Journalof Crystal Growth 81(1987)193—204 193North-Holland,Amsterdam

POLAR-ON-NONPOLAR EPITAXY

HerbertKROEMER

Departmentof Electrical and Computer Engineering, University of California, Santa Barbara, California 93106, USA

One of themost fundamentalproblemsthat must be solvedif device-qualityGaAs is to be grown on Si substratesis that ofsuppressingantiphasedisorder. Recentexperimentalevidenceshows that suchdisordercan be suppressednot only on the (211)orientation,but also on (100), contraryto earliertheoreticalexpectations.A detaileddiscussionis given of themechanismby whichthis suppressiontakesplace,througha combinationof slight misorientationand a high-temperaturesurfaceanneal,which lead to thepairingof all Si surfacestepsinto a particularkind of double-heightsteps.A recentmodel by Aspnesand Ihmexplainstheenergeticpreferencefor this kind of stepby postulatinga drastic reconstructionof the atomic configuration at the stepedge through theformationof a ir-bondedchainrunningalongthestep.Another unexpectedpuzzleis posedby therecentobservationthat on a givenSi(100) surfaceantiphasedisorder-freegrowthwith both possibleGa—As sublatticeallocationscanbeachieved,dependingon initialnucleationconditions.A new detailednucleationmodel is proposedthat explainstheseobservations,by drawing heavilyon earlierconsiderationsof Harrisonetal. concerningtheelectrostaticsof a polar—nonpolarinterface.

1. Introduction probablysafeto say that the principal probleminGaAs-on-Si materials quality is no longer the

When a polar (compound)semiconductorlike, polar-on-nonpolarproblem,but themisfit disloca-say, GaAs, is grown on a nonpolar (elemental) tion problem,a discussionof which lies outsidesubstrateof similar crystalstructure,like Ge or Si, the scopeof this paper.Theemphasisof the paperat least three new problems arise that are not will be on the antiphasedisorderandthe interfacepresentin the more conventionalheteroepitaxial neutralityproblem,bothbecausethey arethe mostgrowth of oneIll/V compoundupon another:(a) fundamentalonesfrom a purely scientificpointofthe problem of antiphasedisorder on the com- view, andbecausethe writer’s own work hasbeenpound side of the interface; (b) the problem of concentratingon these.The cross-dopingproblemlack of electricalneutralityat the interface;(c) the is not totally ignored; it is strongly interrelatedproblem of cross-doping.Although the existence with the the interface neutrality problem, andof these problemshas been recognizedfor some some comments about it will be made in thattime, they haveassumedcentral importancewith context.the recent progressin the epitaxial growth ofGaAs on Si substrates,a developmentof poten-tially very largepractical significance. 2. Suppressionof antiphasedisorder

Although the polar-on-nonpolarproblems areby no means the only problems that must be 2.1. Theproblemsolvedif high-quality GaAs-on-Siepitaxy is to beachieved — the misfit dislocation problem is at The diamond structure in which Si and Geleastas severe— a solution of theseproblemsis a crystallize consistsof two interpenetratingface-necessarystep without which this goal could not centeredcubic sublattices.The two sublatticesdif-be achieved. fer from eachotheronly in the spatialorientation

The presentpaperreviewstheprogressthathas of the four tetrahedralbonds that connecteachbeenmaderecently both in the understandingof atom to its four nearestneighbors(which are onthe problemsof polar-on-nonpolarepitaxy,andin the other sublattice). For example,in fig. 1 thetheir solution. As a result of this progress,it is atomswith thebond orientationsindicatedas“A”

0022-0248/87/$03.50© ElsevierSciencePublishersB.V.(North-HollandPhysicsPublishingDivision)

194 H. Kroerner/ Polar-on-nonpolar epitaxy

surfacewill alwaysexhibit steps.At any steponly

k oneatomic layerhigh(or an odd numberof layershigh) the sublattice site allocation of Ga and As

on oppositesidesof the step is interchanged(fig.

B 3),andan APB results.The APBs are structural defects,and we have

little reasonto expect that theymight turn out toFig. 1. Two sublatticesin a Si crystal,distinguishedonly by be the first benign defectsin the history of semi-

bondorientationin space. conductor technology. Antiphase boundariesinGaAs contain Ga—Ga and As—As bonds. Such

and “B” belong to different sublattices.There is bonds representelectrically charged defects: Ano distinction betweenthe two sublatticesother- comparisonof the number of bonding orbitalswise; both are occupied by the same atomic with the numberof valenceelectronsavailabletospecies. fill them shows that Ga—Ga bondsact as accep-

In the zincblende structure, in which GaAs tors, and As—As bondsas donors, with effectivecrystallizes, the two sublatticesare occupied by charges±q/2 per bond. In general,an APB willdifferent atoms,in the caseof GaAs one by Ga contain roughly equalnumbersof both charges,atoms,the otherby As atoms.In a crystalwithout thus acting as an extremelyhighly compensatedantiphasedisorderthe sublattice allocationis the dopingsheetwith very little netdoping.The situa-samethroughoutthe crystal. But if this allocation tion is least bad for an APB that follows exactlychangessomewhereinside the crystal (fig. 2), the anly {11O} plane, as in fig. 2: In that caseGa—Gainterface between domains with opposite sub- and As—As bondswill alternatewithin eachcrys-lattice allocation forms a two-dimensionalstruct- tallographic unit cell, leading to perfect localural defect called an antiphaseboundary (APB). chargecompensation.But for the deviationsfromThe domainsthemselvesare called antiphasedo- this idealizedarrangement,the lack of exact localmains (APDs). chargebalancewill lead to potential fluctuations

