F

CCM

a

ARRAA

KSAMSIS

1

rtfmtposaTPwfthib

h0

Applied Surface Science 407 (2017) 223–235

Contents lists available at ScienceDirect

Applied Surface Science

jou rn al h om ep age: www.elsev ier .com/ locate /apsusc

luence-dependent sputtering yield of micro-architectured materials

hristopher S.R. Matthes, Nasr M. Ghoniem ∗, Gary Z. Li, Taylor S. Matlock, Dan M. Goebel,hris A. Dodson, Richard E. Wirzechanical & Aerospace Engineering Department, University of California, Los Angeles, 420 Westwood Plaza, Los Angeles, CA 90095, United States

r t i c l e i n f o

rticle history:eceived 29 October 2016eceived in revised form 6 February 2017ccepted 16 February 2017vailable online 23 February 2017

eywords:putteringrgon plasmaicro-architecture

a b s t r a c t

We present an experimental examination of the relationship between the surface morphology of Mo andits instantaneous sputtering rate as function of low-energy plasma ion fluence. We quantify the dynamicevolution of nano/micro features of surfaces with built-in architecture, and the corresponding variationin the sputtering yield. Ballistic deposition of sputtered atoms as a result of geometric re-trapping isobserved, and re-growth of surface layers is confirmed. This provides a self-healing mechanism of micro-architectured surfaces during plasma exposure. A variety of material characterization techniques areused to show that the sputtering yield is not a fundamental property, but that it is quantitatively relatedto the initial surface architecture and to its subsequent evolution. The sputtering yield of textured molyb-denum samples exposed to 300 eV Ar plasma is roughly 1/2 of the corresponding value for flat samples,

elf-healingon–surface interactionurface roughness

and increases with ion fluence. Mo samples exhibited a sputtering yield initially as low as 0.22 ± 5%,converging to 0.4 ± 5% at high fluence. The sputtering yield exhibits a transient behavior as function ofthe integrated ion fluence, reaching a steady-state value that is independent of initial surface conditions.A phenomenological model is proposed to explain the observed transient sputtering phenomenon, andto show that the saturation fluence is solely determined by the initial surface roughness.

© 2017 Elsevier B.V. All rights reserved.

. Introduction

The physics of plasma–material interactions (PMI) has a wideange of applications across various technologies, including elec-ric propulsion, plasma processing, microelectronic fabrication, andusion energy devices. It is now appreciated that in PMI, the unifor-

ity of plasma interaction with surfaces is the exception ratherhan the norm. Spatially non-uniform interaction leads to twootentially interesting areas of study. The first one is the possibilityf fabricating surfaces with self-organized nano structures, and theecond intriguing aspect is the potential for the surface to heal itselfs a result of immediate deposition of some of the sputtered atoms.he two areas are complementary, since spatial non-uniformity ofMI can destabilize even flat surfaces, and lead to self-organization,hile an initially structured surface may tend to become more uni-

orm via ballistic atom deposition. The present study focuses onhe second facet of PMI non-uniformity, in an attempt to explore

ow an initially structured surface can be designed to partially heal

tself. Additionally, we wish to quantify the dynamic relationshipetween surface morphology evolution and the sputtering erosion

∗ Corresponding author.E-mail address: ghoniem@ucla.edu (N.M. Ghoniem).

ttp://dx.doi.org/10.1016/j.apsusc.2017.02.140169-4332/© 2017 Elsevier B.V. All rights reserved.

rate as function of ion fluence. Understanding how surface structureand roughness affect the rate of atomic sputtering, and vice-versa,is a key consideration in engendering longevity and resilience tomaterials used in plasma devices, for example in electric spacepropulsion applications.

Utilization of energetic ion beams to produce surface nano-patterns has been the subject of intense interest for a coupleof decades. Experimental and theoretical efforts have focused onunderstanding the influence of such parameters as the ion energy,flux, fluence, angle of incidence, and sample surface temperatureon the formation of specific surface patterns. Makeev et al. [1]and Behrisch and Eckstein [2] have provided in-depth reviews ofsuch studies that have focused on explaining the physics of plasmasputtering. Early insights into the relationship between surfacestructure and the sputtering rate were first presented by Sigmund[3], who formulated the theory of atomic sputtering from a flatsurface. The theory was later expanded to consider the influenceof surface roughening [4]. The basic idea is as follows: a collisioncascade is formed downstream of an obliquely incident ion justbeneath the surface, and the energy from the cascade causes ejec-

tion of surface atoms when the energy exceeds a critical value(binding energy). If the surface is not locally flat (rough), surfacepeaks will tend to eject less atoms (because the cascade is some-what farther away), while valleys will tend to eject more atoms. This

2 urface

f[ii

itdinceahbtohtp

samatsmitambimo

sCttf2plts[tdyetcrorltibedrrb

24 C.S.R. Matthes et al. / Applied S

undamental idea was further advanced by Bradley and Harper (BH)5], who showed that surface instability ensues, and that roughen-ng and patterning of surfaces bombarded with energetic ions mayndeed be feasible.

The theoretical assumptions of these sputtering models havenspired the examination of the influence of surface roughness onhe sputtering rate through experimental efforts. Sputtering yield,efined as the ratio of ejected atoms from the surface to incident

ons, has been recently shown to be dependent on the surfaceano- and micro-structure [6]. Surface atoms that receive suffi-ient energy from the ion collision cascade that exceeds the bindingnergy will be removed, and will either escape or interact with andjacent surface feature. In the latter case, the ejected atom mayave a chance to be deposited on a nearby surface feature, andecome part of the evolving surface itself. Thus, the surface struc-ure is a dynamically evolving system, where atoms can be removedr deposited in a way that depends on the surface evolutionistory. Such geometric re-trapping of atoms offers an oppor-unity to design sputter-resistant surfaces that have self-healingroperties.

The relationship between the surface structure and the rate ofputtering erosion has been experimentally explored. Rosenbergnd Wehner first examined the influence of surface roughness on aaterial’s sputtering yield. They compared the sputtering yield of

smooth nickel rod bombarded by Ar+ ions from 70 to 600 eV tohat of a threaded rod with a pitch of 0.45 mm [7]. The experimenthowed that surface roughness produces an overall decrease in theeasured sputtering yield as a result of increased geometric trapp-

ng of sputtered atoms. This finding inspired future explorationso understand how varying roughness parameters quantitativelyffect the sputtering yield. Huerta et al. [8] have used a view factorodel to further examine the sputter deposition behavior observed

y Rosenberg and Wehner. Their model found significant decreasesn net sputtering yield for surface pitch angles beyond approxi-

ately 45◦, due to forward-biased sputtering that favors depositionf sputterants into surrounding surfaces.

