ContentslistsavailableatScienceDirect
Acta Materialia
journalhomepage:www.elsevier.com/locate/actamat
Full length article
On factors defining the mechanical behavior of nanoporous gold
Birthe Zandersons
a, Lukas Lührs
a, Yong Li
a,b, Jörg Weissmüller
a,b,∗aInstitute of Materials Physics and Technology, Hamburg University of Technology, Hamburg, Germany
bInstitute of Materials Research, Materials Mechanics, Helmholtz-Zentrum Hereon, Geesthacht, Germany
a rt i c l e i nf o
Article history:
Received 30 September 2020 Revised 3 May 2021 Accepted 5 May 2021 Available online 18 May 2021 Keywords:
Nanoporous gold Dealloying
Mechanical properties Scaling laws Young’s modulus Strength
a b s t r a c t
Nanoporousgold (NPG)madebydealloying takesthe formofanetworkofnanoscalestruts or“liga- ments”.Itiswellestablishedthatthematerial’smechanicalbehaviorisstronglyaffectedbyitsligament size,Landbyitssolidvolumefraction,ϕ.WeexplorethemechanicalbehaviorofNPG,withanempha- sisonestablishingaconsistentdatasetwithcomparableLbutcoveringasignificantrangeofinitialϕ. SpecimensarepreparedfromAg-Aumasteralloys withtheirAuatomfraction,x0Au intherange0.20–
0.35.Sincedealloying replacesAgwithvoids,ϕ may beexpectedtoagreewithx0Au.Yet,spontaneous plasticdeformationeventsduringdealloyingcanresultinmacroscopicshrinkage,decouplingϕfromx0Au. Thisraisesthequestion,howdoϕand x0Auseparatelyaffectthemechanicalbehavior?Weconfirmtwo recentsuggestions,namelyi)amodifiedRoberts-Garboczi-typescalinglawforYoung’smodulusversusϕ ofthematerialinitsas-preparedstateandii)therelevanceofanapparent“load-bearingsolidfraction”
forYoung’smodulusaswellasstrength.Yet,remarkably,wefindthatstiffnessandstrengthoftheas- preparedmaterialshowamuchbettercorrelationtox0Auascomparedtoϕ.Thiscanbeunderstoodasa consequenceofthemicrostructuralchangesinducedbyshrinkage.Studyingthemicrostructureevolution duringannealing,wealsoconfirmthesuggestionthatcoarseningentailsanenhancedlossinstiffness for sampleswith lessersolidfraction. Thisfindingconfirms concernsabout the notion ofself-similar coarseningasageneralbehaviorofdealloying-madenetworkmaterials.
© 2021TheAuthor(s).PublishedbyElsevierLtdonbehalfofActaMaterialiaInc.
ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)
1. Introduction
Nanoporousmetalsmadebydealloyingaremacroscopicmono- lithic bodies that consist, at the nanoscale, of a homogeneous network of struts or “ligaments”. The materials, and specifically nanoporous gold (NPG), are under study as model systems for small-scale mechanicalbehavior [1]. As-preparedNPG takes liga- mentsizes,L,ofafewtensofnanometers[2–5].Annealinginduces coarseningthatlets Lincreaseuptovaluesofseveral100nm[6–
9].
Much of the interest in using the material as a model sys- tem stemsfromthe notionofself-similar coarsening.Size-effects on the mechanical behavior could then be studied in isolation, while the architecture of the nanoscale network remains invari- ant.Thefindingofpower-lawrelationsbetweentheyieldstrength,
σ
y,andLanditsdiscussionintermsofsize-inducedstrengthening exemplifies this approach [10–12]. Experimental tomographic re-∗Corresponding author at: Institute of Materials Physics and Technology, Ham- burg University of Technology, Hamburg, Germany.
E-mail address: weissmueller@tuhh.de (J. Weissmüller).
construction [7,13–15]andseveralnumerical studies[16–19]sup- port the notion of self-similar coarsening. Yet, evidence against that notion has recently been accumulating. Studies of the evo- lutionof thestiffness ofNPGcan only be understoodif thenet- workisallowedtoprogressivelydisconnectduringcoarsening[20]. Furthermore, large-scale atomistic simulations of coarsening ex- ploringan intervalofparameter spacerelevanttoexperimentdi- rectly evidence the disconnection [21]. The disagreement about self-similarity appears to originate in a pronounced dependency ofthetopologyevolutionduringcoarseningonthesolid(volume) fraction,
ϕ
. While networkswithϕ
0.3 tend to maintaintheir connectivityduringcoarsening,inkeepingwiththenotionofself- similarity,networkswithϕ
0.3tendtodisconnect[21].Asstud- iesofNPGhavebeenconsideringnetworkswithϕ
oneachsideofthethresholdvalue0.3,thedifferentfindingsmaysimplyoriginate in slightlydifferent solid fractions in the individual experiments.
Thisadvertisesaninterestinexperimentsprobingtheevolutionof connectivityduringcoarseningasthefunctionof
ϕ
.Liuetal.[22]varied
ϕ
byinterruptingthedealloyingprocessof AgAuPtatvariousstatesofAgdissolution.Theyfoundthedecrease ofthestiffnessuponcoarseningsubstantiallymorepronouncedathttps://doi.org/10.1016/j.actamat.2021.116979
1359-6454/© 2021 The Author(s). Published by Elsevier Ltd on behalf of Acta Materialia Inc. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ )
lesser
ϕ
,andtheyattributethelossinstiffnesstovariationsinthenetwork connectivity. Thisiswell compatiblewiththe aboveno- tion,yetthesystematicvariationofthecompositionwith
ϕ
intheexperimentofRef[22]complicatesthediscussionoftheirobserva- tionsintermsofacorrelationbetweenthelossofconnectivityand
ϕ
alone.Asystematicstudyofthelinkbetweenϕ
andthestiffnessvariationduringcoarseninginelementalnanoporous goldhasnot beenreported.