Such APBs can be expectedto form when that will affect the electronicproperties.InasmuchGaAs is grown on Si or Ge, especially on a as the initial Si surfacestepsare not likely to have(100)-orientedsubstrate, the most widely used the exactorientationwithin the surfaceplanethatcrystallographic orientation for MBE and would leadto comparativelybenignperfect-{110}MOCVD growth. Inasmuchas As forms strong APBs, the APBs actually resulting from thebonds with Si, whereasGa does not, the firstatomiclayerbondingto the Si substrateshouldbeexpectedto be an As layer. Now, any real(100)

Fig. 2. Antiphaseboundary(APB) formationin the zincblendestructure,containing(in thecaseof GaAs) both Ga—Ga andAs—Asbonds.The configurationshownis thesimplestpossible Fig. 3. Mechanismof APB formation during polar-on-non-one,a perfectly (110)-orientedAPB, with alternatingGa—Ga polar growth dueto the presenceof single-heightstepson the

andAs—Asbonds, substratesurface.

H. Kroemer / Polar-on-nonpolar epitaxv 195

nucleationon a real surfacemust be expectedto sites, displacing Ga atoms to the single back-exhibit local chargefluctuationswith largeampli- bondedsites. Usingcrystallographicetchingtech-tude, andhencebe harmful. niques, we were able to demonstratethat the

There are at least two approachestowardsthe GaAslayersare indeedfree of APBs,andthat theessentiallycompleteavoidanceof APBs: Onein- sublatticeallocationis as statedabove[1—4].volves a switch to a different crystallographicorientationon which APBs do not form in princi- 2.3. Stepdoublingon (100) surfacesplc, even in the presenceof steps, like the (211)orientation employedby us [1—4]and described Most investigators working on GaAs-on-Sibelow. The other is to somehowenforcea perfect growth have preferredto continueto work withdoubling of the height of all surface steps, an the conventional(100) orientation,or with wafersapproachthathasproventenablesinceearly1985, deliberately misoriented from the (100) orien-contraryto all prior expectations. tation by a few degrees,relying on step doubling

for the suppressionof APBs.2.2. The (211) solution When a step on a Si(100) surface is an even

numberof atomic layershigh, the two sublatticesIn our own work we haveusedthe (211)orien- on the GaAs side are in registry again, and an

tation for thegrowth [1—4],ratherthan relying on APB will not occur at this step. Unfortunately,itperfectstep doublingon the (100) surface.On a is well establishedexperimentally that for “as-(211) surface the atomic sites of the two sub- polished” exactly (100)-orientedSi surfacesthelatticeshavea different numberof back bondsto mostcommonstepheight is oneatomic layer [5,6]the Si substrate(fig. 4): Oneof the sublatticeshas and there is in fact ample evidence[7] that nottwo back bonds,the otherhas only one, and this only the growth of GaAson Si, but the growthofdifferenceremainseven in thepresenceofsteps.As otherIll/V compoundssuch as GaP,on exactlya result, the two sublattice sites are no longer (100)-orientedSi or Ge substratesusually exhibitsenergeticallyand chemically equivalent,and dc- copiousAPBs.mentary bonding energy considerationssuggest It was first indicatedby HenzlerandClabes[5]that the As atoms, which have a much stronger that on misorientedSi(100) surfacesthereis tend-tendencythan Ga atomsto bond to Si, will seek ency towards step doubling with increasing an-out themorestronglybindingdoublyback-bonded nealing temperature.This was subsequentlyfol-

lowed up in careful detail by Kaplan [6], who~ reportedthaton Si surfacestilted by a few degrees

from the(100)plane towardsthe(011)plane, moststepsare two atomshigh. Inasmuchas the stepdensity on deliberately misoriented surfacesismuch higher than for accurately oriented (100)surfaces,a certain amount of step doubling is tobe expected,and the step doublingmight be ex-tensiveif thereis a simpleenergeticpreferencefordoublestepsoversinglesteps.However, unlessthenumberof remainingsingle-heightstepsis drasti-cally reduced,such tilting would not aid in thedrastic suppressionof APBs. In any event, it ishardto seehow APBs couldbe avoidedcompletely

Fig. 4. A (211)-orientedpolar-on-nonpolarinterface.Onsucha over theentireareaof an entirewafer: In order tosurface, APBs do not form even in the presenceof steps, .

becausethe two sublatticesdiffer in the way they are back- achieveAPB-free growth, it is necessarythat allbondedto thesubstrate,leading to sublatticecontrolby chem- stepsbe two atomshigh, notjust the majority of

