Electronic structure changes of Si „ 001 ...-„ 2 Ã 1 ... from subsurface Mn observed by STM

The deposition of Mn atoms onto the Si 001 2 1 reconstructed surface has been studied using scanning tunneling microscopy STM and first-principles electronic structure calculations. Room-temperature deposition of 0.1 ML monolayer of Mn gives rise to a disordered surface structure. After in situ annealing between 300 and 700 °C, most of the Mn is incorporated into three-dimensional manganese silicide islands, and Si dimer rows reappear in the STM images on most of the substrate surface. At the same time, rowlike structures are visible in the atomic-scale STM images. A comparison with calculated STM images provides evidence that Mn atoms are incorporated into the row structures in subsurface interstitial sites, which are the lowest-energy position for Mn on Si 001 . The subsurface Mn alters the height and local density of states of the Si dimer atoms, causing them to appear 0.6 Å higher than a neighboring Si dimer with no Mn below. This height difference that allows the detection the subsurface Mn results from a subtle interplay of geometrical and electronic effects.


I. INTRODUCTION
In recent years, several aspects of the interaction of manganese atoms with silicon surfaces have attracted attention.It has been speculated that incorporation of Mn on substitutional sites of the Si lattice using nonequilibrium growth techniques could turn silicon into a dilute magnetic semiconductor, 1,2 similar to Mn incorporation in Ge, 3,4 or could be used to achieve ␦ doping of Si with possible applications as spin valves. 3,5An alternative route to spintronics applications focuses on heterostructures formed by magnetic metal layers on silicon.Such heterostructures could be used as Schottky diodes for injection of a spin-polarized current, a concept that has been successfully demonstrated already for nonmagnetic tunnel contacts. 6Theoretical considerations suggest that epitaxial MnSi thin films should order ferromagnetically on Si͑001͒. 7This has lead to the expectation that MnSi films on Si could be used to fabricate magnetic heterostructures by molecular-beam epitaxy.
A number of studies have been devoted to the growth of manganese silicides on silicon ͑see Ref. 8 and references therein͒.It has been found that ultrathin films on Si͑001͒ tend to form three-dimensional manganese silicide clusters upon annealing, 9 and these clusters undergo a ripening process at annealing temperatures of 300-700 °C. 8No closed and flat films of Mn or MnSi on Si͑001͒ have been reported to date.However, the ripening process that takes place during annealing involves material exchange between the clusters.If the sample is quenched to room temperature, a finite concentration of mobile Mn adatoms present between the three-dimensional islands can be retained.Recent calculations by some of the present authors using density-functional theory ͑DFT͒ predict that the lowest-energy position for Mn atoms on the Si͑001͒ surface is a subsurface interstitial site. 7,10However, images that show the effect of these interstitial Mn atoms have not been presented.
In this paper, the Si͑100͒ surface is imaged with scanning tunneling microscopy ͑STM͒ after deposition of a small amount of Mn atoms and subsequent annealing, and rowlike structures on the Si͑001͒-͑2 ϫ 1͒ reconstructed surface with a local c͑4 ϫ 2͒ periodicity are found.By comparing the experimental images with DFT calculations, these structures are identified as c͑4 ϫ 2͒-periodic arrangements of Mn atoms occupying subsurface interstitial sites.

II. EXPERIMENT
The experiments were carried out in a combined molecular-beam epitaxy and scanning tunneling microscopy ͑STM͒ multichamber ultrahigh-vacuum ͑UHV͒ facility. 11ommercially available Si͑001͒ wafers ͓B doped, ͑1.75± 0.75͒ ϫ 10 −3 ⍀ cm͔ were diced into 10ϫ 2 mm 2 rectangular pieces and cleaned ex situ in an ultrasonic acetone bath.They were clamped in sample holders made of Ta, which had been previously cleaned by heating a sacrificial silicon sample at approximately 1250 °C for a period of 24 h under UHV conditions.The samples were introduced into UHV through a load lock.A conventional in situ preparation process of annealing and flashing the sample was used in order to create a clean, flat Si͑001͒-͑2 ϫ 1͒ reconstructed surface, which was confirmed by in situ STM imaging. 11The sample was annealed using direct current heating, and the temperature was measured using an infrared pyrometer.
One-tenth of a monolayer of Mn was deposited using an electron-beam evaporator while the substrate was at room temperature ͑23 °C͒.The deposition rate was measured using a quartz microbalance, and the chamber pressure was below 5 ϫ 10 −10 mbar during the deposition.To observe the effect of annealing, some samples were annealed at 650 °C for 15 min and then cooled to room temperature and transferred under UHV conditions to the STM, while other samples were transferred to the STM without any postdeposition annealing.The STM images were acquired at room temperature with a constant tunneling current of about 0.1 nA and a tip bias of about 2.5 V.These tunneling parameters were found to give the best image quality, and systematic bias dependent imaging was not performed.The base pressure in the STM chamber was below 3 ϫ 10 −11 mbar.

