Ohmic contacts to 2D semiconductors through van der Waals bonding

4.5 MB·

by  M. Farmanbar

Ohmic contacts to 2D semiconductors through van
der Waals bonding
6
1
0
2 Mojtaba Farmanbar,∗ and Geert Brocks

n
a
J FacultyofScienceandTechnologyandMESA+InstituteforNanotechnology,Universityof

9
Twente,P.O.Box217,7500AEEnschede,TheNetherlands
]
l
l
a
h E-mail: [email protected]
-
s
e
m
. Abstract
t
a
m
HighcontactresistanceshaveblockedtheprogressofdevicesbasedonMX (M=Mo,W;
2
-
d
X=S,Se,Te)2Dsemiconductors. InterfacestatesformedatMX /metalcontactspintheFermi
n 2
o
level, leading to sizable Schottky barriers for p-type contacts in particular. We show that (i)
c
[
one can remove the interface states by covering the metal by a 2D layer, which is van der
1
v
Waals-bonded to the MX layer, and (ii) one can choose the buffer layer such, that it yields
3 2
6
1 a p-type contact with a zero Schottky barrier height. We identify possible buffer layers such
2
0 as graphene, a monolayer of h-BN, or an oxide layer with a high electron affinity, such as
.
1
0 MoO3. ThemostelegantsolutionisametallicM(cid:48)X(cid:48)2 layerwithahighworkfunction. ANbS2
6
1 monolayeradsorbedonametalyieldsahighworkfunctioncontact,irrespectiveofthemetal,
:
v
whichgivesabarrierlesscontacttoallMX layers.
i 2
X
r
a
Introduction
Layered transition metal dichalcogenides MX , M = Mo,W, X = S,Se,Te, are widely explored be-
2
cause of their unique properties and their potential for applications in electronic devices.1,2 MX
2
∗Towhomcorrespondenceshouldbeaddressed
1
monolayers are direct band gap semiconductors with band gaps in the range of 1-2 eV, which
haveappealingelectronicandoptoelectronicproperties.3,4 MX layerscanbeobtainedviamicro-
2
mechanical cleaving,5 by chemical vapor deposition (CVD),6,7 or even by spin coating precursor
molecules.8 Important for applications in devices is the ability to have both electron (n-type) and
hole (p-type) transport in these 2D materials. Charge carrier transport in MX field-effect tran-
2
sistors (FETs) is usually dominated by electrons; p-type transport has only been demonstrated in
WSe .9,10
2
A major challenge for p-type transport is that MX forms a large Schottky barrier (SB) for
2
holes with metals commonly used for making electrical contacts. A standard way to reduce a
metal/semiconductorcontactresistanceistoheavilydopethesemiconductorinthecontactregion,
which effectively decreases the SB width. Local doping of a 2D semiconductor is however very
challenging;sofarmosttechniquesusedfordoping2Dmaterials,suchassubstitutionaldoping,11
adsorbed molecules,9,12,13 or electrolytes,14,15 have a limited spatial resolution. Alternatively one
tries to decrease the SB height, essentially by covering the metal by a thin layer to increase its
work function. Oxides have shown their potential for p-type contacts in organic photovoltaics and
light-emitting diodes,16,17 and have also been tested in MoS FETs.18,19 Oxides have also been
2
appliedsuccesfullytoreducetheSBheightforn-typecontactstoMoS .20
2
Common metals generally give n-type contacts with substantial SB heights, leading to high
contact resistances. Although MX monolayers are free of dangling bonds, nevertheless they in-
2
teract with low work-function metals to form a density of interface states with energies inside
the MX band gap, which is sufficiently large to pin the Fermi level and cause a sizable SB for
2
electrons.21–23 We show that high work-function metals yield high SBs for holes by a similar
mechanism.