Irrespectiveoftheissueofself-similar ornon-self-similarevo- lution duringcoarsening,network solidsmadeby nanoscaleself- organization– as,forinstance,bydealloying– exhibitasystematic correlationbetweentheir
ϕ
andtheconnectivityalreadyintheas- prepared state.Thisis reflectedby compilations, inRefs[22] and [23],ofliterature data[5,12,20,22,24–36]forthe stiffnessofNPG.Specifically,thecompilationssuggestthatthevariationofeffective Young’s modulus,Yeff,with
ϕ
inas-preparedsamplesisnotcom- patible withtheexpectation forstructures withconstantconnec- tivity, asembodied inthe Gibson-Ashbyscaling lawfor thestiff- ness.Astudysystematicallyvaryingϕ
indealloyedFe-Cr[37]con-firmsthisobservation.
Soyarslanet al.[23] proposea physicallymotivatedmodel for the microstructure ofNPG, whichlinks therelation betweenYeff and
ϕ
toquantifiablemicrostructurecharacteristicsandspecifically to a scaledtopological genus.Thatstudyproposesa scaling rela- tion that accountsfora lossofpercolation– and,hence,forme- chanical disintegration– at a finite solid fraction. While thelink betweentheconnectivityofas-dealloyedNPGanditssolidfraction iswellconsistentwiththeexistingbodyofresearch[1],asystem- atic study,based on aset ofelemental nanoporous gold samples andwithconsistentpreparationandcharacterizationconditions,of thislinkandofthescalingrelationbySoyarslanetal.hasnotbeen reported.Here, we study the mechanical behavior of as-dealloyed and coarsenedNPGspecimenspreparedfromAg-Aumasteralloyswith their goldatomfraction,x0Au,intherange0.20–0.35.Sincean ide- alized dealloying process replaces all less-noble atoms(here Ag) with voids, values of
ϕ
may be expected in that same range.Ourstudydoes,however,accountforshrinkageduringdealloying, which canmodify
ϕ
,independently ofx0Au.Carefulcontrol ofthe preparation conditionsinour studyaffords a consistentligament size, L,independent ofϕ
or x0Au. We confirm the findings by Liu et al.andbySoyarslan etal.andspecificallythe notionofvaria- tions in connectivitybetween sampleswithsameϕ
. Remarkably, our results point towards a systematicinfluence ofthe composi- tionofthemasteralloyonthemechanicalbehavioroftheporous material.Thus,studiesofthemechanicalbehaviorofNPGneedto accountfortheimpactofx0Au,ontopofthatofϕ
.2. Scalinglaws
2.1. Constantconnectivity
As a basis for discussing the relation between stiffness and strength of nanoporous solids and their solid fraction,it is con- venient to compilethe mostimportant scalinglaws.Afrequently usedbenchmarkforYeff(
ϕ
)instructureswithfixedconnectivityis theGibson-Ashbyscalinglaw[38],Yeff=C1Y0
ϕ
2 (1)withY0 Young’s modulus ofthe massivematerial, andC1 a con- stantthattakesthevalue1foropen-cellfoamsingeneral[38]and specifically for NPG [39]. Similarly, the Gibson-Ashby scaling law forthemacroscopicyieldstrength,
σ
y,linksϕ
forNPGtothelocalligamentstrength,
σ
y0,asσ
y=C2σ
y0ϕ
3/2 (2)withC2=0.3forNPG[39].
2.2. Accountingforconnectivityvariationindealloyednanoporous metals
Thescalingbetweenmechanicalcharacteristicssuchasstiffness orstrengthandthesolidfractionofporousmaterialsisstronglyaf- fectediftheconnectivityvariessystematicallywith
ϕ
andspecifi-callyifapercolationthresholdismetatfinite
ϕ
[23,40–43].Foras-preparedNPGwith
ϕ
<0.5,ithasbeendemonstratedthatamod- ifiedversion [23] oftheRoberts-Garboczi scaling lawforrandom porousmicrostructures [42]isin excellentagreementwithlitera- turedatafortheYoung’smodulus.ThemodifiedRoberts-Garboczi lawtakestheform[23]Yeff=C3Y0
ϕ
−ϕ
P1−
ϕ
P m( ϕ
<0.5)
(3)with
ϕ
P thesolidfractionatthepercolationthreshold.Theunder- lyingmodelisaleveledGaussianrandomfield,originallyproposed forearly-stage spinodallydecomposedstructures [44].Here,ϕ
P= 1−erf(2−1/2)≈0.159 [23].Fitting the parameters to as-prepared NPGresultsinC3=2.03±0.16andm=2.56±0.04[23].NotingthatNPG’sexperimentalvaluesofYeffareregularlysub- stantiallylowerthanthepredictionsoftheGibson-Ashbylaw,Refs [20,22]proposetheintroductionofanapparentload-bearingsolid fraction,
ϕ
lb,whichisalwayslessthantheactualsolidfraction,ϕ
.Inthatapproach,
ϕ
lbisdeterminedfromtheexperimentalstiffness ontheassumptionthatEq.(1)holdsfortheload-bearingstructure.Inotherwords,
σ
y=C2σ
y0( βϕ )
3/2 (4)where
β
denotesanapparentload-bearingfractionfactor,whichis definedsothatβ
=ϕ
lbϕ
, (5)andwheretheapparentload-bearing solid fractionis determined fromtheexperimentalYeffvia
ϕ
lb= Yeff Y0 1/2. (6)
Experimentalvaluesof
σ
yagreewellwithEq.(4).It may be emphasized that the approaches underlying Eqs.
(3)and (6)are not mutuallyexclusive, they justobtain informa- tiononthenetworkconnectivityfromdifferentsources.Thelatter equationisbasedonanempiricalcharacterizationofthenetwork basedonitselasticbehavior,whereastheformerassumesaphysi- callymotivatedsystematiccorrelationbetweenthenetwork’scon- nectivityanditssolidfraction.
2.3. Accountingforshrinkageduringdealloying
Motivatedbytheexperimentalresultsofthepresentstudy(see below),weconsideryetanothermodificationofthescalingbehav- ior.Thismodificationisdesignedtoaccountfortheshrinkagethat isobservedtoaccompanydealloyingunderawiderangeofexper- imentalconditions [45–49].Whileourargumentmayapply more generally,wedesignatethelessnobleandmorenobleelementsas AgandAu,respectivelyforsimplicity.
