more prone to spontaneous changes and thus more illustrative for our study. For the sake of completeness, we performed measurements in which we contacted a single atom deposited on the surface and observed both fluctuations when in contact and the permanent modification of electroluminescence as compared before and after the measurement. This experiment is described in the next section ( Fig. S4 and Fig. S5). Overbias emission condition U = 1 V.
In the experiments displayed in Fig. S4 we deposited a single Au atom (see Fig. S5), established a contact of 1 G0, and then retracted the tip until loss of contact (rupture). Simultaneously, we monitored the electroluminescence integrated intensity (middle column in Fig. S4) and the overbias emission spectrum (right column in Fig. S4). No current feedback was used throughout the experiment. The measurements were performed in an extremely gentle fashion, each time the tip was continuously retracted (in steps of 3.5 pm) by 350 pm in total. If the contact did not break, the tip was retracted by another 350 pm. We found that typically the contact broke after a total retraction of approx. 450 pm. After each experiment, we rescanned the surface and repeated the experiment on the same atom following the same protocol.
This procedure was reproducible and the approach-retract routine could be repeated on the same atom multiple times, as shown in Fig. S4. Similar to the measurements presented in Fig. 2 and Fig. 3 of the main text, the luminescence can evolve in both gradual and step-wise manner which can be correlated with the changes in the conductance (left column in Fig. S4). Again, the spectral shape remains unchanged as displayed in the right column of Fig. S4. The tip was stabilized for 15 h before the series of experiments was started. Fig. S5 shows STM images recorded before ( Fig. S5(a)) and after ( Fig. S5(b)) accomplishing all experiments presented in Fig. S4, the investigated atom (deposited from the tip as described above) is marked by an arrow. We find that at the end of the experiments the deposited adatom had moved by 1 nm in total but no additional atom had been transferred, which demonstrates the reproducible experimental conditions.
In contrast, the apex of the tip has evidently changed: First, the deposited atom and the defects on the surface are imaged as features with smaller extension in Fig. S5(b) than in Fig. S5(a) indicating that the tip apex has become sharper. Second, there is a decrease in the light intensity by 8 % percent (Fig. S5(c)).

7
These observations further confirm that a minute change at the tip structure, even without transferring an atom to the surface, may induce a modification in the plasmonic properties of the junction.  In Fig. 2e of the main text, we observe thermal tip elongation due to the power dissipated in the junction.

Local heat dissipation in the junction
Since the main part of the applied voltage drops across the tip-substrate contact region, we can estimate the dissipated power at the junction to be close to U 2 G0  77 W. About half of the power will be 9 dissipated on the tip and substrate each. However, since heat transport in the substrate occurs over a hemispherical (2) region in the macroscopic Au crystal while the tip is sharp and conical, the strongest contribution to thermal expansion will be due to heat dissipation in the tip. Coarse modeling of heat flow at a conical gold tip suggests a temperature of the order of 30 K to 100 K (see Fig. S6) which may, however, be surpassed in close vicinity to the atomic junction. One can also define the upper temperature bound to be well below a significant fraction of the melting temperature of Au. If such high temperatures were to arise, we would observe a permanent instability of the tunnel junction due to rapid atomic diffusion on much faster time scales than actually observed ( Fig. S3 and S5). The simulation demonstrates that the conical tip exhibits much higher temperatures due to the reduced effective dimensionality in comparison to the half-space available for heat transport in the substrate even when equal shares of the total dissipated electric power are fed to tip and sample.
In laser-based plasmon-enhanced spectroscopies (as e.g., TERS or TEPL) the thermal input by the illumination can become comparable to the power dissipation by the electric current and higher ambient temperatures will play an even greater role. While our study tries to limit heat input by using low voltages and cryogenic temperature, the thermal reorganization will be most relevant at higher conductance (4-5 G0) since here the heat dissipation at the junction is increased by the respective prefactor.