Surface chemistry associated with plasma etching processes

Applied Surface Science 192 (2002) 72–87

Surface chemistry associated with plasma etching processes David B. Graves*, David Humbird Department of Chemical Engineering, University of California at Berkeley, Berkeley, CA 94720, USA

Abstract We present our progress towards an accurate simulation model of plasma etching of silicon. A study of the interactions of energetic argon ions with silicon surfaces using molecular dynamics (MD) simulations is reported. A dynamic balance between ion-induced damage and recrystallization of the surface is detected. By manipulating ion energy, argon ions are able to both create disordered regions near the surface, and recrystallize these disordered regions. Silicon atoms in this amorphous region are readily mixed by argon ions. Limited mixing in the crystalline layer is observed. Fluorine adsorbed on the silicon surface does not mix into the layer with argon ion impact. When an energetic Fþ impacts a silicon surface, the uptake and apparent subsurface mixing of F is much greater than Arþ-induced mixing. However, a closer examination shows that the F impacts have primarily increased the Si surface area by creating crevices and cracks, and that the F remains mainly on the surface of this layer. A similar situation results when SiF3 þ impacts the surface. # 2002 Published by Elsevier Science B.V. Keywords: Surface chemistry; Plasma etching; Silicon; Molecular-dynamics; Ion-bombardment

1. The future of plasma etching Plasma etching is central to many manufacturing processes that employ silicon and other thin film devices for electronic, display or related applications such as microelectromechanical systems (MEMS) devices. The primary focus of the present paper is on silicon device processing, but in many cases this is closely related to other applications as well, in part because the unprecedented success of silicon processing has been adapted for other uses. The basic planar process, first devised in the late 1950s at Fairchild Semiconductor, leading to the development of the integrated circuit by Noyce a year later, has remained in many ways unchanged for 40 years [1]. Chips are generally made up of sets of interconnected solid state transistors, or switches. The solid state transistors, most commonly metal-oxide-semiconductor field * Corresponding author. E-mail address: [email protected] (D.B. Graves).

effect transistors (MOSFETs), consist of source, gate and drain. In manufacturing these devices, a complex set of processes is used, involving lithographic pattern transfer, etching, implantation, diffusion, thin film growth, deposition, polishing, and surface cleaning. Over 300 individual steps are typically used in current leading edge device manufacturing processes [2]. The characteristic dimensions of these features (ca. 2001) are below 100 nm or 0.1 mm. In etching, one must ensure that features are controlled to within this dimensional constraint. For example, the nominal gate length for the Intel Pentium 4 MOSFET is reported to be about 70 nm, and this width must be controlled to within about 5 nm. Intel has projected shrinking this to 20 nm gate lengths by 2007, corresponding to 20 GHz clock speeds. The gate length in 2009 is ˚ . Control of dimensions projected to be 16 nm or 160 A on the feature length scale is one of the most important tasks faced by equipment and process engineers. Although major challenges over the course of the next decade and beyond will be related to lithography

0169-4332/02/$ – see front matter # 2002 Published by Elsevier Science B.V. PII: S 0 1 6 9 - 4 3 3 2 ( 0 2 ) 0 0 0 2 1 - 1

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

and the ability to print features below 50 nm, it is a fact that lithography and etch act together to form the feature. Etch must keep up with, and can even be used to enhance, lithography. An example of the latter is a technique to use the plasma to laterally ‘trim’ the photoresist mask so that a narrower gate electrode can be etched than the printed photoresist feature. By a continuing series of evolutionary improvements in manufacturing technology, the semiconductor industry has managed to reduce the cost per computing and information storage capacities by factors of on the order of thousands since the 1950s [3]. Perhaps not as well known has been the gradual increase in materials complexity, especially in the past 5–10 years. Table 1 lists some of these materials, along with the approximate year of their introduction and/or exploration. This has implications for processing, including plasma etching, and the trend to more complex and additional materials is unlikely to abate. If anything, the expectations are that it will only increase. The motivation for the development of new materials, as well as new device designs, is to keep the semiconductor industry on its historic ‘Moore’s law’ growth pattern. This empirical ‘law’ states that microprocessor and memory device performance doubles about every 18–24 months. Implicit in this growth pattern is that manufacturers found ways to increase the ratio of performance/manufacturing cost. The cost of manufacturing chips has risen steadily, but productivity has grown as fast or faster. For example, the shrink planned for the Intel MOSFET-based microprocessor noted above will ultimately reach limits, which will eventually require new device, packaging or system-level designs. In addition, the capital (and therefore depreciation) costs of manufacturing equipment, today the single largest part of the selling price Table 1 Materials in common use, or being studied for use, in semiconductor manufacturing [4] Year

Material

1970 1980 1990

Al, SiO2, Si, Si3N4, poly-Si PtSi2, CoSi2, TiSi2, MoSi2, TaSi2, W Al2O3, Ta2O5, ZrO2, ZrSixOy, BST, TiN, TiN/Ti, Cu/TiN, Cu BN, SiOxFy, organic polymer, RuO2, Pt, IrO2, Y1, PZT, PLZT

2000

73

of a chip, cannot increase faster than the value of the products. All of these manufacturing, materials, device and economic factors bear on the question of the future of plasma etching in semiconductor manufacturing. It is likely that most of the basic manufacturing technologies, including plasma etching, will continue to be used for several more decades at least. However, the demands made on the technology, from new materials and device architectures to etch equipment design and control, can be expected to continue to increase as they have for the past several decades. There are two parts of the problem of understanding and controlling plasma processes. The first part is control of the species created in the plasma and their spatial uniformity and the stability of the conditions within the equipment. The second is to understand the effect these species have when they impact the surface. It is the second part of the problem that we focus on here. The thesis advanced here is that technological progression will be greatly aided by advancing plasma processing science, especially the ways in which the plasma alters surfaces. This includes understanding the effects of the plasma on materials to be etched, materials that should not etch, and materials within etch tool chambers that should remain impervious to the plasma. Of course, plasma–surface interactions associated with etching is an enormous subject and can only be partially addressed here.