ical bondingpreferences. steps.At first glance, such a propositionappears

196 H. Kroemer/ Polar-on-nonpolar epitaxy

hopeless.Yet it hasbecomeclearsinceearly1985 story, each single-height step on the originalthat such a perfectstep doubling can indeedbe surfacewould pair at randomeitherwith the stepachieved, leading to perfectly APB-free epitaxial to its right or the step to its left. In such a modelgrowth of GaAson Si(100): thereis a 50% probability that the stepto the left(a) Recently, Fischer et al. [8,9] have reported first pairsup with the nextstepevenfurther to thegrowth on deliberately misoriented substrates, left, and a 50% probability that the step to thewhich does indeed appearto be free of APBs, right first pairsup with the next stepeven furtherjudging from the anisotropicetching patternsof to the right, leadingto a (50%)2 = 25% probabilitydevice structureson the epitaxial layers. Aniso- that neither of thesenearest-neighborsteps aretropic etching is one of the simplest and most available for pairing up first with the step ofpowerful techniquesto test for APDs. We have interest,leaving the latterunpaired.In sucha casehadanopportunityto investigateoneof the layers anAPB would form on the averageat every fourthgrown by this group, using our own etch pit initial step, and a misorientationwould increasetechnique[1—3],and we confirm the absenceof the density of APBs rather than decreaseit, de-APDs in that layer. spite the energeticpreferencefor doublesteps.(b) Similarly convincing evidence of APD-free In order to explain the long-rangesuppressiongrowth, basedon an anisotropyof the RHEED of APBs, the left/right randomnessof the steppatternsthat was uniform over the entire wafer pairing mustsomehowbe eliminated.In the pres-area, was presentedby Nishi et al. [10]. The Si enceof double-heightstepsthe surfacelayersonwafers in that work were not deliberatelymison- all terracesbelongto thesameSi sublattice;henceented,but probablyhad a small amount of acci- theremust clearly be a preferencefor one of thedental misorientation. Similar results had been two Si sublatticesover the other. Inasmuchas thereportedearlier by the samegroup for MOCVD- two differentkind of sublatticeplanesdiffer fromgrown GaAson Si [11]. RHEED evidencesimilar eachother only by a 90° rotation in space,theto that of Nishi Ct al., but less direct, had been preferencemechanismcan only resideinside theearlierpresentedby Wang[12], andby hindsight it atomicarrangementat the edges.appearslikely thatWangalso had achievedAPB- At this point it is important to realizethat thefree growth. dangling bond configuration at a step edge de-(c) Perhapsthe most convincing direct evidence pendsnot only on the direction of the step butforperfectstepdoublingalreadyon thepre-growth also on which of the two sublatticesformsthe topSi(100) surfaceis containedin the stunningrecent of the step.The differencebetweenthe two differ-work by Sakamoto and Hashiguchi [13], who ent sublatticesis most pronouncedfor stepsthatshowedthat a nominally (100)-orientedSi surface run along oneof the two KOll) directionswithinwould go from a singly-steppedsurfaceto a dou- the (100) plane,preciselythe kindsof stepsgener-bly-steppedsurfaceduring a prolongedhigh-tem- atedby a tilt aboutoneof thosedirections. It wasperatureanneal(20 mm at 1000°C),with all step pointed out alreadyby Kaplan [6] that thereareterracesbelongingto the samesublattice! then two different kinds of terrace-and-edgecom-

binations possible: (a) “type-A terraces” (my2.4. Stepdoublingmechanism terminology),on the surfaceof which the dangling

bonds of the Si atoms point parallel to the stepTheempiricalobservationthat a Si(100) surface edge,and (b) “type-B” terraceswhose dangling

transformsitself into a single-domainsurfaceupon bondspoint perpendicularlyto the step edge(fig.annealingraisesthe question:How is the possible? 5). Supposenow that there is — for whateverA little reflection shows that perfect long-range reasons— a strongenergeticpreferencefor oneofstep doublingcan never be explainedby the ele- the two bond configurations,say, for that of amentarypropositionthat double stepsare simply type-A step. It was recently pointed out by thisenergeticallypreferred over single steps, without writer [14] thatKaplan’s LEED datastronglysug-any additional assumptions.If that were the whole gestedjust such a preference[15], andthat under

H. Kroemer / Polar-on-nonpolar epitaxy 197~

Fig. 5. Two kinds of bond configurations at [0111-orientedsingle-heightatomic steps.For whatwe call a type-A step, the a

0/2danglingbondsrun parallelto thestep,for the type-Bstep they

runperpendicular. a s : ~

suchconditions,andat sufficiently high tempera- -

tures,atomsfrom type-B edgeswould diffuse to-wards the type-A edges, until the former had ~ ~‘.

simply disappearedby forming double-heightsteps ~, ..~.

that are boundedby edgesof type A. The result -

would be a perfectly doubly-steppedSi surface, [“ol —~‘~ ‘~ [110]

with all terracesbelonging to sublatticeA, andleadingto GaAsgrowth freeof APBs. Fig. 6. Atomic reconstructionproposedby Aspnesand Ibm

The only difficulty with this hypothesisin its [16] for the type-A doublestep,explaimngtheenergeticprefer-ence for double-height type-A steps over all other kinds of

onginal form is that it is hard to see how asteps.Top: unreconstructededge. Bottom: a ir-bondedatomicsufficiently largeenergydifferencecould ansebe- chainis formedalongthestepedges.

tween the two kinds of step edges. Single-bondargumentsfail to predict any energydifference,and it is clear that a significant energetic dif- It thusappearsthat the Aspnes—Ihmmodel is theferemcerequiressome sort of reconstruction,but bestmodel proposedso far to offer an explanationthe writer wasunableto give a specificreconstruc- of why single-domaingrowth of GaAs on Si cantion model. This difficulty hasrecentlybeenover- be achieved.comeby Aspnesand Ihm [16], who havepointedout that at a type-A double-heightstep theatomic 2.5. The role of temperatureandmisorientationconfigurationcanlower its energysignificantly, byabout 40 meV per edge atom, through a drastic An essential ingredient of the step-doublingreconstructionduring which i,--bomdedchainsare hypothesisis a surface temperaturesufficientlyformed, similar to the ir-bonded chainsthat are high for a sufficiently long time to permit diffu-believedto bepresentin the 2 x 2 reconstruction sion of Si atomsfrom the energeticallyunfavora-of the Si(100) surface(fig. 6). The Aspnes—Ihm ble stepsto the favorableones.In Kaplan’s LEEDmodel differs from the earliermodel of ref. [14] in work, theseconditionswere clearly met: the Sithat the energeticpreferenceis not one for type-A surfaceshadbeenheatedto 1100°C,andbecausesingle stepsover type-B single steps,but one for of the deliberatelargemisorientationthe distancedouble-heighttype-A steps only, over all other betweensteps was small. A study of the recentkind of steps,includingsingle-heighttype-A steps. papersreporting reasonablyconvincing evidenceBut the ultimateconsequencesregardingAPB-free of a single-domain surface indicates that in allgrowth are of coursethe sameunderbothmodels, casesthe pre-growthSi surfacewassubjectedto a

