Molecular and atomic manipulation mediated by electronic excitation of the underlying Si(111)-7x7 surface

We report the local atomic manipulation properties of chemisorbed toluene molecules on the Si(111)-7x7 surface and of the silicon adatoms of the surface. Charge injected directly into the molecule, or into its underlying bonding silicon adatom, can induce the molecule to change bonding site. The voltage dependence of the rates of these processes match closely with scanning tunnelling spectroscopy of the toluene and adatom species. The branching ratio between toluene molecules which are moved to a neighbouring site, or those that travel further is invariant to voltage, suggesting a common final manipulation step for both injection into the molecule and into the bonding adatom site. At low temperatures the rate of silicon adatom manipulation matches that of toluene manipulation, further suggesting that all these manipulation processes are driven by electronic excitation of the underlying silicon surface. Our results therefore suggest that a common non-adiabatic process mediates atomic and molecular manipulation induced by the STM on the Si(111)-7x7 surface and may also mediate similar manipulation induced by the laser irradiation of the Si(111)-7x7 surface.

(Some figures may appear in colour only in the online journal) Atomic manipulation using the tip of a scanning tunnelling microscope (STM) to control individual atoms and molecules is the extreme limit of nanotechnology. It is now nearly routine to use the STM tip as a local tool (within the tunnel junction) to artificially assemble structures out of individual atoms on metals like the quantum stadium [1] and the quantum mirage [2]. It can be used to dissociate [3,4] and rotate [5] individual molecules or to induce the controllable switching of macromolecules [6][7][8]. Key to all these reports is the identity and properties of the target atoms and molecules.
Recently new ideas have emerged that have exposed the contribution of the underlying surface to the manipulation process, most strikingly in nonlocal manipulation where targets some distance remote of the STM tip are manipulated [9][10][11][12][13][14][15][16][17][18][19]. In these experiments the charge is transported across the surface and so the surface properties are critical to the molecular manipulation. We have shown, for the Si(111)-7x7 surface, the connection between the electron injection site and the measured spatial distribution of nonlocal manipulation [15]. We reported on the ultrafast coherent [18] and incoherent [17] electron (or hole) dynamics that transport the injected charge from the injection site to the distant molecules. Here we address the final manipulation step, the connection between the hot-electron that arrives at the molecular site and the actual molecular manipulation process itself.
The Si(111)-7x7 surface has been extensively used in studies of molecular adsorption, desorption and more particularly STM induced manipulation both of atomic and molecular adsorbates and the silicon surface itself [20][21][22][23][24][25]. It has also demonstrated photo-induced silicon atom desorption through laser irradiation [26][27][28]. Here by precise and extensive single-molecule manipulation experiments, both at room temperature and low temperature, we show that our STM induced molecular manipulation is driven by the same process that drives the manipulation of individual silicon adatoms of the Si(111)-7x7 surface. Furthermore, we find parallels and connections to the laser-induced silicon desorption work and suggest that a common surface electronic excitation drives all these dynamical processes on the Si(111)-7x7 surface. Figures 1(a) and (b) present large-scale STM images of the Si(111)-7x7 surface taken before and after a series of single molecule manipulation events as marked by boxes. At our imaging sample bias of +1 V Si adatoms image as 'bright' spots, whereas chemisorbed toluene molecules appear as 'dark' spots [4,17,18,29,30]. Figures 1(c) and (d) present high resolution images taken before and after charge injection (+1.9 V, 750 pA) into a single target molecule at the precise location marked with the× symbol. It is evident that the toluene (dark spot) has been removed by the charge injection since the underlying bright silicon adatom is recovered.