III. CALCULATIONS
For comparison with the STM measurements, we performed DFT calculations of atomic configurations that are likely to occur during the initial stages of growth.A comprehensive summary of all the structures considered is published elsewhere. 10The adsorption of Mn atoms on Si͑001͒ is described using a slab model of the surface.DFT totalenergy calculations are performed using the PBE functional for electronic exchange and correlation and the ͑L͒APW + lo approach to electronic structure calculations, as implemented in the WIEN2k computer code. 12The uppermost three Si layers of the slab were fully relaxed until the forces on each moveable atom were smaller than 0.05 eV/ Å. STM images were simulated within the Tersoff-Hamann approximation 13 by integrating the electronic density of states in an energy interval between the Fermi energy ⑀ F and ⑀ F − 1 eV, which corresponds to filled-state images.
We took the energy interval of states contributing to the tunneling current narrower than expected from the overall experimental voltage drop between tip and sample, assuming that part of the applied voltage drops in the sample ͑e.g., due to band bending͒, rather than between the tip and the sample surface.Subsequently, the height above the surface for an isocontour surface of 10 −5 eV/ bohr 3 was calculated and plotted.Using a smaller energy window of 0.5 eV to define the density of states contributing to the tunneling current had little effect on the computed images.Likewise, empty-state images were computed by integrating the electronic density of states in an energy interval between the Fermi energy ⑀ F and ⑀ F + 1 eV.However, these images were found to give little additional information and are not shown here.

IV. RESULTS
The STM images taken after deposition of 0.1 ML ͑mono-layer͒ of Mn at room temperature onto the clean Si͑001͒-͑2 ϫ 1͒ reconstructed surface shows a rather unstructured surface with a random distribution of small clusters, as displayed in Fig. 1͑a͒.In addition, this image displays a single height step of the underlying Si͑001͒ surface running vertically up the image.The STM images taken from samples annealed after deposition at temperatures between 300-700 °C display the formation of three-dimensional islands, ͑not shown͒, while between the islands, the dimer rows of the Si͑001͒-͑2 ϫ 1͒ reconstruction become visible again as displayed in Figs.1͑b͒-1͑f͒. 8In these images, the defect density in the flat areas between the islands is significantly higher than prior to the deposition.In addition to the defects, bright rowlike structures can be observed that run in the ͓110͔ direction, perpendicular to the underlying dimer rows of the Si͑001͒-͑2 ϫ 1͒ reconstruction.Two or more of these rows sometimes form adjacent to each other, resulting in two-dimensional structures with a local c͑4 ϫ 2͒ periodic-ity as highlighted by a rectangular box in Fig. 1͑e͒.The bright rows are about 1.1 Å higher than the adjacent atoms as indicated by the height cross section shown in Fig. 1͑g͒, which is taken from the dashed line in Fig. 1͑f͒.
Several computed STM images, using the Tersoff-Hamann approximation, are displayed in Figs. 2 and 3.The clean Si͑001͒-͑2 ϫ 2͒ reconstructed surface showing the buckled Si dimers running in the ͓11 ¯0͔ direction is displayed in Fig. 2͑a͒.Three images for ͑2 ϫ 2͒ arrangements of Mn atoms in three different sites on the Si͑001͒ surface, subsurface, substitutional, and on surface are displayed in Figs.2͑b͒-2͑d͒, respectively.The Mn subsurface location alters the surface of constant charge density by making the rows run in the ͓110͔ direction, perpendicular to the original dimer row direction, while the other two Mn locations do not alter the row direction.In the computed empty-state images, the zigzag rows characteristic of the clean Si͑001͒-͑2 ϫ 2͒ surface are inverted and show less corrugation, as has been observed earlier. 14For Mn in the substitutional site, the Mn atom is less visible in the empty-state image than in the filled-state image.For both the on-surface and the subsurface position of Mn only, very little changes are observed between computed filled-state and empty-state images.
Comparison of computed images of c͑4 ϫ 2͒ reconstructed surfaces are displayed in Fig. 3.A image of the clean Si͑001͒-c͑4 ϫ 2͒ reconstructed surface showing the buckled Si dimers running in the ͓11 ¯0͔ direction is displayed in Fig. 3͑a͒.In addition, the computed surface showing the effect of Mn incorporated into the subsurface site arranged in a c͑4 ϫ 2͒ periodicity is displayed in Fig. 3͑b͒.The image with subsurface Mn is clearly different from the clean c͑4 ϫ 2͒ reconstruction with rows running in the same direction as the buckled dimers.While the bright features in the image of the clean surface in Fig. 3͑a͒ occur at the positions of the lower Si atom of the Si dimers, the bright features in Fig. 3͑b͒ are centered above the Si dimers with Mn atoms in the subsurface position below.
A comparison between the experimentally obtained STM images and DFT computed surfaces of constant density of states is displayed in Fig. 4. For reference, a high-resolution image of the clean Si͑001͒-͑2 ϫ 2͒ reconstructed surface taken at 77 K showing the buckled dimers along with the DFT computed image showing the same morphology is displayed in Figs.4͑a͒ and 4͑b͒, respectively.A high-resolution STM image of one of the rowlike features observed after Mn deposition and annealing and the computed image of the c͑4 ϫ 2͒ arrangement of Mn in the subsurface position are displayed in Figs.4͑c͒ and 4͑d͒, respectively.In these images, similar morphologies can be observed within the rows along with the out-of-phase arrangement between the adjacent rows.
A height cross section taken from the STM image shown in Fig. 1͑f͒ across three of the rows observed after Mn deposition is plotted along with a similar height cross section taken from the DFT image shown in c͑4 ϫ 2͒ and displayed in Fig. 5.Both plots show three peaks separated by the same distance with the center peak about 0.6 Å above the other two.