198 H. Kroemer / Polar-on-nonpolar epitaxy

high-temperature“heat-cleaning”stepof onekind accidentallyor deliberately.The stepscreatedbyor another,although usually at a lower tempera- such a tilt must necessarilycontain sections thatture and/or for less time than in the case of cannotbenefit from the energy lowering due toSakamotoand Hashiguchi[13]. Although the util- the Aspnes—Ihm mechanism, or whateverotherity of high-temperatureheat treatmentfor surface mechanismmight be present.In such casesonecleaningpurposesis well established,our model shouldexpect that the step geometrythat formssuggeststhat a secondand possibly more im- on the surfacewould run in a zigzag direction,portantfunction is to permit the stepdoubling to with the longerportionsbeing of the energeticallytakeplace.In fact, to achievesucha stepdoubling favored type-A double-stepkind, the shorterkindit may be necessaryto perform the high tempera- of type-B double steps or possibly type-A andture treatmentevenon perfectlycleansurfaces. type-B single steps in close proximity (fig. 7). If

Our model suggestsfurther,andthe experience the stabilizationenergyof the type-Adoublestepsof Nishi et al. [10] and of Akiyama et al. [17] is sufficiently large, APBs would still be sup-confirm it, that a major deliberatemisorientation pressedeven for very large deviation of the tiltis notreally necessary.Its principal benefitwould axis from a (011~axis, exceptpossibly for smallbe to decreasethe distancebetweensurfacesteps regions between any pair of single steps. Em-and therebyto decreasethe time and/ortempera- pirically, this is apparently what happens:ture requiredto achievethe desiredstepdoubling.But it would appearthat, given a sufficiently highannealingtemperaturefor a sufficiently long time, _______________evena small accidentalmisorientationalwayspre-sent might be sufficient to achieve the desiredgoal.

Anotherimportantquestionconcernsthe direc-tion ratherthan the magnitudeof the surfacetilt.In our above discussionwe had assumed,forsimplicity, that the misorientationof the surfaceaway from the exact [100] orientationcorrespondsto a rotation about one of the two (011) direc-tions, thus leadingto stepsthat canline up paral-lel to that direction. It is this assumptionthat ledto two different kindsof stepswith different kindsof dangling bond configurations. The situationwould be quite different for a rotation abouteitherthe [010] or the [001] direction. In that caseboth kinds of steps havedangling bondswhoseprojectionsupon the [100] planerunsat the sameanglerelative to the edgedirection(differing onlyin the sign of that angle), andwhich are energeti-cally equivalentby symmetry.For sucha tilt thereis no mechanismenforcing a coherentstep dou-bling with no remainingsinglesteps,andit hasin [010]

fact been reported[17,18] that GaAs growth on ________ ________ [100]

suchsurfacesleadsto copiousAPBs.In practice,the exactdirectionof tilt will rarely . . . . . .

Fig. 7. Zigzagedgemodel for surfaceswith a tilt axis deviatingagreewith an ideal exact <011) rotation, but will from ~011). Sections of type-A double steps alternatewith

be somewherein betweena favorable<011) and shortersectionsthat might be type-Bdoublestepsor regularly

the unfavorable[010] or [001] orientation,either or irregularlyshapedpairsof single steps.

H. Kroemer/ Polar-on-nonpolar epitaxy 199

Akiyama et al. [17] haveperformedGaAs-on-Si to the Si substratewere a perfect As plane, thegrowth on lens-shapedsurfaces,which presenta two back bonds per As atom would imply acontinuum of both direction and magnitudeof donor-like defectchargeof + q/2 per atom,or atilt, andthey havereportedthat APBs occur only chargedensityof + q/a2, where a is the latticein a narrow band of tilt directionsneara perfect constant.Thisis a verylargecharge(about3 x

10i4

[001]and[010]tilt. donors/cm2).If not neutralized,it would support

A final point that requiresdiscussionconcerns an electric field of about4 X iO~V/cm insidethethe need for a high anneal temperaturebefore growing GaAslayer! On a macroscopicscale,thisnucleation.If a doubly-steppedsurfaceis energeti- chargewould of coursebe neutralizedby mobilecally preferred,thenit shouldbepossibleto create electronsin the conduction band. However, thesucha surfaceby meansotherthana high-temper- neutralizingchargewould extend over an appre-atureanneal, for exampleby a sufficiently weak ciable distanceinto the semiconductor,and thechemicaletch, which might attackthe Si surface field seen by the atoms at the interface itselfonly at the weak steps, but leaves the stronger would bealmostundiminished.stepsalone, until the entiresurfaceconsistsonly As HKWG point out, sucha largefield wouldof the morestablesteps.In fact, the literatureon leadto massiveatomicre-arrangementsduringtheGaAs-on-Si growth contains many reports of high-temperaturegrowthitself, attemptingto neu-seeminglyAPB-free growth following heat treat- tralize the interfacecharge.Onepossibility — notments much gentler than those reported by consideredby HKWG — would be the formationSakamotoet al. [13], suggestingthat a certain of a very largeconcentrationof negativelychargedamountof step doublingmight alreadytakeplace antisite defects (Ga atoms on As sites) on theduring the chemicalpolish and/oretchtreatment GaAsside.Harrisonet al. themselvesproposethatcurrently employed.This is clearly a fertile field the re-arrangementis one of the GaAs/Si inter-for futureexperimentation. faceitself, in sucha way thata significant fraction

of the Si atomsis removedfrom the top Si layerand replacedby Ga atoms,whosebackbondsto