Asthefirststepofdealloying,weconsidertheformationofan initialstructure thatisfreeofshrinkage.Forthisinitialstructure, weproposethat1.)allAgisremovedsothattheinitialsolidfrac- tion,
ϕ
ini,agreeswiththeAuatomfractioninthemasteralloy,x0Au; 2.)themodelofRef [23]applies,includingspecificallyEq.(3).As the next step,we allow forshrinkage.Thereby, thesolid fraction increases fromϕ
ini to the final solid fraction of the as-preparedsample,
ϕ
.Here,weproposethat 3.)theconnectivityremains in- variantduringtheshrinkage.The above propositions lead to the following suggestions:
Firstly,Yeffof theinitial structure (no shrinkage) obeysEq.(3),yet withx0Au substitutedfor
ϕ
.Inotherwords,Yeff=C3Y0
x0Au−ϕ
P1−
ϕ
P m. (7)
Secondly, since the shrinkageis atfixed connectivity, Eq. (1)ap- pliestomaterialsthathaveundergoneshrinkageinthesensethat Yeff scales withthe shrinkage-induced increase in
ϕ
2. ShrinkagewillthenintroduceacorrectiontoEq.(7)intheform Yeff=
v
2C3Y0 x0Au−ϕ
P1−
ϕ
P m. (8)
The parametersC3 andm inEqs.(7) and(8)agreewiththose in Eq.(3).Thedensificationparameter
v
isgivenbyv
=ϕ
x0Au; (9)
itrepresentsavolumeshrinkageratio,inotherwords,theratioof as-preparedNPGvolumeovermasteralloyvolume.
Equation(8)isequivalenttousingtheGibson-Ashbyscalinglaw Eq.(1)with
ϕ
correctedbytheapparentload-bearingfractionfac- torβ
=C3
1 x0Au
x0Au−ϕ
P1−
ϕ
P m/2. (10)
Note, that
β
is independent of whether or not there has been shrinkage. Note also, that the relations discussed in the present subsectionbreakwiththenotionthatYeffisuniquelydetermined byϕ
.Instead,theyallowforx0Au asanadditionalparameter.3. Procedures 3.1. Samplepreparation
MasteralloysAg-Auwithx0Au= 0.20,0.25,0.30,and0.35were preparedasinRef [35].Inbrief, arcmeltingwasfollowedby ho- mogenization via annealing for five days at850 ◦C in an evacu- atedandthensealedglasstube.Wires0.95mmindiameterwere drawnandcutintocylinders1.9mminlength.Vacuumannealing for 3hat600 ◦C followed forrecovery.Todealloy theprecursors produced in this way,we used the two mostcommon methods:
electochemicalcorrosionandfreecorrosion.
Electrochemicaldealloyingusedathree-electrodesetupin1M HClO4 with a Ag wire counter electrode and a Ag/AgCl pseudo- reference electrode. We found that producing crack-free, mono- lithicsamplesrequiredaninitialcurrentdensityof5mA/cm2(re- ferredtotheareaofthemacroscopicsamplesurface).Ineachcase, the dealloying potential, ED wasadjusted tomeet that condition.
ED wasdetermined froma linearpotential sweepat ascan rateof 5 mV/son aseparate precursorsamplewithknownsurfacearea.
Afterdealloying,residualsilverwasremovedbymaintainingapo- tential 0.1 V above ED until the currentfell below 15 μA. After- wards,40potentialcyclesfrom0.1to1.6Vvs.SHEat5mV/swere applied,endingat0.8Vsoastoensureanadsorbate-freesurface.
Thesampleswerethenrinsedinwateranddriedinair.Theresid- ual silver content of all electrochemically annealed sample types was∼1at%.
We found the required ED toincrease from1.13 to 1.28V vs.
SHE for master alloys as x0Au increasedfrom 0.20 to 0.30. While samples with x0Au = 0.20 and 0.25 could be dealloyed at ambi- enttemperature,avoidingcrackformationinprecursorswithx0Au= 0.30requiredthedealloyingtemperaturetoberaisedto50◦C.For
x0Au=0.35wewereunabletoestablishanelectrochemicaldealloy- ing protocol that wouldyield fully dealloyed and crack-free NPG samples.
Forfreecorrosion,themasteralloysweredealloyedin65wt.%
HNO3 at ambient temperatureand the resulting NPGthen rinsed inwateranddriedinair.Forthisprocess,after25daysthex0Au= 0.35masteralloysevolvedtoNPGwitharesidualsilvercontentof
∼3at.%.Prolonged exposurein acidof up to sixweeksgave no furtherreductioninresidualAg.Masteralloyswithx0Au=0.30and 0.25yieldNPGwith∼1at%silverafter4and2daysoffreecorro- sion,respectively.Formasteralloyswithx0Au=0.20,freecorrosion wasnotapplicable,sincethespecimensdisintegratedintopowder upondealloying.
Toevaluate the possibleeffect ofadsorbates on free corroded NPG,we subjected some ofthe free-corrosionsamples to 20po- tentialcyclesbetween0.1Vand1.6Vvs.SHEatscanrate5mV/s, stopping at 0.8 V. This mimics the protocol at the end of elec- trochemical dealloying, andleaves the samplesurface clean,free ofadsorbate. We confirmedthat this conditioningtreatment had noimpactonthemechanicalresponse.Themechanicalcharacteri- zation showninthe figures ofthiswork were obtainedwiththe straightforward free corrosion protocol without the conditioning treatment.
3.2. Structuralcharacterization
On average, the nanoporous samples had a length of 1.77± 0.12 mm anda diameter of 0.94±0.06 mm. By measuring geo- metricaldimensionsand mass,
ϕ
wasdetermined. Forthe deter- minationofϕ
duringdeformation,seeSection3.3below.The ligament size, L was determined as the diameter of the struts in the nanoporous samples, as obtained by analysis of scanning-electron-microscope(SEM)images.Foreverysample,di- ameters ofmore than50 ligamentswere measured fromeach of atleasttwoimagesoffracturesurfaces.Lisspecifiedasthemean, and the standard deviation is specified as the uncertainty (error bar).Theresidualsilvercontentwasdeterminedbyenergydisper- siveX-Rayanalysis,basedonaveragesfromatleasttwosamplesof everyindividualdealloyingtechniqueandelementalcomposition.