2. Uniqueness of plasmas for surface processing Non-equilibrium, partially ionized, reactive plasmas are extraordinarily powerful surface modification technologies. It might even be argued that this technology is uniquely powerful and fortuitously well suited to large-scale, rapid and relatively inexpensive surface-specific processing. Let us consider the basic features of non-equilibrium plasmas. The term ‘nonequilibrium’ here means that the charged species (electrons and ions) generally have a much higher average kinetic energy than the neutral species. A partially ionized gas is a mixture of neutral and charged species. The plasmas are usually created by application of electric and magnetic fields to a relatively low pressure gas. In most cases, the gas pressure is between a few milliTorr and several Torr. The fields

74

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

can range from dc to microwave frequencies. Under the right conditions the neutral gas can be ‘broken down’ and becomes conductive—mainly due to the creation of a sufficient concentration of free electrons in the gas. By passing currents through the gas, the charged species are heated, thereby sustaining the plasma. The peak neutral gas temperature can vary from near room temperature to over 2000 K, but even when the peak gas temperature is high, the relatively low heat capacity of a gas at low pressure means that the gas rapidly cools near walls. Electron average kinetic energy can approach 10 eV per electron (equivalently, over 100,000 K), and positive ions near the boundary of the plasma can reach energies on the same order. Ions accelerated by the relatively large electric fields in the electrical boundary layer of the plasma (i.e. the ‘sheath’) can reach as high an energy as the applied voltage before impacting the surface. Typical conditions in plasma processing are listed in Table 2. The reason for the big difference in neutral and charged species energies is that charged species are lost to walls faster than they are able to transfer via collisions to neutrals the energy they receive from the applied fields. At higher neutral gas pressures, this is no longer the case, and the neutral gas can become quite hot. Non-equilibrium plasmas are unique surface processing environments due to the combination of high fluxes of reactive radicals at low gas temperatures, and the effects of ion bombardment at surfaces. The synergistic effects of ions and neutrals have been widely noted. Ions impacting surfaces with energies on the order of 10 s to many hundreds of eV per ion profoundly affect surface chemistry. Chemical bonds are on the order of electron volts, and are readily broken when thermal energies are on the same order or higher. Ions impacting surfaces deposit their energy Table 2 Typical conditions in non-equilibrium plasmas Gas pressure Gas temperature Degree of ionization (Nplasma/Nneutral) Average electron energy Average ion energy in plasma Average ion energy impacting surface Typical plasma dimension

1–1000 mTorr 300–2000 K (0.03–0.15 eV per atom) 106 to 102 1–10 eV 0.05–1 eV 10–1000 eV 0.1–1 m

very close to the surface, and over a very short period of time. In addition, the impacts are isolated in time and space, as a quick calculation shows. The area of the surface in the vicinity of the ion impact influenced by ˚ the impact is a circle with radius approximately 25 A for a 200 eV ion (denote this area as Aimpact). This corresponds to about 1017 m2. The approximate time between impacts for a region of the surface influenced by a single impact is 1/(ion flux  Aimpact ). Maximum ion flux is on the order of 10 mA/cm2, or 100 A/m2. This corresponds to about 1021 ions/m2 s. The minimum time between impacts is therefore about 104 s or 100 ms. Molecular dynamics (MD) simulations show that the energy deposited from a single ion impact dissipates as heat to the substrate within 1012 s at most. We conclude that the energy from one impact does not overlap with other impacts. The fact that each of the very intense individual ion impacts is isolated means that the energy deposited has time to dissipate to the entire large volume of the wafer. This keeps the surface temperature low. Most of the effects of the impact therefore occur within the several tens of Angstroms that are associated with the depth profile of energy deposition. MD results shown in Fig. 1 illustrate this point, with ion energy remaining plotted versus depth for Arþ impacting Si at normal incidence and various energies. Note that at 200 eV, most of the ˚ . If ion energy has been deposited within about 20–25 A we assume the energy from the ion dissipates within 1012 s, then the peak power density from a 100 eV ion is about 108 or 1012 W/cm2. This is an extremely high power density. By comparison, the power density associated with photons leaving the sun’s photosphere (outer surface) is estimated to be somewhat less than 108 W/cm2. The average power deposited by plasma ions at an exposed surface is just the ion current times the ion energy. Assuming an upper limit for ion current of 10 mA/cm2, and ion energy of 100 eV, the average power is 1 W/cm2 or 104 W/m2. Power of 1 W/cm2 can easily be removed by backside wafer cooling, keeping the wafer near room temperature. The ion impact can initiate dramatic surface chemistry from the peak power deposited, but the wafer as a whole need not be heated. To the authors’ knowledge, there is no other technology that can deliver such high peak powers right at the surface without vaporizing the surface. Photons of comparable energy, for example, have much higher penetration depths than ions and would not deposit

