Atomic-level spatial distributions of dopants on silicon surfaces: toward a microscopic understanding of surface chemical reactivity

.::~:~:!::::~.:.:.-.-. . . . . . . .

~.:~:::~:[

applied surface science ELSEVIER

Applied Surface Science 107 (1996) 25-34

Atomic-level spatial distributions of dopants on silicon surfaces: toward a microscopic understanding of surface chemical reactivity Robert J. Hamers *, Yajun Wang, Jun Shah Department of Chemistry, University of Wisconsin-Madison, 1101 University Avenue, Madison, WI 54706, USA

Received 13 October 1995; accepted 3 December 1995

Abstract We have investigated the interaction of phosphine (PH 3) and diborane (B2H6) with the Si(001) surface using scanning tunneling microscopy, infrared spectroscopy, and ab initio molecular orbital calculations. Experiment and theory show that the formation of P - S i heterodimers is energetically favorable compared with formation of P - P dimers. The stability of the heterodimers arises from a large strain energy associated with formation of P - P dimers. At moderate P coverages, the formation of P - S i heterodimers leaves the surface with few locations where there are two adjacent reactive sites. This in turn modifies the chemical reactivity toward species such as PH 3, which require only one site to adsorb but require two adjacent sites to dissociate. Boron on Si(001) strongly segregates into localized regions of high boron concentration, separated by large regions of clean Si. This leads to a spatially-modulated chemical reactivity which during subsequent growth by chemical vapor deposition (CVD) leads to formation of a rough surface. The implications of the atomic-level spatial distribution of dopants on the rates and mechanisms of CVD growth processes are discussed.

1. Introduction In is widely recognized that the rates and mechanisms of chemical vapor deposition (CVD) reactions on semiconductor surfaces, as well as the morphology of CVD-grown thin films, are often strongly affected by the presence o f even small amounts o f dopants [1-6]. As semiconductor devices shrink in size, scaling laws indicate that dopant concentrations must increase, but at high concentrations dopant

* Corresponding author. Tel.: -t-1-608-2626371; e-mail: rjhamers @facstaff.wisc.edu.

concentrations precipitation can occur. In recent years, there has been increased interest in bypassing thermodynamic constraints on chemical composition and structure through the use o f non-equilibrium methods for growth and doping. In principle, elements from group V can adsorb on or replace the outermost layer of Si atoms on the dimerized Si(001) surface; because the resulting surface has no partially-filled 'dangling bonds', they are chemically passive and have different surface free energy compared with the clean dimerized Si(001) surface. The use o f arsenic and antimony for surfactant-mediated growth has already been demonstrated [7,8]; since phosphorus is isoelectronic with both As and Sb,

0169-4332/96/$15.00 Copyright © 1996 Published by Elsevier Science B.V. All rights reserved. PII S0169-4332(96)00505-3

26

R.J. Hamers et aL /Applied Surface Science 107 (1996) 25-34

phosphorus might also be expected to exhibit surfactant-like behavior. Another form of non-equilibrium growth making use of dopants is the formation of non-equilibrium dopant structures. In the particular case of boron, it is widely known that under proper conditions it is possible to confine the boron atoms to a two-dimensional sheet, and that in this manner it is possible to grow boron-doped structures in which the boron concentration exceeds the bulk solubility limit by several orders of magnitude [9-12]. Because of the importance of dopants in modifying both the surface energies and morphologies of CVD-grown films and the potential for using dopants to form unique surface structures, we have utilized scanning tunneling microscopy (STM), Fouriertransform infrared spectroscopy (FTIR), and ab initio molecular orbital calculations to investigate the adsorption, dissociation, and energetics of molecular precursors used for doping Si(001) during CVD growth. Using this combination of techniques we are able to establish, at an atomic level, some of the mechanisms by which boron and phosphorus modify the surface structure, energy and local chemical reactivity [13-18].