We suggest a practical way to solve the p-type contact problem and tune the SB height by
inserting a monolayer of a 2D material between the metal substrate and the MX semiconductor,
2
seefigure1. Thebufferlayersuppressesthemetal/MX interfacestates. 2Dmaterialshavecertain
2
uniquepropertiesnotfoundinbufferlayersof3Dmaterials.5 Astheinterlayerbondingisvander
2
Waals,neitherthestructureofthe2Dbufferlayer,northatoftheMX semiconductor,isperturbed
2
significantly by stacking them. The 2D buffer layer need not be lattice matched to the metal or
to the MX layer, and the structure of the multilayer will in general be incommensurate. Van der
2
Waals interface bonding also promises the absence of interface states. Covering the metal by an
adsobant layer such as graphene, a monolayer of hexagonal boron nitride (h-BN) or T-MoS , has
2
proved to be beneficial for making n-type contacts.21,24 We show that a 2D buffer layer can be
selectedtoobtainazeroSBheightforholes.
A h-BN monolayer is a buffer layer that can be used for making n-type contacts, because
adsorption of h-BN on a metal decreases its work function by up to 1-2 eV.28 For instance, Co/h-
BNandNi/h-BNarepredictedtoformzeroSBheightn-typecontactstoMX semiconductors,see
2
figure 2.21 A decrease of the work function is unfavorable for making p-type contacts. Metal/h-
BN gives a p-type contact to MX only if the metal work function is sufficiently high, and the
2
MX ionization potential is sufficiently low. We find a zero SB height for Pt/h-BN, and Au/h-BN
2
contacts to MoTe and a low SB for Pt/h-BN/WSe , see figure 2. Alternatively, one can use a
2 2
graphenebufferlayer,29,30 whosebehaviorisqualitativelysimilartothatofah-BNmonolayer.
To make more universely applicable p-type contacts, one needs a buffer layer that effectively
increases the metal work function. Oxides such as MoO are an option. The MoO structure
3 3
consistsofbilayers,makingitconceivabletocoverametalwithasingleMoO bilayer,seefigure1.
3
We find that the electron affinity of MoO is sufficiently high to make carrier transport through its
3
conductionbandpossible,sothatabilayerofMoO doesnotpresentatunnelbarrier.
3
Averyinterestingoptionforcreatingp-typecontactsistousemetallicM(cid:48)X(cid:48) bufferlayers,such
2
as NbS or TaS .31 Their structure is similar to that of semiconducting MX , they are chemically
2 2 2
stable, and have work functions close to 6 eV. We show that a monolayer NbS adsorbed on a
2
metal gives a SB with zero height for contacts to all MX . The initital work function of the metal
2
substrateisirrelevant;Au/NbS andAl/NbS essentiallygivethesamecontact.
2 2
3
Results and Discussion
Van der Waals bonded contacts
The obvious way to make a p-type contact to a semiconductor is to use a metal with a high work
function. The calculated work function of Pt is 5.91 eV, suggesting that this metal should give a
zeroSBtoallMX ,exceptMoS . Inpracticethisisnottrue,asMX interactswithPttogivestates
2 2 2
at the interface whose energy is within the band gap of MX . Initially it was thought that MX
2 2
could escape the formation of interface gap states (IGS), as, unlike conventional semiconductors
such as Si, MX has no dangling surface bonds that interact strongly with the metal surface.32
2
Nevertheless, even a relatively weak interaction yields IGS that pin the Fermi level in the gap,
whichresultsinanappreciableSB.21
As an example, figure 3(a) gives the band structure of the Pt(111)/MoTe interface. The direct
2
interaction between MoTe and the Pt surface perturbs the band structure of MoTe significantly,
2 2
the valence bands in particular. The perturbation is accompanied by the formation of IGS inside
theMoTe bandgap,figure3(b),whichpintheFermilevel. TheSBheightisdefinedas
2
Φ =E −E , (1)
p VB F
withE theFermienergy,andE theenergyofthetopofthevalenceband(measuredasdistances,
F VB
i.e., positive numbers, from the vacuum level). The SB to an electronically perturbed overlayer is
of course not extremely well-defined. One can estimate the SB by aligning the core levels of the
adsorbed MoTe layer to those of a free-standing MoTe layer, which gives Φ = 0.49 eV. With
2 2 p
E =5.04 eV (figure 1) this givesW =4.55 eV as the work function of of Pt covered by
VB M|WTe
2
a MoTe monolayer. As the calculated work function of clean Pt(111) is 5.91 eV, it implies that
2
adsorbingMoTe onPtcreatesalargepotentialstepattheinterfaceof1.36eV.