3.3. Mechanicaltesting
Single loading and load/unload compression tests with engi- neering strain rates of
ε
˙=10−4s−1 where evaluated from a to- tal of 60 samples, using the uniaxial compression test setup in- troducedinRef.[50].Digital ImageCorrelationwiththesoftware DaVis8.2.0 by LaVisionwasused to determinethe strain by vir- tualstraingagesonthesamplesurface,asdescribedinRef.[50,51]. Sincethedeformationinandtransversetotheloaddirectionwas known,themacroscopicdimensionsofthesamplescouldbedeter- minedduringtheentiremeasurement.Withthisinformation,the evolutionofϕ
duringcompressionwasevaluated.We determined Yeff in various states of deformation as the slopeofstraightlinesofbestlinearfittotheentireunloadregime oftherespectiveload/unloadsegment.Inotherwords,Yeffisspec- ifiedasthesecantmodulus.
When subjected to compression, as-prepared NPG exhibits an early yield onset and a continuous elastic-plastic transition [1,33,34].In theinterestofameaningfulseparationofelasticand plastic deformation, the yield strength,
σ
y was measured as the stressat∼1%plasticstrain.Thisisconsistentwithearlierstudies [5,52].TheinitialvalueofYeffwasdeterminedfromasubsequent unload segment, at ∼ 1.5% plastic strain. We found (see Section 4)that thevariationofflow stressaswell asYeff issmallinthat strainrange.Table 1
Properties of as-dealloyed nanoporous gold samples that were used for load/unload compression tests. Au atom fraction, x 0Au, densification parameter, v, solid fraction of the uncompressed samples, ϕ, ligament size, L , Young’s modulus at a plastic strain around εpl= 0 . 015 , Y eff, and apparent load-bearing fraction factor, β. At least five samples of each type were investigated as the basis for the data of this table.
x 0Au v [-] ϕ[-] L [nm] Y eff[GPa] β[-]
electrochemical corrosion 0.20 1.43 ±0.04 0 . 29 ±0 . 01 33 ±6 0 . 35 ±0 . 07 0.23 ±0.03 0.25 1.06 ±0.06 0 . 26 ±0 . 02 32 ±8 0 . 38 ±0 . 07 0.26 ±0.02 0.30 1.02 ±0.01 0 . 31 ±0 . 01 29 ±8 2 . 25 ±0 . 42 0.55 ±0.05 free corrosion 0.25 1.22 ±0.07 0 . 31 ±0 . 02 49 ±13 0 . 50 ±0 . 11 0.26 ±0.04 0.30 1.02 ±0.02 0 . 31 ±0 . 01 31 ±7 1 . 73 ±0 . 33 0.48 ±0.06 0.35 1.02 ±0.01 0 . 36 ±0 . 01 46 ±9 6 . 53 ±0 . 42 0.80 ±0.03
3.4. Atomisticsimulationofrelaxation
The atomistic simulationof relaxationused procedures ofour previousworkforgeneratingrealisticnanoporousinitialstructures [21,53], for implementing the relaxation by molecular dynamics (MD)[34]andforanalyzingtheconnectivity[21].Wehereprovide a briefsummary;details areprovided intheSupporting Informa- tion.
The initial structures–with solid fractions 0.20, 0.25, 0.30 and 0.35 andwithidentical ligamentsize,3.3 nm–weregeneratedby a variantofJohn Cahn’sleveledwave model[44],modifiedtoaf- fordperiodicboundaryconditionsin3D [23].Aface-centered cu- bic (fcc) crystal lattice wasinscribed into the solid phase, using the stress-freebulk latticeparameter(a=408nm)oftheMDin- teratomic potential. The simulation boxes hadan edge length of 200aandcontainedbetween6.4and11.2×106 atoms,depending on
ϕ
.The relaxation by MDused the open-source softwareLammps [54] with an embedded-atom method (EAM) potential for Au [55]andperiodicboundaryconditionsin3D.UsingaNosé-Hoover thermostat andbarostat [56,57],the structures were relaxedfirst athermallyandthenthermallyat300Kfor2ns.Theopen-source softwareOvito[58]wasappliedforvisualization.
The net topological genus, G, of structures with volume V was determined by the open-source code CHomP [59]. G repre- sents the numberof connections in the sample. The topology of a microstructure may be described by a scaled value, g, of the genus,amaterial-specificandsize-independentquantitythatspec- ifiesthenumberofconnectionsinarepresentativestructuralunit [13,14,23,60–62]. The scaled genus of as-prepared NPG has been experimentally quantified [13,14,61] and its variation with
ϕ
aswell as its impact forYeff(
ϕ
) emphasized [1,5,20,22,23,61,63,64]. Furthermore, experimental and modeling data for the variation of gduringannealingare available[13–17,19,21,65].We identified the sizeofthe representativestructuralunitswiththemeandis- tance,˜L,betweenthecentersofneighboringligaments[23],hence g=GL˜3/V. The value of L˜ is implied by the wavelength used in constructingtheleveled-wavestructureandbytheoverallvolume changeduringtherelaxation.4. Results 4.1. Microstructure
Figure 1 shows representative scanning electron micrographs (SEM) forall sampletypes.Theresultingligamentsizesarelisted inTable1.Electrochemicalcorrosionprovidesforparticularlycon- sistent L. Yet, in spite of the larger scatter in the free corrosion samples,alldataisconsistentwithourdealloyingprotocolsresult- inginsensiblysimilarLforeachsampletype,around30–40nm.