V. DISCUSSION
The surface morphology observed in Fig. 1͑a͒ after Mn deposition onto the Si͑100͒ surface is consistent with a Volmer-Weber ͑islanding͒ growth mode.This is in agreement with conclusions drawn from our DFT calculations, 7,10 which show that the formation of a very thin film of MnSi on Si͑001͒ would be thermodynamically unfavorable.In addition, the formation of the larger MnSi islands after annealing is consistent with previous observations for Mn deposition and annealing on this surface. 8,9The rowlike structures observed in the STM images between the islands displayed in Fig. 1 are not typically observed on clean Si͑001͒ surfaces.Therefore, we conclude that these structures must arise from the Mn deposition and annealing.It has been previously determined that the large Mn-Si islands undergo an Ostwald ripening process, 8 which necessitates the exchange of material between islands, and thus these rowlike structures could possibly result from the freezing of these mobile Mn adatoms during the ripening process.
The STM images alone do not indicate the type of species that exist in the observed structures.However, much insight can be gained by comparing these images with both the theoretical STM images presented here and previous studies in the literature about defects or islands on the Si͑001͒ surface.It is well known that even in the absence of Mn, the Si͑001͒ surface can form numerous defect structures.However, defects cannot account for the unique rowlike structures observed in the STM images.5][16][17] Most frequently, missing dimer defects have been observed, which are also largely accountable for the black regions ͑holes͒ in Fig. 1͑c͒.Missing dimer defects can also be present on a clean, wellordered Si͑001͒ surface, but with much lower probability.
Likewise, pure Si adatom islands on Si͑001͒ have been reported to form rowlike structures in the early stages of homoepitaxy. 18However, the monolayer height of these is-lands is typically 1.3 Å, different from the 1.1 Å height of the rowlike structures observed in the STM images in Fig. 1 and the height cross section displayed in Fig. 5.Moreover, the bright protrusions in Figs.1͑b͒-1͑f͒ could not be observed on clean Si͑001͒ surfaces without depositing Mn.In addition, the c͑4 ϫ 2͒ reconstruction of clean Si͑001͒ would give rise to STM images which significantly differ on the atomic scale from the Mn-induced structure with the same local periodicity.
Hence, we conclude that the bright row structures that are being formed perpendicular to the rows of the underlying surface are due to incorporated Mn atoms into the next layer of Si.Three arguments can be given to confirm that these Mn atoms occupy subsurface interstitial sites.First, the subsurface interstitial site is calculated to be the most stable site, lower in energy than any on-surface or bulk interstitial site. 7,10,19Second, in all other periodic structures we considered, the Mn-induced electronic states give rise to an appearance of the simulated STM images different from those for subsurface incorporation ͓compare Figs.2͑b͒-2͑d͔͒.In particular, the acquired STM images are incompatible with the hypothesis of rows of substitutional Mn atoms ͓Fig.2͑c͔͒.Third, there is quantitative agreement ͑as discussed below͒ between the corrugation in the experimentally obtained STM images and in the row structures simulated for subsurface interstitial sites as displayed in Fig. 5.
The DFT calculations tell us that the stability of Mn in the subsurface interstitial site is related to its high coordination.In order to achieve this, the Mn atom establishes chemical bonds to the neighboring Si atoms on the expense of the Si-Si surface bond that is considerably weakened, indicated by an increase of 0.4 Å of the Si-Si dimer bond length compared to the clean Si͑001͒ surface.At the same time, the surface Si atoms above the Mn move outward from the surface.
Interestingly, the DFT calculations show that the physical position of the Si dimers with Mn underneath are topographically higher.However, this height difference is not fully reflected in both the calculated and measured STM images.When the Mn atom is present, electrons are pulled into the newly formed Mn-Si bonds below the surface, making the surface Si atoms, in the presence of Mn, appear less bright.Comparison of the absolute height scales for the two DFT images in Figs.2͑a͒ and 2͑b͒ reveals that Fig. 2͑b͒ is overall 0.2 Å lower than the image of the clean surface, Fig. 2͑a͒.This occurs despite the fact that the positions of the Si atoms with the Mn below are higher than normal for a clean Si dimer.
This electronic effect overcompensates the geometric effect in both the computed as well as the measured images.The Si dimers without Mn below them adjacent to those with Mn below them appear even darker, as evidenced by the grayish parts between the bright spots of Fig. 2͑b͒.This is attributed to the electrons in the dangling bonds of these Si dimers moving into the Mn-Si bonds at the other, neighboring dimer.This electronic effect explains why the middle Si dimer with incorporated Mn does not appear 1.3 Å above the terrace, but only about 1.1 Å ͓cf.Fig. 1͑g͔͒.This effect is also responsible for the height difference of 0.6 Å between the Si dimers with Mn below and without Mn below in the same terrace as displayed in Fig. 5.As explained above, these "normal" Si dimers appear to be lower than expected ͑although geometrically, they are at their usual height͒ due to electronic effects, since part of their charge has been pulled away into the Mn-Si bonds in the neighboring dimer.
Our results are in qualitative agreement with calculated STM images of isolated Mn atoms on Si͑001͒. 19Moreover, our calculations show that the subsurface Mn atoms widen the Si lattice around the interstitial site.As a consequence of these distortions, two Mn atoms in the same Si dimer row interact repulsively.For adjacent interstitial sites, the interaction energy is 0.4 eV/ Mn.Therefore, only half of the available sites are occupied by Mn, giving rise to a local Mn coverage of a quarter of a monolayer with either a ͑2 ϫ 2͒ or c͑4 ϫ 2͒ local arrangement of the Mn atoms.While the formation energy of these two structures is the same within the accuracy of our calculations, only the c͑4 ϫ 2͒ arrangement has been observed in the experiment when two rows were observed next to each other.

