REVIEW ARTICLE

PUBLISHED ONLINE: 20 NOVEMBER 2015 | DOI: 10.1038/NMAT4452

Electrical contacts to two-dimensional

semiconductors
Adrien Allain1, Jiahao Kang2, Kaustav Banerjee2* and Andras Kis1*
The performance of electronic and optoelectronic devices based on two-dimensional layered crystals, including graphene,
semiconductors of the transition metal dichalcogenide family such as molybdenum disulphide (MoS2) and tungsten diselenide
(WSe2), as well as other emerging two-dimensional semiconductors such as atomically thin black phosphorus, is significantly
affected by the electrical contacts that connect these materials with external circuitry. Here, we present a comprehensive treat-
ment of the physics of such interfaces at the contact region and discuss recent progress towards realizing optimal contacts for
two-dimensional materials. We also discuss the requirements that must be fulfilled to realize efficient spin injection in transition
metal dichalcogenides.

S
emiconducting electronic devices utilize fine control over depletion and transfer lengths. In the 2D limit, the properties of
the flow of charge carriers, which are injected into the semi­ the interface — the chemical interaction between the metal and the
conducting material through electrical contacts. The quality of SC — govern everything. Substitutional doping, a common strat­
the electrical contacts — quantified through contact resistance — is egy adopted to decrease the contact resistance in bulk SCs, is not
as important to the proper functioning of the entire device as the applicable here because it would modify both the 2D material and
semiconductor (SC) itself. Since the early 1990s, researchers have its properties. In addition, the pristine surface (that is, no dangling
explored a wide variety of electronic devices based on nanostruc­ bonds) of a 2D material makes it difficult to form strong interface
tures with different dimensionalities, ranging from one-­dimensional bonds with a metal, thereby increasing contact resistance.
(1D) carbon nanotubes1, SC nanowires2 or 2D materials, starting The quantum limit to the contact resistance (RCmin) is determined
with graphene3. by the number of conducting modes within the SC channel27,28,
Advances in the realization of electronic devices4–6, including the which is connected to the 2D charge carrier density (n2D), yielding
recent demonstration of a super-steep-slope atomically thin chan­ RCmin = h/(2e2kF) = 0.026/√n2D ≈ 30 Ω μm at n2D = 1013 cm–2 (ref. 29) —
nel interband tunnelling transistor 7, optical characterization8,9 and a value three orders of magnitude below the typical contact resist­
material preparation10 have revived scientific interest in the sin­ ance to monolayer MoS2. Here, h is Planck’s constant, kF is the Fermi
gle- and few-layer form of layered transition metal dichalcogenides wavevector and e is the electron charge. There is plenty of room for
(TMDCs), a material family that has a wide range of electrical prop­ improvement in this respect, making it all the more important to
erties. This class of materials has a common chemical formula MX2, study the detailed physics of contacts between metals and 2D SCs.
where M stands for a transition metal (most commonly Mo, W, In this Review, we first discuss the geometry and nature of
Nb, Ta or Ti) and X is a chalcogen atom (S, Se or Te). Compounds the interfaces between 2D materials and metal contacts. We then
based on Mo and W are the best-known examples of semiconduct­ describe the charge-injection mechanisms that occur at the con­
ing TMDCs. In their single- and few-layer form they show many tacts and summarize what is known about ‘Schottky barriers’ to
attractive features such as atomic-scale thickness and large band­ 2D SCs. We then discuss contact resistance and how it scales, and
gaps (1–2 eV), leading to a high degree of electrostatic control11 and present an overview of current research in this area. Finally, we
scalability for nanoscale transistors12, exquisite sensing capabili­ discuss the influence of contact resistance on spin injection into
ties13, high breakdown voltages14, tunable optical properties8,9,15–17, a 2D SCs. Most research into devices based on 2D SCs is carried
high degree of mechanical flexibility 18 and the possibility of engi­ out using MoS2, owing to its wide availability. Concepts and limi­
neering new materials through the realization of van  der  Waals tations that we outline here can, however, be readily extended to
hetero­structures19. Other semiconducting 2D materials such as other 2D SCs.
phosphorene20 and silicene21 are also attracting growing interest.
One of the most common electronic devices, in both the research Interface geometry
and industrial environments, is the field-effect transistor (FET). There are two fundamental interface geometries (topologies)
Low contact resistance in 2D SC-based devices is critical for achiev­ between bulk (3D) metals and 2D materials: top contact (Fig. 1a)
ing high ‘on’ current, large photoresponse22 and high-frequency and edge contact (Fig. 1b), which perform differently.
operation23. However, the major issue for 2D SC FET transistors is Pure top contacts can be made by simply avoiding contact
the existence of a large contact resistance at the interface between between the metal and the edges of the 2D material30. However, the
the 2D SC and any bulk (or 3D) metal, which drastically restrains formation of a pure edge contact using standard lithographic tech­
the drain current 24–26. Contacting 2D SCs presents a certain number niques is difficult in a single- or few-layer 2D material, owing to its
of experimental and conceptual challenges. The theoretical concepts atomically thin body; so far, only one example has been reported31 of
that underlie our understanding of conventional metal–SC contacts a purely edge-contacted graphene monolayer. In most experiments,
break down in the limit where the SC thickness is smaller than the contacts to 2D materials are a combination of these two geometries.