A number of efforts have been made to observe the reducedputtering yield of rough or structured surfaces. Ziegler et al. usedhemical Vapor Deposition (CVD) to cover tungsten surfaces withungsten whiskers, which were up to 80 �m high. They foundhat the whiskers dramatically reduced the sputtering coefficientor various ion energies, using 2–3 keV He+ ions to a dose of.8 × 1023 m−2 [9]. CVD has also been used by Hirooka et al. toroduce textured molybdenum surfaces characterized by dome-

ike and faceted features on the order of 1 �m in diameter. Theextured surfaces showed little damage compared to a polishedample, each under 40 keV He+ ion irradiation dose of 3 × 1022 m−2

10]. Exposure of beryllium surfaces to 1000 eV H+ ions has showno produce high-angle, closely packed nano-cones at irradiationoses near 7.3 × 1021 m−2 [11]. As the cones develop, the sputteringield of the surface was found to be reduced by 30–40%. More recentfforts confirm these findings [12,13], in which beryllium exposedo 100 eV heavy hydrogen plasma at a fluence dose of 3 × 1025 m−2

auses a needle-like morphology to emerge at higher ion fluence,esulting in a measured reduction in the sputtering yield by a factorf ∼2. These investigations provide a basis for the effect of surfaceoughness and texturing on a material’s sputtering yield. However,ittle is known about the evolution behavior of material surfaceshat have nano- and micro-features and the corresponding sputter-ng rate during prolonged plasma exposure. We aim to extend thisody of knowledge by quantifying the relationship between surfacevolution and the sputtering yield of surfaces with purposefully-

esigned surface architecture. The current research broadens ourecent findings [6] that nano- and micro-architectures result ineduced sputtering yield, with a focus on the transient nature ofoth surface structure and sputtering yield.

Science 407 (2017) 223–235

The objective of the present work is to observe the dynamicsof surface nano- and micro-structure as function of ion fluence,and to examine its corresponding influence on the sputtering yieldunder prolonged low-energy Ar ion bombardment. We thus focuson the time (ion fluence) dependence of concurrent surface mor-phology and sputtering yield to reveal the interconnection betweenthem. We employ a CVD process to fabricate refractory metal sur-faces with unique surface architectures, and present observationsof their temporal evolution as function of plasma ion fluence. Thesechanges are the result of sputtering erosion, deposition of ejectedsurface atoms onto adjacent structures, and atom transport by sur-face diffusion. A Quartz Crystal Microbalance (QCM) is used toprovide quantitative measurements of the sputtering rate at dif-ferent ion fluence. In the next section, we present a description ofthe experimental facility and diagnostic tools used to characterizesamples as function of plasma exposure fluence. We give an expla-nation of the CVD process used to fabricate surfaces with uniformmicroarchitecture. Section 3 presents observations of erosion anddeposition of dendritic surfaces that have micropillar architectureunder low fluence irradiation. We also examine the self-healingqualities of the surface, and report on the fluence dependent natureof sputtering yield in view of an erosion-ballistic deposition mech-anism. In Section 4, the behavior of the observed sputtering yieldaccording to the cumulative fluence exposure is mathematicallymodeled in a way that characterizes rough surfaces. Lastly, theresults are discussed and summarized in Section 5.

2. Experiments

2.1. Plasma source and diagnostics

The UCLA Plasma–materials interaction (UCLA Pi) test facilityuses a hollow cathode to generate a plasma that is confined by aseries of solenoidal magnets. The plasma terminates approximately2 m downstream of the cathode on a negatively biased materialsample at a spot size of approximately 1 cm diameter. For the resultreported herein, an argon plasma was used. The target surface canbe differentially biased to provide normally incident ion bombard-ment at a desired ion energy. Tests were performed at a workingpressure of 8 × 10−5 Torr, calibrated for argon. This results in a meanfree path of the ions on the order of meters, which exceeds the scaleof the plasma sheath, and ensures normal incidence of the plasmarelative to the sample surface. Matlock et al. present a detaileddescription of the Pi facility and the plasma operating conditions inreferences [14,15]. The Pi facility is equipped with multiple diag-nostics for measuring the properties of the plasma and sputteredatoms. An extensive description of the Pi facility in-situ and ex-situdiagnostics is provided in reference [16], so only a brief descriptionof the QCM measurement system and techniques used in this studyis given here.

The QCM (setup shown in Fig. 1) provides time-resolved sput-tering rate of the target sample. As incoming sputtered atoms getdeposited on the surface of the sensor, the resonance frequencyof the quartz crystal decreases. In turn, this correlates with thecollected mass from the target sample, and hence the number ofsputtered atoms from the target area. This measurement allowsthe differential sputtering rate at a given location to be measuredover a period of collection time. Polar measurements are obtainedby mounting the QCM to a rotary stage with a moment arm of25.4 ± 0.2 cm. The sputtering distribution is assumed to be axisym-metric because of the cylindrical nature of the incident plasma

column, which allows the differential sputtering rate, measured in[ng cm−2 s−1], to be considered a function of the polar angle only.The measurements in this study concern the overall sputtering rateof the entire surface. Therefore, the orientation of surface features

C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235 225

itTamst

totSQms

2

prpailrscd

mstshsoget

mtaTb

Table 1Material sample descriptions.

Sample no. Architecture Composition Feature size

Sample A Micro-spears Re 10 �mSample B Capped dendrites Mo-coated Re 5 �m

Fig. 1. Sketch of the QCM set-up and coordinate system.

s not relevant in considering the cumulative effect of the sput-ering distribution, and the assumption of axisymmetry is valid.his approximation is further discussed in Section 3.3. To perform

polar scan, the QCM is rotated from ̨ = 90◦ to ̨ = 30◦ at 5◦ incre-ents, pausing for a 1-minute duration at each angle to provide

ufficient collection time, and thus reduce measurement uncer-ainty.

The differential sputtering rate provided by the QCM was usedo obtain a differential sputtering yield, which was then integratedver the hemisphere in front of the target to obtain the total sput-ering yield. This value was averaged over the QCM scan period.ince the time-scale of transient erosion is much longer than theCM scan duration (�erosion � �QCM), the presented QCM measure-ents are shown to have good accuracy in resolving the transient

puttering yield.

