Figure 2.
Department of Synchrotron Radiation Research
Institute of Physics, Lund University
Box 118, S-221 00 Lund, Sweden
With a scanning tunnelling microscope (STM), even very small amounts of an adsorbate are accessible for studies, which is not the case for most surface science techniques. The STM study presented here is concentrated on the very initial stages of the adsorption of Li on Si(001) addressing the question whether the Li prefers to sit on defects and step edges or on top of the dimer atoms [6]. Our results show that the half missing dimer defect (C-type defect) act as the first adsorption site. Once the C-type defects are occupied, Li adsorb on top of the dimer atoms in agreement with earlier STM studies [6,7], and forms small clusters.
The Li atoms and clusters stabilise the buckling of the dimers, resulting in a local c(4x2) periodicity. This leads to the occupation at the Fermi level of an empty surface state, associated with the c(4x2) reconstruction, which we observe with angle-resolved ultraviolet photoelectron spectroscopy (ARUPS).
The substrates used for the experiments were cut from (001) oriented silicon wafers [14]. We have used different methods of preparation before transfer into vacuum: (1) Degreasing in methanol, (2) degreasing and etching in an HF:H20 solution and (3) a wet etching treatment described by Ishizaka and Shiraki [15]. Pre-treated samples where then transferred into the UHV-chamber, outgassed at 300-400°C for 12 h and, finally, repeatedly flashed at 1150-1200°C for 5-10 s by direct current resistive heating until a well ordered and clean surface was obtained. After each flash the temperature was decreased rapidly to 800°C, and finally the sample was cooled to room temperature over a period of about 5 minutes. No variance in the final result due to different pre-treatments was discerned. Both p-type Si(001) (B, 0.02-0.03 W cm) and n-type Si(001) (P, 0.2-0.4 W cm) have been used for the measurements.
The evaporation of Li was performed with a well outgassed getter source (SAES Getters S.p.A.). During the evaporation the pressure in the chamber never exceeded 3x10-10 mbar. The coverage was estimated by measurement of the change in work function determined by secondary electron cutoffs in the photoemission experiment and by inspection of the STM topographs in the STM experiment.
The STM measurements presented here of the Li adsorbed Si(001) surface is concentrated to a coverage of about 0.005 ML (1ML here is defined as the Si atom density of a full monolayer, i.e. 6.8x1014 atoms/cm2). Applying a negative tip bias with respect to the sample immediately resulted in a transfer of Li atoms from the sample to the tip and the loss of clean tunnelling conditions. This behaviour has been reported for K/Si(001) [19], and is very likely due to the electropositive character of Li. Therefore only topographs and tunnelling spectra of the filled states are possible.
Figure 2(a) shows a typical filled states topograph of an Li adsorbed Si(001) surface. The type-A defect and its derivative multiple-vacancy defects are still present but the type-C defects seems to have disappeared. A careful inspection shows that the C defects are not missing but they now appear as protrusions in the topograph indicating that they now have a higher density of filled states. We interpret this to be due to Li occupying the defect. Different types of Li clusters and a second adsorption site on top of the dimer atoms (as proposed in Ref. [6]) are clearly observed in Fig. 2(a). An enlarged view of two types of small clusters, denoted K1 and K2, is shown in Fig 2(b). Taking into account the metallic radius of Li (1.52 Å) these two clusters are likely to be two successive stages of Li clusters with K2 having one more Li atom than K1. A highly buckled dimer is seen next to the smallest cluster K1, which is incorporated in the larger cluster, K2. Hasegawa et al. [6] reported that Li stabilises the buckling of the dimers into small patches of c(4x2) and p(2x2) symmetries around the Li clusters, which is also observed here. On p-type samples topographs tended to have much lower resolution than on n-type samples, and a characteristic trapping noise could be observed when the tip was over the Li sites, as shown in Fig. 2(c). This will be discussed in greater depth below.