The Si(111)-7x7 reconstruction is well described by the dimer-adatom stacking fault model [31], whereby the unit cell comprises 12 adatoms with 4 distinct types depending on their position within the unit cell: faulted middle (FM), unfaulted muiddle (UM), faulted corner (FC) and unfaulted corner (UC). Figures 1(e) and (f) show schematically one half of the Si(111)-7x7 unit cell. Adatoms are represented by circles and restatoms by crosses. Toluene molecules image and exhibit near identical thermally induced diffusion and desorption properties as for benzene and chlorobenzene on the Si(111)-7x7 surface [32][33][34]. It follows that toluene binds to the surface in an identical di-σ bonding fashion as those molecules [35] forming one bond to an adatom site and a second bond to a neighbouring restatom site. It has been previously shown that molecules attached to FM adatoms react most readily [32]. Therefore, all local experiments discussed in this paper were performed by injecting into molecules adsorbed to FM sites. Due to the di-σ bonding, a molecule bonded to an FM site as in figure 1(c) has two possible and crystallographically equivalent neighbouring restatoms to form the second molecule-surface bond as shown schematically in figures 1(e) and (f). Figure 2(a) shows an example of the tip-height trace and the tunnelling current recorded during a charge injection (+1.9 V, 750 pA) with a desorption event at 0.58 s after the start of the injection. All experiments were performed with the feedback loop engaged, thereby keeping the STM tip at the desired manipulation parameters for a set amount of time. The instant at which the manipulation event (desorption of the toluene molecule) occurs is recorded both as a spike in the tunnelling current in figure 2(a) and as a corresponding steplike change in tip height as the (dark) molecule leaves the surface and reveals the underlying (bright) Si adatom. Our automated experiments perform, on average, ∼100 individual manipulation experiments for each set of injection parameters. To analyse the results, instead of the usual procedure of statistically binning the survival times into time brackets [6,36], we aggregate the raw survival times to produce a time-dependent population of molecules that still retain their original positions. Figure 2(b) presents the number of molecules that retain their initial position N(t) after a time t out of an initial population (number of experiments) N 0 with injection parameters of +1.9 V and 750 pA. The measured time-dependent outcome is not, as is usually the case, related to a single exponential decay. Instead we find two time-scales, one for a fast process and one for a slower process. For identical molecules, if there are two competing routes to manipulation, then the rate dependence for that process would still conform to a single exponential decay (with a decay constant which is the sum of the two routes). This clearly does not fit our results. Instead we propose we have two distinct molecular populations, A and B, that image in an identical fashion in our passive scans, but react differently to the injected charges. Mathematically therefore we can define the overall population time-dependence as where N(t) is the total number of molecules that remain in their original position after an injection time t at a current I. N 0 is the total number of manipulated molecules from an initial population of N 0 A and N 0 ) with a probability of manipulation per electron given by a A and a B respectively and e is the charge of an electron.
The population (pre-exponential factors N 0 A and N 0 B of equation (1)) for the fast and the slow events, averaged over all experimental data, are equal within error. The binding energies of the two populations must therefore also be similarly equal. The schematic diagrams of figures 1(e) and (f) show that for a single FM adatom there are two possible and crystallographically equivalent binding sites of the attached molecule since there are two crystallographically equivalent silicon restatoms for the molecule to bond to in its di-σ bonding configuration. We find identical double-exponential decay behaviour for single molecule manipulation of benzene on the Si(111)-7x7 surface. Therefore we can rule out any extra degree of freedom in the binding geometry due to the CH 3 unit of the toluene molecule. Moreover, thermal desorption spectroscopy and high resolution electron energy loss spectroscopy experiments have shown that both toluene [37,38] and benzene [39,40] adsorb and desorb from the Si(111)-7x7 surface molecularly. Thus, we firmly expect each of the black spots in our STM images to correspond to a single chemisorbed, and intact, toluene molecule. Instead, we propose that the origin of the two 'populations' is due to our injections being precise, that is we inject at the same position relative to the bonding adatom, but not accurate in that we do not inject exactly atop the darkened silicon adatom site. As can be seen in figure 1(d) and later in figure 4(a), the injection location, given by the× in both cases is indeed slightly offcentre of the targeted adatom site. It is this off-centre injection, relative to the adatom, that differentiates between the crystallographically equivalent binding sites, resulting in the measured two-population manipulation dynamics.