.2. Material preparation

Micro-architectured material samples investigated here areroduced by Ultramet Inc. using a CVD process for a wide range ofefractory metals. Fabrication details of micro-architectured sam-les are outlined by Ghoniem et al. [6]. The CVD fabrication methodllows for high purity levels and controlled material parameters,ncluding composition, density, and coating thickness. By control-ing the pressure and temperature conditions of the CVD process,efractory metal dendrites, or nano-rods, may be grown from aubstrate surface. The size parameters of these dendrites can beontrolled to produce features of variable length, diameter, surfaceensity, and vertical taper.

There exist three main categories of micro-architecturedaterials: reticulated, fractal, and dendritic surfaces. Reticulated

tructures are characterized by a network of interconnected fibershat comprise a porous or foam structure. Metallic foams havehown to exhibit favorable thermal management because of theirigh surface area-to-unit volume ratio, <20,000 m2/m3 [17]. Theecond category, fractal geometries, possess a significant amountf folds in their surface structures. These types of structures offerreat potential in their heat transfer and sputter resistance prop-rties, the latter of which will be experimentally demonstrated inhis study.

Dendritic surface structures remain as the last category oficro-architectured geometries. Examples of specific geometries

hat fall within the dendritic classification include micro-spears,s shown in Fig. 2a, and capped nodules, as seen in Fig. 2b.he improved behavior of architectured dendritic surfaces maye attributed to several factors. Thermal stresses are reduced,

Sample C Dense fractal clusters Mo 100–200 �mSample D Sparse fractal clusters Mo 100 �m

resulting from the capability of fine surface features to withstanda greater level of distortion. Secondly, net sputtering erosion fromthe ion bombardment is minimized because of geometric trappingof re-deposited atoms. Further, the higher surface area promotesbetter heat distribution over the sample, preventing localizedoverheating. These properties greatly enhance the longevity andresilience of materials used in space EP devices, which experienceextreme plasma environments during regular operation.

Dendritic rhenium is the most extensively produced sampletype by Ultramet, as it is used widely in aerospace and semicon-ductor applications, however any other refractory metal can bedeposited by CVD. Rhenium deposited onto a substrate results inthe formation of dendrites with hexagonal symmetry in the cross-section. The CVD conditions allow for these to be grown uniformlyor with a taper to form spear-like structures. Rhenium dendritescan then be used as scaffolds and coated with tungsten or molyb-denum to produce layered nano-rods. These nano-rods can vary indiameter from 100-1000 nm, and possess aspect ratios of 5–20.Molybdenum is another material that has been used to createunique samples with fractal geometry, which have been used inthis study to examine the transient yield properties of texturedmaterials.

3. Results

3.1. Evolution of surface morphology

The purpose of the experimental study was to examine thebehavior of nano- and micro- architectured surfaces relative to thatof planar samples. Various material samples were examined, andeach underwent a variety of tests and characterization techniquesto understand the physical mechanisms that determine surfaceevolution in plasma-facing materials. The following sections out-line the results from this experimental study, beginning with lowfluence plasma exposures, leading to later experiments monitor-ing sputter yield. Table 1 catalogues the four samples that weretested for this study with a description of their key characteris-tics, including surface architecture, material composition, and theaverage feature size for their respective architecture (i.e. spears,dendrites, clusters).

Testing began with low fluence (�t) exposures on textured sam-ples with smaller features to observe the early-stage plasma effectson the material surfaces. Two different samples, designated as “A”and “B”, each possessed clearly distinct surface structures and wereexposed in the Pi facility to incremented levels of plasma fluence.Sample A is pyramidal Re, and Sample B is dendritic Re scaffoldingcoated with Mo. Fig. 3 displays SEM images of both of these samplesat 1000× magnification prior to exposure. It can be seen that Sam-ple A (Fig. 3a) exhibits sharp, pyramid-like micro-spear structuresprotruding in different directions. Alternately, Sample B (Fig. 3b)is characterized by cylindrical dendrites topped with smooth, uni-form prismatic flat caps. The differing architectures is the result ofcontrolling the speeds of nucleation and growth during the CVD

process used to manufacture the samples.

Both samples were exposed to equal levels of fluence at vari-ous increments. Ion fluence testing began at a very low level, withan initial exposure of 5 × 1021 m−2. The fluence was incremented

226 C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235

Fig. 2. (a) SEM of micro-spear texturing, 5000×. (b) SEM of capped nodule surface structures, 5000×.

ple A,

fdffflotitolIad

F

Fig. 3. (a) SEM images of Sam

or each successive exposure. Because of the prismatic caps on theendrites of Sample B, there was little noticeable change in the sur-ace structure at low levels of fluence. The caps provided protectionrom ion bombardment and displayed little erosion at these loweruences. Sample A, however, exhibited changes on the surfacesf each of the dendrite facets with the emergence and aggrega-ion of small atom clusters (islands), as the plasma fluence wasncremented. Fig. 4 shows the nucleated structures that formed onhe dendrite surfaces. This observation demonstrates the growthf nucleated atomic clusters as fluence is increased, indicating bal-

istic deposition of sputtered atoms on adjacent dendrite surfaces.t appears that these islands have nucleated from self-depositions sputtered material from adjacent microspears is deposited ontoendrite facets. It will be shown later that EDS analysis on the

ig. 4. Higher magnification SEM images of Sample A at 20,000×. (a) Pre-exposure, (b) ion

1000×. (b) Sample B, 1000×.

sample before and after each exposure indicate the absence of sur-face contamination by impurities.

The development of the dendrite structures was analyzed acrossthe various levels of fluence to quantify the effect of plasma expo-sure on the observed sputter deposition. The SEM images of thedendrite faces were analyzed using ImageJ processing software[18] to count the number of “islands” that were visible on the sur-faces at each fluence level. The results are plotted in Fig. 5, wherethe density of the islands is calculated by dividing the number ofislands by the effective area of the image. The results show that

the density levels out as the total fluence increases after an initialrapid increase in the number or islands. This may be attributedto a rapid island nucleation rate, followed by a slower growthrate.

fluence: �t = 5 ×1021 m−2, (c) 1.5 × 1022 m−2, (d) 6.5 × 1022 m−2, (e) 1.15 × 1023 m−2.

C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235 227

Fig. 5. Density of atomic clusters (islands) on dendrite faces as function of ionfluence.

Ffla

aspwAsm

Fd

ig. 6. Average radius and spacing of surface clusters (islands) as functions of ionuence. Note that the surface is completely regenerated (covered) at a fluence ofpproximately �t = 1023 m−2.