Electrical Engineering Institute, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland. 2Department of Electrical and Computer
1

Engineering, University of California, Santa Barbara, California 93106, USA. *e-mail: kaustav@ece.ucsb.edu; andras.kis@epfl.ch

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1195

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLE NATURE MATERIALS DOI: 10.1038/NMAT4452
a Density functional theory (DFT) has shown that edge contacts lead
to shorter bonding distance with stronger hybridization (orbital
Metal 2D material Metal overlap) than top contacts, and transport simulations show that the
incorporation of additional interfacial species (such as oxygen) can
help to improve bonding and increase transmission31. The reported
contact resistance for a Cr edge contact with monolayer graphene
is about 150 Ω μm, which is in good agreement with the value of
b 118 Ω μm predicted from simulations31. For TMDC SCs, edge con­
Metal Metal tacts can also be advantageous compared with top contacts. This has
2D material been verified by DFT simulations for both monolayer 32,34 and multi­
layer TMDCs35. The main reasons are the stronger orbital overlaps
and the reduction of tunnel barriers.
Edge contacts are particularly relevant to multilayered 2D mate­
rials36 owing to the large conductivity anisotropy of 2D materials
Figure 1 | Interface geometries of metal–2D contacts. a, Top-contact between the in- and out-of-plane directions. A model accounting
configuration. b, Edge-contact configuration. for both top and edge contacts to multilayer graphene37 showed that,
in this case, edge contacts significantly reduce the overall contact
Contrary to the case of bulk (3D) SCs (Fig.  2a,b), the pristine resistance36. Another model, based on a resistor network and con­
surfaces of 2D materials do not tend to form covalent bonds. The sidering the back-gate screening effect 38, has been used to compare
interfaces between metals and 2D materials in the top-contacted top- and edge-contacted multilayer graphene devices. Using this
configuration can therefore only be formed by a van der Waals model, the extracted edge contact resistance from experiments is
(vdW) gap in most situations (Fig.  2c). As shown in Fig.  2d, the 150–360 Ω μm for each graphene layer, which is relatively small com­
vdW gap in such top-contact interfaces acts as an additional ‘tun­ pared with the tunnelling resistances between each layer 38. Similar
nel barrier’ for carriers, before the inherent Schottky barrier (SB)32. work has been reported39 in which the currents flowing through the
The tunnel barrier greatly reduces the charge injection from metals, graphene surface and edges were theoretically and experimentally
which results in higher contact resistance. investigated by patterning graphene under the contact metal with
One option to overcome this vdW gap is to take advantage of different perimeter-to-area ratios.
edge contacts. In fact, edge contacts to monolayer graphene have Nevertheless, as mentioned previously, most practical contact
been modelled33 and shown31 to perform better than top contacts. structures involve both the edge and the top surface of the 2D

a Metal–bulk SC contact c Metal–2D SC contact e Metal–2D SC contact

(with vdW gap) (with hybridization)

Metal Metal Metal

A A 2D SC A 2D SC
B C Bulk SC B B’ C B B’ C
C
Metal Au Ti

Ti
Ti
Au
vdW gap Ti
S S Alloy
Mo Mo
S S
Bulk SC Covalent bonds MoS2 Covalent bonds MoS2

b d f
TB
–
e SB
SB e– e–
SB
EF EF EF
EC EC EC
Metal Metallized
Metal 2D SC Metal
Bulk SC 2D SC
2D SC
A A B A B
EV vdW EV EV
B B’ B’
C gap
C C

Figure 2 | Different types of metal–SC junction and their respective band diagrams. a,b, Schematic (a) and corresponding band diagram (b) of a typical
metal/bulk SC interface. c,d, Metal/2D SC interface with vdW gap (for example, Au–MoS2 contact). e,f, Metal/2D SC interface with hybridization (for
example, Ti–MoS2 contact, where MoS2 under the contact is metallized by Ti). EF, EC and EV represent the Fermi level of the metal, and the conduction and
valence bands of the 2D SC, respectively. TB and SB indicate the tunnel and Schottky barrier heights, respectively. A, B, B’ and C represent different regions
in the current path from the metal to the SC. The blue arrows in b, d and f represent the different injection mechanisms. From top to bottom: thermionic
emission, thermionic field emission and field emission (tunnelling). In d, only thermionic emission is available.