Figure2plotsthesolidfraction
ϕ
versusx0Au.Thefigureisbased on sets of 9 samples of x0Au= 0.20, 22 electrochemical corrodedTable 2
Properties of annealed nanoporous gold samples. Listed are the Au atom fraction, x 0Au, the densification parameter, v, the solid fraction of the uncompressed samples, ϕ, the ligament size, L , Young’s modulus at a plastic strain around εpl= 0 . 015 , Y eff, the apparent load-bearing fraction factor, β, and the number of tested samples. At least four samples of each type were investigated as the basis for the data of this table.
x 0Au v[-] ϕ[-] L [nm] Y eff[GPa] β[-]
0.25 1.15 ±0.10 0.29 ±0.03 116 ±28 0.19 ±0.05 0.17 ±0.02 0.30 1.03 ±0.02 0 . 31 ±0 . 01 115 ±37 1.36 ±0.29 0.43 ±0.04 0.35 1.03 ±0.01 0 . 36 ±0 . 01 106 ±28 5.75 ±0.19 0.75 ±0.01
and14freecorrodedsamplesofx0Au=0.25,28electrochemicaland 10free corrodedsamplesofx0Au = 0.30and29samplesof x0Au = 0.35,showingmeanvalueandstandarddeviationineachset.The twoparameters agreeforthehighersolidfractions,indicatinglit- tleshrinkage.Yet,thereisatrendfor
ϕ
ofsampleswithlesserx0Autobeincreased.Inother words,sampleswithlesserx0Au aremore pronetoshrinkage.Atx0Au=0.25,theshrinkageismoreseverefor thefreecorrosionsamples.Masteralloyswithx0Au=0.20exhibited particularly severe densification (
ϕ
substantially larger than x0Au) when dealloyed electrochemically, while they disintegratedupon freecorrosion.Thebehaviorisconsistentwiththenotionthatthe percolationlimit ofthenetworksolidisreachedatϕ
≈0.16[23], whichisonlyslightlybelowthevalueof0.20.Annealingin airat300 ◦C suppliedsampleswith coarsermi- crostructures. Annealing times where 15 min for samples with x0Au=0.25and30minforthosewithx0Au=0.30and0.35.There- sultingligamentsizesarelisted inTable2. Thesizesherescatter around110nm.
4.2. Stiffnessofas-dealloyednanoporousgold
Figure 3 plots the stress-strain graph of an exemplary load/unloadcompression test onan as-dealloyedx0Au=0.25sam- ple. The figure illustrates that values of Yeff in various stages ofdeformation were determined assecant moduli by analysisof unload/reload segments. Figure 4(a) illustrates how these values evolve during the compression ofrepresentative samples.As de- scribedinSection3.3,wetakethevalueofYeffataplasticstrainof around 0.015asrepresentative ofthestiffnessinthe as-prepared state.
Figure5 presentsanexemplary map ofthedisplacementfield onthesamplesurface, hereforasamplewithx0Au=0.20atcom- pressiveengineeringstrainof0.30.Itisseenthatthedisplacement gradient isuniform overthe entiresample. The observationsup- portsthe homogeneityof the samplesand theuniformity of the deformation.
Figure4 (b)plotsYeffforas-prepared samples,asobtainedby averaging the experimental results forseveral specimens ofeach sample type. These values are also listed in Table 1, where the properties ofall samples used forload/unload compression tests
Fig. 1. Scanning-electron-microscope images of fracture surfaces with identical scale. Upper row: nanoporous gold made by electrochemical corrosion of Ag 1−xAu xwith (a) x = 0 . 25 , (b) x = 0 . 30 , (c) x = 0 . 20 . Lower row, samples made by free corrosion of Ag 1−xAu xwith (d) x = 0 . 25 , (e) x = 0 . 30 , (f) x = 0 . 35 .
Fig. 2. Solid fraction, ϕ, of as-prepared nanoporous gold versus Au fraction, x 0Au, in master alloy. Mean value and standard deviation for sets of at least 9 samples of each type. Dealloying method is indicated in legend. Dashed line: identity, applica- ble when there is no densification during dealloying.
are summarized. Within the range of solid fractions covered by our sampletypes,Yeff can be seento varyby morethan one or- derofmagnitude.Yet,thegraphofYeffversus
ϕ
doesnotrevealastrongcorrelation,nordoesitsuggestagreementwithanyone of thescalinglawsEq.(1)or(3),whicharerepresentedbysolidlines inthefigure.
Remarkably, the experimental Yeff exhibit a much better cor- relation withthe Aufraction,x0Au in themaster alloy.This isap- parent inFigure 4(c), whichreveals aconsistent trendforYeff to increase with x0Au. The solid linein that figure isthe scaling law ofEq.(7),inotherwords,Eq.(3)withx0Au substitutedfor
ϕ
.Thisgraphagreeswiththetrendinthedata,suggestingthatthemas- teralloycompositionisrelevantforthemechanicalbehavior,even whenthesolid fractiondeviatesfromx0Au asaresultofshrinkage duringthedealloying.
As an unbiased measure for correlation, we have computed Spearman’s rankcorrelation coefficient for theexperimental data inFig.4(b)and(c).Wecomparedthecorrelationoflog(Yeff)with
Fig. 3. Engineering stress, σ, vs. engineering strain, εeng, of an load/ unload com- pression test of dealloyed Ag 75Au 25. Inset: details of a load/unload segment; red line indicates linear fit. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
ϕ
tothatwithx0Au.Thecorrelationcoefficientevaluatesto0.88forϕ
andto0.97forx0Au.Thisconfirmsthesignificantlybetter corre- lationbetweenthemechanicalbehaviorandthemasteralloycom- position,asopposedtothesolidfraction.A notable exception fromthe agreement betweenEq. (3)and experimentisthematerialwithx0Au=0.20,whichissubstantially stifferthan predicted.Thisdeviationcoincideswiththeverysub- stantial shrinkage of that sample. This might be a consequence of the gold content in the precursor being close to the percola- tion limit, which might lead to extended restructuring processes alreadyduringdealloying.
4.3. Strengthofas-dealloyednanoporousgold
As abackground forourdiscussion ofthe plasticdeformation behavior,Fig. 6(a)showsthestress-strainresponseofrepresenta- tive specimensin single-loadingcompression tests. Weobserve a
Fig. 4. Effective Young’s modulus, Y eff, determined by unload regimes of load/ un- load compression tests of nanoporous gold made by dealloying Ag-Au master alloys with a molar Au fraction of x 0Au= 0.20, 0.25, 0.30 and 0.35. Color code and legend in (b) and (c) identify x 0Auand dealloying method, respectively. (a) Y effvs. engineer- ing stain, εengof representative samples. (b) Y effat a plastic strain around 0.015 vs.
solid fraction, ϕ; averages over several samples of same type with error bars indi- cating the standard deviation. Solid lines: gray, Gibson-Ashby scaling law, Eq. (1) , and red, modified Roberts-Garboczi scaling law, proposed by Soyarslan et al., Eq.
