Heterogeneous nucleation of pits via step pinning during Si(100) homoepitaxy

As a baseline for studies in this paper, we start by describing nanoscale traits of ideal epitaxial Si growth surfaces over the temperature transition (450 ≲ T ≲ 750 °C) from islanding to step-flow growth produced in our homebuilt system. For Si device applications, growth temperatures are typically chosen in this range [2].

Figure 1 shows STM images of growth surfaces as a function of T. We see only generally anticipated epitaxial features that are common to all of our thin films: (1) for ${T}\lesssim \,650\,^\circ {\rm{C}}$ C, 2D adatom island nucleation is routine, figures 1(b)–(d), while (2) for T ≳ 650 °C, we repeatably observe only step-flow growth, figure 1(e), characterized by periodic trains of single monolayer atomic steps. Our results are qualitatively consistent with results described in many previous studies and growth models [16, 38]. The temperature for step-flow growth depends weakly on miscut (terrace width). Generally, adatom islands become sparse with increasing temperature and vanish as surface diffusion lengths become large with respect to the distance between adjacent atomic steps [10, 16].

There are several studies of clean Si(100) homoepitaxy (see review by Voigtländer [16] and references therein) that show the same structural trends that we report in figure 1, and define standard structural nomenclatures (see footnote 4) that we use throughout the paper. We have reproduced clean Si(100) epitaxy results to contrast against defective pitted films in subsequent parts of the paper.

A controlled air leak into the growth chamber during Si deposition induces a new feature, nanosized pits, in the growth surface, figure 2. The pitted morphology was obtained in the homebuilt system in growth cycles interleaved with the clean epitaxy in figure 1.

Zoom In Zoom Out Reset image size

The pits are [110]-oriented four-fold symmetric openings comprised of dense step bunches. The pit edges run approximately parallel to the [110]-oriented dimer rows of the 2 × 1 reconstruction. The smallest and shallowest pits are discernible from regular equilibrium surface vacancy defects when they exceed 2–3 atomic layers depth and roughly 1–2 nm on a side. The deepest pits approach but do not exceed the film thickness, suggesting that the pits are caused by step-pinning at some defect site introduced during film deposition.

Figure 2 shows the increasing pit density versus air pressure $\gt {10}^{-8}$ Torr. The pit density diminishes and vanishes for T ≳ 650 °C, as step-flow growth becomes dominant. The T-dependence of pit density suggests that there is a correlation between 2D island nucleation and pit formation.

By contrast, leaking high purity N₂ into the growth chamber at equivalent pressures does not induce nearly as large a density of pits. Figure 3 compares depositions performed with air versus high purity N₂ leaks. It is likely that some other constituent of air, besides N₂, is responsible for pit formation. By comparing the RGA data in figure 3(c), the only detectable significant differences are partial pressures of O₂, and Ar (mass 32 and 40). At the growth temperatures here, molecular oxygen has a sticking coefficient, ${{S}}_{o}=0.02\mbox{--}0.05$, and well-characterized oxidation reactions with the surface [39–41].

Zoom In Zoom Out Reset image size

We find that oxygen-contaminated flux from Si sublimation sources also induces growth pit formation. Residual oxides in or on a sublimation source provide a detectable flux of oxygen-bearing species, causing both pits and bulk oxygen contamination.

Through a few years of experience, it has become clear that our commercial Si cell frequently (not always) produces pitted thin films. Figures 4(a) and (b) show examples of typical growth surfaces produced by the commercial source⁵,⁶. The pitted morphology occurs only up to around T ∼ 650 °C, giving way to smooth step-flow growth at higher temperatures, figure 4(b). The variation between pitted versus smooth growth morphology suggests that pitting is driven by a secondary uncontrolled variable, which we will demonstrate is most likely residual oxygen contamination.

Zoom In Zoom Out Reset image size

If degassed inadequately, our homebuilt source produces thin films with pitted surfaces too, but the pit density diminishes as a function of cumulative deposition time as the Si source material is conditioned. Figures 4(c) and (d) show growth surfaces produced by the homebuilt source versus source burn-in time. Figure 4(c) shows a typical pitted growth surface produced after roughly 17 h of source conditioning via sublimation (∼1100 °C, sublimation of a few micrometers of Si), while (d) shows a smooth pit-free growth with fourteen more hours of conditioning.

The correlation between the pitting in figure 4(c) and oxygen-bearing contaminants is elucidated by RGA spectra sampling the growth flux. Typical RGA spectra are shown in figure 4(e). After 17 h of degassing, we see Si peaks (mass 28–30) along with a variety of peaks consistent with additional residual oxygen, including small atomic (16) and molecular oxygen peaks (mass 32), elevated water (18) and SiO (masses 44–46). Whereas, after 31 h of source degassing, there is no longer detectable molecular oxygen, SiO, or additional water above the UHV background.

Consistent with oxygen-rich flux gas, both of our sources tend to produce oxygen-rich thin films as the substrate growth temperature is reduced. According to secondary ion mass spectroscopy for growth temperatures T ≲ 300 °C, both the commercial and homebuilt system typically produce epitaxy with residual oxygen concentration on the order of 10¹⁹ cm⁻³.