1196 NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials

© 2015 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOI: 10.1038/NMAT4452 REVIEW ARTICLE
material, with the top surface having a greater contribution due a
to the large surface-area-to-edge-area ratio of the contact. Hence, ac-GNR
other than optimizing the edge contacts, the contact resistance can (semiconducting)
be improved by reducing the tunnel barrier at the top surface. This CNT
can be achieved by hybridization between atoms of contact met­ (metallic)
als and 2D SC surfaces. DFT simulations show that specific met­
als can form covalent bonds to 2D SC surfaces (Fig. 2e), and hence
eliminate the vdW gap; for example, Ni for graphene40, Ti for MoS2
b
(refs 34,41), Pd for WSe2 (ref. 34), Mo/W for MoS2/WSe2 (refs 32,42)
and Ti2C (a 2D metallic material) for MoS2 (ref. 43).
Strong hybridization can also distort the properties of 2D ac-GNR
SCs below the top contacts (especially for a monolayer)42. This (semiconducting)
can modify the sheet resistance of 2D SCs below the contacts
(ρ2D
contact), thus causing a change in contact resistance (the calculation Wide graphene
of which will be discussed in a later section). It should be noted (metallic)
that hybridization can both increase and decrease ρ2D contact. DFT pre­
dicts that selected metals such as Ti and Mo create non­localized
overlap states in the original bandgap of MoS2 (ref.  32), which c
effectively turns MoS2 under the contact into a new metallic com­
Semiconducting TMDC
pound (Fig. 2f). In such a situation, ρ2D contact reduces. On the other
hand, if the monolayer is partially metallized, ρ2D contact may increase
owing to localized states.
It is worth noting that the DFT predictions for cases of strong
hybridization (Fig. 2f) are based on the assumption of perfect inter­
faces. In practice, close-to-perfect interfaces require the removal or
prevention of surface impurities (such as resist residues), as well as
an annealing process. For example, in graphene, during annealing,
the carbon atoms can dissolve into the contact metal (Ni or Co) and Metallic TMDC
thus form strong covalent bonds, which contribute towards a much
smaller contact resistance44.
For multilayer TMDC SCs, only the top layer can be hybrid­
ized by metal top contacts, and thus only the vdW gap between the
Figure 3 | Various ‘seamless’ contact schemes. a, CNT–GNR contact.
metal and the top layer of the TMDC is eliminated, as predicted by
b, Graphene–GNR contact. c, Metallic TMDC–semiconducting TMDC
DFT35,42. The vdW gaps between the bottom layers still exist. Hence,
contact. ac, armchair.
to improve the carrier injection into the bottom layers, it is preferable
to have edge contacts with all layers.
Although a number of selected metals form strong hybridiza­ The concept of seamless contacts from such all-graphene circuits
tion at interfaces and therefore suppress the tunnel barrier, Fermi can also be adapted to other 2D SCs (Fig. 3c). In recent studies49,50,
level pinning occurs due to the work function of the metal layer at contacts between metallic 1T-MoS2 and semiconducting 2H-MoS2
interfaces changing (into the metal–MoS2 alloy’s work function)32 as have been fabricated using phase engineering 49,51 (changing 2H
well as the creation of gap states from the weakened intralayer S–Mo phase into 1T phase). The resulting contact resistance of 0.2 kΩ μm
bonding 45. Such effects can significantly impact the Schottky barrier is the lowest ever reported for this material. Another option would
height (SBH). be to grow metallic and semiconducting TMDCs in sequence by
Apart from 3D metals, one must also consider the possibility chemical vapour deposition. Further theoretical studies are needed
of contacting 2D SCs using other (or the same) low-dimensional to explore this seamless contact scheme on 2D materials other
materials. ‘Native’ chemical bonds are expected at such interfaces. than graphene.
For example, because the carbon family includes both metallic allo­
tropes (such as metallic carbon nanotubes (CNTs), graphene and Charge-injection mechanisms
wide graphene ribbons) and semiconducting allotropes/structures There are two mechanisms by which charge can be injected into a
(such as semiconducting CNTs, graphene nanoribbons (GNRs) and SC: thermionic emission over the SB, and field emission (tunnel­
vertically biased AB-stacked bilayer graphene), one can first fabri­ ling) across the SB (Fig.  4a). The thermionic emission–diffusion
cate semiconducting (or metallic) carbon and then tune one side theory 52 describes the current–voltage characteristics of a metal–SC
to be metallic (or semiconducting). Relevant theoretical studies junction as a function of SBH. Carrier recombination can also be
include the CNT/graphene interface46 (Fig. 3a), the graphene/GNR a current-limiting process if an inversion layer is present near the
interface46,47 (Fig. 3b) and the monolayer/multilayer graphene inter­ contact. This is mostly the case in low-bandgap SCs (easy to form an
face48. The bonds at such interfaces are native sp2 carbon–­carbon inversion layer) such as Ge nanowires53, and could be significant in
bonds — the same as the bonds inside both the metallic and the black phosphorus (which has a small bandgap of 0.3 eV).
semiconducting sides — resulting in a ‘seamless’ contact between In TMDCs we deal mostly with thermionic emission at low dop­
the two. ing, with thermionic field emission starting to contribute as doping
In a recent study 47, an all-graphene circuit scheme based on increases (Fig. 4a). This is similar to the case of small geometry silicide
seamless contacts was proposed and evaluated by numerical simula­ contacts in advanced complementary metal–oxide–semiconductor
tion. The reported seamless contacts, in which both contacts/inter­ (CMOS) technologies54.
connects (wide graphene) and transistors (GNRs) are envisioned to Contrary to the bulk case (Fig. 2a,b), where the diffusion region B
start from a single sheet of graphene, greatly reduced the contact extends both laterally and vertically into the SC, in a metal–2D SC
resistance (down to 0.1 kΩ μm) and improved circuit performance junction with no hybridization (Fig. 2c,d and Fig. 4a), the position
in terms of noise margin, speed and power consumption. of the bands vary only laterally, so that charge carriers injected far

