Nuclear Instruments and Methods in Physics Research B 180 (2001) 99±104

www.elsevier.nl/locate/nimb

Angular dependence of the sputtering yield from a cylindrical track

E.M. Bringa *, R.E. Johnson
Engineering Physics, University of Virginia, Charlottesville, VA 22903, USA

Abstract

The dependence of the sputtering yield on the incident angle, H, is determined using molecular dynamics (MD)
simulations for a cylindrical track produced by a fast ion. For a `small' spike radius and for the mean energy in the
track, Eexc , smaller than the binding energy, U, a …cos H† 1:7 dependence is found, close to the linear collision cascade
(LCC) result and to some thermal spike models. On the other hand, when Eexc > U , the incident angle dependence is
…cos H† 1 . For a larger spike radius we obtain a …cos H† 1:6 dependence for both high and low energy densities.
Analytic spike models based on di€usive transport are shown not to give satisfactory results. In addition, at low energy
densities we see correlated atom ejection ignored in analytic models. Applying the MD results to the experimental data
for electronic sputtering of solid O2 at large excitation densities suggests that the e€ective spike radius is larger than the
initial Bohr adiabatic radius indicating that energy is rapidly transported from the initially narrow track. Ó 2001
Elsevier Science B.V. All rights reserved.

PACS: 61.80; 71.15.D; 79.20.R

Keywords: Sputtering; Energy transport; Thermal spikes; Oblique incidence

1. Introduction ejected per ion incident), but they also are used to
predict the dependence of the yield on the angle
Ejection of atoms and molecules from a solid of incidence and the angular and energy distri-
by ion bombardment (sputtering) is a process bution of the ejecta. In this paper, we present
which can test our ability to describe non-equi- results from molecular dynamics (MD) simula-
librium, molecular scale processes in solids. Be- tions of the ejection of atoms from an energized
cause of their simplicity, equilibrium models like track in an atomic solid. These are used to test
the thermal spike model, are often used to the predictions from spike models for angular
describe the sputtering of metals and insulators distribution of the ejecta and the incident angle
[1±7,20]. Such models are used to calculate the dependence of the sputtering yield from a track
average yield (the number of atoms or molecules produced by a fast ion.
Earlier we compared spike results to MD sim-
*
Corresponding author. Tel.: 1-804-924-4344; fax: 1-804-
ulations of the dependence of the yield on the en-
924-3104. ergy deposited per unit path length, …dE=dx†, and
E-mail address: [email protected] (E.M. Bringa). the ejected atom energy distribution for normal

0168-583X/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 8 - 5 8 3 X ( 0 1 ) 0 0 4 0 2 - 5
100 E.M. Bringa, R.E. Johnson / Nucl. Instr. and Meth. in Phys. Res. B 180 (2001) 99±104