3. Interface atomic sfructiwe and neutrality the Si substratehavethe oppositechargeimbal-anceandhenceneutralizethe As—Si bond charge

In our original discussionof the initial nuclea- [20]. Electrical neutrality would be reachedwhention of GaAs on a Si(100) surfacewe madethe the numberof Ga—Si bondscreatedin this break-simplifying assumptionthat the last Si planewas up of thelastSi planeequalsthe numberof As—Sian unbrokenplane, in which casechemicalbond- bonds, and the authors proposethat the recon-ing argumentsled to the conclusionthat the first struction proceedscloseto this stage.atomic plane on the GaAs side should be an Harrisonet al. considertwo limiting casesofunbrokenAs plane, as inside bulk GaAs. How- idealized atomic arrangements,both of whichever, it was pointed out by Harrison, Kraut, would restorea perfectlyneutral interface,shownWaldrop, and Grand (HKWG) [19] already in in fig. 8. In the first of these,all Si atomsbroken1978 in the context of a GaAs/Geinterfacethat out from the top Si plane(referredto asplaneNo.an atomic configuration composedof unbroken 0 in what follows) are removedfrom the vicinitybulk planesat a polar—nonpolar(100) interfaceis of the interface, to the surface of the growingenergeticallyhighlyunfavorable[4]. Thiscaneasily epilayer.Interfaceneutralityis thenachievedwhenbe understoodby argumentsmoreadaptedto our one-halfof the Si atomsare replacedby eitherGaneeds than those given by HKWG, as follows, or As atoms. In the second arrangement,allRecall that Ga—GaandAs—As bondsare changed broken-outSi atomsremainin plane1, the planedefects,eachcarrying a charge±q/2. The same directly atop the original Si surface.In this case,argumentapplies to Ga—Si and As—Si bonds,ex- neutrality is achievedalreadywhenone-quarterofcept that the defect changeis only half as large, the Si atoms are broken out. The authorspoint±q/4 perbond.If the first atomicplaneadjacent out that the first of thesearrangements,while free

200 H. Kroemer/ Polar-on-nonpolar epitaxi

One would certainly expect that the As-firsthypothesisremainsvalid undernucleationcondi-tions underwhich the Si is first exposedto an Asflux, permitting the formation of copious As—Sibonds, before turning on the Ga flux. But theoutcomeis far less clear if the initial exposureofthe Si surfaceis to Ga rather thanAs, as advoc-atedby us for the (211) orientation.The questionclearly calls for an experimentalanswer.

The questionis readily testedby the etchpat-

tern geometrygeneratedby anisotropicetchedondeliberatelymisoriented(100)surfaces.The orien-tation of the etchpits relativeto the rotation axisdependson the sublattice allocation. Such testshavebeenperformedby Fischeret a!. [8,9], usingtwo different nucleationconditions,by depositingeitheran As prelayeror a Ga prelayerbeforetheactualgrowth. They found that APB-freegrowthcould beachievedin bothcases,but the sublatticeorderingdependedon the natureof the prelayer.Evidently a sublatticeswitch does take place! Intheir most recent work [22], Fischer et al. show

Fig. 8. Two models of atomic re-arrangementproposedby that the sublattice allocation also changeswithHarrison et al. [19] to achieve electrical neutrality at the nucleationtemperature,underwhat are implied tointerface,both involving theremoval of Si atoms from thetop be otherwiseunchangedconditions: For nuclea-Si layer. Top: single transitionlayer model, requiring removal

lion at low temperaturesof 450—500°C they findof one-half the Si atoms,without their re-incorporationintothe GaAsnear the interface.Bottom: doubletransitionlayer APB-free growth with one particular sublatticemodel, requiring the removal of one-quarterof the Si atoms, ordering. There follows a temperaturerange (ofand their re-incorporationinto the first atomic layer on the unstatedwidth) inside which copious APBs are

GaAsside; thisconfiguration hasthelower energy. observed.Above a certain(unstated)temperature,

APB-free growth is again achieved,but with asublatticeorderingoppositeto thatat low temper-atures.

of a net electric charge, still carries a residual Theauthorsinterprettheir observationsin termselectric dipole, whereasthe secondarrangementis of a model in which the first atomicplanefollow-freeof both, andhencerepresentsa stateof lower ing an unbrokenSi surfaceis eitheran unbrokenoverall electrostaticenergy. As planeor an unbrokenGa plane, dependingon