3. Results and discussion

3.1. Phosphorus / Si(O01)

Phosphine interacts quite strongly with the Si(001) surface. At high fluxes ( > 10 -6 Torr), both STM [14] and infrared measurements [18] show that a substantial amount of the phosphine adsorbs molecularly. STM images, as in Fig. la, reveal that at low coverages PH 3 molecules tend to align along a given dimer row, while at higher coverage (Fig. lb) the correlations between rows lead to a c(4 × 2) configuration (0.25 monolayer P coverage). As will be shown below, infrared spectra of these samples show peaks due to both PH 3 and PH 2, demonstrating that some dissociation occurs already at 300 K on the time scale of the experiment. In a previous study [18] we showed that the ratio of PH3(ads) to PH2(ads) increases with increasing flux (pressure) for a fixed total exposure, supporting the conclusion that disso-

2. Experimental For S T M measurements, samples of N-type Si(001), < 0.1 f~ cm resistivity, were cleaned in methanol and degassed in ultrahigh vacuum (UHV); annealing to 1400 K at a pressure of approximately 2 × 10 -1° Torr leaves a clean, well-ordered Si(001) surface as described previously [19]. For infrared measurements an identical cleaning procedure was used, except that the samples were double-polished, B-doped, 11-25 12 cm resistivity; the 0.5 mm edges of the samples were cleaved at a beveled angle and polished optically flat. A Mattson RS-1 FTIR spectrometer was directed into the sample, making approximately 40 reflections from the surface via total internal reflection before exiting and being directed onto an InSb detector. Sample temperatures were measured using an infrared pyrometer or by measuring the power applied to the sample during resistive heating, and later calibrating this using a thermocoupie directly attached to the sample face. Ab initio calculations, as described below, were using the Gaussian 94 Molecular Orbital Program (Ganssian).

Fig. 1. (a) STM image of Si(001) surfaceexposedto 0.6 L PH3 at 300 K. (b) STM image of Si(001) surface exposed to saturation coverage of PH3, showing local ordering of molecularly-adsorbed PH3 into c(4 X 2) symmetry.

R.J. Hamers et aL /Applied Surface Science 107 (1996) 25-34

ciation of PH 3 on Si(001) occurs on the time scale of minutes at low coverage but is stable for hours or days at high coverage. When a surface saturated with PH 3 is annealed to temperatures between 575 K (at low coverage) and 675 K (at high coverage), the phosphorus-containing molecular fragments completely decompose to atomic P and adsorbed H over the temperature range from and the STM images reveal a dramatic change in the surface morphology. This step edges on the surface become extremely rough, and surface is decorated with a large number of islands. STM investigation of these surfaces show that they contain a large number of S i - P heterodimers in the outermost layer, as will be discussed in more detail below. The island morphology, however, is easily understood as demonstrating the phosphorus atom on top of the Si(001) surface are quite unstable and easily replace a Si atom in the outermost (dimerized) surface layer; the ejected Si atoms then diffuse and agglomerate into islands. Because molecular PH 3 adsorbs into a c(4 × 2) arrangement at 0.25 monolayer, a single sequence of room temperature adsorption followed by annealing cannot produce a surface which is completely passivated by phosphorus. Such a surface, however, can be prepared by exposing the Si(001) surface to PH 3 at an elevated temperature of approximately 800 K, above the hydrogen desorption temperature. Fig. 2 shows a Si(001) surface produced by exposure to 10 -7 Torr PH 3 for 200 s (20 L exposure) at 825 K (imaging is done after cooling to room temperature). At low resolution, Fig. 2a shows that the surface is now broken up by a large number of line defects, which appear as narrow dark lines extending diagonally, most readily visible just above and to the left of center. (Due to a slight tip asymmetry, these are not easily visible on all terraces.) We have addressed the symmetry and possible atomic structures of these defects in a previous publication [13]. High-resolution STM images, as in Fig. 2b now clearly show P - P dimers; each phosphorus atom has an occupied 'lone-pair' orbital, and STM images of the P-terminated surface clearly show the individual 'lone pair' orbitals associated with each P atom. Because the coverage is not exactly 1 monolayer, some of the surface atoms are Si atoms; these appear as bright spots, much higher than the P atoms, because the electron in the 'lone pair' lies near the

27

Fig. 2. STM image of phosphorus-terminated Si(001) surface. (a) Large-area image, showing strain-related line defects (above and left of center). Imagedimensions 800 × 600 A. (b) High-resolution image (92 A X92 A) showing individual P atoms (dark protrusions) and Si atoms (bright protrusions), and strain-induced line defects (bottom center and other locations).