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1197

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLE NATURE MATERIALS DOI: 10.1038/NMAT4452

a Metal contact b
e– 2D SC EA

Flat-band condition
(1)
qϕB0 +
+ (2)
+ (3)
+
+
+ +
EF
EC
(1) thermionic emission qϕB0
(2) thermionic field
emission
vdW gap (3) field emission
(1) EV Vg
EC
qϕB0 EA
EF
Flat-band condition c
qϕB0
EV
(1) EC
EA EC∞ – EC0
qϕB0
EF

EV

Decreasing n-doping VDS = 0 V VDS

or Vg

Figure 4 | Charge-injection mechanisms and extraction of the SBH. a, Here we consider the case of a vdW gap (that is, without strong hybridization).
Shown are energy diagrams at the end of the charge-injecting source contact (without the B region shown in Fig. 2d), going from high to low channel
doping (decreasing gate voltage), as indicated by the vertical black arrow. In the ‘off’ state, the energy barrier to overcome (EA) is larger than the SBH
(qϕB0). By increasing the doping level, a flat-band condition is met when EA = qϕB0. When the channel is doped beyond the flat-band condition, charge
injection is mediated by a combination of thermionic emission and tunnelling. b, Method of extracting the SBH. Using transport measurements, EA is
determined for each value of the gate voltage Vg (blue circles). The flat-band condition is realized at the point where EA stops depending linearly on Vg
(dashed line). EA is then equal to the SBH qϕB0. c, SBH lowering at finite source–drain bias VDS due to image forces leads to an underestimation of the SBH.
The error can be decreased by extracting the SBH at finite bias (blue circles) and then extrapolating it to zero bias (dashed line).

from the contact edge first encounter the flat-band region B before of an n‑type SB FET; Fig. 4a) consumes most of the voltage drop and
the diffusion region Bʹ32 (Fig. 2c,d). In this case, the relative contri­ dominates the transistor behaviour. The current density injected
butions from thermionic emission and tunnelling become difficult through a reverse-biased SB is:
to predict.
Because charge injection in 2D SCs strongly depends on the SBH, qV
qϕB0
knowledge of its value and how to alter it would allow the process to J = A*T α exp – 1– exp – (1)
be optimized. In the ideal case, the SBH qϕB0 between a metal and kBT kBT
a SC is determined by the difference qϕB0 = q(ϕm – χ) between the
metal’s work function qϕm and the SC’s electron affinity qχ, which where A* is the Richardson constant and can be derived for 3D and
is also referred to as the Schottky–Mott rule52 (q is the elementary 2D SCs, ϕB0 is the SBH, α is an exponent equal to 2 for bulk SCs and
electric charge). In reality, however, the Fermi level at the metal/SC 3/2 for 2D SCs,V is the applied bias52,56, T is temperature and kB is
interface is often pinned. We can quantify this by inspecting the SBH the Boltzmann constant. For qV >> kBT, equation (1) simplifies to:
dependence on ϕm, given by S = dϕB0/dϕm, with S = 1 corresponding
to the ideal case or Schottky limit, and S ≈ 0 corresponding to that
qϕB0
of a pinned Fermi level. The origin of this pinning in metal/bulk SC J = A*T α exp – (2)
interfaces is the presence of metal-induced gap states55. For metal kBT
contacts to 2D SCs, as discussed in previous sections, the presence
of a metal–MoS2 alloy with a different work function32 and the crea­ The transistor behaviour in the sub-threshold regime (small gate
tion of gap states from the weakened intralayer S–Mo bonding 45 voltage, Vg) (Fig. 4a, bottom) therefore follows57:
contribute to the pinning. The resulting reduced tunability of the
SBH reduces the efficacy of engineering ohmic contacts by choice of EA
the contact metal (or work function) alone. J = A*T α exp – (3)
kBT
Extraction of the Schottky barrier height
The most common way of extracting the SBH is to measure the where EA = qϕB0 + EC∞ – EC0 is the total activation energy that charge
activation energy in the thermionic emission regime. In a SB FET carriers must overcome to access the channel, and EC∞ – EC0 is the dif­
geometry, two Schottky diodes connected back-to-back are situated ference between the conduction band minimum in the bulk and at
at the source and drain. The reverse-biased contact (the source side the interface (Fig. 4a), due to upward band bending. The so-called

1198 NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials

© 2015 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOI: 10.1038/NMAT4452 REVIEW ARTICLE
Pt Contact length l
MoS2
0.20 Ref. 24 (2–12 nm) Metal Transfer length LT
Ref. 60 (13 nm) Au
Ref. 70 (bulk)
Schottky barrier height (eV)

Ref. 74 (bulk)
0.15 Ni Injected current
rC
2D SC
W Co

0.10

2D 2D
ρcontact ρchannel
J
0.05 Ti

Sc Figure 6 | Current crowding at the contact edge region. The resistor

S = 0.09 network illustrates the transmission line model. The orange arrows depict
the current density, J. rC represents the resistivity of the metal/SC interface.
3.5 4.0 4.5 5.0 5.5 contact and ρchannel represent the SC’s sheet resistivity under the contact and
ρ2D 2D

Metal work function (eV) in the channel, respectively.