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

75

Fig. 1. Plot of ion remaining energy vs. depth for Arþ impacting Si at normal incidence. The four curves (top to bottom) are plotted with data obtained from averaging many impacts at 200, 100, 50 and 20 eV, respectively. The symbol (^) denotes the depth of the amorphous– ˚ , leaving the regions below this crystalline transition for each ion energy. For example, 200 eV Arþ creates an amorphous region above 15 A crystalline. If average ion energy is below about 12 eV, the silicon remains crystalline.

their energy so near the surface. The net result is that very intense, highly localized individual ion impacts deposit energy at surfaces exposed to low pressure, non-equilibrium plasmas. The time-averaged effect results from averaging these individual impacts over experimental time scales. The chemical and structural modifications at the very near-surface region that result from the cumulative effects of these isolated individual impacts cannot be achieved, in general, with any other technology. In addition to the unique attributes of plasma– surface interactions just enumerated, the plasma offers important advantages over the use of ion and radical beams under high vacuum conditions. Although it would be possible to replicate the effects of plasma– surface treatment with beams under vacuum, it could not be done at the same rate and cost. Ions generally

must impact the surface without suffering collisions with gas phase species in order to promote etch anisotropy. Thin boundary layers called ‘sheaths’ form at walls, retarding electron motion out of the plasma and accelerating positive ions towards surfaces. The fact that plasmas form sheaths on the order of hundreds of microns means that ions can be accelerated across a nearly collisionless sheath to impact a surface with little angular scattering at pressures of tens to hundreds of milliTorr. A relatively high gas pressure means that the surface experiences a much larger neutral flux than would occur under high vacuum conditions, thereby raising process rates. Also, for etching applications, the rate at which etch products can be removed depends on the pumping speed (proportional to the size of the pump) and the gas number density or pressure. If the gas pressure were

76

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

to be low enough to allow an ion beam to reach a surface without collisions (that is, under high vacuum conditions), the pumping rate of etch products would be very slow unless the pump were enormous. Large pumps greatly increase the capital and operating costs of the process, and slow pumping and slow etching reduce wafer throughput and increase cost-of-ownership of the tool. It is generally possible to find a set of conditions such that the plasma etch process meets the sometimes conflicting demands of the process: rate, uniformity, anisotropy, critical dimension control, low damage and contamination, infrequent cleans, little process drift, etc. It is, however, getting more difficult and challenging to meet the aforementioned increasing needs for future devices and materials. The largely empirical plasma etch process development needs to be augmented by better plasma processing science.

3. The nature of plasma–surface interactions A corollary of the points made above about the effects of the plasma on surfaces is that the plasma modifies the chemical nature and behavior of the regions near the surface influenced by ion impact. An example of this has been demonstrated by Layadi et al. [5]. These authors clearly show that the halogencontaining plasma has strongly altered the chemical and physical characteristics of the top several nanometers of the silicon. Similar arguments apply to plasma deposition of thin films, as well. Any fundamental understanding of plasma–surface chemistry must come to grips with the profound changes induced at surfaces by the plasma. This complicates studies of plasma–surface chemistry since every time the plasma environment changes, the surface changes. It is argued that this fundamental fact renders plasma–surface chemistry unusually difficult, and it has a correspondingly complicating effect on simulations of reactive ion–surface interactions. 3.1. Ion–surface interactions in plasma processing: MD simulations The complexity of events occurring in the vicinity of an ion impact, coupled with the extremely small distance and very short time duration of the event, has made direct experimental observation of ion–surface

interactions very difficult. However, some of the same characteristics, combined with vast increases in computer power, have made it possible to use computer simulation to investigate the phenomenon. Several simulation techniques have been employed, including several variations on Monte Carlo simulations, but the technique described here is MD. One motivation for MD studies of ion–surface interactions is to develop a systematic understanding of plasma–surface interactions. Reactive free radicals created in the plasma are one important part of the chemistry induced by plasmas at surfaces being processed. However, the extraordinary peak power density deposited by ions is at least very important. It happens, fortuitously, that a single ion impact under conditions most commonly relevant to plasma–surface interactions can be treated with a collection of atoms numbering no more than several thousand, on a time scale usually less than 5 ps. In some cases, a few hundred atoms followed for 1 ps or less is sufficient. With semi-empirical interatomic potentials, this means that thousands of impacts can be simulated on inexpensive computers relatively quickly. The combination of increasingly inexpensive computing, the development of reasonably accurate interatomic potentials for some materials of common interest in semiconductor manufacturing, and the fact that plasma processing mainly involves the cumulative effects of individual ion impacts, has made technologically relevant computer simulation studies of reactive ion–surface interaction feasible at modest cost. MD has been used in sputtering simulations for several years, starting with Harrison [6], and later by Garrison and co-workers [7–10]. Stillinger and Weber developed interatomic potentials for the silicon–silicon, fluorine–fluorine, and silicon–fluorine [11–13], which Schoolcraft and Garrison used to perform a first study of F atom etching of silicon [14,15]. Barone and Graves also used these potentials to study higherenergy impacts of fluorine ions on silicon, and removal of products from a fluorinated silicon layer by argon ions [16,17]. The Stillinger–Weber (SW) potentials were later corrected by Weakliem and co-workers, and used in a study of fluorinated silicon surfaces [18,19]. In this article, we present our recent progress toward a realistic simulation of silicon etching in a halogen– argon plasma. In many ways, our simulations were carried out in the spirit of Barone and Graves [16,17]

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

77

and Feil et al. [20], but with a focus on new, large length-scale phenomena that have been observed using faster computers to simulate larger systems of atoms.