Figure 3(a) shows ARUPS normal-emission spectra of the Si(001) valence band for increasing Li deposition. Here we have defined the coverage corresponding to a work-function change of -1.8 eV as one monolayer, following earlier photoemission studies [3-5]. The peak denoted S corresponds to the filled dangling bond state. This peak shifted downward in energy and became broader with increasing coverage as reported earlier [4]. In addition, an extra peak denoted A in Fig. 3(a), became visible at the Fermi level on the Li adsorbed surface. The intensity of peak A reached a maximum at a coverage of 0.1 ML and then decreased with further Li deposition, and could not be observed for coverages >0.4 ML. It should be emphasised that peak A was observed in the same coverage regime on both p-type and n-type samples. Our ARUPS measurements revealed that the A peak is highly localised (to within a few degrees of normal emission) around the Gamma point in k-space, which implies that this state is somewhat delocalized in real space.
Figure 3(b) shows tunnelling spectroscopy from a Li site and a Si dimer at a Li coverage of 0.005 ML. Both spectra are dominated by a peak at 0.6 V, which corresponds to the ¹-bonding dimer state, S, seen in ARUPS. However there is a clear difference between the spectra at voltages between -1.0 V and -1.5 V, where the Li site has a higher local density of states. We would not expect to observe this difference in the ARUPS spectra because the extra intensity only occurs at the Li adsorption site and the Li concentration is too low.
Analogous observations were made earlier [20,21] on clean, highly n-doped Si(001), where a state appeared at the Fermi level as a result of excess electrons on the surface that partially filled an empty surface state. Similarly, an inverse photoemission study [22] showed the existence of two dispersive empty surface bands in the bandgap, one of which had a minimum close to the Fermi level at the Gamma point. It was argued that this state was associated with areas of correlated, asymmetric dimers which formed a c(4x2) reconstruction. Calculations of this Si(001)-c(4x2) structure [23,24] do indeed show a minimum in the empty surface state band at the Gamma point. On the other hand, in an STM study [18] it was suggested that the peak in Refs. [20] and [21] originates directly from states localised on the C defects.
Our results suggest that a combination of the two interpretations is appropriate. The C defects induce or stabilise buckling of the dimers around them, and thus create a patch of c(4x2) reconstructed Si. All the Li adsorption sites and clusters have a similar effect, although their influence extends over a larger area, as would be expected for an adsorbate which actively injects charge into the surface. We suggest that the peak A appears when the Li density is sufficiently high that a substantial proportion of the surface is c(4x2) reconstructed, and sufficient charge has been transferred from the Li atoms to the substrate. If this is the case, similar effects would be expected upon adsorption of low concentrations of other alkali metals as has indeed been observed in studies of K/Si(001) [25,26] and Cs/Si(001) [27].
We therefore interpret the peak A as originating from a stabilisation and partial occupation of an empty surface band by the Li. Unfortunately it is not possible to say how much the appearance of the c(4x2) state is due to charge injection from the alkali metal and how much it is due to localised stresses around the adsorption site, but a satisfactory theoretical explanation should consider both effects.
The effects discussed above were observed on both n-type and p-type substrates. However, on p-type silicon further interesting electronic effects were observed in the STM experiments. This includes the trapping noise seen over the Li sites in Fig. 2(c), as well as negative differential conductance (NDC) seen in tunnelling spectra taken on both the Li sites and clean Si dimers.
Figure 4 shows differential conductance curves (dI/dV) acquired at different positions on the p-type surface. The region of NDC, centred at -2.0 V, is very clear for the spectrum taken with the tip directly over a Li site. However, the second curve in Fig. 4 shows that a weaker region of NDC also occurs when tunnelling out of a Si dimer. Such NDC was observed for all the dimers on the p-type surface, even those outside the patch of asymmetric dimers surrounding the Li site.
Earlier STM studies have reported NDC for systems such as oxidised silicon [28], boron-Si(111) [29,30], and germanium-Si(001) [31], where it was either explained in terms of the filling of electron traps by tunnelling electrons [28] or resonant tunnelling between localised states on the tip and the surface [29-31].
We believe that the NDC observed here can not be explained by tunnelling between localised states. Because the NDC is observed when the tip is over a silicon dimer, any explanation involving localised states must explain why we do not observe NDC when using the same tip and the same silicon sample before any Li is deposited. Also, we have observed the NDC using several different tips and several different samples, so an explanation involving a particular tip geometry, as has been proposed to explain the B-Si(111) results [32], also seems unlikely.