Figure 5 | SBHs for different multilayer MoS2 devices reported in be employed for TMDCs include internal photoemission72,73 and
the literature. C–V measurements30.
Figure 5 plots SBH values between multilayer MoS2 and different
flat-band condition EC∞ – EC0 = 0 is realized when Vg = VFB (Fig. 4a, metals24,60,70,74 against the corresponding metal work functions. In
centre band diagram), where VFB is the flat-band voltage. As long as the limit of 1–3 layers, the SBH is expected to depend strongly on
Vg < VFB, the activation energy EA depends linearly on Vg: the number of layers because the bandgap increases as the thickness
is decreased. Figure 5 therefore only compares values of SBH that
EA = qϕB0 – (1 + Cit / Cox)–1 (Vg – VFB )(4) were obtained on thicker (from 2 nm to bulk) layers. As was shown
in ref. 24, in this range there is no significant dependence of the SBH
where Cox is the gate oxide capacitance and Cit is the interface on MoS2 thickness. It can be seen from Fig. 5 that the Fermi level is
trap capacitance24. pinned to around 100  meV below the conduction band, and that
Increased doping pushes the system from the flat-band condi­ the SBH depends weakly on the metal’s work function (S = 0.09).
tion into a regime where tunnelling contributes (Fig. 4a, top). In this This is in contradiction with early results from X-ray photoelectron
regime, the temperature dependence should deviate from the above spectroscopy measurements, which indicated that the SBH for bulk
equations and EA will no longer depend linearly on Vg. MoS2 is in the Schottky limit (S = 1)66.
To extract the SBH, we identify the voltage at which EA stops A recent scanning tunnelling microscopy investigation offers
depending linearly on Vg (Fig. 4b), given by EA = qϕB0. This method one possible explanation for this discrepancy 75. The MoS2 surface
was used in ref. 24 to extract the SBH between MoS2 and different hosts a high density of metallic defects that are likely to be clusters
metals. A variant of this method takes care (to first order) of SB of sulphur vacancies. These defects are associated with variations
lowering by using 58: in the Fermi level as large as 1  eV and are believed to shunt the
pristine regions and thus strongly influence electrical properties.
q
J = A*T α exp – (ϕ – V/n) (5) The SBH values derived from electrical measurements can there­
kBT B0 fore be significantly lower than those measured with ultraviolet
and X-ray photoelectron spectroscopy. These observations are also
where the ideality factor n accounts for barrier lowering due to in line with previous observations of widely varying SBH in Au
image charges. The activation barrier at zero bias is then linearly nanoclusters evaporated onto MoS2 (ref.  71). SBH inhomogenei­
extrapolated from the activation energies obtained at finite biases ties can also explain76 the apparent temperature dependence of
(Fig.  4c). Several studies have used this second method for MoS2 SBH and ideality factor in graphene–MoS2 (ref. 77) and In–WSe2
(refs 58–61). (ref. 78) contacts.
When using this method, it is important to remember that the WSe2 was also investigated by scanning tunnelling micros­
activation energy and the SB are not the same thing. They can be copy 79 and, although similar structural defects were present, no
identified only when the flat-band condition is met, and not for metallic defects were found. This could explain why contacts to
every value of Vg. The channel resistance62 contributes signifi­ WSe2 are often much more resistive. Ultraviolet and X-ray photo­
cantly to the observed temperature dependence. When the MoS2 electron spectroscopy analysis has revealed that, like MoS2, the
channel is gated beyond the metal–insulator transition, its resist­ SBH for WSe2 approaches the Schottky limit 67–69 — a fact supported
ance lowers with decreasing temperature63. The total (channel plus by the high sensitivity of the SBH to graphene’s work function in
contact) resistance can then also decrease — this should not be graphene–WSe2 contacts64.
misinterpreted as a negative SB. Decoupling the metal from the SC surface decreases the amount
The SBH can also be extracted in the thermionic field emission of Fermi level pinning and has a number of positive effects. It was
regime. This method was used in refs 64,65 and seems to describe recently reported that an intervening oxide58,60,61,80 or graphene81
quite well the injection of electrons into WSe2 from ionic-liquid- layer between the metal and MoS2 significantly reduces the contact
gated contacts64. Other methods that have been used to extract resistance and SBH.
the SBH in 2D SCs include ultraviolet and X-ray photoelectron Fermi level pinning is also critical because it impacts the possi­
spectroscopy 66–69, which measures the band bending, and scan­ bility of realizing FETs with both polarities. FETs made of (naturally
ning tunnelling spectroscopy of Au nanoparticles deposited on top n‑doped) mono- and multilayer MoS2 exhibit n‑type behaviour
of MoS2 (refs 70,71). Finally, some techniques that could in future with all metals, with the notable exception of MoOx contacts to

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1199

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLE NATURE MATERIALS DOI: 10.1038/NMAT4452
a line model88,89, which gives the following expression for the contact
106
Ref. 35 (Au/Ti) Ref. 93 (Au/Ti) resistance RC (in Ω m):
Ref. 42 (Au/Mo) Ref. 94 (Au/Ni) 2.0
Ref. 49 (Au) Ref. 95 (Au/Ni)
Ref. 62 (Au/Ti) Ref. 96 (Graphene) ρ2D
Contact resistance (Ω µm)