At this point an interesting questionarises.If nucleationconditions.They suggestthat the switchthe HKWG re-arrangementtowardsanessentially from As to Ga for the first layer is simply aneutral interfaceshould indeedgo to completion, consequenceof the loss of As by evaporationator at least near-completion,then therewould be highertemperatures.no longer any energeticpreferencefor this first We would not wish to rule outan unbrokenAsplane after the original Si surface to be an As plane model for the As-dominatedcase,despiteplane, and hence not for a specific sublattice the electrostaticargument that speakagainst it.allocationfor Ga andAs on the GaAsside! Which But the formation of a simpleunbrokenGa planeof the two sublatticeswould be which would then bondedto anunbrokenSi planeas a result of Asbe decidedby the kinetics of the nucleationpro- loss by evaporationis extremely unlikely, on thecess. purely chemical grounds of the very different

H. Kroemer/ Polar-on-nonpolar epitaxv 201

strengthof Ga—Si and As—Si bonds.In our own [011] Side View:

work on the growth of GaP on Si we found thatPlane #

during the thermal decompositionand desorptionof a GaP film the last Ga would evaporatelong ,,~—I’

before the last phosphorous[23]. One would ex- 0

pect the sameto be true for GaAs on Si, andrecentwork by Bringanset al. [24] strongly sup-ports this expectation.Hencethe formationof anunbrokenGa plane bonded to an unbroken Sisurfaceis extremelyunlikely so long as there is [100] Downward View:

any As presentat all.Theobservationsof Fischeret al. [8,9,22]clearly o 0 0

call for a different explanation.In the next sectionwe propose a mechanismfor the Ga-dominatednucleationmode that leadsto the observedfinal 0 0result, but from diametrically oppositeinitial as-sumptions.

0 0 0 0 C)4. Proposednucleation model

4.1. Ga-dominatednucleation.’ the As—Si site ex- Si Ga Plane #1

changepostulate 0 Si • As Plane #0

We makethe following two initial postulates: . .

Fig. 9. Proposedfirst stage of the nucleationof GaAs on Si(a) Arnving Ga atoms bond to Si atoms only under Ga-rich conditions: an incoming As atom interchanges

when at least one Ga—As bond can be formed sites with a Si atom in plane 0 (the original top Si plane),along with every Ga—Si bond. The idea behind simultaneouslybonding two “waiting” Ga atoms.The ejected

this postulateis that the 1/4 electronexcessof the Si atom is placedon an adjacentsite in plane 1. Top: [011]

Ga—As bond is transferredto the Ga—Si bond, view; bottom: [100](i.e. downward)view.

whereit helps forming thelatter bond.(b) Even the formationof As—Si bondsis facili-tatedif at the sametime Ga—Si bondsare formed, Note that in this initial nucleationstep twoto takeup the electronexcessof the As—Si bond. As—Si bonds and two Ga—Si bondsare formed;

Thesepostulatesleadto the ideathat the initial hencethe nucleusis an electricallyneutralobjectnucleationof GaAson Si doesnot simply takethe in the senseof the work of HKWG. The numberform of eitherAs or — much less likely — Ga first of Si—Si bonds does not changeduring the sitebondingto the original Si surface, especiallynot exchange.in the presenceof a sufficiently largenon-bonded If we maketheplausibleassumptionthat the Siand hence mobile Ga concentrationon the Si atom ejectedfrom plane0 is simply placedon onesurface. Instead, we therefore make the central of the adjacentsitesin plane1, the site in plane2postulate that: that hasbackbondsto both this Si atom and to(c) the initially arrivingAs atomswill undergoan the adjacentGa atom,will form a naturalbondingexchangereaction during which a Si atom from site for anotherAs atom,andan As—Gapair willthe last Si plane(plane0) exchangessiteswith an connect between the remaining bond of the Siarriving As atom, with two Ga atoms simulta- atom in plane 1 and a Si atom in plane0, asneously bonding to both the As atom and two shown in fig. 10. During this secondnucleationadjacentSi atomsof the original surface,asshown step again two As—Si and two Ga—Si bonds arein fig. 9. formed.

202 H. Kroemer/ Polar-on-nonpolar epitaxy

(011] Side View: did take place, the resulting composition of thePlane # differentplanesadjacentto the interfacewould be

2 exactlywhat is demandedby the HKWG 2-layer1 reconstructionmodel.0

4.2. TheAs-dominatedcase

The experimentsof Fischeret al. [8,9,22]makeit clear that under lower-temperatureAs-do-

[100] Downward View: minatedconditionsthe As—Si site exchangepos-

tulatedfor Ga-dominatednucleationdoesnot oc-

o cur. Thereasonfor this is probablyoneor bothofthe following: eitherthe presenceof an excessof

~ • I.,J Ga may be necessaryto drive the exchange,or theo o temperatureis too low for the exchangereaction

to overcomean energeticreactionbarrier that is

very likely present,for examplethe As2 dissocia-

o o o o tion barrier. Quite possiblyboth mayplay a role.Nor is it clear whether or not the HKWGinterfacereconstructionprocesstakesplaceat the

As Plane #2 lower temperaturesand, if not, exactly how elec-trical neutrality is subsequentlyestablished— if

Si Ga Plane #1 indeedit is. Quite possibly the HKWG mecha-

• Si • As Plane #0 nism occurs even then, but with the Si atomsejectedfrom plane0 now replacedby Ga rather

Fig. 10. Proposedsecondstageof thenucleationof GaAson Si than As atoms,becauseof theprior formationof aunder Ga-rich conditions: two As atoms bond to the Si atom tightly-bondedpartial As coveragein plane1. Anin plane1, and a Gaatom closesthebondingloop of theouter alternatepossibility would be that the top Si plane

of thetwo As atoms to theSi surface. . . .

remainsintact, but is neutralizedby the formationof a very high concentrationof Ga-on-As site