2
The changes in the electronic structure of the adsorbed MoTe layer are visualized in fig-
2
ure 3(b). Starting from the total density of states (DoS) of the Pt(111)/MoTe system, and sub-
2
4
tractingtheDoSofthecleanPt(111)slab,onecancomparetheresulttotheDoSofafree-standing
MoTe layer. The comparison shows considerable differences in the MoTe band gap region,
2 2
whicharedirectevidencefortheformationofIGS.
ThepatternoftheseIGSdependsontheparticularmetal/MX combination. Wehavehowever
2
notfoundanelementalhighworkfunctionmetalthatdoesnotgiveIGS.Forallhighworkfunction
metal/MX contacts we have studied, IGS are formed that pin the Fermi level and yield a sizable
2
SB. The same problem has emerged previously for low work function metals and n-type SBs.
Introducing a buffer layer can break the interaction between MX and the metal and eliminate the
2
IGS. This layer must be sufficiently thin, such that it does not form a large barrier for the charge
carriers. Inaddition,theinteractionbetweenMX andthebufferlayermustnotcreatenewIGS.
2
A single atomic layer of graphene or h-BN obeys these criteria. Such a layer presents only a
thinbarrierthatessentiallyallowsformetallictransportthroughthelayer.33 Insertinggrapheneor
ah-BNmonolayerbetweenametalsurfaceandMX disruptsthemetal-MX chemicalinteraction,
2 2
anddestroysanymetal-inducedIGS.Grapheneorh-BNbindtoMX viavanderWaalsforces. One
2
does not expect such an interaction to create new IGS. This is illustrated for Pt(111)/h-BN/MoTe
2
in figures 3(c) and (d). Inserting a monolayer of h-BN restores the electronic structure of WTe ,
2
where the projected bands are essentially those of free-standing MoTe . The DoS of Pt(111)/h-
2
BN/MoTe minus that of Pt(111) is essentially identical to the DoS of a free-standing MoTe
2 2
layer, in particular in the gap region. In other words, it shows no sign of IGS generated by any
h-BN/MoTe interaction.
2
The concept also works if one uses graphene to cover the metal. The interaction between
metal/graphene or metal/h-BN and MX is van der Waals, so the electronic structure of any MX
2 2
is preserved, and IGS are absent. A serious drawback however is that adsorption of graphene or
h-BNgenerallydecreasesthemetalworkfunctionconsiderably,e.g.,by0.6-1.1eVforPtandAu,
see table 1. The reduction originates from a dipole layer that is formed at the metal/graphene or
metal/h-BNinterface,wherePauliexchangerepulsiongivesandominantcontribution.34
The reduction is partly canceled by a potential step ∆V formed at the graphene/MX or h-
2
5
Table 1: Calculated work functions W of metal/graphene, metal/h-BN and metal/MoS sub-
2
strates; potential steps ∆V at graphene/MX , h-BN/MX and MoS /MX interfaces; SB heights
2 2 2 2
Φ (eq. ??). All values in eV; all MX in the H-structure. The calculated work functions of clean
p 2
Pt(111)andAu(111)surfacesare6.04and5.58eV,respectively.