(3) . (c) plots the data of (b) vs. x 0Au. Red line: Eq. (3) with x 0Ausubstituted for ϕ. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
trend for the strength andflow stress to increase withx0Au. This agreeswithearlierresultsfromnanoindentationtests[11,67].
Figure 6(b)exploresthevariationof
σ
y withϕ
.Thedatarep-resentsaverages andstandard deviationsofseveralspecimensfor each sampletype;x0Au isindicated bylabels.AsforYoung’s mod- ulus,
σ
yincreaseswithϕ
,yetthecorrelationisagainpoor.Inpar- ticular,sampleswithlowerx0Au areseentoexhibitlowstrengthin view oftheircomparativelyhighϕ
.Thegraylineshowsthemod-ifiedGibson-Asby lawproposed byRef. [37].This modelisbased onEq.(2)forwhichthesolidfractionisreplacedbytheapparent
Fig. 5. Homogeneity of deformation. Photograph of NPG, electrochemically deal- loyed from Ag 80Au 20, in situ at 30% engineering strain. Axial (projected on the load axis) displacement component from digital image correlation is indicated by color code. Load surfaces can be seen at top and bottom of image. Bright spots: speckle pattern, applied for enhanced contrast, as in [66] .
Fig. 6. Strength and flow behavior. Color code designates Au atom fraction in mas- ter alloy, see labels in (b). Legend in (c) specifies corrosion method. (a): Engineering stress, σ, vs. engineering strain, εeng, for representative samples. (b): Strength, σy, obtained as σat 1% plastic strain, vs. solid fraction, ϕ. Average values are given for sets of several samples of the same type, with standard deviations indicated by er- ror bars. Gray line: scaling model proposed by Ref. [37] . (c): σyas in (b), plotted versus Au atom fraction in master alloy, x 0Au. Gray line: Eq. (4) combined with Eq.
(10) .
Fig. 7. (a) Yield stress at ligament level, σy0, vs. plastic strain, εpl, of nanoporous gold made by dealloying Ag 1−xAu xwith x = 0.25, 0.30 and 0.35 (see color code and legend in (b)). σy0was obtained by combining the Gibson-Ashby scaling law, Eq.
(2) , with the reloading yield stress, σ, of load/unload compression tests,. Triangular symbols: Eq. (2) with the true solid fraction, ϕ; pentagonal symbols: Eq. (2) with the apparent load-bearing solid fraction, ϕlbof Eq. (6) . Note the much better agree- ment in the latter case. (b) Symbols: experimental σ vs. ϕlb. Line: Gibson-Ashby law Eq. (2) , using σy0= 1 . 7 GPa from (a).
load bearingsolid fraction
ϕ
lb (Eq. (6)). Forthe requiredYeff weusedthemodifiedRoberts-Garboczilaw,Eq.(3).
Here,asforthestiffness,itisagaininstructivetoplottheme- chanical characteristics against x0Au as opposed to
ϕ
. That plot,Fig. 6(c), shows a substantially improved correlation. Spearman’s rankcorrelationcoefficientishere0.97forlog(
σ
y)withx0Au,acon- siderablystrongercorrelationthanwhatissuggestedbythevalue, 0.70,forthecorrelationoflog(σ
y)withϕ
.Section2.3introducesahypothesisabouttheimpactofshrink- ageduringdealloyingontheconnectivity,leadingintoEq.(10)as a prediction forthe apparent load-bearing fraction factor,
β
.Thesolid line in Fig. 6(c) shows the prediction of Eq. (4), using Eq.
(10) for
β
. This representation is seen to match the experimen- tal trendwell. The observation confirmsthat x0Au, rather thanϕ
,is the moreappropriate materials characteristics determiningthe mechanicalpropertiesofas-preparedNPG.
4.4. Apparentload-bearingsolidfraction
Experimental dataforthe yieldstrength ofNPGmaybe com- bined with scaling equations, such as Eq. (2), to evaluate the strength,
σ
y0, orflow stress atthescale ofthe ligamentsof NPG.Thatstrategyisofinteresttoourworkbecausetheligamentsizes aresensiblyidenticalforallsamples,soweexpectthe
σ
y0toagree forallsamples,independentoftheirsolidfraction,connectivityor deformationstate.Whetherornotthisexpectationisbornoutpro- videsaconsistencycheckforthescalinglawinquestion.Figure 7(a)showsthe evolutionof
σ
y0 withplasticstrain,ε
pl, forselectedsamplescoveringthefullrangeofx0Auofourstudy.TheFig. 8. Apparent load-bearing fraction factor, β, vs. Au atom fraction in master alloy, x 0Au. Data points represent averages over several samples of the same type, with error bars indicating standard deviation. Note overlap of two data points at x 0Au= 0 . 25 . Red line: βcalculated by Eq. (10) .
opensymbolsarebasedonexperimentaldatafor
σ
andonEq.(2),with
ϕ
determinedfromtheaspreparedsolidfractionalongwiththelongitudinalandtransverseplasticstrainsduringcompression.
Here,thepredictionfor
σ
0variesstronglywitheachofthesetwoparameters,x0Au and
ε
pl.Inviewoftheexpectationthatσ
y0 should takeonauniversalvalue,thedataconfirmsthatEq.(2)initsorig- inalformdoesnotappropriatelydescribethestrengthofNPG.The situation is quite different when the apparent load-bearing solid fractionϕ
lb issubstitutedforϕ
in Eq.(2).The closedsymbolsinFig. 7(a)represent the
σ
0 obtainedin that way. Here,ϕ
lb is de-terminedfromtheexperimentaldataforYeff alongwithEq.(6).It isseen that the data isnow inmuch better agreement withthe expectationofconstant
σ
y0.Thisconfirmsϕ
lbasaparametergov-erningthestrengthofNPG,inagreementwiththesuggestionsin Refs[20,22].