incidence. We showed that spike models could be 2. MD simulation

used to understand aspects of the MD simulations
of the yield at low excitation densities but failed MD simulations were performed to follow the
dramatically at high excitation densities as melting response of an atomic solid to the excitation of a
and the radial pressure pulse controlled the energy cylindrical region. These simulations were carried
transport. In spite of the complexity of the non- out using a Lennard±Jones (L±J) (6±12) potential
6 12
linear energy transport processes [19,22] we V …r† ˆ 4e‰…r=r† …r=r† Š with a cut-o€ radius
showed that the yield for ®xed spike radius had a rcut ˆ 2:5r to describe an fcc atomic solid. This
simple, nearly linear dependence on …dE=dx† at cut-o€ radius includes up to ®fth nearest neighbors
high excitation densities. Here we show that the for a density n ˆ 1:046=r3 (a total of 78 neigh-
incident angular dependence can di€er in these two bors). More details on our previous L±J calcula-
regimes and it di€ers from spike model predictions tions can be found in [11±13].
in both regimes. Since the equations of motion can be fully
A problem of particular interest to us is the scaled using the L±J parameters the results also
sputtering of weakly bound condensed gas solids. scale with these parameters, r for length and e for
In astrophysics, this process can be a source of the energy. The mass will only change the timepscale 
ambient neutral atoms and molecules around icy through the dimensionless time, t ˆ r m=e.
objects [8]. Whereas low energy ions deposit most Since the binding energy of a fcc solid, U  8e, and
of their energy in knock-on collisions, which di- the mean particle spacing, l ˆ n 1=3 , are propor-
rectly produce sputtering, in condensed-gas solids tional to the L±J parameters, they can replace r
incident fast light ions produce a `track' of ion- and e in the scaling of our results. Earlier we
izations and excitations. The subsequent non-ra- showed that the yield versus energy density de-
diative relaxation processes release energy which posited in the track also scales with U and l for
leads to particle ejection in these weakly bound more realistic potentials [13] and discussed the
solids. The dependence of the yield on the angle of di€erences for amorphous materials.
incidence has been measured both for refractory In order to calculate the energy transport and
materials, in which knock-on sputtering domi- sputtering from a cylindrically excited region of
nates, and for MeV ion sputtering of solid O2 [9] radius rcyl , each non-radiative de-excitation event
and solid CO [10]. was simulated by giving an atom a kinetic energy
In this paper MD simulations have been carried Eexc in a random direction. We call …dE=dx†eff the
out for an atomic, condensed gas solid. Due to total kinetic energy release per unit path length in
scaling, these results are applicable to a broad the cylinder. This is usually presumed to be pro-
range of materials, including molecular solids portional to the true stopping power of the mate-
[12±14]. Here we calculate the incident angle rial but, of course, does not have to be [14].
dependence of the yield and the ejecta angular Therefore, in our fcc solid …dE=dx†eff is equal to
distribution from a cylindrically energized track at Nexc Eexc =d, where d is the interlayer spacing, Nexc is
both high and low excitation densities. As we have the average number of excited particles per layer in
shown earlier, this energized `track' can be repre- the track of radius rcyl . For a large radius or an
sentative of either knock-on or electronic sputter- amorphous material this becomes …dE=dx†eff 
2
ing of a solid. We compare the results to nprcyl Eexc with Eexc the average non-radiative en-
predictions from models for knock-on sputtering ergy release per atom in the track.
and to spike models for sputtering. We will refer to The dependence of the yield on the angle of the
standard spike model results. By this we mean ion beam respect to the surface is studied here. We
models in which only a di€usion equation with a use H as the incidence angle measured respect to
well-behaved heat conductivity is used to obtain the surface normal. The beam is assumed here to
the temperature pro®le, numerically or analyti- be in the xz plane coming from x > 0 towards
cally, and then use this pro®le to calculate the loss x < 0 and the intersection of the track with the
of surface material. surface is centered at x ˆ 0 with the surface at
 E.M. Bringa, R.E. Johnson / Nucl. Instr. and Meth. in Phys. Res. B 180 (2001) 99±104 101

z ˆ 0. A yield Yi is calculated for each simulation.