MoSe WSe MoTe WTe
2 2 2 2
W ∆V Φ ∆V Φ ∆V Φ ∆V Φ
p p p p
Pt/Gr 4.86 0.21 0.42 0.21 0.21 0.28 0 0.28 0
Pt/h-BN 5.00 0.18 0.31 0.18 0.10 0.24 0 0.24 0
Pt/MoS 5.32 0.14 0.03 0.14 0 0.18 0 0.18 0
2
Au/Gr 4.80 0.21 0.48 0.21 0.27 0.28 0 0.28 0
Au/h-BN 4.88 0.18 0.43 0.18 0.22 0.24 0 0.24 0
Au/MoS 5.05 0.14 0.30 0.14 0.09 0.18 0 0.18 0
2
BN/MX interface, see figure 4. Although the weak interaction between h-BN and MX does
2 2
not give IGS, it does lead to an interface potential step, which originates from the Pauli repulsion
between the electrons from the h-BN layer and those originating from the MX layer.34 As the
2
chalcogenideatomsformtheouteratomiclayersinMX ,itisthennotsurprisingthat∆V depends
2
on the chalcogenide species, but not on the metal species of the central atomic layer, see table 1.
For graphene/MX and h-BN/MX interfaces ∆V is positive toward MX i.e., it decrease the SB.
2 2 2
Thestepsareactuallyquitemoderate,i.e. intherange0.2-0.3eV,seetable1,sotheydonotcancel
outtheworkfunctionreductionsdiscussedinthepreviousparagraph.
One obtains a zero SB height only with graphene- or h-BN-covered metals with a high work
function, such as Pt and Au, in contact with MX that has a sufficiently low ionization potential.
2
TheSBheightcanbecalculatedfromthenumbersgivenintable1andfigure1
Φ =E −W −∆V. (2)
p VB
AsbydefinitionΦ ≥0,anegativenumberindicatesazeroSBheight,Φ =0. Onlythetellurides
p p
MoTe and WTe satisfy this criterion. WSe gives a small SB of 0.10 eV with Pt/h-BN but the
2 2 2
other selenide monolayers give appreciable SB heights in the range 0.2-0.5 eV. The sulfides (not
shownintable1)havelargeSBswithheights∼1eV.
OnecanhoweverexpectthatthesituationbecomesmorefavorableforMX multilayersasthe
2
6
bandgapofmultilayerMX issmallerthanthatofaMX monolayer. ForMoS ithasbeenargued
2 2 2
thatthebandgapreductionismonotonicinthenumberoflayersandthatitisequallydividedinto
an upward shift of the valence band and a downward shift of the conduction band.35 This is likely
toholdmoregenerallyforallMX compounds. IndeedanexplicitcalculationofbilayerWSe on
2 2
Pt/h-BNgivesaSBthatiszero. ItimpliesthatvanishingSBstomultilayerWSe arepossiblewith
2
graphene-orh-BN-coveredhighworkfunctionmetals.
Adding additional h-BN layers on top of a h-BN/Au or h-BN/Pt substrate does not change the
SB height, as the potential steps formed at the interfaces between the h-BN layers are negigibly
small. By itself h-BN is an insulator forming SBs for holes with heights 0.9-1.1 eV and 1.1-1.3
eV with Pt and Au, respectively.28 Therefore, h-BN acts as a tunnel barrier between Pt or Au and
MX . A single h-BN layer forms only a thin barrier that is very transparent,33 but the contact
2
resistanceisexpectedtogrowexponentionallywiththenumberofh-BNlayers.36
The principles outlined above can be extended to other buffer layers besides graphene or h-
BN. For instance, adsorbing a single MoS layer on a high work function metal such as Au or Pt
2
reduces its work function by 0.5-0.7 eV. That still leaves us with a substrate that has a sufficiently
high work function to yield p-type contacts to the tellurides with a zero SB height, and a small to
zero SB height to the selenides, see table 1. As the Fermi level for MoS adsorbed on Au or Pt is
2
close to the middle of the MoS band gap,21 the MoS layer then acts as tunnel barrier between
2 2
the metal and the MX layer. Because a MoS monolayer is thicker than graphene or h-BN, one
2 2
expectsittopresentmoreofabarrier,andyieldahighercontactresistance.