Our results are similar to those obtained on oxidised silicon where both NDC and trapping noise have been observed when the STM tip is positioned above a trapping site in the oxide. In that case the trapping noise results when near-Fermi level traps are filled by thermally excited electrons, whereas the NDC is associated with the filling of different traps, 1 eV or more away from the Fermi level, by electrons tunnelling from the tip [28]. In contrast, on this surface both effects seem to be caused by a single, long lifetime trap localised on the Li adsorption site.
When the STM tip is over a Li site the trapping noise seen in Fig. 2(c) occurs. If the tip is held stationary the tunnel current is seen to jump suddenly as shown in Fig. 5(a). Both STM of oxide traps [33,34] and macroscopic investigations of metal-oxide-semiconductor field-effect transistor (MOSFET) devices [35] show a similar switching, where the current flow in the vicinity of the trap is reduced when the trap is filled as the occupying electron repels nearby conduction electrons.
In the case of the thermally activated oxide traps close to the Fermi level the probability of the trap being occupied could be varied by applying an electric field and thus changing the potential of the trap with respect to the surface Fermi level. In the STM this results in a change in the average time taken to fill and empty the trap, and the traps' occupation statistics follow a Poisson distribution as a function of tip bias [34]. In Fig. 5(b) we plot the logarithm of the mean time that a typical Li trap remains empty as a function of tip bias. The data fit a straight line well, which is good evidence for thermal activation of traps whose energy with respect to the Fermi level varies with the electric field in the tunnel gap.
Unfortunately the precise mechanism by which the trap energy changes is not clear. It is possible that a state associated with the Li atom experiences a Stark shift with respect to the surface Fermi level, or that tip induced band-bending moves all the surface states with respect to the bulk Fermi level. Evidence that the second mechanism might occur is provided by surface photo-voltage measurements on the clean surface [36] where tip induced band bending was clearly demonstrated.
A more complete analysis of the trapping noise as in Refs. [33] and [34] would be welcome, but the much longer trap lifetimes make this difficult. If the STM feedback loop is left on while the current is measured it changes the distance between the tip and the surface in reaction to the jumps in the current (which explains the obvious slow variation seen in Fig. 5(a)) and therefore affects the electric field on the trap. On the other hand, if the loop is frozen for long enough to take statistically significant data, a tip crash is almost inevitable.
Nevertheless, from the measurements of the NDC and the trapping analysis the following interpretation arises. The adsorbed Li creates a trapping state close to the surface Fermi level. As the tip bias varies the occupation statistics of the trap also change. I-V spectra represent the average tunnel current and therefore show a region of NDC where the trap becomes progressively more likely to be filled. Because the trapping state is physically on the surface it has a direct effect on the surrounding surface states and so the filled trap can block current even at some distance from the trapping site, which is why we observe NDC over the whole surface. Finally, we do not observe any of these effects on the n-type substrate because electrons from the bulk ensure that the trapping state is permanently occupied.
Figure 1.
Filled states topograph of the Si(001)-2x1 reconstructed surface.
(Vs = -1.8V, It = 0.3nA, 20x20 nm2)
Denoted in the figure are the defect types A ( missing dimer defect) and C ( half missing dimers).
STM topograph of the Si(001)-2x1 with a coverage of 0.005 ML of Li.
a) Vs = -1.5V, It = 0.2nA, 47x47 nm2, n-type.
b) Vs =
-1.5V, It = 0.25nA, 8x7 nm2, n-type.
c) Vs = -1.2V, It = 0.1nA, 11x11 nm2,
p-type.
a) Normal emission ARUPS spectra measured on the Li deposited Si(001)-2x1 surface for increasing Li coverage.
b) Normalised
conductivity curves on the Li exposed Si(001) surface at a coverage of 0.005
ML.
The dimer spectra is averaged over 542 individual IV curves and the
lithium spectra is averaged over 135 individual IV curves.
Differential conductivity curves (dI/dV )
acquired on Li-Si(001)-2x1.
The dimer and Lithium spectra are averaged over
92 and 72 individual IV's respectively.
a) Current trace for a typical trapping event with the tip positioned above a Li site.
b) Average time analysis for the time
spent out of a trap vs. the applied gap voltage.
The straight line is a
least-squares fit to a Poisson distribution.