4. Description of MD simulations The general theory and methods of MD simulations are discussed in several texts [21–23]. In this section, we will summarize these basic ideas and procedures as they apply to simulating the impact of an ion onto a surface at energies between 5 and 1000 eV. 4.1. Impact simulation as a model for ion bombardment In a typical impact simulation, an energetic species (an ion) is introduced at a random location above a collection of atoms (the surface, also referred to as the ‘cell’, or ‘layer,’), and given some kinetic energy and incident angle. The trajectories of the ion and all atoms in this model system are computed over the short collision that ensues. With accumulation of these impacts, atoms sputter from the surface, the surface composition changes, and morphology may develop. An example of this model surface is illustrated in Fig. 2. 4.2. MD calculations MD calculates the trajectories of interacting particles by solving Newton’s equations of motion. This requires that all the particles be classical, rather than quantum. The repulsive and attractive interactions between atoms are modeled by interatomic potential energy functions. This function is essential to MD calculations. It describes the potential energy surface that results from the proximities of every atom to every other atom. The calculations presented here used the SW [11–13] interatomic potentials to describe interactions between silicon and fluorine atoms, and Molie`re-type potentials [24] for interactions between rare gas atoms and all other atoms. It should be noted that while energetic species are often referred to as ‘ions’, it is actually assumed that these species are neutralized just before impact, and are therefore interacting with the surface as neutrals.

˚ on a side. The Fig. 2. Side view of a MD cell, approximately 30 A blue spheres are silicon and the white spheres are fluorine. The sides and bottom of the cell are defined by lines to guide the eye. The lateral boundaries are periodic and the bottom layer of silicon is fixed in space. The top surface is exposed to individual impacting ions and/or neutral species to simulate exposure to the plasma. The silicon atoms are disordered, having been displaced from their crystalline lattice positions by a series of ion impacts. The F atoms were placed on top dangling bond sites.

Any charge contributions to the interatomic potentials are thus ignored. The system’s potential energy surface is given by summing over all unique atomic interactions in the system. The negative gradient of the potential energy surface with respect to the position of an atom gives the three-dimensional force acting on that atom. Given this force, Newton’s equations of motion may be integrated numerically to compute the atom’s trajectory. Numerical integration is carried out using the velocity form of the Verlet algorithm—a central-difference method that assumes atoms’ accelerations are constant over a small time interval [25]. This interval must be chosen carefully. To assume that an atom has constant acceleration over the time interval is to assume that the potential energy surface itself does not change over this interval. This is of particular issue in impact simulations, where some atoms may move very fast relative to others. A rule of thumb is to choose a time step that is as large as possible, but still conserves the total energy of the system. (A time step that is too large will introduce fluctuations in total energy as

78

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

atoms become unphysically close to one another.) For this reason, and for simplicity, a time step of 1 fs was used throughout the work presented here. 4.3. Boundary conditions The cell acts as a small control volume of material. The top of the cell is open to allow addition and removal of atoms, while the bottom atoms are fixed to prevent bulk motion of the cell in space. The cell becomes a small part of an infinite system when periodic boundaries are imposed in the lateral dimensions, implying that atoms may interact with neighboring atoms or their images, and that if an atom’s trajectory takes it outside the cell through a periodic boundary, it must re-enter from the other side. In this way, the periodic boundaries create a closed system, from which energy cannot escape. Instead, the energy added to the cell with each impact is removed by a scheme proposed by Berendsen et al. [26], which essentially couples the cell to an external heat bath by scaling all the particles’ velocities at every time step. 4.4. Time between impacts A typical collision cascade from an energetic ion subsides in under a picosecond, and another event will not occur in the same space for roughly a millisecond. But while simulation of the picosecond process is tractable with MD, simulation of the millisecond interval is not. Therefore, when accumulating impacts during a simulation, this time is not included. Instead, it is assumed that nothing happens on the surface during this interval except for dissipation of excess kinetic energy and desorption of any rare gas atoms. Because the long intervals between impacts are not simulated, the length of clock-time represented by a simulation is undefined. For this reason, ion quantities are reported in terms of fluence, rather than flux. Fluence is normalized by the monolayer (ML), defined to be a [1 0 0] plane of silicon atoms, about 7  1014 atoms/cm2. 4.5. Impact statistics One is generally interested in simulating the effects of many ions over surfaces of macroscopic

dimensions, so in order to obtain statistically significant results, impact points are selected at random and tens to hundreds of impacts are simulated. In some cases, the surface is allowed to change after each impact; in other cases, it is more appropriate to use the same initial conditions for each trajectory. 4.6. About molecular graphics Impact simulations take on a new dimension when one has the proper tools to visualize the results. We note that the graphics in this paper were created with VMD [27] or Raster3D [28].