High electron affinity oxide layers
A buffer layer that effectively decreases a metal work function limits the scope for using it to cre-
ate p-type contacts. A buffer layer that increases the work function would be more advantageous,
whichrequiresalayerthatacceptselectronsfromtheunderlyingmetal. OxidessuchasMoO and
3
WO are known for their potential as p-type contacts in organic photovoltaics and light-emitting
3
diodes,16,17 andhavealsobeentestedinfield-effecttransistorsbasedupnMX .18,19 Morespecifi-
2
7
cally,α-MoO (thethermodynamicallystablephaseofMoO )isalayeredmaterial,whichconsists
3 3
of covalently bonded bilayers, see ôR´ˇrG˘figure 1, that are van der Waals stacked. MoO is a large
3
band gap material with an experimental band gap of 3.0 eV. The electron affinity of this material
is however an exceptionally high 6.7-6.9 eV.16,17 Therefore, MoO is predicted to be an electron
3
acceptorwithrespecttocommonmetals.
Adsorbing a MoO -bilayer on a common metal leads to a transfer of electrons from the metal
3
to the MoO , and sets up a dipole layer that effectively increases the work function, provided that
3
the adsorption process does not destroy the structure and integrity of the MoO overlayer. For
3
instance, the calculated electron work function of Cu(111)/bilayer-MoO is 7.08 eV, as compared
3
to 4.98 eV for the clean Cu(111) surface. Moreover, the density of states of the MoO conduction
3
band is sufficiently high such as to pin the Fermi level at the bottom of the conduction band, see
figure5.
A MoO bilayer has no dangling bonds. A MX layer adsorbed on MoO is therefore likely
3 2 3
to be van der Waals-bonded. Indeed from our calculations we find a small MoO -MoS equilib-
3 2
rium binding energy of 0.17 eV per MoS formula unit, and a large equilibrium bonding distance
2
(between to top layer of oxygen atoms and the bottom layer of sulfur atoms) of 3.1 Å. The weak
MoS /MoO bondinghaslittleinfluenceontheelectronicstructureofeitherlayer. Anymetalthat
2 3
can be covered by bilayer-MoO without disrupting its structure should then give a work function
3
W thatissufficientlyhightogiveazeroSBheighttoallMX ,seefigure2. AsW >E electrons
2 VB
aretransferredfromMX toMoO ,therebypinningtheFermilevelatthetopoftheMX valence
2 3 2
band,aswellasatthebottomoftheMoO conductionband,seefigure5.
3
The MoO layer does not act as a tunnel barrier, as transport of charge carriers takes place
3
throughtheconductionbandoftheoxide. IfoneaddsadditionalMoO layers,oneexpectsthatfor
3
thicknessesbelowthemeanfreepathofthechargecarriersthecharacteristicsforinjectionfromthe
metal into the MX layer remain the same. The contact resistance does increase for thicker MoO
2 3
layers,however,aschargecarriermobilitiesinoxidelayersaretypicallysubstantiallysmallerthan
inmetals.18
8
Inprincipleanyoxidethathasalayeredstructurecanbeusedthisway. Onecancoverametal
withanoxidemonolayer;iftheoxidelayerbondstoMX throughvanderWaalsinteractions,and
2
ifitselectronaffinityissufficientlyhigh,itshouldgiveap-typecontact. Incontrasttographeneor
h-BN, the metal species is not very important, as the conduction band of the oxide pins the Fermi
level. BesidesMoO ,forinstanceV O hasalayeredstructureandahighelectronaffinityofupto
3 2 5
7eV.16 Adrawbackofusingsuchoxidelayersarethattheyarestrongoxidizingagents,whichcan
react with molecules in the environment, or with the metal electrodes. For instance, the Cu/MoO
3
interface is metastable and oxidized Cu ions can diffuse into MoO .16 Other metal/MoO inter-
3 3
faces,suchasAu/MoO ,arestable,however.
3
Metallic M(cid:48)X(cid:48) buffer layers
2
A buffer with a high electron affinity that is less reactive than an oxide would be very convenient.