Fermi energy and is therefore more accessible to tunneling than the low-lying phosphorus lone pair electrons. In order to better understand the origin of these defects, we performed studies of the surface structure over a range of P coverages. In contrast to the large number of line defects produced at high phosphorus coverage, Fig. 3a shows that at intermediate P coverages these defects are completely absent. The surface has a mottled appearance, and high-resolution im-

28

R.J. Hamers et aL /Applied Surface Science 107 (1996) 25-34

ages as in Fig. 3b show that the surface now consists of a large number of P - S i heterodimers. In such heterodimers, the Si atoms appear high (brigh0 and the P atoms lower because electrons can readily tunnel into and out of the Si atom dangling bond [131. The observation of large numbers of P - S i heterodimers is surprising because typically it is assumed that because heterodimers have a formal 'dangling bond' on the Si atom, it will be therrnodynami-

cally favorable for the surface to disproportionate into regions of 'clean' Si=Si dimers and regions of P - P dimers, via the surface disproportionation reaction 2(P-Si) ~ (P-P) + (Si=Si). Yet, experiments show that just the opposite is true. By counting the number of P-P, Si=Si, and Si-P dimers on the surface, we determined that the equilibrium constant for this reaction is 0.07. The effective temperature of the distribution can only be estimated, but we believe that a value of approximately 500 K is reasonable. Then, since AG = - R T In Keq, we find that the free energy change in the disproportionation is about 0.1 eV, or 0.05 eV/dimer.

3.2. Ab initio molecular orbital calculations f o r P / Si(O01)

Fig. 3. STM image of Si(001) surface with less than 1 monolayer phosphorus. (a) Large area image, 900 A×900 A; (b) high resolution image, 100 A× 100 A, showing individual Si and P atoms. Note the absence of line defects at this phosphorus coverage,

The unexpected stability of Si-P heterodimers is also predicted from molecular orbital calculations. The starting point in these calculations is the Si10H14 cluster shown in Fig. 4a. A complete energy minimization is performed in which the positions of all Si and H atoms are allow to vary to minimize the total energy of the cluster. Removal of the top Sill 2 group leaving the positions of all other atoms fixed produces a Si9H12 cluster with an exposed, unreconstructed (001) surface as shown in Fig. 4b. In this and subsequent calculations, the positions of silicon and phosphorus atoms are always allowed to completely adjust to minimize the total energy. Simultaneously, the positions of the H atoms which terminate the edges of this Si9H12 cluster can be held fixed (to simulate a rigid crystal lattice), or can be allowed to relax (to simulate a soft crystal lattice). The relaxed configuration is always lower in energy, and the difference between the constrained and unconstrained energy minimizations represents a strain energy, as discussed below. Similar calculations were then performed in which one or two of the Si atoms in the outermost layer were replaced with phosphorus. All calculations were performed using the Ganssian 94 program with the 6-31 + G* basis set and the Becke3LYP density functional to account for electron correlation/configuration interaction effects. Fig. 4 c - e shows the results of the calculations for the Si= Si dimer, the Si-P heterodimer, and the P - P

R.J. Hamers et aL /Applied Surface Science 107 (1996) 25-34

dimer, respectively. First, we note that the overall energy change for 2 S i - P ~ S i = S i + P - P was calculated to be + 0 . 2 7 eV, or about + 0 . 1 3 eV per S i - P dimer if we use the constrained optimization (rigid crystal lattice), or about 0.15 eV (0.07 e V / d i m e r ) if we use the unconstrained optimization (soft crystal lattice). These numbers are at least qualitatively in agreement with the value of about 0.05 e V / d i m e r obtained from analysis of the STM images. Most importantly, both theory and experiment agree that the S i - P is the energetically favorable chemical form of phosphorus on the surface, despite the fact that it leaves the Si atom with a coordination number of only three. The likely origin of the stability of the S i - P heterodimer is the lattice strain associated with formation of a P - P dimer. The strain energy associated with a P - P dimer will depend on whether the adjacent dimers are also P - P dimers or whether they are S i = S i dimers; this comes into play through the choice of constrained versus unconstrained boundary conditions for the cluster. For example, if the entire surface consists of P - P dimers, then atoms in layer 2, 3 and 4 (in Fig. 4a above) can only relax perpen-