Further quantification is displayed in Fig. 6. ImageJ was usedgain to measure the average radius of the islands, and the totalpacing between islands was also determined. These results arelotted in the same figure to determine the coalescence fluence,

here the average island radius equals the inter-island spacing.t this point, the islands have completely covered the dendriteurfaces. The figure shows that this coalescence occurs at approxi-ately �t = 1 ×1023 m−2.

ig. 8. SEM images of Sample B. (a) Wide angle SEM of a dendritic cluster (flower) at 1000amage due to plasma exposure at 10,000×.

Fig. 7. Possible deposition processes on the surface of a micro-architectured mate-rial. (1) Ballistic deposition from sputtering of adjacent surfaces, (2) Plasma-mediatedredeposition.

3.2. Self-healing mechanism

Identifying the source of deposition is important in understand-ing the behavior of surface patterning that results from plasmaexposure. Fig. 7 illustrates two potential routes to the redistributionof surface atoms that have been impacted by ion bombardment.The first process shown (labeled as “1”) is characterized as “ballis-tic deposition”, and results from line-of-sight trapping of sputteredatoms. In this case, ejected atoms sputtered from one surface fea-ture are deposited onto adjacent features. Surface clusters thenform as a result of diffusion, nucleation, and growth processes. Thisballistic self-healing process (“1”) is especially relevant in the case ofmicro-architectured materials. Alternatively, deposition of ejectedatoms may be classified as “plasma-mediated redeposition,” andschematically labeled “2” in the figure. In this case, sputtered atomsdue to plasma exposure reenter the plasma stream, where theyare ionized and re-deposited back onto the material surface. Thisprocess typically occurs under high plasma flux conditions. Theexposures performed in the Pi facility used a very low ion fluxthat creates conditions insufficient to experience plasma-mediated

redeposition. The analysis of Matlock et al. shows that the ratio ofredeposition flux to sputtering flux is at most 10% for the test condi-tions in this experiment. For details on the re-deposition calculationand analysis see Refs. [14–16].

×, (b) undamaged cluster dendrites at 10,000×, (c) dendrites that have experienced

228 C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235

F first

3

iampbteem

ehiat

F

ig. 9. Sample A, 1000× (a) Surface after low fluence tests, 6.8 × 1023 m−2, (b) After.6 × 1024 m−2.

In consideration of these two possible routes of self-healing, its determined hare that ballistic deposition, which can be gener-lly achieved through material design, is the dominant mode ofaterial regeneration. On the other hand, plasma-mediated rede-

osition can be enhanced through plasma device design, which iseyond the scope of the current investigation. As demonstrated byhe low fluence results of Sample A, micro-architectured materialsxhibit geometric retrapping of material sputtered through plasmaxposure, which in turn achieves mass retention that extends theaterial lifetime.The geometric nature of material architecture has some influ-

nce on its resistance to sputtering erosion; for example Sample B

as a distinct dendritic geometry compared to Sample A. At local-

zed regions on the surface, overgrown dendritic clusters (flowers)re observed, as shown in Fig. 8. While these protruded forma-ions do not represent the general characteristics of the surface,

ig. 10. Average relative area increase of the spear footprints and flattened tips.

higher fluence exposure, �t = 2 ×1024 m−2, (c) 2.5 × 1024 m−2, (d) 3.1 × 1024m−2, (e)

important insight can be garnered from the effects of ion bom-bardment on these unique structures. They can be used as surfacemarkers to observe the dynamics of one or several dendrites asthey are affected by plasma bombardment. Low fluence testing pro-duced little change in the surface of Sample B, as the prismatic capsprotected the dendrites from experiencing significant wear. How-ever the flower formations exhibited some slight changes. Fig. 8shows the effects of low fluence exposure on the dendritic clusters.Fig. 8a shows a wide angle image of a cluster protruding from themore uniform surface, while Fig. 8b displays a magnified view ofcluster dendrites that have remained relatively undamaged afterlow fluence plasma exposure. Fig. 8c shows a grouping of theseprotruded dendrites that have experienced damage after exposureto a fluence of �t = 6.5 × 1022m−2.

In order to observe changes in the larger-scale surface featuresafter the initial low fluence tests, plasma exposure continued tohigher fluence levels for both samples A and B. Fig. 9 shows thechanges in a particular formation found on Sample A over severalexposures, where specific points on the feature have been identifiedwith numbers to allow detailed observations. This location on thesurface of Sample A is a larger cluster of spears (flower feature) thatlikely formed at a low-energy site on the surface during the CVDprocess. Examining this formation reveals the result of the depo-sition observed during the low fluence tests, but on a larger scale,both in size and fluence. The image in Fig. 9a shows the cluster afterthe low fluence tests, but prior to this higher level fluence testing.It can be seen that the overall appearance of the cluster is relativelyunaffected by the initial round of tests, as the dendrites have main-tained a spear-like structure with clear edges. Successive plasma

exposures show the development of the cluster as it erodes andexperiences self-healing effects. From the first test, it can be seenin Fig. 9b that the tips of the spears have begun to erode, result-ing in a flattened top surface. In addition, the base of the dendrites

C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235 229

-heali

cgvtsst

tsft

Fig. 11. Schematic diagram of the self

an be compared to the previous image, Fig. 9a, to see that materialrowth has occurred as the perimeter of the structures enlarge, andarious crevices have filled in with material. This may be attributedo the geometric re-trapping of sputtered material through line-of-ight deposition. The features continued to develop in this way foruccessive exposures, as seen in Fig. 9c–e. EDS measurements showhat no contamination or oxidation is contributing to these results.

The images in Fig. 9 were analyzed using ImageJ to quantifyhe growth of the surface features. The software was used to mea-

ure the base footprint of each of the spears seen in the surfaceeature at each of plasma exposure. In addition, the area of the flat-ened region at the tip of the dendrites was measured to examine

Fig. 12. Sample B feature development at increasing levels of fluence. (a) �t =

ng mechanism observed in Sample A.

the erosion of the spear tips. Fig. 10 displays the average growthof the spear footprints and exposed tips, respectively, relative totheir original size prior to exposure. It can be seen that the relativeincrease of the exposed tips is more rapid than that of the footprints.After the fourth exposure, the footprint areas had grown an aver-age of approximately twice in size, while the area of the exposedrhenium at the tips had grown about three times in size. The errorbars indicate the standard deviation among each of the dendrites,which increases as the dendrites expand and the tips of the spears

erode.

The self-healing mechanism is schematically illustrated inFig. 11, whereby material is lost at the spear tips and sputtered

4.2 × 1023m−2, (b) 9.2 × 1023m−2, (c) 1.4 × 1024m−2, (d) 1.9 × 1024m−2.