105
Ref. 92 (Au/Ti) Ref. 97 (Au)
1.8 RC = ρ2D rC coth l rC
(6)

Optical bandgap (eV)

Ref. 98 (Au/Ti)
1.6
104 BV doping where l is the contact length. It can be seen that the dependence
CI doping 1.4
of RC on l is nonlinear because of current crowding. For contact
2H phase lengths much larger than the transfer length LT = √(rC/ρ2D), where
3
10 LT is the average distance that an electron (or hole) travels in the
1.2
SC beneath the contact before it enters the contact, the expression
1T phase for the contact resistance becomes RC = √(ρ2DrC) (in units of Ω m)
102 1.0 and is no longer dependent on the contact length. In some reports
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
the ‘as-measured’ resistance is referred to as the ‘contact resistance’,
Number of layers
in which case the so-called contact resistivity is expressed in units
b of Ω m. The quantity ‘resistance × contact area’ (in units of Ω m2) is
108 1L
WS2 sometimes also used to characterize contacts, especially when con­
ML 1L WSe2 /WS2
1L tacts exhibit a significant dependence on contact length l (when l
10 7
ML MoS2 and LT are of the same order of magnitude)90.
2H In graphene, the diffusive approximation breaks down due to the
Ref. 99 (1L)
1T
1L
Ref. 97 (1L)
longer electron mean free path, which calls for a ballistic treatment of
106 WSe2
ML the contact resistance91. In TMDCs, the much lower mean free path
Contact resistance (Ω µm)

Ref. 92 (1L)
ML Black phosphorus implies that the transmission line model can be applied, provided
1L MoS2
105 Ref. 96 (4L) at 300 K
Ref. 97 that the resistivity of the portion of semiconducting material under
Ref. 93 (1L)
WSe2
Ref. 128 (19 nm) (2–5L)
the contact (ρ2D contact), and not the resistivity of the semi­ conducting
104
with electrolyte Ref. 100 (1L) ML MoS2 channel (ρ2D channel), is used in place of ρ
2D
in equation (6). Note, how­
Ref. 96 (4L) at 100 K ever, that this approach cannot accurately model metal contacts to
Ref. 94 (8L)
Ref. 119 (8L)
Ref. 49 (3L) multilayers owing to the greater impact of edge contacts36,37.
103 Graphene The transfer length has been studied in monolayer 92 and 2L–6L
Ref. 64 (4–12L)
Ref. 95 (5–7L) (ref. 62) MoS2. A value of LT = 600 nm was found92 in the monolayer
102 Ref. 81 (16 nm) Low standby dissipation
case, assuming ρ2D contact  =  ρchannel. Using a more elaborate measure­
2D
ML MX2 High performance l Low power
+ molecular
m tota
doping L =22 n 0% of ITRS requirements
ment similar to a four-terminal Kelvin resistor scheme described
62
g
RC
=2 for 2015 in ref. 87, one can accurately determine both ρ2D contact and rC. A value of
101 LT = 20–70 nm was found for bilayer MoS2 with Ti contacts and up
102 103 104 105 106
ρ2D (Ω –1
) to 200 nm in 6L MoS2 (ref. 62). This discrepancy indicates that the
channel
assumption ρ2D contact = ρchannel fails in the case of atomically thin chan­
2D

Figure 7 | Contact resistance for top-contacted 2D SCs. a, Minimum

nels owing to the strong influence of the metallic electrode on the
contact resistance as a function of the number of atomic layers in several
channel underneath.
studies on MoS2 (refs 35,42,49,62,92–98). The red dashed line depicts the
optical bandgap measured in ref. 9. b, Contact resistance for different values
Experimental review of contact resistances
of channel sheet resistivity in several different studies on TMDCs, black
Figure 7 shows the contact resistances of 2D SCs found in the lit­
phosphorus and graphene49,64,81,92–96,99,100,119,97,128. The dashed line depicts the
erature. From an experimental point of view, the contact resistance
criterion RC = 20% of total resistance in a transistor with a 22-nm-long gate,
depends mainly on three parameters: contact metal, ρ2D contact and

Lg. The dash–dot line is a guide to the eye: if a line runs parallel to it, its RC
number of layers. This makes it difficult to compare contact resist­
scales as √ρ2D
ance values found in the literature because available data sets often
channel. The different star symbols represent the target values for
2015 as indicated by ITRS 2012 for different types of transistor, namely high
differ by more than one parameter. Although results obtained using
performance, low power and low standby dissipation.
different metals are available, from this meta-analysis it is difficult
to draw clear conclusions as to which metal yields the best contact
to any given material. We nonetheless indicate the contact metal
multilayer MoS2 (ref. 82). In WSe2 FETs the situation is more con­ used in each study. Figure  7a shows the minimal RC values from
trasted, with reports of clear n‑type transport in monolayers with several studies on MoS2 as a function of number of layers35,42,49,62,92–98.
Al, Ti, Ag, In (ref. 6) and Au (ref. 83), and clear p‑type transport Despite the scatter in the data, there is a clear trend of decreasing
with Au (ref. 84), Pd (ref. 85) and Ag (ref. 86). Finally, owing to its RC with increasing thickness. This comes as no surprise because
very high work function, MoOx seems to be the best hole injector the larger bandgap in thinner flakes (red dashed line in Fig. 7a) is
for both multilayer MoS2 and WSe2 (refs 79,82). expected to give rise to larger SBs.
A better comparison involves contact resistances measured
Contact resistance scaling at similar sheet resistances. We do not have access to the values
The top contact is the most common contact geometry in use today. of ρ2D
contact, but using the channel resistivity ρchannel (Fig.  7b) instead
2D