29

dicular to the surface, not parallel. For an isolated P - P dimer on an otherwise Si(001) surface, however, these atoms can relax both parallel to and perpendicular to the surface. Since our calculations permitted relaxation in both directions, these numbers represent lower bounds on the strain energies; a complete overlayer of P - P dimers will then have a strain energy larger than 1.13 eV/dimer. Note, however, that the strain energy associated with a S i - P dimer is only slightly larger than the strain associated with the reconstructed Si(001) surface, imparting an additional strain of about 0.2 eV/dimer. However, the formation of a P - P dimer imparts about 0.6 e V / d i m e r strain. Thus, if we only consider the lattice strain, the change in strain energy associated with 2 S i - P ~ Si= Si + P - P is thermodynamically uphill 0.12 eV. This represents a very significant fraction of the overall energy change of about 0.27 eV associated with this disproportionation. Thus, we conclude that strain is an important, and most likely the dominant, force responsible for the unusual stability of S i - P dimers on Si(001). We note that the cluster used here is small (one dimer); however, STM experiments have shown that

b)

E--O

c)

d)

P.

e)

E= -1.651 eV (con) E= -2.184 eV (rel)

E= -1413.029 eV (con) E=-1413,801 eV(rel)

E= -2824.141 eV (con) E= -2825.272 eV(rel)

Estrain = 0.53 eV

Estrain = 0.77 eV

Estrain = 1.13 eV

Fig. 4. Clusters used in ab initio calculations. (a) Starting Sil0H14 cluster; (b) unreconstructed Si(001) surface exposed by removal of S i l l 2 group from (a); (c) dimerized Si(001) surface with energies for constrained (con) and unconstrained, or relaxed (rel) energy minimization procedures; (d) Si-P heterodimer with constrained and relaxed energies; (e) P-P dimer duster with constrained and relaxed energies.

R.Z Hamers et al./Applied Surface Science 107 (1996) 25-34

30

dimer tilting typically induces neighboring dimers in the same dimer row to tilt in an alternating manner, while correlations between adjacent rows are weaker [20]. This correlated tilting presumably occurs to further reduce lattice strain. While the single unit cell used here permits dimer tilting, the absence of longer-range dimer correlation is absent. However, this effect of including this longer-range interaction would be to further stabilize the P - S i heterodimer at the expense of Si=Si and P = P dimers. We also note that examination of the STM images does not show any clear alternation of P - S i and S i - P dimers, suggesting that, while dimer tilting does occur for the P - S i dimer, the long-range correlation of dimer tilt directions does not appear to significantly modify the surface energetics. Therefore, we believe that the Si 9 cluster captures the primary features of the S i - P and P - P dimers, at least at a qualitative level.

3.3. Consequences of atomic-level dispersal of phosphorus on surface chemical reactivity The formation of S i - P heterodimers instead of Si=Si and P - P dimers has several consequences. First, we note that the 'dangling bond' on the Si atom can be terminated by a hydrogen atom. Fig. 5 shows FTIR spectra for PH 3 adsorbed on Si(001) at 300 K and subsequently annealed to temperatures as indicated. Immediately after exposure at room tem-

~10_3''o"

...........

perature, a mixture of molecular adsorption (giving rise to the three peaks at 2268, 2280, and 2290 cm -1) and dissociative adsorption (giving rise to P - H peaks at 2239 to 2260 cm -1 and S i - H peaks near 2090 and 2070 cm -1). As the sample is annealed, the FTIR spectra show the dissociation of molecular PH 3 into PH 2 (550-600 K) and several S i - H vibrations. The S i - H stretching mode at 2091 cm-1 arises from the S i - H stretch of dimers of the form d b - S i - S i - H (where 'db' represents one electron in a dangling bond). At higher temperatures, the H atoms diffuse and pair onto the dimers, forming dimers of the 'monohydride' phase, HSi-SiH; this gives rise to symmetric and asymmetric modes at 2087 and 2096 cm -]. Finally, at high temperatures, only a single S i - H peak is observed near 2110 cm-~; this peak arises from H atoms adsorbed onto P - S i heterodimers, forming species of the form P Sill. This last observation is particularly important, as the persistence of this vibrational mode at higher temperatures demonstrates that the presence of surface phosphorus increases the binding of hydrogen on the surface through the formation of P - S i l l 'hydrided heterodimers'. This conclusion is also supported by quantum chemistry calculations of energies and vibrational frequencies of these clustersl A Gaussian 94 calculation of the energetics of hydrogen desorption from modified Si 9 clusters as described above indicates the following reaction energies: H - S i - S i - d b ~ Si=Si + H(gas) P - S i - H ~ P-Si-db + H(gas)