Averaging – at1% plastic strain – over the ligament strength represented by the closed symbols in Fig. 7(a) suggests
σ
y0= 1.7 GPa for our material, which is in accordance with Ref. [20]. Thisvalue underlies thesolid linewhich representsthemodified Gibson-AshbybehaviorinFig.6.Thesignificanceof
ϕ
lbisalsoapparent inthegraphoftheef-fective,macroscopicflow stress versus
ϕ
lb,Fig.7(b).Representa- tivespecimensofeachsampletypeareincludedinthisplot.Ade- tailedviewoftheindividualsampleswithx0Au=0.30and0.35re- vealsadeviationfromthequadratic,Gibson-Ashby-type variation thatemergeswhenEq.(2)isevaluatedwithϕ
replacedbyϕ
lb.Yet,sampleswithx0Au =0.25andsmaller,especiallytheelectrochemi- callycorrodedones,dofollowthatvariationindividually.Moresig- nificantly,theoveralltrendfordataspanningmorethanoneorder in
σ
is consistent with the Gibson-Ashby-type law. Thereby, our observationconfirmsthesuggestedscalingofσ
withϕ
lb.Weem-phasize,however,that anyapparent agreementbetweendataand scalinglawmaydeceive:substantialdifferencesinthemicrostruc- tureandinthe mechanicalbehavior aresimplyparameterized by
ϕ
lb.Atthepresentpointinourdiscussionofthestrength,thatpa-rameterisempirical,governedbytheratioofYeffandY0,tobede- terminedinaseparateexperiment.Inotherwords,thegoodagree- mentdoesnotimplyanaprioripredictivepoweroftheapproach.
Anapproachtowardspredicting
ϕ
lbhasbeendiscussedinSec-tion 2.3, whereit isembodied in Eq.(10)for theapparent load- bearingfractionfactor,
β
.Fig.8verifiesthatequation,bycompar-Fig. 9. Molecular dynamics simulation of relaxation. Scaled topological genus, g, versus solid fraction, ϕ. Solid line: theory [23] for the leveled-wave initial struc- tures. Symbols: evolution of gand ϕ during relaxation of structures with differ- ent initial solid fractions, as indicated in legend. Each structure starts sensibly on the theory line and then evolves towards increasing ϕas the relaxation time pro- gresses. All structures were relaxed under identical conditions and for identical time periods, see Supporting Information for details. Some data points overlap. Dashed horizontal lines represent guides to the eye for constant values of inital g. Insert:
exemplary rendering of simulation volume, here for initial ϕof 0.35, after relax- ation.
ing itspredictions to
β
valuesdetermined fromtheexperimental Yeff.TheunderlyingexperimentaldatabaseisshowninFig.4and listedinTable1.Figure8confirmsthepredictionthatβ
increasessystematicallywithincreasingx0Au,andtheexperimentalvaluesare indeedseentoscatteraroundthetheoryline.
Itisofinteresttofocusonthesampleswithx0Au=0.25,which exhibit substantial differences in their initial
ϕ
, as apparent inFig.4(b).Inspiteofthosedifferences,the
β
valuesarenearlyiden-tical.Inotherwords,weobservepracticallynoimpactofshrinkage on
β
. That isconsistent with the notion that the connectivityis established early-on duringthe dealloying, andremains invariant duringlatershrinkageevents.4.5. Relaxation,shrinkage,andconnectivityevolution
Molecular dynamics simulations ofthe relaxation of NPG ex- plored the interrelation between volume shrinkage and the con- nectivity, asparameterized bythe scaled genus g. Thesimulation studied the same four
ϕ
as the experiment, namely 0.20, 0.25, 0.30, and0.35. The initial structures were constructed by means of theleveled-wavemodel. Thatmodel,whichalso underliesour analysisofthestiffnessandspecificallyEq.(3),hasbeenfoundto quantitatively agree with experiment on as-prepared NPG in re- spect to connectivity, Young’s modulus, and strength [23,68]. In agreement withthe experiment, the results(Fig. S2a inSupport- ing Information) show progressively largervolume contraction asϕ
is reduced.We found that the relaxation brings practically no relative increase inthe netgenus Gforlargerϕ
.Yet,G increasessubstantiallyatlesser
ϕ
(Fig.S2b).Figure 9 summarizes the findings for the relaxationin a plot of gversus
ϕ
, following theevolution of each ofthe four modelstructures duringtherelaxation.Thefirstdatapoint(atlowest
ϕ
)foreachstructureisseentoberightonthetheorylineforthelev- eledwave model,validatingourimplementationofthatmodel.At the largest
ϕ
value,0.35, thetracesofg(ϕ
)are sensiblyhorizon- tal, confirmingthat thedensification hereleavestheconnectivity unchanged. Astheinitial valueofϕ
isreduced, theincrease inϕ
is enhanced and at the same time the traces of g(
ϕ
) slope up- wards,indicatingthattherelaxationwillenhancetheconnectivityFig. 10. Exemplary depiction of a reconnection event in nanoporous gold with an Au content of 0.20 as a result of relaxation. Images show molecular dynamic simu- lations of a nanoporous structure before (upper image) and after relaxation (lower).
Numbers refer to initially individual ligaments that connect during the relaxation event.
ofstructures withlesser initial solid fraction.Wefound that this trendisparticularlystrongforthestructurewith
ϕ
=0.20.The observationsfromMD alsopoint towardsan explanation:
many “dangling” ends of disconnected ligaments were observed in the low-density structure. The shrinkage can bring the dan- glingends incontactwithneighboringregions ofthesolidstruc- ture, thereby establishing new connections. Several examples of this process are marked in Fig. 10. That process is less likely in thedenserstructures,becausetheystartout wellconnected,with muchfewerdanglingends.
4.6. As-dealloyedvs.coarsenednanoporousgold
We now discuss the evolution of the mechanical behavior in response to coarsening, focusing on sample types that have un- dergonenegligible shrinkageduringpreparation,so thattheir as- preparedstatehas
ϕ
≈x0Au.Figure11(a)plotsthemeanvalueand standarddeviationofYeffofthesampletypesfromthissetversusϕ
.As introducedin Tables1 and2,the initialandfinal ligamentsizesareconsistentlyL≈40and110nm,respectively,forallsam- pletypes.Thedataconveysthat annealingcanintroduceadensi- fication,particularlyforsampleswithsmallx0Au.InrespecttoYeff, thefigureillustratesagaintheenhanced complianceatreduced
ϕ
inthe initial state.Moreover, the arrowsillustrate that annealing reducesYeff.The remarkableobservationis that thisreduction is smallforsampleswithx0Au=0.35,whereasthereisapronounced decreaseinYeffforsampleswithlowerx0Au.ThereductioninYeffis relativelylarger forsamples withsmaller
ϕ
.Thatisobviousfromthelarger downwardshiftofthedatapointsforsmaller
ϕ
,alongwiththelogarithmicordinatescaling.