In the weak-coupling limit, the contact resistance is a combination should provide a more reliable comparison between contacts. Each
of the metal/SC interface resistivity rC (expressed in ohms square curve in Fig. 7b corresponds to a back-gate voltage sweep, wherein
metre; Ω m2) and the SC sheet resistivity ρ2D (expressed in ohms per the back gate dopes both the channel and the contact areas. We
square or Ω ☐–1)87. If the contact is diffusive — that is, if the charge assembled data points from a set of studies reporting both contact
carriers are scattered many times within the SC before being kicked resistance and total transistor resistance (in some cases four‑probe
out of the SC and into the metal — then the interface can be mod­ resistance), and in which L >> LT, so that contact length is irrelevant.
elled as a resistor network (Fig. 6). This is the so-called transmission The channel resistivity was extracted by subtracting the contact

1200 NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials

© 2015 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOI: 10.1038/NMAT4452 REVIEW ARTICLE
a b 8
FĹ SC FĹ Graphene

Magnetoresistance ratio ΔR/R (%)

RĹ RSC RĹ WSe2
6 WS2
Parallel MoS2
MoSe2
RĻ RSC RĻ
Lchannel =
4
2 µm for graphene
FĹ SC FĻ 100 nm for TMDCs
RĹ RSC RĹ 2
Antiparallel
RĻ RSC RĻ 0
102 103 104 105 106
Contact resistance RC (Ω µm)

c d

Optimal contact resistance RC (kΩ µm)

108 5
1L WS2 
1.2
Contact resistance RC (Ω µm)

7
Ref. 99
10
4 1.0
ΔR/R > 2%

(ΔR/R)max (%)
106 6
0.8
T=73 K
1L WSe2 (holes) 3 T=150 K

ΔR/R (%)
5 Ref. 100 T=210 K 4
10 1L WSe2 (electrons) T=300 K
0.6
Ref. 100 2 2
104 0.4
5–7L Cl-doped WS2 1 2 3 4 5
0
10 10 10 10 0.2
103 Ref. 95 RC (Ω µm)
0
104 105 106 107 50 100 150 200 250 300
ρ2D
channel
(Ω –1
) Temperature (K)

Figure 8 | Spin injection and contact resistance. a, In devices composed of SCs contacted with ferromagnetic (F) contacts, the magnetoresistance
signal vanishes when the resistance of the nonmagnetic material is much higher than that of the contacts. b, Magnetoresistance as a function of contact
resistance calculated from equation (9) (Box 1) for parameters listed in Table 1. The filled and empty squares indicate where the contact resistance of the
ferromagnet–graphene contact lies with and without a tunnel barrier, respectively116. c, Contact resistance as a function of channel sheet resistivity in WSe2
and WS2. The shaded area corresponds to ΔR/R > 2% in WSe2 (region above the dashed line in b). The blue line corresponds to a back-gated monolayer
WS2 device99. The red and blue dots were obtained in electrolyte-gated 1L (monolayer) WSe2. The square is a multilayer WS2 with molecular doping95.
d, Calculated maximum magnetoresistance as a function of temperature in monolayer WS2 and corresponding optimal contact resistance calculated using
experimentally determined temperature-dependent mobilities99 and spin lifetimes124. Inset: Magnetoresistance as a function of contact resistance.