j/<--

i---q --

i

i

S __.J 2000

,'5,~:~---~]..~1 ,.300 K I 2100 2200 2300 2400 Frequency (cm "1)

Fig. 5. Infrared spectra of Si(00l) exposed to PH 3 as described in text, and then annealed to the indicated temperatures. Note appearance of new S i - H feature at high temperature.

+3.53 eV, + 3.67 eV.

This in turn indicates that H atoms are more stable by approximately 140 meV when bonded on P - S i l l dimers rather than Si-SiH dimers. Although the kinetics of hydrogen desorption are complicated by the fact that desorption typically occurs through molecular H 2 and not atomic H [21,22], the infrared and ab initio calculations suggest that the presence of phosphorus in CVD processes might reduce the surface reactivity in part through the formation of P - S i l l heterodimers in which the strong binding of H makes these sites less reactive toward incoming reactants. A second consequence of the formation of such heterodimers is the manner in which they affect CVD reactions. Of importance here is the fact that

R.J. Hamers et al. /Applied Surface Science 107 (1996) 25-34

"T

~1~

i [7 ,

5X 1 O-4 tl

~ P

~ . o~ o4

bY , ./" 2000

Op=O.6 I

j,v

o ¢,. d _ L . . ~

/

o , ¢ ¢

,-.._._

j

.H H ~

~'"ti' ["

.f

k~i

Op=0.2

,VH

~,-~, '~' ', I~ ,.ft.~i-~i

21 O0 2200 2300 Frequency (cm -1)

2400

Fig. 6. Infrared spectra of PH 3 interacting with Si(001) having different degrees of surface phosphorus enrichment, as indicated. Note the inhibition of PH3 dissociation at high phosphorus coverages. Fig. 6e shows the spectrum obtained after annealing Fig. 6d to 700 K; note the very large peal(at 2110 cm -1. FTIR spectra, showing stability of H as P-Sill hydrided heterodimer at higher temperatures. most CVD precursors are believed to require two adjacent reactive sites in order to completely dissociate. The implications of this can be seen from the data in Fig. 6, which shows FTIR spectra after exposure of PH 3 to Si(001) surfaces containing increasing amounts of surface phosphorus in the form of P - S i heterodimers. Fig. 6a shows the FTIR spectrum after a clean Si(001) surface was exposed to PH 3 at 300 K. The sample was then annealed to 800 K to decompose the PH 3, producing P - S i heterodimers and gas-phase H 2. After cooling to 300 K, the surface w a s exposed to 5 L PH3, yielding the spectrum shown in Fig. 6b. This procedure was then repeated two more times, yielding the spectra in Fig. 6c and d. Because the saturation coverage of PH 3 on Si(001) at 300 K is about 0.2 monolayer [14], each exposure-anneal cycle replaces approximately 20% of the surface Si atoms with P atoms (most as P - S i heterodimers). Thus, each cycle of the expose-anneal-cool sequence in Fig. 6 a - d increases the surface P coverage, so that the sequence shown in Fig. 6 a - d represent adsorption of PH 3 at 300 K onto surfaces containing approximately 0, 0.2, 0.4, and 0.6 monolayer P, respectively. Finally, the surface shown in Fig. 6d was annealed to 700 K for 1 rain,