5. MD results and discussion 5.1. Arþ impacts on silicon One of the simplest illustrations of the effects of ion bombardment at surfaces is for a rare gas ion such as Arþ, impacting silicon surfaces at normal incidence and a range of energies. After repeated impact, an initially crystalline Si surface will have its top layers rendered amorphous as shown in Fig. 3 for 100 eV Arþ. After about 0.5–0.8 ML fluence, the steady state ˚ is reached for amorphous layer thickness of about 9 A the case of 100 eV ions. The interface between the crystalline and amorphous layers appears to act almost like a kind of phase boundary. Animations made of the impacts show that atoms in the crystalline layer near the boundary will occasionally be disordered. However, several impacts later, this region will be recrystallized. The depth of the amorphous layer appears to be controlled by some dynamic balance. The thickness of the amorphous layer is a function of ion energy: the higher the energy, the thicker the amorphous layer. This is summarized in Fig. 1, showing how the amorphous–crystalline transition depth appears to coincide with ions having about 10–12 eV remaining. These results suggest that a previously amorphous region near the boundary may be recrystallized if the deposited energy is not too high or too low. Fig. 4 illustrates the process. On the left axis is plotted the thickness of the amorphous layer as a function of ion fluence. Snapshots of the side view of the layer are shown at different fluences. At zero fluence, the layer is crystalline. On the right axis is

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

79

Fig. 3. Plot showing approach to steady state of the amorphous silicon layer thickness. Silicon surface under normal incidence 100 eV Arþ bombardment. Initial and final configurations are shown.

plotted the ion energy. From 0 to about 1.5 ML, 200 eV ions impact the surface, and the amorphous layer thickness approaches its steady state thickness of ˚ . From 1.5 to about 4.25 ML, the ion about 15–20 A energy is reduced to 100 eV. The amorphous layer ˚ under 100 eV ion thickness reduces to about 8–10 A bombardment. In other words, formerly amorphous silicon is recrystallized when the ion energy is reduced. This process is continued to 50 eV, then to 20 eV. After 20 eV, the amorphous layer thickness is ˚ . Essentially all but the top reduced to only about 1–2 A surface atoms are recrystallized by gradually lowering ion energy to 20 eV. In a related simulation, we found that by starting with 200 eV ions and an amorphous ˚ , then immediately reducing the layer depth of 15 A ion energy to 20 eV, the layer recrystallized much

more slowly. Both simulations suggest that ions must deposit enough energy at the amorphous–crystalline boundary to allow atoms to rearrange in the epitaxial pattern. In the case of drastic energy reduction, only excursive ions have enough energy to penetrate to the boundary. The plot in Fig. 1 illustrates the key result: the amorphous–crystalline boundary is very near the same location for each case—the depth at which the ion has about 10–12 eV remaining. For 200 eV ions, the impacting ion has 10–12 eV remaining at a depth ˚ . For 100 eV ions it occurs at 9 A ˚ . For 50 eV of 15 A ˚ , and at 20 eV, at about 1–2 A ˚. ions it occurs at 3–4 A Therefore, the simulations show that if a relatively thick amorphous region is impacted with an ion of the proper energy, the thickness of the amorphous layer can be reduced as long as the ion deposits less than

80

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

Fig. 4. Plot showing silicon amorphous layer thickness vs. Arþ fluence as ion energy changes. The approximate amorphous layer thickness ˚ as the ion energy (right axis) is reduced stepwise from 200 to 10 eV. The inset side views of the silicon (left axis) declines from 15 to about 2 A layer illustrates that the ‘recrystallized’ regions are not completely free of defects.

about 10–12 eV in the vicinity of the boundary. The new amorphous–crystalline boundary will occur at the depth corresponding to the ion having 10–12 eV remaining. This is known to have an application in ion beam-assisted silicon epitaxy: by firing an energetic Si into the deposited amorphous layer every so often, its final thickness is reduced, and annealing times are shortened [29,30]. Detailed analysis of this result must await further study, but some preliminary thoughts and observations might guide this study. The amorphous–crystalline boundary location results from competitive, dynamic processes. When an ion impacts a surface, there is a rapid deposition of kinetic energy in the near-surface region, starting at the surface, and decreasing with depth. The atoms in the layer rapidly get heated up,

and become more mobile. The energy is transferred rapidly to adjacent atoms via collisions, and the atoms’ energy decreases, eventually returning to room temperature or something close to it. There is a driving force for atoms in the amorphous region to attain their epitaxial position above the crystalline layer, but the atoms in the amorphous layer are rapidly losing energy, and may become frozen into the amorphous configuration before they have a chance to recrystallize. Only if the atoms above the crystalline layer have enough energy (but not too much) for enough time to find the proper low energy configuration (the crystalline configuration) will recrystallization occur. There may be other constraints as well, since the atoms above the recrystallizing atoms will influence the atoms below them. There are complex, collective

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

processes occurring, and temporal and spatial gradients in energy and atomic configuration probably play important roles. 5.2. Mixing and mass transfer due to ion impact The results presented in Section 5.1 illustrate how rare gas ions can amorphize and recrystallize silicon near-surface regions. It is therefore related to energetic species-assisted epitaxial film growth, but it also has important implications for our understanding of the effects of ions during etching and deposition, and other plasma-assisted processes. Visual inspection of animations of impacts show that Si atoms in the amorphous region are considerably more mobile than atoms in the crystalline layer. The role of near-surface layer atomic

81

mixing by impacting ions has long been thought to be important for etching. We have therefore attempted to understand the mixing and mass transfer processes induced by ions in this relatively simple case. One way to characterize the mixing induced by ions is to compute the mean displacement of the substrate atoms after repeated ion impacts. The displacement from original positions should increase monotonically with fluence. A plot of the mean displacement as a function of position and ion fluence is shown in Fig. 5 for 200 eV Arþ on silicon. Recall that the amorphous– ˚ below the surface for crystalline boundary is 15 A this energy. The mean displacements of atoms in the amorphous region are considerably larger than those in the crystalline layer, but some movement is seen even in the crystalline region. Mixing and therefore

Fig. 5. Mean displacement of atoms under continuous bombardment by argon, 200 eV, as a function of depth. Atoms were labeled by depth before the first impact, and maintained this designation throughout. Atoms that were sputtered before the end of the simulation were not included in the average.