2 urface Science 407 (2017) 223–235

aoToTdmtp

rtlssaltsbboigEa

3

mioeos0em

aasdT

30 C.S.R. Matthes et al. / Applied S

toms are deposited onto the sides of the dendrites. As previouslybserved at low fluence, deposited atoms form surface islands.hese islands collect and coalesce at higher fluence to create layersf deposited material on the outer surface of adjacent dendrites.he mass loss at the tips of the spears results in a shortening of theendrite height and a widening of the top surface. The depositedaterial provided by adjacent faces and the sputtered tips causes

he outward growth of dendrite bases. The result is a widened foot-rint as the base increases in size.

The higher fluence tests performed on Sample B producedesults that were similar to those from Sample A. Fig. 12 showshe development of a surface feature on Sample B at incrementedevels of fluence. After the initial low fluence tests, the structurehown in Fig. 12a had little variation from its unexposed state. Thetructure after the initial higher fluence exposure, seen in Fig. 12b,gain demonstrates the erosion of the dendrite tips and outer Moayer, exposing the inner rhenium core. Additionally, comparinghe images shows that the dendrite stems have enlarged and thepaces between them have filled in. This material growth effect maye due in part to the material from the tip of the displayed feature,ut is also likely due to deposited material eroded from the capsf the smaller uniform dendrites, which are clearly seen to eroden the images. The development of the surface feature continues torow outward as the top erodes for successive exposures. Again,DS confirms the absence of impurity contaminants or oxidationffecting these results.

.3. Sputtering yield measurements

Previous work has produced a range of experimentally deter-ined yield values for the sputtering of planar Mo by Ar+ at various

on energies [19–21]. Fig. 13 summarizes the yield measurementsf previous studies, and includes a fitting curve displaying thempirical equation for sputtering yield at normal incidence devel-ped by Yamamura et al. [22,23]. The test conditions of the presenttudy used Ar plasma at 300 eV, which has shown yields between.39 and 0.58 atoms/ion in previous experimental studies. Low ionnergies have produced notably inaccurate yield values for planaraterials under ion bombardment.Now that we explored the self-healing mechanism associ-

ted with sputtering and ballistic deposition of nano- and micro-

rchitectured materials, two additional micro- architectured Moamples, designated as “C” and “D”, were used to measure theependence of the sputtering yield on cumulative ion fluence.he samples were fabricated by Ultramet on 2-inch diameter disk

Fig. 14. SEM micrographs of samples (a) C and (b) D s

Fig. 13. Summary of previous measurements of the dependence of sputtering yieldon incident ion energy for Ar+ on Mo.

substrates. The samples possessed similar textures, characterizedby coarse fractal surface clusters; however they varied in the fea-ture density. The first specimen, Sample C, was distinguished by adenser covering of coarse surface formations that generally rangedfrom 100 to 300 �m in diameter. The second specimen, Sample D,was covered with similar patterns, but had fewer large features anda greater amount of rough, flat space between the features. Themorphological differences are seen in Fig. 14, where the sampleshave not yet been exposed to plasma bombardment.

It is important to note here that micro-architectured surfaceswill exhibit a rather complex variation in the local angle of ionincidence, which in turn affects the resulting angular sputter distri-bution orientation. This study aims to quantify the cumulative effectof the individual ion impacts over the entire surface, and there-fore considers the average orientation of all impact locations. Withthe generally random nature of the surface facets, an approxima-tion is needed to obtain a global view of the surface behavior. Theaverage angle of ion incidence will be taken to be consistent withthe orientation of the overall surface, relative to the direction ofthe plasma stream. Therefore, the individual sputter distributionsresulting from each ion impact will each contribute to the totaldistribution as measured relative to the sample surface.

Because of the near-Gaussian radial variation of the plasma fluxprofile, the edges of the plasma-exposed portion of the samples dis-played a gradient of erosion [14]. Examining the outer boundaries of

urface morphologies prior to plasma exposure.

C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235 231

Fig. 15. Sample C morphological changes at increasing fluence, from (a) to (d). A

Table 2EDS measurements for surface composition (weight percentage) of each samplebefore and after plasma exposure.

Sample no. Elements Weight % Weight %Pre-exposure Post-exposure

Sample A C 6.65% 2.55%O 3.95 0.33Re 89.41 97.12

Sample B C 4.68% 0.52%O 2.91 2.45Mo 91.96 71.83Re 0.45 25.20

Sample C C 0.0% 0.0%O 7.58 1.93Mo 92.42 98.07

Sample D C 0.0% 0.0%

tmtettcatiTeh

dbooftsTppnwtodcsi

one azimuthal plane over the entire sample surface, as discussedearlier in this section. It can be seen that there exists a transientregion where the sputtering yield increases as the surface changes.

O 5.43 1.41Mo 94.57 98.59

he plasma distribution allowed for characterization of the surfaceorphology at various levels of plasma exposure. Fig. 15 shows

he changes in the surface features, starting outside the plasma-xposed region and moving inward. Each location moving towardhe center experienced a greater dose of fluence throughout theesting period, as the flux was most concentrated at the center. Itan be seen that as fluence increases, surface structures erode andre flattened. It can be noted in Fig. 15b that the structures appearo have enlarged when compared to those in Fig. 15a, correspond-ng to the findings of the exposures performed on samples A and B.he fully eroded region is shown in 15d, at a location that is closenough to the center, where nearly all recognizable surface featuresave been flattened.

To ascertain the quality of measured sputtering yield, energy-ispersive X-ray spectroscopy (EDS) was performed on all samples,efore and after ion irradiation. This data allowed for the presencef contaminants or oxidation to be easily detected. The purposef the composition tests was to verify that the changing surfaceeatures came about with the absence of large-scale contamina-ion. This result demonstrates that the observed deposition wasputtered material, rather than oxidation or other such processes.he surface composition data is presented in Table 2, which dis-lays the weight percentage composition of sample surfaces beforelasma exposure and after testing was complete. Surface contami-ants that were generally detected were carbon and oxygen, whichere present on the surface prior to testing as a result of manufac-

uring and material handling. It was necessary to track the presencef these elements to assure the absence of oxidation or carboneposits affecting the development of the surface morphology. It

an be seen that in general, both the carbon and oxygen levels wereignificantly reduced after exposure as compared to prior to test-ng. This indicates that no contamination occurred to the sample

s cumulative fluence increases, surface features are eroded and flattened.

surface throughout plasma exposure tests, and that surfaces wereactually cleaned of these elements at low plasma exposure fluence.