Metallic M(cid:48)X(cid:48), M(cid:48) =V,Nb,Ta, X(cid:48) =S,Se compounds have a layered hexagonal structure similar
2
to that of semiconducting MX . The latter compounds contain a group VI transition metal M,
2
whereas the former compounds contain a group V transition metal M(cid:48). The electronic structure
of these two compound groups is basically similar, but in MX the topmost valence band is com-
2
pletelyfilled,whereasinM(cid:48)X(cid:48) itisonlyhalf-filledbecauseM(cid:48)hasoneelectronlessthanM.31This
2
makestheM(cid:48)X(cid:48) compoundsmetallicwitharelativelyhighworkfunction. Forexample,NbS and
2 2
TaS monolayershavecalculatedworkfunctionsof6.22and6.12,respectively.
2
Most common metals have a lower work function. If they are covered by a M(cid:48)X(cid:48) monolayer,
2
electronsaretransferredfromthemetaltotheM(cid:48)X(cid:48) layer,effectivelyincreasingtheworkfunction,
2
as in the case of an MoO overlayer. As the density of states at the Fermi level of M(cid:48)X(cid:48) is high,
3 2
the transferred electrons will hardly modify the Fermi level, which is therefore effectively pinned
bytheM(cid:48)X(cid:48) layer. Forinstance,theAu(111)andAl(111)surfaceshavecalculatedworkfunctions
2
of 5.43 and 4.24 eV, respectively. Adsorbing a NbS monolayer gives work functions of Au/NbS
2 2
and Al/NbS surfaces of 6.11 and 6.06 eV, respectively. In other words, the work function differs
2
by(cid:46)0.15eVfromthatofafree-standingNbS monolayer,irrespectiveofthemetalsubstrate.
2
9
This remarkable property make metallic M(cid:48)X(cid:48) compounds viable buffer layers for making p-
2
dopedcontactstoMX semiconductors. WeenvisionthatM(cid:48)X(cid:48) layerscanbedepositedonametal
2 2
substrate in a similar way as MX layers. A MX layer can then be deposited on M(cid:48)X(cid:48) by van
2 2 2
der Waals epitaxy, for instance.6,7 The interaction at the M(cid:48)X(cid:48)/MX interface is van der Waals,
2 2
which hardly perturbs the electronic structure of either layer. The potential step at the M(cid:48)X(cid:48)/MX
2 2
interface,∆V ≈0.1eV,issmallandhardlyaffectstheM(cid:48)X(cid:48) workfunction.
2
Figure 6 shows the calculated band structure of the Au(111)/NbS /WSe multilayer. It illus-
2 2
tratestheperfectp-typecontactformedinthiscase,withtheFermilevelcoincidingwiththetopof
theWSe valenceband. AtthesametimetheFermicutsthevalencebandofNbS ,confirmingthat
2 2
it acts as a conducting layer. Indeed, the local DoS calculated at the Fermi energy shows a state
that is delocalized over the whole multilayer, see figure 6. We expect that similar p-type contacts
canbeformedwithotherMX layersusingNbS orTaS monolayersasabuffer.
2 2 2
Adding additional NbS or TaS layers, the charge carrier injection into MX should remain
2 2 2
the same as long as the thickness of the buffer layers is below the mean free path of the charge
carriers. For thicker layers the contact resistance starts to depend on the thickness of the buffer
layers.
Conclusions
We propose a particular technique to construct p-type contacts with zero SB heights to MX (M =
2
Mo,W; X = S,Se,Te) 2D semiconductors. Using first-principles DFT calculations we show that a
directmetal/MX interactionleadstointerfacestatesthatpintheFermilevelintheMX bandgap,
2 2
giving a sizable SB. Inserting a well-chosen buffer layer between the metal surface and the MX
2
layer breaks thisinteraction, and unpins the Fermi level, if MX is bonded to thebuffer layer with
2
vanderWaalsforces.
A monolayer of h-BN or graphene constitute only a thin barrier for transport. Adsorbing h-
BN on a high work function metal such as Pt or Au, gives a zero SB height for contacts to the
10