Figures 1(e) and (f)
show schematically an example of an offcentre injection as depicted by the red star. In one bonding configuration the injection will be directly atop the molecule, whereas in the other configuration it will 'miss' the molecule and inject primarily into the location of the bonding adatom. Figure 3(a) presents the voltage dependence, between +1.2 and +2.1 V, for manipulation of toluene molecules. Below +1.2 V we find, within the 8 s time-fame of our injection experiments, no evidence of STM induced manipulation. At the two lowest voltages used, +1.2 and +1.4 V, the decay curves exhibit a single exponential decay. At all higher voltages the double-exponential dependence is evident. Figure 3(b) shows the probabilities per injected electron of inducing a manipulation event determined from these curves. The single exponential data and the 'slow' process results have been scaled by a factor of 5 to aid clarity. Above +1.5 V the voltage dependences of the fast and slow processes follow each other. At +2.1 V the nonlocal channel [15,17,18] for manipulation opens with an associated dramatic increase in the probability of manipulation per injected electron. At all the voltages probed we find that the manipulation process is a single-electron process and for the parameters probed here is invariant to the electric field between tip and molecule [41].
We also measure the manipulation outcome itself defining two categories, one which we label as 'diffusion' where the molecule breaks the adatom bond, but retains its bond to the restatom, and migrates to a neighbouring middle or corner adatom site. The other which we label as 'desorption' where the molecule is found further away or completely desorbed and hence both the bond to the adatom and the restatom must have been broken. The inset to figure 3(b) shows the ratio of molecules that diffuse to those that desorb. We see that the desorption outcome is the most likely and that this ratio, i.e., the branching ratio, is relatively invariant to the injection bias voltage. The uniformity of the outcome across all voltages probed suggests a common manipulation mechanism, that is, the fast and the slow channels appear to have the same final manipulation step. Furthermore the lack of voltage dependence of the outcome branching ratio implies that the precise identity of the excited state that triggers manipulation is the same. We speculate that, as for nonlocal manipulation, the injected charge undergoes ultrafast energy relaxation within an electronic state before inducing manipulation through a non-adiabatic dynamics induced by electronic transition process [42][43][44].
The silicon adatoms of the Si(111)-7x7 can themselves be manipulated by the STM tunnel current [9,45]. They can hop to a position atop a neighbouring silicon adatom leaving behind an atomic vacancy. Figures 4(a) and (b) show high resolution STM images of the clean Si(111)-7x7 surface taken at 77 K. Charge was injected at location× inducing the FM adatom to hop on top of the adjacent adatom site, thereby creating a bright feature. At room temperature this state is metastable and within » 600 ps (taking a 0.18 eV activation energy [45]) the adatom returns to its original site [9]. This annealing effect at room temperature precludes any direct measurement of such adatom manipulation due to the limited time-resolution of the STM [46]. At low temperature the metastable, i.e., hopped, state is long lived and can therefore be imaged, allowing the measurement of local manipulation of individual FM silicon adatoms. Figure 4(c) shows for two voltages, +2.5 and +2.8 V, the manipulation decay curves of adatoms and toluene molecules at 77 K. (See [47] for a video of time-lapse-STM images taken at 77 K and +0.5 V (passive voltage) of the Si(111)-7x7. Between each image an 'active' image was recorded at +2.3 V inducing a degree of adatom hopping.) We observe three possible outcomes for the adatommolecule assembly after injection at 77 K: the molecule has desorbed revealing a clean adatom underneath; the molecule has desorbed and the adatom that it was attached to has hopped on top of a neighbouring adatom; and the molecule undergoes incomplete desorption becoming physisorbed atop its original chemisorption binding site. All three possible outcomes have the same initial step, namely the breaking of the molecule-surface covalent bond. There is some evidence that these outcomes are manifested in steps of different heights that are recorded in our tip height traces [39]. However, we presently have insufficient statistical data to allow us to clearly differentiate between these outcomes. At room temperature we cannot differentiate between these outcomes at all due to the short lived lifetimes of the hopped adatom state and of the molecular