Since Sample B had Mo-coated Re micro spears, the EDSspectrum showed that, as expected, post-exposure compositioncontained a significantly higher presence of Re, as the outer Mocoating was removed. Fig. 16 illustrates the EDS spectrum for Sam-ple B. It can be seen that prior to exposure, oxide and carbidelevels are present, which decreased in amount through subsequentplasma exposure. Additionally, the presence of Re in the spectrumbecomes apparent as the Mo coating is sputtered away and the Rescaffolding is exposed. Sample C was pure Mo, so ideally the surfacecomposition should not change through plasma exposure. The EDSspectrum showed little change in the surface composition otherthan a drop in the oxygen peak as the surface oxygen layer wasremoved at low fluence. This is denoted in Fig. 17.

Sample C was irradiated up to a total cumulative ion flu-ence of �t = 1.61 × 1026 m−2 at an average flux of approximately� = 2.3 × 1021 m−2 s−1 using 300 eV Ar+ plasma. QCM measure-ments were taken periodically throughout the exposure in order todetermine the sputtering yield as a function of fluence. The connec-tion between sputtering yield and fluence is attributed to changesin the surface morphology by the erosion and deposition processes,clearly observed in previous experiments on samples A and B. Itwas expected that at small fluence, ballistic deposition of sput-tered material would be prevalent by geometric re-trapping andsurface diffusion. Fig. 18 shows the QCM data taken for Sample C,prior to any bias error adjustments. “Average Yield” refers to theapproximation obtained by integrating the sputter distribution in

Fig. 16. EDS spectrum for Sample B before and after exposure.

232 C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235

Asy

w��ra

mtcttawQdgtcstmFQ

Ft

so a brief summary is given here for the obtained results. Error due

Fig. 17. EDS spectrum for Sample C before and after exposure.

t a certain fluence level, approximately �t = 1.3 × 1026 m−2, theputtering yield levels off and does not change appreciably. Thisield saturation effect will be explored more fully in Section 4.

The test conditions were replicated on Sample D, whichas also irradiated with 300 eV Ar+ plasma at approximately

= 2.3 × 1021 m−2 s−1 to a total cumulative ion fluence oft = 1.72 × 1026 m−2. The results again showed a transient yield

egime, leading up to a fluence around �t = 1.4 × 1026 m−2, wheregain the yield no longer increases.

The QCM measurements were compared to mass scale measure-ents of weight loss over the course of testing. It was determined

hat the mass scale results indicated a greater weight loss than thatalculated from the QCM data. This difference may be attributedo calibration errors and assumptions made about the experimen-al setup, which affects calculations of the mass loss. Specifically,symmetry in the sputtering distribution can cause inaccuracieshen the results are integrated over the hemisphere. Since theCM probe only sweeps over one plane, an anisotropic sputteringistribution would result in an underestimated mass loss if there isreater sputtering in other directions than the measured plane. Thisype of bias error attributed to the experimental setup would beonsistent across all measurements, and the transient trend of theputtering yield at lower fluence would be unaffected. Assuming

hat the true mass loss value to be reflected by the scale measure-

ent, the magnitude of the bias may be estimated for each test.or Sample D, the mass scale showed a loss of 2060.9 mg, while theCM estimated 2360.7 mg, leaving a bias of about 13%. The mass

ig. 18. QCM sputtering yield measurements for micro-architectured Sample C prioro bias error correction.

Fig. 19. QCM sputtering yield measurements for micro-architectured Sample Dprior to bias error correction.

scale had an accuracy of 1 mg, which accounts for a 0.3% errorin the discrepancy between the two methods. For Sample C, thebias error was estimated to be on the order of 8%. Accounting forthis discrepancy allows for a closer comparison of the transientregion, where any differences might be attributed to the surfacearchitecture alone Fig. (19).

Applying these corrections and plotting the data for both sam-ples together in Fig. 20, it can be seen that there are differencesbetween the samples in the yield magnitude over the transientregime. The steady-state yield notably appears at approximatelythe same cumulative fluence, but the values leading up to thatfluence are clearly dependent on the surface architecture differ-ences between the two samples. This difference can most likely beattributed to the morphological variations, where the less densearrangement of surface features on Sample D results in a loweramount of ballistic deposition and a higher rate of lost materialthan Sample C. Previous tests with the same experimental setupperformed on planar molybdenum samples exposed to 300 eV Ar+

plasma have produced yields of approximately 0.4, which cor-responds to the steady-state bound observed in these tests onmicro-architecture Mo surfaces.

A full error analysis for the QCM data can be found in Ref. [16],

to noise was accounted for by calculating the standard deviation ofthe QCM rate measurements across the 1-minute collection timeat each individual angle. In general, deviation in the measurements

Fig. 20. Sputtering yield as function of ion fluence, measured by the QCM for samplesC and D after correction of the bias error.

C.S.R. Matthes et al. / Applied Surface Science 407 (2017) 223–235 233

F encey

wtysntlsbodmiisctffis

ihuclottitt

4r

ieClKrIis

ig. 21. Relationship between surface roughness and sputtering yield in the low fluield as function of ion fluence.

as found to be under 3%. Additional uncertainty can be attributedo assumptions regarding the angular distribution of the sputteringield across the target. Since setup restrictions prevented mea-urements from being taken at angles below 30◦, estimations wereecessary at low angles to obtain an integrated yield number acrosshe plane. Yamamura developed an empirical formula for the angu-ar distribution of sputtering yield that indicates a reduced rate ofputtering at lower polar angles [24,25]. Therefore, there exist twoounds for the yield distribution. The first bound is assuming a yieldf 0 normal to the target, and the second assumes that there is noecrease in the yield at lower angles by setting the yield at nor-al equal to that measured at 30◦. The data presented in the plots

s an averaged value of the upper and lower bounds. This resultsn an uncertainty of approximately 5%. The angular distribution ofputtered material on micro-architectured samples is more fullyonsidered in Li et al. [16]. Here it is shown the differential sput-ering yield of a micro-architectured surface deviates substantiallyrom a flat surface at the beginning of plasma exposure. The pro-le then eventually tends to the traditional butterfly profile of a flaturface (Yamamura’s fit) after long plasma exposures (17 hours).