To get insight into the atomistic origin of the pits, we have performed a T-dependent study of the early stages of film growth that reveals a correlation between 2D adatom island nucleation and pit formation. For both air-induced pits, and pits correlated with contaminated source flux, the pit density diminishes rapidly with growth temperature, figures 2 and 4. Pits vanish at the step-flow transition. In this study, we used the homebuilt source in an inadequately degassed (16 h at 1200 °C) condition, that produced epitaxy similar to figure 4(c). The various pit sizes in figure 4(c) suggests that pits nucleate at various times during the evolution of the film, and we found it necessary to grow at least 2 nm thick Si (∼16 monolayers) in order to see the first easily identifiable few-monolayer pits in a reasonably sized STM image (we chose 500 × 500 nm²).

The results of the growth study are shown in figure 5. At growth T = 250 °C, surface roughness is so dominated by small 3D island stacks that it is hard to identify definitive pits. This characteristic island-upon-island stacking mode is a special consequence of growth dynamics on the 2 × 1 surface, where islands in the topmost layer nucleate at the growth APBs in the layer below [9, 15, 16]. For T = 360 ^◦C, isolated pits with depth 2 atomic layers are clear. It is clear that the pit density diminishes rapidly with increasing T, reaching zero for 610 < T < 660 °C.

Zoom In Zoom Out Reset image size

We have plotted the pit density versus T in figure 5(b), along with the estimated island density. For $T\lt 610\,^\circ {\rm{C}}$, pit and island densities diminish roughly exponentially with increasing T. The pit density diminishes as $\exp (0.35\pm 0.1\,{\rm{eV}}/{{k}}_{{\rm{B}}}{T})$, while the islanding density is $\exp (0.9\pm 0.1\,{\rm{eV}}/{{k}}_{{\rm{B}}}{T})$. We assumed a simple exponential fit since nucleation and roughening processes are typically thermally activated [11–13]. Future work will elucidate the origin of the slopes in the fit. Both the pit and island densities diverge from the exponential trend and vanish abruptly for 610 < T < 660 °C. Islands disappear as step-flow growth becomes dominant. The simultaneous extinction indicates that pit formation is related to 2D islanding, consistent with findings of Xu et al [33].

This section describes the evolution of individual pits, from nucleation, then subsequent pit growth by step pinning and bunching, and the consequent evolution of the surface as pits become dense.

Higher resolution images of nascent pits 2, 6, and >20 atomic layers deep in figures 6(a)–(c) show that pits evolve following a well-known route of step-flow pinning against a point defect [27]. Figures 6(a)–(b) reveal the size of the pit nuclei, as well as the subsequent mode of pit growth. We have plotted atomic layer height contours, ${\rm{\Delta }}z=1.36\,\mathring{\rm A}$, to show the pit cross-section in subsurface layers. The pit lengths and widths range from 0.8 to 1.2 ± 0.2 nm in their deepest resolved layer. The 2 × 1 unit cell measures $0.77\times 0.38$ nm², so it is evident that the pits must start at an atomic-scale defect.

Zoom In Zoom Out Reset image size

The steady-state evolution of a single pit is step-pinning and bunching with an apparent tendency toward pairing to form lower-energy D_B steps, figure 6(c), as the pit sidewalls become steeper due to the crowding of S_B steps on their downhill S_A neighbors [42–45]. D_B steps are favored as the angle of step bunches becomes larger than 2°–4° [42, 43, 46].

In later stages of growth, two trends become apparent, illustrated by figure 6(d). First, clusters of pits form as multiple steps wrap around more than a single pit at once. Second, an intuitive smoothing effect on remaining ledges between the pits becomes apparent [33]. As pits become more dense, the remaining ledges between pits diminish to sizes where growth is dominated by single nucleation events on ledges. As we will explain, this suppresses new pit formation because new pit nucleation sites occur at APBs between adatom islands, which promote oxygen uptake. Steps flow to the pit edges producing relatively step-free regions between pits. Such an effect could be useful for step-flow engineering of atomically flat structures.

We find that our pits are a metastable feature. Annealing a pitted sample smoothes the surface. Figure 7(a) shows a 8 nm thick $T=550\,^\circ {\rm{C}}$ growth surface, dense with few-nanometer deep pits, that after annealing for 5 min at $T=750\,^\circ {\rm{C}}$ becomes smooth and uniformly stepped, figure 7(b). In relaxed homoepitaxial Si, pit formation costs the additional free energy of each step edge bounding each pit, building-in a driving force for surface diffusion mediated smoothing to fill-in the pits in an anneal that desorbs or dissolves the oxide nuclei. UHV annealing an oxidized Si(100) surface at $T\gtrsim 700\,^\circ {\rm{C}}$ for a few minutes will remove nanometer-thick oxide layers [47].