physisorbed state. Hence to be consistent, for the low temperature measurement we take the time to manipulation to be at the moment of the first step in each injection trace and amalgamate all the data for a given set of injection parameters. What is apparent from figure 4 is the near identical behaviour of adatoms and toluene molecules for +2.5 V injections. The decay curves for both species have single exponential decay character and, as shown in figure 4(d), have within error the same decay constant, i.e., probability per electron of manipulation. Therefore we can associate the slower process found for molecular manipulation not with excitation of the molecule, but of the underlying silicon adatom. At low temperature we can observe the adatom manipulation itself. At room temperature it is the signature of molecular manipulation that reveals the excitation of the underlying silicon surface. At the higher voltage of +2.8 V the toluene manipulation exhibits the two-exponential decay behaviour, with its slower decay constant matching that of the adatom manipulation at +2.8 V. This single-exponential to double-exponential change for the molecule mimics the results at room temperature, but shifted by ∼1 V. This shift could be due to a combination of unpinning of the Fermi level due to the highly doped sample used at low temperatures, non-equilibrium tunnelling into the Si(111)-7x7 surface and band-bending [48]. The probabilities per electron closely match those previously reported for adatom manipulation at 3.0 V and 79 K [45].
To identify the origin of the room temperature manipulation voltage thresholds in figure 3 we compare the voltage dependence of the fast and the slow processes with the density of states, as measured by STS, of a toluene molecule bonded to an FM adatom ( figure 3(c)) and of a clean FM adatom itself ( figure 3(d)). We find for each two peaks in the range probed, one at +1.2 V and a higher lying peak at +2 V. We identify, as before [17,18], the state at +2 V with the back-bond state of the adatom [49]. For the slow process (injection into the bonding adatom site) the voltage dependence follows that of the adatom back-bond state peaked at +2 V. Note also that the large energy width of the state matches the large energy range of the slow process.
The voltage dependence of the fast process (direct injection into the molecule) also appears to follow the voltage dependence of the +2 V state but with one difference. There is a much more prominent feature at +1.2 V for the molecular STS measurements. For chlorobenzene the lowest unoccupied molecular orbital (LUMO) was found at +0.9 V, therefore it is possible that this large peak at +1.2 V is the LUMO of our toluene molecules. (In STS below +1 V we find only the remnant of the low lying dangling bond peak at +0.5 V.) The lack of the fast process for direct injection into the molecule below +1.4 V suggests that the molecularly derived +1.2 V state is manipulation inactive. That is, tunnelling into the molecule dependent +1.2 V state does not lead to excitation of the underlying silicon adatom.
We conclude that for charge injection directly atop a toluene molecule, but with bias voltage below +1.4 V, no manipulation takes place as the tunnel current does not populate the silicon back-bond +2 V state. At higher voltages the tunnel current can efficiently populate the back-bond state leading to a large probability of manipulation. Whereas injecting slightly off the molecule and more directly into the bonding adatom leads to population of the +2 V state at all (c) Time-dependent decay of initial population of adatoms (stars) and toluene molecules (circles) for both +2.5 and +2.8 V injections (750 pA). The adatom data is fitted with a single exponential decay. The molecular decay is fitted with a single decay at +2.5 and a double decay at +2.8 V. (d) Probability per electron of manipulation for molecules and adatoms at 77 K. Note the adatoms data has been slightly offset (+0.02 V) in voltage to aid clarity. voltages probed, but with overall a lower probability than the direct molecular injection. The molecule appears to simply act as a conduit to enhance the fraction of the tunnel current that can lead to manipulation, where that manipulation step is the same step no matter the voltage, or the position of the injected charge. Of course, given the small probability of manipulation most electron injection events do not lead to manipulation, but may well excite some level of vibrational excitation of the molecule (presumably still through the +2 V adatom excitation process). If a second electron arrives before the system has fully relaxed, then that second electron may well induce further manipulation, as observed for intra-molecular bond breaking for chlorobenzene on Si(111)-7x7 [4].