Because of the complexity of micro-architectured materials, its important to note the critical influence the local sputtering yieldas on the overall surface evolution. The orientation of the individ-al surface points relative to the incoming ions may result in certainrystalline facets exhibiting preferred sputtering, which in turneads to non-uniform erosion demonstrated by the developmentf plateaus observed in our SEM images. These changes diminishhe geometric retrapping capabilities of the surface, and result inhe transient sputtering yield that is observed. Rather than provid-ng a full physical examination of these processes, this study aimso offer a description of the resulting sputtering yield behavior ofhe total surface.

. A phenomenological model of transient sputtering fromough surfaces

In the low fluence regime, it can be seen that the average sputter-ng yield increased in a linear fashion. In the early stages of surfacevolution, up to a fluence of approximately 5 × 1025 m−2, Sample

was characterized between short exposures to track its morpho-ogical changes. Profilometry was performed on the sample using aeyence VHX-1000 digital microscope to measure the RMS surface

oughness, ε, of the sample at each incremented level of fluence.n addition, the QCM provided corresponding yield measurementsn the fluence range that was examined. Digital profilometry waselected due to its ability to produce 3-dimensional height data of

regime. (a) Evolution of the RMS surface roughness for low fluence. (b) Sputtering

the surface as a non-contact method. Scanning probe microscopyproved to be ineffective for high roughnesses, such as those char-acteristic of micro-architectured surfaces. Fig. 21 shows the trendsof the surface roughness and sputtering yield across the examinedrange of fluence. It can be seen that as the micro-architectured sur-face was eroded due to plasma exposure, the roughness decreasedand the yield increased in an approximate linear fashion. At higherfluence, the yield growth is no longer linear and the yield valuesaturates. This result in the low fluence regime demonstrates themorphological dependence of the sputtering yield, where the yieldis clearly a function of surface roughness.

The experimental results suggest that the rates of change of bothroughness and sputtering yield are dependent on the instantaneousion fluence. With this observation, we formulate a phenomenolo-gical model of surface and sputtering yield evolution. The purposeof this model is to provide further description of the experimen-tal results, which demonstrate a relationship between the surfaceroughness and the measured sputtering yield. Let the roughnessbe denoted as ε, the sputtering yield as S, the ion fluence as � = �t,where � is the ion flux. Furthermore, we take the simplest relation-ship between variations in S, ε and � to be linear. Thus,

dε = �εd� = dS

K. (1)

Here � and K are two phenomenological constants that characterizethe rate of change with fluence (�), and the influence of surfacemorphology on sputtering (K). The relationships expressed in Eq.(1) lead to a differential equation that may be solved to express theroughness as a function of fluence.

dε

d�= −�ε ⇒ ε = ε0e−�� (2)

The assumption that the change in the sputtering yield dS is propor-tional to the change in roughness leads to an expression mediatedby a constant K, which can be differentiated with respect to thefluence and solved to produce an equation for the sputtering yield.

dS = Kdε (3)

dS

d�= K

dε

d�= K

(−�ε0e−��

)(4)

∫ S

S0

dS = −�Kε0

∫ �

0

e−��d� (5)

( −��)

S = S0 + Kε0 1 − e (6)

Therefore, the sputtering yield as a function of plasma fluence fora particular micro-architectured surface may be modeled as a sat-uration curve.

234 C.S.R. Matthes et al. / Applied Surface

ityct

S

HtFflsw

edsarosm

5

ppaeshi

tioapflhspfe

Fig. 22. Yield data for Sample C and model curve fit.

This model may be used to fit a transient yield curve to the exper-mentally observed data in this study. The saturation curve is fit tohe data by defining K = S∞ − S0, where S0 is the initial sputteringield of the textured surface, and the S∞ is the value at which theurve converges, which corresponds to the yield of a flat surface ofhe material.

(�t) = S0 + (S∞ − S0)(1 − e−�t/�0 ) (7)

ere �0 is an erosion time constant (�0 = 1/�) that uniquely quan-ifies a particular surface morphology. This model is displayed inig. 22 with the plotted yield data points for Sample C at variousuence levels. It can be seen that this curve, calculated using theimulated roughness at the corresponding fluence, can be fit quiteell to the experimental data.

The fitting constant �0 = 1/� is a characteristic relaxation param-ter that is unique to each respective surface texture, and willepend on the scale of its initial roughness and type of surfacetructures. Therefore, tracking the roughness evolution can providen important characterization tool for micro-architectured mate-ials. The quantification of surface evolution for material surfacesffers an avenue for future research that can provide a better under-tanding of how surface architecturing improves the behavior of aaterial under plasma bombardment.

. Conclusions

This study has achieved a number of notable observations thatrovide insight into the various phenomena experienced duringlasma–surface interaction. The fabrication of nano- and micro-rchitectured surfaces through a CVD process has allowed for anxploration of the transient effects of surface morphology on theputtering rate under normal ion incidence. These material surfacesave been shown to have self-healing properties under plasma

rradiation.The process of ballistic deposition has been observed through

he tracking of spear-like and dendritic surface features acrossncremented levels of ion fluence. Sample A showed the appearancef deposition “islands,” which grew in size and quantity throughdditional plasma exposure. The absence of plasma-mediated rede-osition [14,15] indicated that ballistic deposition at the plasmauxes used in these experiments is the dominant mode of self-ealing. This phenomenon was discerned at higher fluence on both

amples A and B, in which larger surface features were shown toossess qualities of both erosion and deposition. The tips of theseeatures were shown to progressively erode with increasing lev-ls of plasma fluence, and the filling-in of crevices and widening

[

Science 407 (2017) 223–235

of dendrite bases supported the hypothesis that the ejected atomsdue to sputtering events were deposited onto adjacent surfaces.

The sputtering yield of materials has been demonstrated tonot simply be a fundamental material property, but determinedsignificantly by surface geometry. While the reduced sputteringyields of textured surfaces has been observed previously [6,7,9],this study established the transient nature of the sputtering yield,and concretely related this behavior to surface evolution. As plasmaexposure affects the surface morphology over time, the sputter-ing yield will be altered according to changes in surface geometry.This transience can be quantified according to the plasma fluence,and has been shown to have different relaxation rate depend-ing on the initial surface structure. Differences in the texturingof Samples C and D showed that the rate of sputtering differedaccording to the density of surface features for the fractal-typetextures used here. Therefore, the rate of yield reduction through-out the transient sputtering regime is determined largely by theinitial morphology of the material surface. A phenomenologicalmodel was presented to relate the roughness evolution of a sur-face to the total fluence exposure. It was asserted that each roughsurface texture can be characterized by a relaxation parameter,�0, which determines the saturation fluence of the correspondingsputtering yield. Future research can provide empirical correla-tions for the value of �0 associated with various surface texturesand roughnesses. Further exploration of other surface geometriesmay produce higher yield reduction, and provide additional insightinto the physical processes present during plasma exposure andself-healing phenomena.