82

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

displacement increases with fluence in both regions, with a clear break seen at the boundary. At higher fluences, the plot shows the mean displacement is highest at the surface of the amorphous layer, but the relative uniformity of the displacement through this layer is evident. The layer seems fairly uniformly mixed. The crystalline region shows much larger gradients in mean displacement with depth. Another way to visualize the mixing is to color or render translucent atoms originating in one region of the layer before a series of impacts, then show the atoms after the impacts. The movements of the highlighted atoms can be clearly discerned. This is done in Fig. 6(a)–(d) for silicon atoms at the top of the

amorphous region, with Arþ impacts at 200 eV. In Fig. 6(a), we see the side view of a layer, already amorphized by 200 eV ion impacts, and with the top Si atoms colored, and the others made translucent to enhance visibility through the layer. In Fig. 6(b), we see the distribution of these atoms throughout the layer after 2 ML ion fluence. An analogous set of images is shown as Fig. 6(c) and (d). In this case, we see a top view of a set of colored atoms arranged as a square columnar section of the top surface in Fig. 6(a). After 2 ML ion fluence, the top view of Fig. 6(d) shows how these atoms dispersed laterally. Atomic displacement, or diffusion, appears to be at least approximately isotropic in the amorphous layer.

Fig. 6. Axial and lateral mixing in the amorphous layer. Argon, 200 eV, normal incidence. (a) The initial top monolayer is shown opaque. ˚ square columnar section of the amorphous layer is shown opaque (b) Atoms from the top monolayer after 2 ML ion fluence. (c) A 10 A (top view). (d) Atoms from the central column in (c) after 2 ML ion fluence.

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

A related question is how much mixing occurs between the amorphous layer and the crystalline layer. Fig. 5 implies that although the crystalline layer atom mobility is lower than that in the amorphous layer, there is some mixing. The images in Fig. 7(a)–(d) are intended to illustrate this comparison. Fig. 7(a) is a side view of the layer with silicon atoms in the amorphous and crystalline layers separately colored. Fig. 7(b) and (c) show the mixing between the layers after 0.25 and 2 ML ion fluence, respectively. A close examination reveals that atoms originally in the crystalline region that move into the amorphous region spread relatively rapidly through the amorphous region, but that atoms originating in the amorphous region that move into the crystalline region have not moved far from the original interface. Fig. 7(d) makes this clearer by rendering the atoms originally in the

83

crystalline layer translucent and those in the amorphous region colored. 5.3. Ion-induced mixing of F into the silicon surface The relatively sharp boundary between the amorphous and crystalline regions, and the pronounced differences in atomic mobility between these regions shows that the material or chemical characteristics of the substrate influences the nature of the ion-induced processes in the near-surface region. Silicon atoms in the amorphous region are in a less stable configuration and are more likely to move around when energy is deposited near them, compared to the atoms in more stable crystalline configurations. However, this chemical specificity is much more pronounced when fluorine is adsorbed at the surface of the silicon layer

Fig. 7. Atom transfer between crystalline and amorphous phases. Argon, 200 eV, normal incidence. (a) Initial cell, showing colors used for amorphous and crystalline phases. (b) After 0.25 ML of impacts. Amorphous phase shown transparent. (c) After 2 ML. (d) After 2 ML, with crystalline phase shown transparent.

84

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

and the effects of ion bombardment are observed. The nominal expectation is that the adsorbed F will be mixed into the sub-surface amorphous region, in more or less the same way that Si atoms are mixed (cf. Fig. 6). However, as seen below, this does not happen to any significant extent. We explore two different approaches attempting to mix F into the sub-surface region of a silicon layer. In the first method, fluorine atoms are allowed to passivate the entire surface of an amorphized silicon surface. That is, dangling bonds at the surface of the silicon layer are completely saturated with fluorine atoms. Next, Arþ at 50 eV impacts the surface at normal incidence. After this impact, any open dangling bonds on the silicon surface are repassivated with fluorine. The total amount of adsorbed F is recorded, and the process is repeated. In the second method, energetic Fþ repeatedly impacts an initially amorphized silicon layer. The fluorine uptake is recorded. The results of these simulations are presented in Fig. 8.