VI. CONCLUSION
Under suitable preparation conditions, two-dimensional rowlike structures could be created by the deposition of 0.1 ML Mn on Si͑001͒ and subsequent annealing, besides the three-dimensional Mn-silicide islands reported previously.
By combining STM images with first-principles calculations, we could identify these rowlike structures as being due to Mn atoms in subsurface interstitial sites arranged in a local c͑4 ϫ 2͒ pattern.No evidence for substitutional incorporation of Mn was found.This makes it unlikely that similar physical concepts could be used to turn Si into a diluted magnetic semiconductor as has been demonstrated for Mn-doped Ge. 3,4 Yet, the observed two-dimensional structures induced by Mn could be useful for achieving ␦-layer doping of Si by Mn.Likewise, the observed interstitial incorporation of Mn is compatible with the atomic structures associated with the predicted ferromagnetism of flat Mn-Si islands or films on Si͑001͒. 7

FIG. 1 .
FIG. 1. STM images of ͑a͒ the Si͑001͒-͑2 ϫ 1͒ reconstructed surface after deposition of 0.1 ML of Mn at room temperature and ͓͑b͒-͑f͔͒ after deposition of 0.1 ML Mn at room temperature and subsequent annealing at 650 °C at different resolutions as indicated.The rows running in the ͓11 ¯0͔ direction consist of Si dimers known from the clean substrate.The bright features running in the ͓110͔ direction include Mn atoms.͑e͒ A contrast enhanced inset of ͑c͒ with a c͑4 ϫ 2͒ unit cell highlighted with the dashed box.͑g͒ Height cross section taken from the region in image ͑f͒ as indicated by the dashed line.

FIG. 2 .
FIG.2.Theoretically computed images using the Tersoff-Hamann approximation ͑surfaces of constant density of filled states down to 1 eV below the Fermi energy͒ for ͑a͒ clean Si͑100͒, ͑b͒ Mn incorporated into subsurface interstitial site, ͑c͒ Mn in a substitutional site, and ͑d͒ Mn in a on-surface site.

FIG. 5 .
FIG. 5. Line scan perpendicular to the rows with local c͑4 ϫ 2͒ structure ͑dashed line͒ along the marked line in Fig. 1͑f͒ and from the DFT calculations ͑solid line͒.