Erosion due to plasma sputtering has been shown for the firsttime to be a dynamic process, in which surface evolution is tightlycoupled with the sputtering rate. A quantitative measurementof the transient nature of a material’s sputtering yield has beenachieved, and has been demonstrated to be related to the evolutionof surface morphology.

Acknowledgements

This material is based upon work supported by the US AirForce Office of Scientific Research (AFOSR), under award num-bers FA9550-11-1-0282, FA9550-16-1-0444, FA9550-14-10317,and FA2386-13-1-3018. We would like to acknowledge Mr. BrianWillaims of Ultramet for providing textured refractory metal sam-ples.

References

[1] M.A. Makeev, R. Cuerno, A.-L. Barabasi, Morphology of ion-sputtered surfaces,Nucl. Instrum. Methods Phys. Res. Sect. B: Beam Interact. Mater. Atoms 197(3) (2002) 185–227.

[2] R. Behrisch, W. Eckstein, Sputtering by Particle Bombardment: Experimentsand Computer Calculations From Threshold to MeV Energies, vol. 110,Springer Science & Business Media, 2007.

[3] P. Sigmund, Theory of sputtering. I. Sputtering yield of amorphous andpolycrystalline targets, Phys. Rev. 184 (2) (1969) 383.

[4] P. Sigmund, A mechanism of surface micro-roughening by ion bombardment,J. Mater. Sci. 8 (11) (1973) 1545–1553.

[5] R. Mark Bradley, J.M.E. Harper, Theory of ripple topography induced by ionbombardment, J. Vac. Sci. Technol. A: Vac. Surf. Films 6 (4) (1988) 2390–2395.

[6] N.M. Ghoniem, A. Sehirlioglu, A.L. Neff, J.-P. Allain, B. Williams, R.Sharghi-Moshtaghin, Sputtering of molybdenum and tungsten nano rods andnodules irradiated with 150 eV argon ions, Appl. Surf. Sci. 331 (2015) 299–308.

[7] D. Rosenberg, G.K. Wehner, Sputtering yields for low energy He+, Kr+, and Xe+

ion bombardment, J. Appl. Phys. 33 (5) (1962) 1842–1845.[8] C.E. Huerta, T.S. Matlock, R.E. Wirz, View factor modeling of

sputter-deposition on micron-scale-architectured surfaces exposed toplasma, J. Appl. Phys. 119 (2016) 113303.

[9] J.F. Ziegler, J.J. Cuomo, J. Roth, Reduction of ion sputtering yield by specialsurface microtopography, Appl. Phys. Lett. 30 (6) (1977) 268–271.

10] Y. Hirooka, T. Imoto, T. Sano, Helium ion irradiation on chemically depositedmolybdenum and molybdenum carbide, J. Nucl. Mater. 113 (2-3) (1983)202–206.

urface

[

[

[

[

[

[

[

[

[

[

[

[

[

[24] Y. Yamamura, N. Matsunami, N. Itoh, Theoretical studies on an empiricalformula for sputtering yield at normal incidence, Radiat. Effects 71 (1-2)

C.S.R. Matthes et al. / Applied S

11] D.M. Mattox, D.J. Sharp, Influence of surface morphology on the low energyhydrogen ion erosion yields of beryllium, J. Nucl. Mater. 80 (1) (1979)115–119.

12] D. Nishijima, R.P. Doerner, M.J. Baldwin, G. De, Temmerman, Erosion yields ofdeposited beryllium layers, J. Nucl. Mater. 390 (2009) 132–135.

13] R.P. Doerner, C. Björkas, D. Nishijima, T. Schwarz-Selinger, Erosion ofberyllium under high-flux plasma impact, J. Nucl. Mater. 438 (2013)S272–S275.

14] T.S. Matlock, D.M. Goebel, R. Conversano, R.E. Wirz, A dc plasma source forplasma–material interaction experiments, Plasma Sources Sci. Technol. 23 (2)(2014) 025014.

15] G.Z. Li, T.S. Matlock, D.M. Goebel, C.A. Dodson, R.E. Wirz, Implementation ofin situ diagnostics for sputter yield measurements in a focused plasma, in:52nd AIAA/ASME/SAE/ASEE Joint Propulsion Conference, 2016, July.

16] G.Z. Li, T.S. Matlock, D.M. Goebel, R.E. Wirz, C.S. Matthes, N.M. Ghoniem, In situsputtering yield measurements of molybdenum surfaces bombarded by argonplasma, Plasma Sources Sci. Technol. (2017), Submitted for publication.

17] S. Sharafat, N.M. Ghoniem, M. Anderson, B. Williams, J. Blanchard, L. Snead,Micro-engineered first wall tungsten armor for high average power laserfusion energy systems, J. Nucl. Mater. 347 (3) (2005) 217–243.

18] M.D. Abramoff, P.J. Magalhaes, S.J. Ram, Image processing with ImageJ,Biophoton. Int. 11 (7) (2004) 36–42.

[

Science 407 (2017) 223–235 235

19] C.H. Weijsenfeld, A. Hoogendoorn, M. Koedam, DSputtering of polycrystallinemetals by inert gas ions of low energy (100–1000 eV), Physica 27 (1961).

20] K.A. Zoerb, J.D. Williams, D.D. Williams, A.P. Yalin, Differential sputteringyields of refractory metals by xenon, krypton, and argon ion bombardment atnormal and oblique incidences, in: 29th International Electric PropulsionConference, 2005, pp. 2005–2293.

21] N. Laegreid, G.K. Wehner, Sputtering yields of metals for Ar+ and Ne+ ionswith energies from 50 to 600 eV, J. Appl. Phys. 32 (3) (1961) 365–369.

22] Y. Yamamura, H. Tawara, Energy dependence of ion-induced sputtering yieldsfrom monatomic solids at normal incidence, At. Data Nucl. Data Tables 62 (2)(1996) 149–253.

23] N. Matsunami, Y. Yamamura, Y. Itikawa, N. Itoh, Y. Kazumata, S. Miyagawa, K.Morita, R. Shimizu, H. Tawara, Energy dependence of the ion-inducedsputtering yields of monatomic solids, At. Data Nucl. Data Tables 31 (1) (1984)1–80.

(1983) 65–86.25] Y. Yamamura, S. Shindo, An empirical formula for angular dependence of

sputtering yields, Radiat. effects 80 (1-2) (1984) 57–72.