These are then averaged to obtain the average
yield Y and the `error' bars in Y represent the
standard deviation.
We studied the yield from a [0 0 1] surface.
Yields from other crystallographic orientations
have been shown to have roughly the same de-
pendence on …dE=dx†eff [12]. Of course, the fo-
cused collision component of the ejecta, discussed
earlier [15] will di€er for di€erent crystallo-
graphic surfaces. The beam was directed along
the h1 0 0i direction in the xz plane. Some simu-
lations were run rotating the sample, so the beam
would not be directed along any low index
crystallographic direction, and the yield did not Fig. 1. Sputtering yield as a function of scaled …dE=dx†eff ,
change within the standard error. Simulations x ˆ …dE=dx†eff …n 1=3 =U †, for normal incidence (H ˆ 0o , up tri-
angles) from [12] and for H ˆ 60o (squares). A line with slope
were carried out for rcyl ˆ 2r and rcyl ˆ 5r. The
x1:3 is shown as a guide to the eye.
yield at normal incidence and large …dE=dx†eff
was shown to scale roughly linearly with rcyl
[12,14], and we expect this to remain valid for The yield, for rcyl ˆ 2r, as a function of the in-
angular incidence. cident beam angle H is shown in Fig. 2, normalized
to the yield at H ˆ 0. In the `threshold' regime the
yield varies steeply with H, approximately as
3. Incident angle dependence of the sputtering yield …cos H† 1:7 . On the other hand, at large …dE=dx†eff ,
where the yield is nearly linear with x, it is seen to
In order to explain the experimental results for vary approximately as …cos H† 1 . Earlier we showed
the sputtering of solid oxygen at high deposited that the threshold regime could be roughly de-
energy density, a spike model with a constant ra- scribed using spike models but the large …dE=dx†eff
1
dius rcyl  2r was used in [7]. For this spike radius, regime could not. On the other hand, the …cos H†
the average MD yield, Y, as a function of scaled dependence found here for large …dE=dx†eff and the
stopping power, x ˆ …dE=dx†eff (n 1=3 =U ) is shown nearly linear dependence shown on track radius [14]
in Fig. 1 for normal incidence (H ˆ 0), from [12], indicates that the yield is determined by the amount
and for H ˆ 60. The yields are seen to exhibit the of material in the surface region attaining a critical
same overall dependence on …dE=dx†eff . There is a energy density prior to the radial cooling of the
threshold regime at the lowest values of x and spike. In this sense it resembles an ablation process
1:7
there is a roughly linear regime at the larger values [16,17]. The reason for the steep …cos H† depen-
x, both discussed extensively before [12,13]. A ®t at dence at Eexc ˆ 0:8U is the large sensitivity of the
large x of the form Y / xn gives n ˆ 1:1  0:2 for surface ¯ux to the surface temperature when the
H ˆ 0 (from [12]) and n ˆ 1:3  0:3 for H ˆ 60, thermal energy is low respect to the sublimation
indicating the dependence of the yield on energy, as discussed in [12,18]. The dependence for
…dE=dx†eff does not change signi®cantly with inci- the Eexc ˆ 4U case is then less steep because at high
dent angle. In the following, two cases are con- energy the above sensitivity is lost, leading to a
sidered in detail, one in the `threshold' regime, …cos H† 1 dependence that is related to the larger
Eexc ˆ 0:8U …x  10†, and another in the large x surface of the track.
regime, Eexc ˆ 4U …x  50†. The Eexc < U case is At normal incidence we showed earlier [14] that
equivalent to the spike criterium for low excitation the calculated yield for a molecular solid has
2
density, ‰…dE=dx†eff =…prcyl nU † < 1Š, with the op- similar scaling laws to that for an atomic solid. We
posite being true for Eexc > U . also expect this to be true when the incident angle
102 E.M. Bringa, R.E. Johnson / Nucl. Instr. and Meth. in Phys. Res. B 180 (2001) 99±104

agreement [9]. However, they both exhibit a stee-

per dependence on incident angle than the results
from our MD simulations, although the analytic
model (Eq. (1)) does approach an inverse cosine
dependence at very large incident angles. This
disagreement with the MD simulations is consis-
tent with our earlier work. That is, spike models
do not apply at large …dE=dx†eff and the results
from the MD simulations di€er from the measured
results for condensed gas solids for the typical
assumptions made about …dE=dx†eff and rcyl which
we will discuss below.
We note that the standard linear collision cas-
cade model (LCC) for describing knock-on sput-
tering [24] also predicts a dependence on incident

Fig. 2. Y …H†=Y …0† for rcyl ˆ 2r, Eexc ˆ 0:8U (open circles) and
Eexc ˆ 4U (solid squares). The lines are a guide to the eye:
…cos H† 1:1 (solid), …cos H† 5=3 (dashed). Experimental points
for 2 MeV Heq‡ (charge equilibrated He‡ ) bombardment of
solid O2 from [11] are also included (solid triangles), as well as
the analytical ®t from the spike superposition model [23] (dot-
ted line).