31

which is sufficiently high to dissociate the PH x species, while leaving the H atoms on the surface; Fig. 6e shows the resulting spectrum. A comparison of Fig. 6 a - d shows that as the number of P - S i dimers increases, the adsorption of PH 3 is decreased, as evidenced by a reduction in all PH x peaks and the S i - H peak: However, it should be noted that the PH 2 peaks at 2239-2260 cm -1 are virtually eliminated, while the spectral features of molecularly-adsorbed PH 3 at 2268, 2280, and 2290 cm -1 are retained, albeit slightly shifted as a result of lattice strain. Thus, it appears that PH 3 can adsorb as an intact molecule, but its dissociation into PH2(ads)+ H(ads) is strongly inhibited by surface phosphorus. A likely explanation for the inhibition of dissociation in the presence of surface phosphorus is that the PH 3 molecule, like many other species of interest in CVD processing, only requires one surface site to adsorb, but requires 2 immediately-adjacent surface sites in order to react. If an adsorbed molecules requires two immediately adjacent bare Si surface sites to react, then P atom segregation into P - P and Si=Si dimers would produce a surface reactivity scaling with surface phosphorus coverage Op as ( 1 - Op), while P atom dispersal into P - S i dimers would produce a surface reactivity scaling as (1 Op) 2. Thus, one implication of the atomic-level P atom dispersal is that it efficiently reduces the surface reactivity by preferentially eliminating one of the two p a i r e d surface sites of S i = S i dimers. This mechanism might also be operative in decomposition of other precursors such as Si2H 6 and Sill 4. In particular, we note that Si2H 6 has two decomposition pathways: Si2H 6 ~ 2Si(~) + 3H2(g ), and Si2H 6 ~ Si(s ) + H2(g ) + SiH4(g). Although the first appears to be the major reaction channel, the second pathway increases in importance in the presence of surface phosphorus [23,24]. Engstrom and co-workers [23,24] have recently obtained evidence that Si2H 6 adsorption might take place by S i - H activation (proceeding through an Si2Hs(ads) surface intermediate), rather than through S i - S i bond cleavage (proceeding through two Sill 3(ads) intermediates). It is likely that adsorbed Si2Hs(aas) can dissociate to Sill 2 + Sill 3 only if two adjacent surface sites are present, while if adjacent empty surface sites are not present dissociation occurs to produce SiH4(g ) + Sill(ads ). This overall mechanisms would predict that

32

R.J. Hamers et al. /Applied Surface Science 107 (1996) 25-34

the Sill 4 channel should increase with increasing phosphorus coverage, in agreement with the experimental measurements of Engstrom and co-workers [23,24]. Indeed, recent infrared measurements [25] show that adsorption of Si2H 6 onto phosphorus-enriched Si(001) shows a substantially increased stability for Sill 3 species, consistent with this overall picture. Thus, we find that the presence of lattice strain favors the formation of P - S i dimers at the expense of P - P and Si--Si dimers, despite the fact that P - S i dimers are formally 'radical' or 'dangling bond' species. These heterodimers affect the surface reactivity in several ways: first, the P atoms are unreactive because they have a saturated 'lone pair' orbital. Second, the Si atom in the P - S i dimers binds strongly to hydrogen, which also helps to passivate the surface to other chemical reactions. Finally, the dispersal of P on the surface leads to a smaller number of surface locations where two adjacent surface sites are available; this in turn inhibits the dissociation of adsorbed species. The last effect is demonstrate here for PH3, but is also likely operative for other reactants such as Si2H 6 and Sill 4.

3.4. B o r o n / Si(O01)

The behavior of boron on Si(001) is in stark contrast to that of phosphorus. As shown in Fig. 7, the decomposition of diborane (B2H 6) on Si(001) produces a new surface feature which is directly associated with the presence of boron. The situation here is quite different from the previous case of phosphorus: the boron atoms phase-segregate on the surface, producing a new reconstruction with a high local boron concentration, separated by regions of clean Si. The formation of a new reconstruction and phase segregation of boron on the surface is likely driven by the fact that typical B - S i bond lengths are typically much shorter than S i - S i length, so that the direct substitution of B into the surface introduces an even larger amount of strain than P does. As a result, it appears to be more efficient for boron atoms to redistribute into an entirely new reconstruction which optimizes the local energy. A similar effect has been previously predicted by Meade and Vanderbilt [10] for As on S i ( l l l ) , where the (Vr3- × f 3 ) A s / S i ( l l l )

