A molecular dynamics study of H radical bombardment of CH₃ : Si(1 0 0)— comparison of simulation and experiment

Joseph J Végh and David B Graves

Department of Chemical Engineering, University of California, Berkeley, CA 94720, USA

¹ Currently at Lam Research Corporation, Fremont, CA, USA.

² Author to whom any correspondence should be addressed.

E-mail: graves@berkeley.edu

Received 5 September 2009, in final form 30 September 2009
Published 27 October 2009

Molecular dynamics (MD) simulations have been utilized for many years to examine the details of plasma–surface interactions of numerous feed-gas chemistries and substrate materials. These systems include, but are not limited to, the interactions of plasma species containing fluorine, oxygen, hydrogen, nitrogen, chlorine, carbon, and/or inert species incident on metal, semiconductor or polymer surfaces [1–8]. In general, these simulations are carried out using a semi-empirical potential energy function (PEF) to describe the interactions of all the atoms in the system. This type of function can approximate the salient reactive and non-reactive interactions while minimizing computational cost. Such PEFs are generally fit to equilibrium properties (i.e. not to reaction transition state properties) and as a consequence there is no guarantee that the pertinent activation energy barrier heights for a given system will be captured with a high degree of quantitative accuracy. In many cases, the predominant interactions in the system are mitigated by high-energy ion bombardment of the surface. In such systems, the energy imparted from incoming ions may `wash-out' any small, spurious energy barriers in the PEF, and reasonable quantitative agreement with experiment can be achieved. However, for some systems of interest (such as spontaneous etch of Si by F radicals at room temperature) the accuracy of the PEF is more critical.

For example, the Stillinger–Weber (SW) PEF for Si and F interactions has been shown in some cases to over-predict the barrier for SiF–F bond formation by ~3 eV, which apparently prevents simulation of spontaneous etching of Si by F at 300 K [9–17]. The improved Si–C–F reactive empirical bond order (REBO) PEF of Abrams, Tanaka, Humbird and Graves was able to overcome the limitations of the SW PEF, and simulations of spontaneous etch of Si by F with the REBO PEF yielded product distributions consistent with experiment [17]. Maroudas and co-workers also utilized a REBO-type PEF to study the interactions of H with Si [18–20] and have shown very good agreement with experiment [18, 19] and ab initio calculations [18–20]. These simulations are a few examples where a semi-empirical PEF has been used to capture thermal processes with reasonable accuracy.

This publication will outline another system, for which a REBO-type PEF for Si–C–H interactions was utilized to examine bombardment of CH₃ : Si (1 0 0) by H radicals at 300 K. We will show that the etch products formed and the reaction mechanisms observed match closely with experimentally reported results. We will also outline some apparent limitations of the PEF and how these may affect studies of more complicated systems based on this same technique.

The basic MD techniques used in this study have been outlined in detail elsewhere, and are essentially the same as those employed in our previous studies of Si-based systems with the Si–C–F potential [21, 22]. These simulations utilize the Si–C–H REBO PEF developed by Beardmore and Smith for the examination of C₆₀ bombardment of Si surfaces, with a few subtle differences [23]. We have employed the same parametrization as Beardmore and Smith (as outlined in [23]), with the following exceptions: where the authors utilize Brenner's `Potential I' for C–C interactions, we have utilized `Potential II' for these interactions, as it was previously shown that this parameter set better captured product distributions in the Si–C–F systems we have previously studied [24, 25]. Further, Beardmore and Smith note (table 7, [23]) that values for the tri-cubic spline function F(2,3,2) and its corresponding derivative were chosen differently than outlined in Brenner's HC potential to allow for more accurate interactions between C₆₀ molecules and various surfaces. Since our study does not involve these types of interactions, we have used the original values determined by Brenner [25]. Beardmore and Smith also offer a discussion of the available parametrization sets for the Murty and Atwater Si–H interaction potential. We have chosen to use the form that allows for Si–H–Si bridge bonding [26].

The starting simulation cell used in this study is shown in figure 1. A base of 1440 Si atoms is covered with a monolayer of CH₃ groups on the (1 0 0) surface. Each simulation was carried out as in our previous studies: a bombarding species (in this case, atomic H) was placed above the starting surface at a random x, y coordinate. For thermal species at 300 K, the initial velocity vector was chosen from a half-Maxwellian distribution (–v_z only, directed at the surface). For energetic species (1 eV and 5 eV H), all impacts occur at normal incidence. All of the H trajectories were tracked until the incoming species scattered, stuck to the surface or reacted to form volatile products. At the end of each trajectory, any products were removed and the cell was cooled back a set-point temperature of 300 K by coupling to a Berendsen-style heat bath [27]. The next impact was then begun. This scheme allows for observation of the surface evolution over several thousand impacts with the species of interest.

Cheng et al carried out a set of experiments in which a CH₃ terminated Si(1 0 0) surface was exposed to H radicals obtained by cracking of H₂ with an 1800 K tungsten filament [28]. Their surface was prepared by an initial exposure of the Si surface to CH₃I, which they report led to a 1 : 1 coverage of CH₃ and I. The initial H bombardment preferentially removed I, leaving a surface with 50% CH₃ coverage. They found that subsequent H bombardment resulted in the removal of the surface CH₃ groups, through the formation of C-containing silane groups. Methylsilane, CH₃SiH₃, was the predominant etch product. CH₄ (i.e. the expected product for direct H abstraction of surface methyl) was not formed [28].