The F-passivated Si surface reaches essentially unity monolayer coverage, and remains at that level, with no fluorine mixed into the substrate. By contrast, energetic F impacts increases F composition of the layer leading to over 5–6 ML of F coverage after a fluence of 14 ML of Fþ. Clearly, energetic F is much more effective in fluorinating the silicon layer than Arþ impacts onto a F-coated surface. This in itself is interesting, but if one looks more closely at the layers impacted with energetic F, one sees that the layer structure has altered considerably. The F has remained mainly at the ‘surface’, but the surface area has increased due to cracks and fissures in the silicon layer. The F has not ‘mixed’ into the sub-surface Si at all, but rather has created more surface adsorption sites by creating deep cracks and fissures in the previously continuous silicon layer. We note that under experimental conditions (implying seconds to minutes of exposure) with fluxes of room temperature F atoms impacting a silicon surface, F will readily diffuse and insert into the silicon layer. F atoms etch

Fig. 8. Fluorine uptake and silicon etch plots for bombardment with Fþ at 50 eV, and for bombardment with Arþ at 50 eV and F passivation. Final configurations for each case are shown. Black atoms are F, gray atoms are Si.

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

silicon spontaneously at room temperature, creating thick SiFx layers while doing so. The simulations reported here are not intended to treat this situation, but rather focus solely on the questions associated with ion-induced processes that occur on picosecond time scales. However, the structures that result from energetic F impact may not be completely different from the thick fluorosilyl layers created by thermal F atoms. In either case, F atoms will saturate silicon bonds, restricting the ability of the silicon to form networking bonds. The interconnections between silicon are reduced, and the surface structure probably becomes more like a series of weakly connected strands of silicon backbones with fluorine terminations restricting cross-linking and the length of the strands. 5.4. Impacts of SiFx þ onto Si We saw in the previous sections that F adsorbed at the silicon surface is not mixed appreciably into the silicon sub-surface either by Arþ or even Fþ

85

bombardment. This is in distinct contrast with the way that silicon itself mixes into the sub-surface region after Arþ bombardment, as seen in Figs. 6 and 7. However, what about the case in which F is already attached to Si when it impacts the surface? We explored this case by impacting the initially pure silicon layer with SiF3 þ at 100 eV and normal incidence. The results of the uptake of Si and F from the impacting ion and the etching of original Si, as a function of ion fluence, is shown in Fig. 9, with the resulting mixed layer shown in the inset. The layer structure, showing a deep crevice lined with F, is similar to the layers resulting after Fþ bombardment. The Si atoms that joined the layer as ions are highlighted and are seen in Fig. 9 to be well mixed through the top region of the layer. The original silicon in the layer moves readily through this top mixed layer, resulting in significant etching. This is an example of how ionized etch products that impact a surface exposed to a plasma will mix into the top surface region, redepositing silicon into the mixed layer, but also etching the underlying silicon.

Fig. 9. Fluorine uptake and silicon etch and deposition plots for bombardment with SiF3 þ at 100 eV. Final atomic configuration is shown in the inset. The small atoms are F, the large atoms are the original Si (dark), and deposited Si (light).

86

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

6. Concluding remarks

Fig. 10. The content of C and F as a function of the 100 eV CF3 þ ion fluence. The narrow curves denote each statistically equivalent data set, while the thickest curve is the average. The dotted curve is ˚ 2) surface, scaled by area, and is not data from a larger (1000 A included in the average [31].

We present as Fig. 10 a similar plot of carbon and fluorine uptake onto a silicon surface from 100 eV normal incidence CF3 þ , taken from the paper of Abrams and Graves [31].1 These authors noted the evident importance of mixing within the SixCyFz layer in promoting silicon from the substrate into the mixed layer, then finally leaving the surface as an etch product. The plot of carbon and fluorine uptake are quite different from the corresponding case for SiF3 þ. The layer comes to a steady state much more quickly, and there is no evidence of the increase in surface area that was seen for the SiF3 þ case. The mixed layer appears to be more continuous and more compact. Carbon appears to promote a more cohesive nearsurface region when fluorine is present than does silicon by itself. 1 It should be noted that the interatomic potentials used in this case were based on the Tersoff–Brenner form rather than the Stillinger–Weber potentials used for the silicon–fluorine system reported here. Although quantitative differences are expected from the use of these different forms for the interatomic potentials, the pronounced qualitative differences seen here are unlikely to be due to the potentials. Abrams and Graves [32] examined the differences between F uptake onto Si from energetic Fþ using the two different forms of potential and found quantitative but not qualitative differences in the results.

The results presented here can be summarized as follows. Bombardment of silicon surfaces with Arþ induces intense mixing and creates an amorphous layer in the top part of the layer. The thickness of this layer depends on the ion energy, ranging from 1– ˚ at 20 eV to 15 A ˚ at 200 eV. It is possible to reverse 2A the amorphization damage by gradually decreasing the ion energy, such that the ion energy remaining at the amorphous–crystalline interface is less than about 12 eV and more than about 0.05–0.1 eV. A previously damaged silicon surface can be recrystallized up to the top surface, although this surface retains some disorder at even the lowest ion energies we tested (20 eV). Although pure silicon mixes readily in the amorphous layer, and considerably less effectively in the crystalline region, if F atoms are adsorbed at the surface of the silicon, it does not mix into the layer upon ion bombardment. Even when energetic Fþ impacts the surface, true sub-surface mixing of F into Si remains unlikely. Rather, the energetic F simply creates cracks and crevices in the silicon, and mostly remains adsorbed on this extended surface. A similar result was obtained when F impacts the silicon surface in the form of SiF3 þ : fluorine shows limited mixing into the sub-surface layer, the surface roughens, and F prefers to stick out from the extended surface. The obvious explanation is that the F experiences repulsive interactions with other species when it is bonded to silicon, and minimizing these interactions is energetically favorable. Only when carbon is present does the layer appear to avoid the roughening observed in simulations with only F and Si. Detailed study of the role of carbon must await future work. Several general observations follow from the results presented here. Carefully controlling rare gas ion energy can reverse ion-induced damage and can be used for recrystallization of previously damaged silicon surfaces, at least in some cases. Although nearsurface mixing induced by ion impact is obviously important in some and probably most systems, it is not universally effective. In particular, we observed that fluorine, once attached to a silicon atom, prefers to be at a surface, minimizing its repulsive interactions with other species. Except when carbon is present to keep the layer together, this tends to cause extensive surface roughening and minimal sub-surface fluorine mixing.