The last-mentioned observation is further illustrated by the graphinFig.11(b),whichshowstheapparent load-bearingfrac- tionfactor
β
,Eq.(5),inadisplayanalogoustoFig.11(a).ThedataFig. 11. Comparison of as-dealloyed and annealed nanoporous gold with varying Au amount in Ag-Au master alloy, x 0Au. (a) Effective Young’s modulus, Y eff, at low strains, normalized with Young’s modulus of massive gold, Y 0= 79 GPa [69] , vs.
solid fraction, ϕ. The stated percentage quote the drop in stiffness due to coars- ening. (b) figures β= (Y eff/Y 0)1/2/ϕ vs. x 0Au. The data points are averages over all samples of the same type, with error bars indicating the standard deviation.
is here plotted versus x0Au in order to emphasize its assignment tothesampleidentity.Thisfigureunderlinesthat thefindings for the variationofYeffinresponse toannealingimplyavariation of theapparentload-bearingfraction,inotherwords,achangeinthe networkconnectivity.Thatvariationisrelativelylargerforsmaller solidfractions.
5. Discussion
The present study investigates the mechanical behavior of nanoporous gold (NPG) made by dealloying Ag-Au, withan em- phasisoncoveringawiderangeofinitialAuatomfractions,x0Auin themasteralloyand,thereby,awiderangeofsolidfractions,
ϕ
intheporousmaterial.We exploittheestablishednotionthatvaria- tions intheeffectiveYoung’smodulus,Yeff,provideasignatureof variationsinthemicrostructure ofthematerial’snetwork ofliga- ments, andspecificallyforvariationsintheconnectivity. Thereby, ourexperimentsonthemechanicalbehavioraffordconclusionson howthe microstructuredependsonx0Au and
ϕ
intheas-prepared state, andon how themicrostructure evolveswhen the ligament size,Listunedbyannealing-inducedcoarsening.Inthe contextofourwork,variationsofthemechanicalprop- erties areonlyindirectlyrelatedtovariationsinthescaled genus g, namelythroughthe apparent load-bearingfraction factor,
β
ofthenetwork.Thatparameterisempiricallyaccessiblefromtheex- perimentalYeff alongwithEq.(6).Qualitatively,anydecreasein
β
canonlybeunderstoodwithaconcomitantdecreaseing.Thislink underliesourdiscussion.
5.1. As-dealloyedstate:variationofconnectivitywithsolidfraction
The data in our work confirms the established notion [1,5,20,22,61,63,64,68] that scaling lawsfor network structures of constantconnectivity,suchastheGibson-Ashbylaws,donotpro- videanadequatedescriptionofthevariationofstrengthandstiff- ness of NPG with its solid fraction.Our results also support the approach ofRefs [20,22],in which theGibson-Ashby law forthe strength,Eq.(1),isappliedtoanequivalentnetworkthataccounts only forthe load-bearing struts. Tothis end, the true solid frac- tionisreplacedwithanapparent,load-bearingone,whichcanbe empiricallydeterminedbasedonexperimentaldataforYeffandon Eq.(6).The agreement withthe experimental strength datasup- portsthenotionofaneffective,load-bearingnetwork.
Thehighsignificanceoftheparameter
ϕ
lbsuggestsan inspec-tionofitsrelationtothenetworkmicrostructure.Oneobviouscon- cept relatesdifferencesbetween
ϕ
lb andthe truesolidfractionϕ
todisconnectednetworkstruts,whichcarrynoload.Thatnotionis consistentwiththeexpectationthat coarsening– whichisalways a part of the microstructural evolution during dealloying – re- quiresligamentstopinchoff.Pinch-off eventscreatedisconnected struts, whichwill persistuntil theirmaterial isredistributed into theload-bearingparts ofthe networkbydiffusion.Yet,theintro- duction of disconnectedstructures is not the only wayin which thestiffnessofanetworkcanbereduced.Relevantchangesinthe topologycanincludedifferencesinthemeannumberofstrutsper node[63],orinthegeometryofload-bearingrings[70].The sig- nificance of those microstructural descriptors for the mechanical behaviorofthe randomnetwork ofNPGremains tobe examined indetail.
As another word of caution, we have pointed out that stud- ies predicting
σ
y based onϕ
lb so farrequired empirical dataforYeff. That approach leaves the issue of a systematic dependence of thescaled genus – and, hence,of the load-bearing solid frac- tion– onthetruesolidfractionand/orthemasteralloycomposi- tionasanopenissue.Astronghypothesisonasystematicdepen- dencebetween g and
ϕ
underlies the modifiedRoberts-Garboczi scaling law of Ref [23], Eq. (3). The law accounts for a system- aticlossinconnectivitywithdecreasingsolidfractioninNPG.Our considerationsontheimplicationofthislast-mentionedapproach for the apparent load-bearing fraction factor,β
are well consis-tent with the data. This suggests that the leveled wave model ofRef. [23] catchesessential aspects of thephysics that leadsto the systematic decrease in stiffness with solid fraction in deal- loyed NPG.That agreement betweenscaling law andexperiment has already been pointed out, with previously publishedexperi- mentaldatafromvariousteamsasthebasis[23].Thepresentwork strengthenstheagreementbyadding,forthefirsttime,adataset withaconsistentsynthesisandcharacterizationprotocol.
5.2. As-dealloyedstate:impactofshrinkageonconnectivity
Shrinkage duringdealloying lets the solid fraction,
ϕ
, enduplarger thanthe non-dissolvedatom fraction,which isideallyx0Au. Asexemplifiedinourstudy,thepresenceornotofshrinkage,and itsnumericalamount,dependonthedealloyingconditions.Specif- icallyourexperimentspointtowardsmoreshrinkageatlesserx0Au, andthisisconfirmedbyourMDstudyofrelaxation.
Remarkably, we find samples that have undergone shrinkage consistentlymorecompliantthansamplesofsamedensitythatdid notexperience shrinkage.Furthermore,wefindthat stiffnessval- uesdofollowsystematictrendswhenplottednot versusthetrue solid fraction,