Our MD simulations reproduce both the product distribution and the proposed reaction mechanism that were reported experimentally for bombardment with thermal H radicals. As illustrated in figure 2, the MD simulation results indicate that the vast majority of carbon (~79% or 57 of the original 72 C atoms on the surface) is removed from the surface as methylsilane, SiCH₆. This is consistent with experiment. All of the C removal in the simulation occurs through the production of SiCH₆ and other silane groups. In further agreement with experiment, the MD simulations showed no CH₄ production.

The experimental studies led to the following proposed mechanism for etch [28]. H will break and insert into Si–Si bonds in the substrate, forming first SiHCH₃, then SiH₂CH₃. At this point, an additional H radical can break the remaining Si–Si bond to form the volatile product, SiH₃–CH₃. The MD simulation also shows this to be the etch mechanism. Figure 3 shows a series of images of the simulation cell as an H radical approaches a surface SiH₂–CH₃ group, leading to the breaking of the final Si–Si bond and the formation of the SiCH₆ product.

The work of Cheng et al also examined other surface reactions through isotopic labelling [28]. A Si surface covered in CD₃ groups was bombarded with H radicals; the surface was held at 300 K. This led to the production of HD, D₂ and H₂, showing that a rapid H/D exchange occurs on the surface. This exchange/abstraction leads to a removal rate for D that was approximately twice as high as the –CD₃ stoichiometry would predict [28]. In this respect, the MD simulations do not agree well with experiment, as we see no abstraction of H from the methyl groups. Removal of the original surface hydrogen and carbon occurs stoichiometrically throughout the simulation (i.e. 3 H for every C), indicating the complete lack of surface H exchange. Although some H₂ is formed, this happens by recombination of incident H on the underlying silicon; none of the original surface hydrogen is abstracted by or exchanged with incoming H radicals (table 1). However, as the incident energy of the bombarding H is increased, some abstraction is seen. We note that experimental characterization of the energy distribution of the H radicals formed from the filament source utilized by Cheng et al is inherently difficult. As such, Cheng et al have not directly measured the absolute flux or energy of the H atoms bombarding the surface in their experiments [28]. It is likely that their H is more energetic than the bulk of the maxwellian-distributed H (centered at 300 K) utilized in our MD. This could be partially responsible for the discrepancies between our work and the experimental results of Cheng et al.

To further examine the effects of radical energy, additional simulation runs were carried out with the incoming H at 1 and 5 eV (normal incidence, single energy—no maxwellian distribution). At both of these energies, the C removal rate is increased compared with the case with maxwellian H at 300 K. This appears to be the result of deeper penetration of the incoming hydrogen into the substrate material, which leads to the breaking of Si–Si bonds further away from the CH₃ groups on the surface. This, in turn, leads to the creation of single products with multiple C and/or Si atoms, most of which contain 2 or 3 Si atoms, though some products formed during 5 eV H bombardment that contained as many as 28 Si atoms. At both of the elevated energies studied, the incoming H abstracts some of the surface H, forming H₂ products. At 5 eV, the incoming H is also seen to strip some of the surface H off the CH₃ groups, which then leaves the surface as free H radicals. The formation of these products is summarized in table 1.

Even though we see a discrepancy in the thermal chemistry for H abstraction/exchange involving the CH₃ groups, we still capture the overall CH₃ removal mechanism correctly. This etch process is dependent on the spontaneous etch of Si by thermal H, which appears to be represented reasonably well by the PEF. By looking at the continued Si etch after the last CH₃ group has been removed, a value for the steady state Si etch yield can be calculated. The simulations give a value of ~0.066 Si atoms per incident H. Additionally H bombardment was also carried out on an amorphized Si surface. The steady state yield on this surface was found to be ~0.075. Chuang and Coburn carried out experimental etching studies with an H radical source on Si, and arrived at a steady state etch yield of ~0.01 [29]. Other experimental studies have estimated the maximum spontaneous etch yield of Si by 300 K H to be ~0.07 [30, 31]. Our simulated yields fall very close to the upper bound of this range, and within an order of magnitude to the lower bound. Further, simulations of thermal H₂ bombardment show no reaction with the Si surface. This is also consistent with experimental evidence that thermal H₂ does not spontaneously etch Si [30, 31].

MD simulations of thermal H bombardment of a methyl-terminated Si (1 0 0) surface give product distributions and an observed mechanism that match well with experiment [28]. However, other experimentally observed chemistry is not correctly captured by this simulation. Experiments showed that significant exchange between the surface and incoming hydrogen atoms should occur. This is not seen in our MD simulations with H radicals at 300 K. While there is some uncertainty in the nature of the energy distribution of the H radicals used in the experimental results, and we do begin to see abstraction of the surface H during bombardment with energetic (1 and 5 eV) H and H₂, even with 5 eV H bombardment in MD, the observed surface H exchange rate is less than that predicted by experiment. This may suggest that a small, spurious energy barrier exists in either or both of the C–H or H–H interaction potentials. The etch mechanism for CH₃ removal relies on spontaneous H etching of Si, and these interactions appear to be captured qualitatively to semi-quantitatively. H induced Si–Si bond breaking leads to the formation of the predominant CH₃ containing product. The spontaneous etch yield of Si by thermal H is also captured reasonably well by the MD simulations.