D.B. Graves, D. Humbird / Applied Surface Science 192 (2002) 72–87

The specific chemical interactions of the species involved significantly affect the nature of the resulting near-surface region. Ions deposit energy and rearrange atoms in the near-surface region in all cases. However, the nature of the rearrangement and the resulting structural and composition profiles depend sensitively on the species present on the surface and the ions impacting the surface. The results presented here are obviously a very small subset of the large combination of chemistries and materials used in etching, deposition and other plasma-assisted processes. Clearly, specific interactions will depend on the system and conditions. However, the simulations reported here and elsewhere suggest that it is possible to greatly extend the control of surface and near-surface properties, in part by utilizing atomistic simulations.

References [1] C.R. Barrett, Scientific American, January 1998, pp. 56–61. [2] S. Wolf, R.N. Tauber, Silicon processing for the VLSI era, in: Process Technology, 2nd Edition, Vol. 1, Lattice Press, Sunset Beach, CA, 2000. [3] D.A. Patterson, Scientific American, January 1998, pp. 86– 90. [4] S. Sze, Semiconductor Devices: Physics and Technology, 2nd Edition, Wiley, New York, 2001. [5] N. Layadi, V.M. Donnelly, J.T.C. Lee, J. Appl. Phys. 81 (1997) 6738. [6] D.E. Harrison Jr., Crit. Rev. Solid State Mater. Sci. 14 (Suppl. 1) (1988) S1–S78. [7] B.J. Garrison, N. Winograd, D.M. Deaven, C.T. Rieman, D.Y. Lo, T.A. Tombrello, M.H. Shapiro, Phys. Rev. B 37 (1988) 7197. [8] R. Smith, D.E. Harrison, B.J. Garrison, Phys. Rev. B 40 (1989) 93.

87

[9] R. Maboudian, Z. Postwa, M. El-Maazawi, B.J. Garrison, N. Winograd, Phys. Rev. B 42 (1990) 7311. [10] D.E. Sanders, K. Prasad, J.S. Burnham, B.J. Garrison, Phys. Rev. B 50 (1994) 5358. [11] F.H. Stillinger, T.A. Weber, Phys. Rev. B 31 (1985) 5262. [12] F.H. Stillinger, T.A. Weber, J. Chem. Phys. 88 (1988) 5123. [13] F.H. Stillinger, T.A. Weber, Phys. Rev. Lett. 62 (1989) 2144. [14] T.A. Schoolcraft, B.J. Garrison, J. Vac. Sci. Technol. A 8 (1990) 3496. [15] T.A. Schoolcraft, B.J. Garrison, J. Am. Chem. Soc. 113 (1991) 8221. [16] M.E. Barone, D.B. Graves, J. Appl. Phys. 77 (3) (1995) 1263–1274. [17] M.E. Barone, D.B. Graves, J. Appl. Phys. 78 (11) (1995) 6604–6615. [18] P.C. Weakliem, C.J. Wu, E.A. Carter, Phys. Rev. Lett. 69 (1992) 200. [19] P.C. Weakliem, E.A. Carter, J. Chem. Phys. 98 (1993) 737. [20] H. Feil, J. Dieleman, B.J. Garrison, J. Appl. Phys. 74 (2) (1993) 1303–1309. [21] M.P. Allen, D.J. Tildesley, Computer Simulations of Liquids, Oxford University Press, New York, 1987. [22] J.M. Haile, Molecular Dynamics Simulation, Wiley, New York, 1992. [23] D.C. Rapaport, The Art of Molecular Dynamics Simulation, Cambridge University Press, Cambridge, 1995. [24] I. Torrens, Interatomic Potentials, Academic Press, New York, 1972. [25] W.C. Swope, H.C. Andersen, P.H. Berens, K.R. Wilson, J. Chem. Phys. 76 (1982) 637. [26] H.J.C. Berendsen, J.P.M. Postma, W.F. van Gunsteren, A. DiNola, J.R. Haak, J. Chem. Phys. 81 (1984) 3684. [27] W. Humphrey, A. Dalke, K. Schulten, J. Mol. Graphics 14 (1996) 33. [28] E.A. Merritt, D.J. Bacon, Methods Enzymol. 277 (1997) 505. [29] B. Strickland, C. Roland, Phys. Rev. B 51 (1995) 5061. [30] H. Hensel, H.M. Urbassek, Phys. Rev. B 58 (1998) 2050. [31] C.F. Abrams, D.B. Graves, J. Appl. Phys. 86 (11) (1999) 5938–5948. [32] C.F. Abrams, D.B. Graves, J. Appl. Phys. 88 (6) (2000) 3734– 3738.