Zoom In Zoom Out Reset image size

By contrast, growth pits form as a thermodynamically spontaneous intrinsic feature of heteroepitaxial growth of both compressive and tensile strained (Si/SiGe or SiGe/Si) thin films under deposition conditions similar to ours [48–51]. For example, compressively strained Si_0.70Ge_0.30 deposition on Si (0.9 Å s⁻¹, T = 550 ^◦C) produces distinct pits that are evident by film thicknesses ∼15 nm [51]. 3D pits provide energetically equivalent strain-energy relief to more widely studied 3D islands (quantum dots) [35]. And in contrast to our annealing results, strain-driven pits actually enlarge during annealing [36, 52].

We have demonstrated that growth pits in Si epitaxial thin films are vastly more probable (1) with increasing oxygen-bearing contamination in the growth environment (figures 2–4), and (2) when 2D adatom islands are present during growth (figure 5). From these observations, we conclude that pits nucleate at oxide defects that occur at surface features unique to the 2D islanding growth mode.

Because the ordered Si(100)-2 × 1 surface is semiconducting, defects such as atomic steps, vacancies, or impurities that induce states closer to E_F are the most probable reaction sites for chemisorption processes [39, 53]. On Si(100)-2 × 1 surfaces with clean step-terrace structure, the oxygen sticking coefficient is $\ll 1$ and oxidation initiates at point defects [39] and step-edges [53]. Beyond obvious additional length of atomic step edge bounding each island, a surface defect distinct to island growth mode is the intrinsic APBs that occur with 50% probability between adjacent Si islands.

An APB occurs at the junction of two surface domains, e.g. islands, whose dimer reconstructions are out-of-phase by 1/2 of the 2 × 1 unit cell perpendicular to their dimer rows [9]. Two islands may converge at either S_A or S_B edges forming A or B-type APBs. B-type APBs rapidly capture Si adatoms and serve as preferential nucleation site for an overlayer island, and they are most often observed to be covered by an overlayer. Hence, in low-T epitaxy, B-type APBs contribute to roughening, and most likely eventual amorphization by capturing adatoms and promoting nucleation. By contrast, A-type APBs do not preferentially capture adatoms and they are frequently observed on surface in the 2D island growth mode, see figure 5 inset at T = 360 °C. In clean Si epitaxy, A-type APBs are entirely a surface defect, and they eventually fill-in during growth with no lasting consequences on the bulk Si structure.

Both in MBE and CVD (disilane), B-type APBs serve as reactive sticking, nucleation, and growth sites [9, 16, 37]. A structural model for B-type APBs, developed from STM images, consists of two S_B edges and a row of split-off dimers, providing relatively high densities of dangling bonds and sites with strained geometry unique from the the surrounding 2 × 1 reconstruction, making the APBs preferential sticking sites [9, 37]. Scanning tunneling spectroscopy measurements show that the S_B steps have enhanced DOS at E_F as compared to S_A steps [39].

Based on our observations about the role of oxygen and 2D Si islands in pit formation, our hypothesis is that pits heterogeneously nucleate at Å-sized oxide step-pinning centers that form preferentially on B-type APBs. Subsequently, passing steps pin against the oxide features. We have not been able to resolve the naked pit nucleus by STM during growth, but it is likely to consist of an Å-sized oxide-complex inserted at a point that stabilizes the local reconstruction, adequately delaying the interlayer bond recoordination required to go from the 2 × 1 reconstruction to a bulk diamond structure. The preferred point for oxygen binding is direct oxygen insertion into the surface dimer bond [54]. Naturally, since Si–O bonding is stronger than Si–Si bonding, we speculate that oxygen insertion in a dimer-bond at an APB adequately delays bulk bonding for a period long enough to allow a step pile-up and pit formation (for our choice of growth conditions).

Our findings are reminiscent of oxide-induced 3D mound formation by step-pinning defects during Si etching by molecular oxygen [40, 41, 53, 55, 56]. Comparison of our results with step-dynamics during oxidation-induced etching processes can be useful since etching is effectively time-inversion of growth. During oxygen-induced etching of both the Si(100)-2 × 1 and (111)-7 × 7 surfaces, the surface roughens by step-pinning at nanosized oxide defects to produce multilayer islands [40, 41, 53, 55, 56] During etching, Si flux is away from the surface and steps are retracting, while during epitaxy steps are advancing, hence step-pinning during etching produces mounds, while during epitaxy it produces pits. Oxygen also drives step pinning, step bunching, and faceting on other surfaces, both with and without growth flux [25].

During Si CVD growth from silanes, it is possible that pits may form by the mechanism described here, although at lower densities since CVD growth surfaces will be relatively protected from oxygen adsorption by partial H passivation under typical growth conditions [37, 57, 58]. The likelihood for pitting will increase with growth T and diminishing growth flux, both of which diminish the H passivation, exposing more oxygen binding sites and increasing the potential for pit formation. STM images of growth surfaces during CVD via silanes are qualitatively similar to MBE surfaces, with islands and APBs, suggesting the possibility for oxide and pit nucleation [37, 58].