is varied. Therefore, the data for electronic sput-

tering of solid O2 bombarded with charge equili-
brated 2 MeV Heq‡ [9] is also displayed.
Assuming rcyl is equal to the Bohr radius this data
is relevant to the large …dE=dx†eff regime. In com-
paring this data to models care should be taken
above H ˆ 60 as neither spike models nor our MD
model allow for backscattering or surface rough-
ness. In addition, the analytic expression

Y ‰HŠ=Y ‰0Š ˆ ‰…4=p†…1= cos H† tan 1 …1= cos H†Š

…1† Fig. 3. Y …H†=Y …0† for rcyl ˆ 5r, Eexc ˆ 0:8U (open circles),
Eexc ˆ 2U (open diamonds) and Eexc ˆ 4U (solid squares). Ex-
for the incident angular dependence predicted for perimental points for 2 MeV Heq‡ (charge equilibrated He‡ )
bombardment of solid O2 from [11] are also included (solid
high …dE=dx†eff by a spike model [23] is also
triangles). Functions of the form 1/cosn H with n ˆ 1:2, 1.6, 2
shown. The results for electronic sputtering of are included as a guide to the eye. Error bars are only shown for
solid O2 and the analytic expression from the spike Eexc ˆ 0:8U for clarity; they are 10% for the rest of the MD
model were shown earlier to exhibit reasonable data.
 E.M. Bringa, R.E. Johnson / Nucl. Instr. and Meth. in Phys. Res. B 180 (2001) 99±104 103

angle like that in the analytic spike model above [23]. ated by a fast ion in a solid. We calculated the
This is due to the variation of the distance from the dependence of the sputtering yield on the angle of
incident particle track to the surface and the trans- incidence of the ions. The sputtering yield as a
port of energy to the surface in both cases. For small function of the e€ective energy deposition,
incident angles (and Mtarget =Mprojectile 6 3 for LCC) …dE=dx†eff , was shown to have roughly the same
5=3
this dependence is approximated as …cos H† in behavior with …dE=dx†eff for both normal inci-
both models which is also shown in Fig. 3. This dence (H ˆ 0) and for H ˆ 60.
dependence is seen to agree, fortuitously, with our At ®xed energy deposition, the variation of the
MD results for the threshold regime even though the yield with incident angle H is 1= cosn H with
sputtering process is non-linear and the dependence n  1:6 for large spike radius (rcyl ˆ 5r). When
at small angles is extrapolated to large angles. In the spike radius is smaller (rcyl ˆ 2r), n  1 for
fact, in both models, as seen for the analytic spike large …dE=dx†eff and n  1:7 for small …dE=dx†eff .
result in Fig. 3, the dependence on incident angle is When describing electronic sputtering the as-
less steep at large angles. Therefore, care should be sumption of a spike radius equal to the Bohr
taken in using the small angle result, cos 5=3 H. radius does not appear to be consistent with the
Claussen [18] obtained an approximate result for a experimental data. However, if the energy ex-
spike having an initial width. He found that, as the pands rapidly as we suggested [14], the spike ra-
energy density decreases, the dependence of the dius increases giving an angular dependence that
yield on incident angle rapidly becomes steeper than is consistent with the experimental results for
5=3
…cos H† . For the parameters used in the MD electronic sputtering of a number of condensed
simulations of the threshold regime Claussen pre- gas solids.
n
dicts …cos H† with n  3:75 for rcyl ˆ 2r, a de- Even though one would like to use simple
pendence not found here. analytic models to describe the sputtering yield
Recently, we suggested [14] that the high energy from cylindrical tracks those available appear to
density deposited within the Bohr radius rapidly fail. At low …dE=dx†eff , small variations in the
expands until the mean energy per particle is close surface temperature distribution caused by the
to the binding energy of solid O2 . If correct, a change in beam angle and correlated emission
radius of about rcyl ˆ 5r should be used to com- a€ect the yield. The discrepancies between our
pare with the experimental data in Fig. 2. This is results and spike models are principally due to
shown in Fig. 3, for several excitation energies. the di€erences in the surface temperature pro®les
The excitation energy in the atomic solid giving the given by the radial di€usion equation from those
same …dE=dx†eff would be Eexc ˆ 2U [14] for obtained in the MD simulation [11]. Therefore,
rcyl ˆ 5r. We see that the angular dependence in although we showed that spike models can de-
both the linear and threshold regime can be scribe aspects of the sputtering in at low
roughly approximated by a cos 1:6 H dependence, …dE=dx†eff , they, apparently, cannot describe the
in reasonable agreement with experiments on O2 dependence of the yield on incident angle in ei-
and CO [9,10]. The steeper dependence occurs ther the high …dE=dx†eff or low …dE=dx†eff re-
because at large track radii the spike is sustained gimes. As a consequence, the rough agreement of
longer so emission from deeper layers contributes the analytic spike model with the data for elec-
signi®cantly to the yield, as discussed in [21]. In tronic sputtering of solid O2 [9] is probably for-
addition, considerable ejection of low energy par- tuitous.
ticles occurs at all …dE=dx†eff for wide tracks.