Fig. 7. (a) STM images of Si(001) exposed to B2tt 6 at 810 K. Note the patchy appearance of the surface due to segregation into boron-reconstructed and 'clean' surface regions. Image dimensions 485 ~, X 485 A, (b) High-resolution image of boron-reconstructed Si(001) surface with c(4×4) reconstruction. Image dimensions 50 A X 65 ~,.

surface is predicted to be more stable than the phase-separated regions of (1 × 1 ) A s / S i ( l l l ) and Si(111)-(7 × 7). In the case of B/Si(001), the boron atoms phase-segregate into clean Si and a c(4 × 4) reconstruction. Although several related boron-in-

R.J. Hamers et aL /Applied Surface Science 107 (1996) 25-34

duced reconstructions are observed, the most common B-induced reconstruction has c(4 X 4) symmetry, as shown in Fig. 7b. Two aspects of the boron-induced reconstruction are particularly interesting. First, it is important to understand the atomic arrangements which are revealed in the STM images. Second, it is important to understand how these boron-induced reconstructions affect the surface reactivity and morphology. Using quantitative Auger electron spectroscopy, we found that the boron coverage in the boron-reconstructed regions corresponds to 0.5 monolayer [15,17]. By analyzing a very large number of STM images as a function of exposure and at different sample-tip voltages, we developed a structural model which we believe accounts for all the experimental observations. In this model [15], boron atoms substitute into the first full atomic layer of the Si lattice, in a square arrangement of 4 boron atoms. This layer is then capped with Si=Si dimers a n d / o r dimer vacancies. The relative numbers of Si=Si dimers and dimer vacancies can vary, resulting in several related reconstructions. At high boron exposure, the c(4 × 4) reconstruction shown in Fig. 7b predominates. These boron-reconstructed regions also affect the surface chemical reactivity toward disilane and affect the growth morphology. At < 0.5 monolayer boron coverage, the surface contains patches of c(4 × 4) B/Si(001), separated by regions of clean Si. If this surface is exposed to disilane, STM images show that the disilane fragments are observed only on the clean Si regions. Complementary infrared studies measuring the intensity of S i - H stretching mode vibrations shows that the overall chemical reactivity toward disilane scales approximately as ( 1 - OB). This result indicates that disilane molecules which impinge onto the B-reconstructed regions recoil back into the gas phase and do not diffuse on the surface to a more reactive region of clean Si. The spatially-varying chemical reactivity induced by the boron-reconstructed regions has dramatic effects on the surface morphology during CVD growth [16]. To investigate the effect of surface B on the growth morphology, we performed a direct comparison between two samples: one sample was clean Si(001), and the second was a Si(001) surface which was exposed to B2H 6 at 810 kelvin so that 5% of the surface area was covered with the B-induced c(4 X 4)

33

Fig. 8. Boron-inducedroughening during CVD growth. Image dimensions 4700 AX3000 A. (a) Surface structure following growth of 10 monolayers Si by thermal decompositionof Si2H6 on clean Si(001) at 810 K. (b) Surface morphologyfollowing growth of 10 monolayers Si by thermal decompositionof Si2H6 on Si(001) surface with 5% of surface area exhibiting boron-induced reconstruction.

reconstruction. Both samples were then exposed to 300 L S i 2 H 6 at 2.5 X 10 -7 Ton" for 1200 s. As shown in Fig. 8a, growth on the clean sample is nearly layer-by-layer, producing large flat terraces. In contrast, growth on the boron-exposed surface (Fig. 8b) produces an extremely rough surface. The surface instead grows large number of pyramidal structures. This dramatic change in surface morphology and roughness occurs because of the spatiallymodulated chemical reactivity which results from the phase-segregation of boron into reconstructed patches. This boron-induced surface roughening arises because the formation of a boron-induced reconstruction creates a spatially-modulated chemical reactivity. As in the case of phosphorus, the ability to observe the spatial distribution of dopant atoms is

34

R.J. Hamers et al. / A p p l i e d Surface Science 107 (1996) 25-34

key to understanding how the dopants modify the surface chemistry and morphology,

National Center for Supercomputer Applications, grant #CHE950012N.