It is evident that this nucleationprocessauto- antisite defects,as mentionedearlier. Or maybematically leads to a placementof all Ga atoms the interfaceremainshighly charged,being neu-into plane 1 andof all As atomsinto planes0 and tralized only by mobile electronsin the conduc-2, theoppositechoicefrom whatwould be present tion band. This is evidently anotherfertile fieldif the As—Si siteexchangedid not takeplace,but for future research.exactly the order observedby Fischeret a!. undernucleationconditionsof sufficientlyhigh tempera- 4.3. Residualdefectsture and Ga flux. There canbe little doubt thatany subsequentlateral spreadingof eachnucleus The high-temperatureGa-dominatednuclea-thusformedwill retainthis sublatticeorder. This tion model presentedaboveis an idealization. Inspreadingcan proceedeither by adding As—Ga practice, one must expect numerousdefects topairs,or by adding two As atomsandperforming occur,especiallythe following three:anAs—Si siteexchange.Becausethe two processes (a) Occasionally,As atomsmay endup in plane1,lead to oppositebond charges,a combinationof by bondingto Si plane0 without undergoingtheboth may be expectedto takeplace,to aid in the site exchangereaction. So long as these atomsestablishmentof electrical neutrality, but it is remainin the minority, they would simply appeardoubtful that perfectneutralizationwill takeplace. as local antisitepoint defects(double donors)on

We notehoweverthat, if perfectneutralization what is otherwisea Ga sublattice,without causing

H. Kroemer/ Polar-on-nonpolar epitaxy 203

actual finite-sizeantiphasedomains. 5. Conclusions(b) As mentionedalready,in those areasof theinterface where the growth proceedsby lateral With the 1985 emergenceof convincing evi-spreadingfrom the initial nuclei, the neutraliza- dencethat APBs in GaAs-on-Si(100) growth cantion of As—Si bondsby Ga—Si bondsformedmay be suppressed— and the understandingof thebe incomplete, leading to a net doping of the suppressionmechanismpresentedhere— the mostinterface, which may be either donor-like or urgentnext problembecomesthat of the suppres-acceptor-like,dependingon the exact details of sion of the propagationof misfit dislocation,espe-theprocess. cially for minority carrier deviceapplicationssuch(c) Finally, some of the Si atoms removedfrom as lasers.For heterostructureFETs, the interfaceplane 0 might not be incorporatedinto plane 1, chargeand the relatedcross-dopingproblem maybut are taken up by the growing GaAs bulk be of similar importance.With bothproblems,theinstead. On purely thermodynamicgroundsone long-term issue is not just to make “good” de-should expect at leastthe GaAslayersclosestto vices, but to do so without having to resort tothe interfaceto be Si dopedto the thermodynamic thick buffer layers.solubility limit. The achievementof excellentAPB suppression

Additional Si atoms may accumulateon the on the (100) orientationmakestheswitch to (211),GaAs surface, from where they are gradually long advocatedby us [1—4],a less urgent one.incorporated into the growing GaAs as bulk However, the (211) orientationnot only remainsadopant. perfectlyviableoption, butwe continue,in fact, to

Becauseof thesevarious kinds of chargedde- believe that the long-term potential of the (211)fects, the GaAs-on-Si(100) interfacewill almost orientationis still the betteronefor many devices.certainlybe one with a residualinterfacecharge Most of the nucleationandinterfacechargeprob-sufficiently large that it cannot be ignored for lems discussedin sections2—4 of this paperper-devicepurposes,combinedwith a heavySi doping tamedalmostexclusively to the problems of theof at least the near-interfaceregionof the crystal. (100) orientation, and are absent on the (211)The extent to which these two effectswill take orientation. Not only is thereno problem aboutplacewill dependstronglyon detailsof the exact the sublatticeallocationsduring nucleation,theregrowth procedure. also is no natural interfacecharge: on a perfect

Up to a point, thesedefectsare largely incon- (211) surface, the numbersof As—Si and Ga—Sisequential, so long as they remain confined to bonds are exactly the same(see fig. 4), leadingwithin a few atomic monolayersof the original naturally to an electrically neutral interface.Theinterface,and do not havea deleteriouseffect on lack of a charge implies the lack of an electricthe quality of subsequentlayers.The defectstruc- field driving any atomic re-arrangement;henceture neartheinterfaceis likely to be dominatedby any takeupof Si by the growing GaAsshouldalsothe very large densityof misfit dislocationsthere, be much weaker.This has indeedbeenobserved:comparedto which the other defectsare a corn- In their work on the MBE growth of GaP on Si,parativelyminor disturbance.Largely becauseof Wright et al. [2] found that the uptakeof Si by athe misfit dislocations,the GaAs/Si interfaceit- (211) layer was much less than that by a (100)self is not likely to be usableas a part of the layer grown side by side. All of these could be“intrinsic” device for most devicesundercurrent sizeableadvantagesof the (211) orientationoverconsideration,and its short-rangepropertiesare the (100)orientationfor applicationsin which thethereforenot of primary concern.Apart from the near-interfacequality, and especiallya low neardislocations,the defect making itself felt farthest interfaceimpurity uptake of the GaAs are im-from the interfaceis probably Si uptake by the portant.growing GaAs, and its suppressionprobably de- Probably the biggest unknown in the (100)servesthe highestpriority afterthe suppressionof versus(211) competitionis the behaviorof misfitthe propagationof misfit dislocations, dislocations on the two orientations.Nothing is

204 H. Kroemer / Polar-on-nonpolar epitaxy

known yet about differences between the two Ericksonand R.Youngman,AppI. Phys.Letters47 (1985)

orientations in this regard, but the differencescould be severe,andthey could decidethe issue. [9] Ri. Fischer, N. Chand, W.F. Kopp, C-K. Peng, H.Morkoc, KR. Gleasonand D. Scheitlin, IEEE Trans.