The Si(111)-7x7 surface has also been the subject of several laser induced manipulation experiments where STM imaging is used as a tool to characterise the extent and probability of laser induced silicon adatom desorption. In short, a 2 eV photon energy threshold was found and ascribed to a possible electronic transition between surface electronic states, notably from the filled dangling bond state (at −0.2 eV) to the unfilled back-bond state (as we identify with thẽ +2 V state in our STS measurements) [26]. The rate of adatom desorption was super-linear with two photons required per adatom desorption event. The kinetic energy of the ejected adatoms was invariant to the energy of the incident photons (2.3 to 4.6 eV) indicating that the final step of the 2-photon process was common across all probed photon energies [27]. (Just as we also surmise from our local experiments, that the injected charge relaxes back to the bottom of the back-bond state sate hence the invariance of the manipulation outcome on electron energy.) A phonon-kick mechanism following two-hole localization on the adatom site was proposed. From the time-dependent experimental data a lifetime of 170 ps for the intermediary state was proposed.
Here we offer an alternative possibility, that the common intermediate state is not a positively charged adatom, but is a manipulated adatom that has hopped onto a neighbouring adatom site. Ultra-fast pump-probe experiments have measured electron lifetimes of ∼100 fs in the Si(111)-7x7 surface at the relevant energies [50] and so it seems unlikely that a charge remains at the surface site for the required 170 ps. The estimated intermediate 'two-hole' lifetime appears to match more closely with the room temperature lifetime ∼600 ps of the metastable manipulated silicon adatom. Therefore we rationalise the two-photon process as the first photon creating an electron-hole pair which presumably explore a large region of the surface just as in our nonlocal STM experiments. It can induce an adatom to hop onto an adjacent site (as we observe in the low temperature STM injections). Here we conclude that a single photon can only induce this hop, not complete adatom-desorption. It seems likely therefore that the measured lifetime of 170 ps for the two-photon desorption experiment reflected the lifetime of the, presumably neutral, hopped state. Such a two-photon process involving an intermediate state, the 'hopped' adatom, would explain the invariance of the kinetic energy of the photo-ejected adatoms as they would originate from the same meta-stable 'hopped' adatom site. During our STM injections at 750 pA the time between tunnelling electrons is typically 210 ps. Therefore one may ask why we do not observe complete desorption of an adatom due to a two-electron process? After the first manipulation event the adatom has moved out of the tunnel junction and so cannot undergo further local manipulation. The atomic movement of the initial manipulation step, simply precludes any second manipulation event. Once past the nonlocal voltage threshold it may be possible to induce a two-electron process, but the time between electron scattering events at the 'hopped' sites will be much reduced from the time between tunnel electrons in the STM junction. Perhaps such a nonlocal method could be a way of indirectly accessing the time between impinging electrons through the nonlocal channel.
In conclusion, we propose that STM molecular manipulation experiments (ours and others'), STM silicon adatom manipulation experiments, and the photo-induced manipulation experiments all have a common origin in their initial manipulation process, namely the electronic excitation at the back-bond site of the adatoms of the Si(111)-7x7 surface. It also seems likely that injections over the adatom manipulation threshold induce adatom hopping, but at room temperature the restricted time resolution of the STM precludes its observation. The Si(111)-7x7 surface therefore cannot simply be thought of as a passive substrate even though that is how it appears in STM images. Instead for any bias voltage above +2 V it should instead be thought of as a seething surface of adatom motion.