4. Summary and conclusions Acknowledgements

In this paper we studied the ejection of atoms The work was supported by the NSF Divisions
from a cylindrical energized region, like that cre- of Astronomy and Chemistry.
104 E.M. Bringa, R.E. Johnson / Nucl. Instr. and Meth. in Phys. Res. B 180 (2001) 99±104

References [12] E.M. Bringa, R.E. Johnson, è. Dutkiewicz, Nucl. Instr.

and Meth. B 152 (1999) 267.
[1] G.H. Vineyard, Radiat. E€. 29 (1976) 245. [13] E.M. Bringa, R.E. Johnson, M. Jakas, Phys. Rev. B 60
[2] R.E. Johnson, R. Evatt, Radiat. E€. 52 (1980) 187. (1999) 15107.
[3] P. Sigmund, C. Claussen, J. Appl. Phys. 52 (1981) 2. [14] E.M. Bringa, R.E. Johnson, Surf. Sci. 451 (2000) 108.
[4] P. Sigmund, M. Szymonski, Appl. Phys. A 33 (1984) [15] E.M. Bringa, Nucl. Instr. and Meth. B 153 (1999) 64.
141. [16] R.E. Johnson, in: T. Baer, C.Y. Ng, I. Powis (Eds.), Large
[5] O. Ellegard, J. Schou, H. Sùrensen, Europhys. Lett. 12 Ions: Their Vaporization, Detection and Structural Anal-
(1990) 459. ysis, Wiley, New York, 1996, p. 49.
[6] O. Ellegaard, J. Schou, B. Stenum, H. Sùrensen, R. Pedrys, [17] L. Zhigilei, B. Garrison, Appl. Phys. Lett. 74 (1999) 1341.
Nucl. Instr. and Meth. B 62 (1992) 447. [18] C. Claussen, Ph.D. thesis, Fysisk Institut, Odense Univer-
[7] R.E. Johnson, M. Pospieszalka, W.L. Brown, Phys. Rev. B sity, Denmark, 1982.
44 (1991) 14. [19] M. Jakas, Radiat. E€. and De€s. in Solids 152 (2000) 157.
[8] R.E. Johnson, Rev. Mod. Phys. 68 (1996) 305. [20] H.M. Urbassek, J. Michl, Nucl. Instr. and Meth. B 122
[9] K.M. Gibbs, W.L. Brown, R.E. Johnson, Phys. Rev. B 38 (1997) 427.
(1988) 11 001. [21] R.E. Johnson et al., Surf. Sci. 179 (1987) 187.
[10] W.L. Brown et al., Nucl. Instr. and Meth. B 1 (1984) 307. [22] M. Jakas, E.M. Bringa, Phys. Rev. B 62 (2000) 824.
[11] E.M. Bringa, R.E. Johnson, Nucl. Instr. and Meth. B 143 [23] R.E. Johnson, J. Phys. C2 50 (1989) 251.
(1998) 513. [24] P. Sigmund, Phys. Rev. 184 (1969) 383.