The simulations presented here serve as another example that semi-empirical PEFs can be used to capture some aspects of the thermal interactions in chemically reactive systems. However, because the PEFs are parametrized with equilibrium bond energies and structure, there is no guarantee that all of the reaction energetics in the system will be accurately represented. Despite these inherent imperfections, with careful and judicious application, this Si–C–H PEF could still provide important insight into the interactions of plasma species with surfaces of interest, for example, SiC or model low-k materials.

[1] 

Ohta H and Hamaguchi S 2001 J. Chem. Phys. 115 6679 
CrossRef

[2] 

Yamada H and Hamaguchi S 2005 Plasma Phys. Control. Fusion 47 A11 
IOPscience

[3] 

Smirnov V V, Stengach A V, Gaynullin K G, Pavlovsky V A, Rauf S, Stout P J and Ventzek P L G 2005 J. Appl. Phys. 97 093302 
CrossRef

[4] 

Smirnov V V, Stengach A V, Gaynullin K G, Pavlovsky V A, Rauf S, Stout P J and Ventzek P L G 2007 J. Appl. Phys. 101 053307 
CrossRef

[5] 

Jang I K and Sinnott S B 2004 J. Phys. Chem. B 108 18993 
CrossRef

[6] 

Pregler S K and Sinnott S B 2006 Phys. Rev. B 73 224106 
CrossRef

[7] 

Ni B, Lee K H and Sinnott S B 2004 J. Phys.: Condens. Matter 16 7261 
IOPscience

[8] 

Abrams C F and Graves D B 1999 IEEE Trans. Plasma Sci. 27 1426 
CrossRef

[9] 

Stillinger F H and Weber T A 1985 Phys. Rev. B 31 5262 
CrossRef

[10] 

Stillinger F H and Weber T A 1988 J. Chem. Phys. 88 5123 
CrossRef

[11] 

Stillinger F H and Weber T A 1989 Phys. Rev. Lett. 62 2144 
CrossRefPubMed

[12] 

Schoolcraft T A and Garrison B J 1990 J. Vac. Sci. Technol. A 8 3496 
CrossRef

[13] 

Schoolcraft T A and Garrison B J 1991 J. Am. Chem. Soc. 113 8221 
CrossRef

[14] 

Schoolcraft T A, Diehl A M, Steel A B and Garrison B J 1995 J. Vac. Sci. Technol. A 13 1861 
CrossRef

[15] 

Weakliem P C, Wu C J and Carter E A 1992 Phys. Rev. Lett. 69 200 
CrossRefPubMed

[16] 

Weakliem P C, Wu C J and Carter E A 1992 Phys. Rev. Lett. 69 1475 
CrossRef

[17] 

Humbird D and Graves D B 2004 J. Chem. Phys. 120 2405 
CrossRefPubMed

[18] 

Ramalingam S, Maroudas D and Aydil E S 1998 J. Appl. Phys. 84 3895 
CrossRef

[19] 

Sriraman S, Agarwal S, Aydil E S and Maroudas D 2002 Nature 418 62 
CrossRefPubMed

[20] 

Sriraman S, Mahalingam P, Aydil E S and Maroudas D 2003 Surf. Sci. 540 L623 
CrossRef

[21] 

Végh J J, Humbird D and Graves D B 2005 J. Vac. Sci. Technol. A 23 1598 
CrossRef

[22] 

Végh J J, Humbird D and Graves D B 2008 J. Vac. Sci. Technol. A 26 52 
CrossRef

[23] 

Beardmore K and Smith R 1996 Phil. Mag. A 74 1439 
CrossRef

[24] 

Tanaka J, Abrams C F and Graves D B 2000 J. Vac. Sci. Technol. A 18 938 
CrossRef

[25] 

Brenner D W 1990 Phys. Rev. B 42 9458 
CrossRef

[26] 

Murty M V R and Atwater H A 1995 Phys. Rev. B 51 4889 
CrossRef

[27] 

Berendsen H J C, Postma J P M, van Gunsteren W F, DiNola A and Haak J R 1984 J. Chem. Phys. 81 3684 
CrossRef

[28] 

Cheng C C, Lucas S R, Gutleben H, Choyke W J and Yates J T 1992 Surf. Sci. 273 L441 
CrossRef

[29] 

Chuang M C and Coburn J W 1990 J. Vac. Sci. Technol. A 8 1969 
CrossRef

[30] 

Dinger A, Lutterloh C and Kuppers J 2001 J. Chem. Phys. 114 5338 
CrossRef

[31] 

Abrefah J and Olander D R 1989 Surf. Sci. 209 291 
CrossRef