Electron DevicesED-33 (1986) 206.[10] 5. Nishi, H. Inomata, M. Akiyama and K. Kaminishi,

Acknowledgements Japan.J. Appl. Phys.24 (1985) L391.[11] M. Akiyama, Y. Kawaradaand K. Kaminishi, Japan.J.

The writer gratefully acknowledgescontribu- Appl. Phys.23 (1984) L843.

tions from many individuals. Dr. B.A. Joyce [12] WI. Wang,Appl. Phys.Letters44 (1984) 1149.

(Philips)wasa discussionpartnerduring the early [13] T. Sakamotoand G. Hashiguchi,Japan.J. AppI. Phys.25(1986) L57.

phasesof this work. Dr. W.I. Wang(IBM) wasthe [14] H. Kroemer, in: Heteroepitaxyon Si Technology,Materi-

first to try to persuademe that APB-freegrowth als ResearchSociety Proc. 1986 Spring Meeting, in the

could indeed be achieved on mis-oriented press.

substrates,followed by ProfessorH. Morkoc and [15] In ref. [6], Kaplanhimself writes: “On relatively low stepdensity,i.e. accuratelycut (100) crystals,regionsterminat-

Dr. R.J. Fischer.(Universityof Illinois). Drs. M.ing on thedifferent sublatticesoccur with equalprobabil-

Akiyama and S. Nishi (Oki Electric Co.) and ity. This should be truealsoof high stepdensity vicinal

especially Dr. T. Sakamoto (ETL-Tsukuba) surfaces,unless the surface energy is highly sensitive to the

completedthat persuasion.To ProfessorMorkoc dangling bond configuration relative to the steps” [emphasis

and Dr. Fischer additional thanks are due for mine]. Kaplandid not follow up this remarkany further;his datashowthattype-A terracesmu.st bepresent,but he

intense discussions,for providing a wafer to test fits his LEED datato a model that assumesboth kindsof

for freedomfrom APBs, andfor making valuable terracesto be present,without stating whetherthe data

information available prior to publication. Dr. could be fitted just as well or better by an “A-only”

D.E. Aspnes(Bellcore)resolvedmanypuzzleswith model. In ref. [14] thepresentwriter arguesthat onecould

a preprint of his work. My formerco-worker,Dr. have concluded already from Kaplan’s data that there

P.N. Uppal,now at Martin Marietta,participated must be a strong energeticpreferencefor type-A stepspresent.in many discussions;he also performedthe tests [16] D.E. Aspnesand J. Ihm, to be published.

for APBs on the Illinois wafer. Dr. E.A. Kraut [17] M. Akiyama, K. Kawarada,S. Nishi and K. Kaminishi,

(Rockwell) served as a patient soundingboard 1986 SpringMeeting, Materials ResearchSociety.

throughoutthis work. Last but not least,much [18] S. Sakai, T. Soga, M. Takeyasu and M. Umeno, 1986

gratitudeis dueto the US Army ResearchOffice SpringMeeting,Materials ResearchSociety.[19] WA. Harrison,E.A. Kraut,JR.WaidropandR.W.Grant,for supportingthis work. Phys.Rev. B18 (1978)4402.

[20] Our discussionhereand in the rest of this sectiongoes

beyond HKWG in specifically assumingthat the firstReferences atomic plane above the original Si surface is, at least

initially, an As plane.The argumentsin Harrisonet al.,

[1] S.L. Wright, M. Inadaand H. Kroemer, J. Vacuum 5~i. apartfrom discussingGaAs-on-Gerather thanGaAs-on-Technol. 21(1982)534. Si, applyindependentlyof whichof thetwo sublatticeson

[21 S.L. Wright, H. KroemerandM. Inada,J. Appl. Phys.55 theGaAsside is which, so long asthe two differentkinds(1984) 2916. of atomsoccurin alternatingplanes.

[3] P.N. Uppal and H. Kroemer, J. Appl. Phys. 58 (1985) [21] R.J. Fischer,W.T. Masselink,J.KIem, T. Henderson,T.C.2195. McGlinn, MV. Klein, H. Morkoc, J. Mazur and J.

[4] For a review, see also: H. Kroemer, Surface Sci. 123 Washburn,J. AppI. Phys.45 (1985) 374.(1983) 543. [22] R. Fischer,H. Morkoc, C. Choi, N. Otsuka, M. Longer-

[5] M. Henzler andJ. Clabes,in: Proc. 2nd Intern. Conf. on boneandL.P. Erickson, to bepublished.Solid Surfaces,Kyoto, 1974 [Japan.J. AppI. Phys.Suppl. [23] S.L. Wright, PhD Dissertation,University of California,2, Part 2 (1974) 389]. SantaBarbara,CA (1982), unpublished.

[6] R. Kaplan, SurfaceSci. 93 (1980) 145. [24] RD. Bringans, R.I.G. Uhrberg, MA. Olmstead, R.Z.[7] For extensivereferencessee Uppal and Kroemer[3]. Bachrach and i.E. Northrup, 1986 Spring Meeting,[8] Ri. Fischer,NC. Chand,W.F. Kopp, H. Morkoc, L.P. Materials ResearchSociety.