STM and sample preparation
Room temperature experiments were performed with an Omicron STM1 in an ultrahigh vacuum (UHV) chamber with a base pressure of less than 1×10 −10 mbar. The experiments at 77 K were conducted on an Omicron UHV qPlus AFM/ STM system at a base pressure of 1×10 −11 mbar. Both machines were operated with a Nanonis control module. Precut Si(111) samples (n-type, phosphorus doped, 0.001-0.002 Ω cm) were prepared in the same way as in reference [51]. Tungsten tips were etched in a 2M NaOH solution and outgased in vacuum before being mounted into the STM head. Toluene was purified by the freeze-pump-thaw technique with liquid nitrogen and checked for purity with a quadrupole mass spectrometer. The clean Si(111)-7x7 surface was covered with small amounts of toluene (∼2 L) dosed through a leak valve at base pressure 5×10 −9 mbar. This corresponds to a typical coverage of ∼3 molecules per unit cell. All images were obtained with purely passive scanning conditions (+1 V, 100 pA at room temperature and +1.5 V, 100 pA at 77 K) [51]. Stability during the injection was ensured by a drift-compensation software, enhanced with a feature locking routine prior to injection. Molecule displacement was confirmed by comparison of the passive scans taken before and after an injection. Typically, the drift compensation values during each injection experiment would range from 100 fm s -1 up to 2 pm s -1 in all three directions (x, y and z).

Automated experiments
Each data point associated with the probability of desorption is obtained from the manipulation of 150-200 individual toluene molecules. Since the mean and the standard deviation of an exponential distribution t -t exp( ) can be approximated to the decay constant τ, a fit to 200 desortpion events would have a standard error of~7 %. The acquisition of such a large data set for each set of manipulation parameters (I and V ) was only possible through a combination of LabVIEW automation and an in-house Matlab computer analysis suite [51]. Each experiment that manipulated 10 individual molecules took 10 to 15 min with the following sequence: (1) acquisition of a large 'overview' image (25 nm´25 nm); (2) automated computer identification of precise location of all adatom and molecule positions within the overview image; (3) random selection of 10-15 molecules located on FM adatom sites and not within the same unit cell as each other.
(4) Coordinates sent to the STM; (5) STM positioned at the co-ordinates of the first molecular site, a small 3 nm´3 nm image recorded and cross-correlated with the overview image to determine the absolute present position of the tip. Correction to the position made to place the tip atop the target site; (6) three high resolution small images (3 nm´3 nm) taken sequentially centred on the injection site, during the second image the tip was halted as close to the centre of the image, the desired atomic position, as possible and, if required, a small adjustment made to ensure the tip was at the absolute centre of the image and then the current injection carried out (see below for details). The tip was then returned to the halted scan position and the scan resumed. Although this scan was halted for typically 8 s to perform the injection, we found as shown in figure 5 little or no spatial offset between the two halves of the image, indicating that there is little or no thermal drift during the injection. (7) Steps 4 and 5 repeated for all identified molecular sites; (8) acquisition of a large 'overview' image (25 nm´25 nm) for comparison with the initial image and for identifying the outcome of the injection.
A time lapse video of a complete experiment can be found online, see [52].

Local injections
After the tip was positioned over the target site (step 6 above) the following procedure was performed: (1) the feedback loop was disabled; (2) the tip was retracted by 1 nm from the surface to ensure no (or minimal) interaction with the target while the injection parameters are set; (3) the bias voltage was ramped to the required value for the injection experiment; (4) for the charge injection experiments, the feedback loop was engaged to attain the injection tunnelling current value; (5) at the end of the injection time (typically 8 s) the feedback was disengaged and the tip was retracted by 1 nm; (6) the tunnelling parameters were adjusted back to the passive imaging conditions, the feedback loop was re-engaged and imaging recommenced.
This procedure allows accurate control and measurement of all the tunnelling parameters, height, voltage and current. It prevents any high current transients due to the rapid ramping of the voltage and can easily be adapted to perform other local experiments, for example, spectroscopy.

Scanning tunnelling spectroscopy
Spectroscopy experiments were carried out using the same automation as above, but instead of charge injection at the target site STS was performed. The STS spectra were acquired directly with a software based lockin amplifier. The bias was modulated at 521 Hz with an amplitude of 20 mV. Spectra were acquired in the range from 0 to 2 V. From an initial stabilisation voltage of +1 V, the tip was pushed closer to the surface by 100 pm V -1 to amplify the signal at low bias [53].
Acknowledgments PAS gratefully acknowledges support from the EPSRC grant EP/K00137X/1. KRR was funded by a University of Bath studentship.