4. Summary and conclusions

References

We have applied scanning tunneling microscopy, Fourier-transform Infrared Spectroscopy, and quantum chemistry techniques to investigate the atomiclevel spatial distribution of phosphorus and boron on Si(001) surfaces. The structures adopted by these dopants are a reflection of the energetics associated with the lattice strain, since B - S i and P - S i bonds are typically shorter than Si-Si bonds. Phosphorus, with an optimal bond length only slightly shorter than silicon, can substitute into the Si lattice, but the resulting strain favors formation of Si-P heterodimers, while boron's short bond length leads it to form an ordered reconstruction with a vastly different structure from that of the Si(001) surface. In the case of phosphorus, an understanding of the atomic-level spatial distribution has lead to a detailed understanding of the ways in which phosphorus modifies the reactivity of Si(001): (1) by preferentially reducing the number of surface locations where there are two adjacent Si atoms by formation of P - S i heterodimers; (2) by stabilizing surface hydrogen on P - S i l l heterodimers; and (3) by simple site-blocking, since the P atoms are coordinately saturated and hence unreactive. From these studies, it is clear that the influence of dopants on surface chemical reactivity goes far beyond simple 'site-blocking' mechanisms and involves a complex interplay of lattice strain, bond hybridization, and the energetics of surface reconstructions.

Acknowledgements This work was supported in part by the U.S. Office of Naval Research and the National Science Foundation. Computer time was provided by the

[1] F. Eversteyn and B.H. Put, J. Electrochem. Soc. 120 (1973) 106. [2] R.F.C. Farrow and J.D. Filby, J. Electrochem. Soc. 118 (1971) 149. [3] C.A. Chang, J. Electrochern. Soc. 123 (1976) 1245. [4] B.S. Meyerson and W. Olbricht, J. Electrochem. Soc. 131 (1984) 2361. [5] B.S. Meyerson and M.L. Yu, J. Electrochem. Soc. 131 (1984) 2366. [6] W.-C. Wang, J.P. Denton and G.W. Neudeck, J. Vac. Sci. Technol. B 11 (1993) 117. [7] M. Copel, M. Reuter, E. Kaxiras and R.M. Tromp, Phys. Rev. Lett. 63 (1989) 632. [8] M. Copel, M.C. Reuter, M.H. von Hoegen and R.M. Tromp, Phys. Rev. B 42 (1990) 11682. [9] B.S. Meyerson, F.K. LeGoues, T.N. Nguyen and D.L. Harame, Appl. Phys. Lett. 50 (1987) 113. [10] R.D. Meade and D. Vanderbilt, Phys. Rev. Lett. 63 (1989) 1404. [11] B.E. Weir, L.C. Feldman, D. Monroe, H.-J. Grossmann, R.L. Headrick and T.R. Hart, Appl. Phys. Lett. 65 (1994) 737. [12] B.E. Weir, D.J. Eaglesham, L.C. Feldman, H.S. Luftman and R.L. Headrick, Appl. Surf. Sci. 84 (1995) 413. [13] Y. Wang, X. Chen and R.J. Hamers, Phys. Rev. B 50 (1994) 4534. [14] Y. Wang, M.J. Brouikowski and R.J. Hamers, J. Phys. Chem. 98 (1994) 5966. [15] Y. Wang, R.J. Hamers and E. Kaxiras, Phys. Rev. Lett. 74 (1995) 403. [16] Y. Wang and R.J. Hamers, Appl. Phys. Lett. 66 (1995) 2057. [17] Y. Wang and R.J. Hamers, J. Vac. Sci. Technol. A 13 (1995) 1431. [18] J. Shan, Y. Wang and R.J. Hamers, J. Phys. Chem. 100 (1996) 4961. [19] Y. Wang, M.J. Bronikowski and R.J. Hamers, Surf. Sci. 311 (1994) 64. [20] R.J. Hamers, R.M. Tromp and J.E. Demuth, Phys. Rev. B 34 (1986) 5343. [21] J.J. Boland, Phys. Rev. B 44 (1991) 1383. [22] J.J. Boland, Phys. Rev. Lett. 67 (1991) 1539. [23] L.-Q. Xia, M.E. Jones, N. Maity and J.R. Engstrom, J. Chem. Phys. 103 (1995). [24] N. Maity, L.-Q. Xia and J.R. Engstrom, Appl. Phys. Lett. 66 (1995) 1808. [25] J. Shan and R.J. Hamers, in preparation (1996).