resistance from the total resistance and renormalizing with the that this effect is partially due to decreased channel resistivity. Yet
channel’s aspect ratio: ρ2D channel = (Rtotal – Rcontacts)W/L, where L and W this cannot be the only effect, otherwise data before and after dop­
are the length and width of the channel, respectively ing would fall on a straight line. This reflects the enhanced contribu­
We plot the results obtained on different TMDCs and black phos­ tion from tunnelling at the contact edges as the SB is thinned down,
phorus, mono- and multilayer, before and after molecular or electro­ which seems particularly strong in WSe2 (refs 64,100).
lyte gating of the channel, as well as the 2015 silicon-on-­insulator By far, the lowest contact resistances have been obtained using
target values according to the 2012 International Technology phase engineering on few-49 and single-layer 50 MoS2. The local con­
Roadmap for Semiconductors (ITRS). The significance of this repre­ version of MoS2 from the semiconducting 2H phase to the metal­
sentation is evident if one compares the data from ref. 92 and ref. 93: lic 1T phase under the metal contacts results in seamless contacts
in Fig. 7a, the minimum RC (monolayer case) seems to be lower for with resistances as low as 200 Ω μm in very resistive devices, thus
ref.  93, but Fig.  7b reveals that this is only due to higher doping. pushing the performance of TMDC-based transistors closer to the
Actually, for a given value of sheet resistivity, RC is lower in ref. 92. quantum limit 27–29.
Figure  7b reveals several interesting things. First, it confirms Graphene is a promising candidate to contact 2D SCs64,101–103, and
that, at constant sheet resistivity, the contact resistance decreases it can be used to realize purely 2D circuits103. Another advantage
with increasing MoS2 thickness97. We can also see that in many cases of graphene is the tunability of its work function by electrostatic
(thick MoS2, WS2, WSe2 and black phosphorus), RC scales as √ρ2D channel doping, which makes it a versatile contact metal64. The interface
(dash–dot line), which indicates that the transmission line model between MoS2 and graphene was recently found96 to yield very low
(equation (6)) applies reasonably well even if we use ρ2D channel instead of RC, as well as reducing RC with decreasing temperature in ultraclean
contact. Interestingly, few-layer MoS2 (refs 93,97) displays a stronger
ρ2D MoS2 samples. Once again, this result can be understood in the
dependence on ρ2D channel, compared with multilayers
94,97
. This could framework of the transmission line model: in these high-mobility
indicate that, in this particular case, the assumption ρ2D contact ≈ ρchannel is
2D
devices, the sheet resistivity drops by several orders of magnitude
not valid because of possible MoS2 band structure distortion by metal as the temperature is lowered, which in turn affects the contact
orbitals and Fermi level pinning under the contacts. In comparison, resistance. Figure 7b shows the data from ref. 96 at T = 300 K and
monolayer WS2 with Au contacts99 displays a much higher RC but it T = 100 K. The two data sets can be roughly linked by the dash–dot
follows roughly √ρ2D channel, which may indicate a weaker coupling to the line, which indicates that equation (6) describes the decreased con­
metal compared with MoS2. tact resistance. When compared with other metals at comparable
Molecular and electrolyte doping of the channel have often been sheet resistivity 97, however, graphene performs well but is not signif­
used to decrease the contact resistance94,95 (Fig. 7a). Figure 7b shows icantly better. One intrinsic limitation of graphene–TMDC contacts

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1201

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLE NATURE MATERIALS DOI: 10.1038/NMAT4452

Box 1 | Magnetoresistance of a 2D SC-based lateral spin valve.

Introducing the resistivity of the ferromagnetic electrodes ρF, as low as 1% in transparent contacts and as high as 35% in MgO
contact resistance rC and the spin asymmetry coefficients β and tunnel barriers on graphene115. Note that γ might be lower in the
γ of the electrodes and tunnel barriers, respectively, defined case of SBs118.
as ρ()
F   =  2ρF*(1  –  (+)  β) and r()
C   =  2rC*(1  –  (+)  γ), where The expression ΔR/R can be adapted for the case of a 2D chan­
ρF = ρF*(1 – β2) and rC = rC*(1 – γ2), the relative magnetoresistance nel using indications given in ref. 113. When the SC thickness t is
ΔR/R is given by 113: no longer comparable to the contact width w, relaxation in the SC
is divided by a factor w/t. We must then modify the expression by
2(βrF + γrC* )2 using rSC = ρSClsfSCw/t = ρ2D
channellsf w. Knowing that rF << rSC, rC
* and
SC
ΔR = 2
L + rSC 1 + rC* (7) using RC* = rC*w = RC/(1 – γ ), where RC is the contact resistance in
2
(rC* + rF) cosh SC sinh LSC Ω m, we finally have:
lsf 2 rSC lsf
and ΔR = 2(γRC* )2
2
=
R ρ 2D
lsf SC
RC*
R = 2(1 – β2)rF + rSC L + 2(1 – γ2)rC* + RC* cosh LSC + channel 1+ sinh LSC
lsfSC l sf 2 ρ2D lsfSC lsf
channel
L
(β – γ) 2rF rC* + rSC (β2rF + γ2rC* ) tanh
2lsfSC (8)
–1

2 (9) 2D
ρchannel lsfSCγ 2 RC* tanh LSC
L 2D 2l sf
(rF + rC*) + rSC tanh ρchannel L Ka2(1– γ 2)RC* + 2
2lsfSC
lsf tanh LSC
SC
RC* + ρchannel
2D
2l sf
where rF = ρF*lsfF and rSC = ρSClsfSC have units of Ω m2, lsfSC and lsfF are
the spin diffusion lengths in the SC and ferromagnet, respectively
(lsf = √Dτsf, where D = μkBT/q is the spin diffusivity, τsf is the spin- This expression is very similar to that derived in ref. 117 for the
flip time and μ is the electron mobility), and L is the length of the case of graphene, except for the contribution from Rchannel ≥ Rcontact,
SC channel. which must be taken into account in our case where lsfSC  ≤ L. This
The spin injection efficiency, given by γ = (J – J)/(J + J), expression is valid for a ‘confined’ geometry, in which the chan­
depends on both the polarization of the ferromagnet’s density of nel does not extend laterally beyond the electrodes. This has been
states and the amount of spin-flip during tunnelling. This can be discussed in detail in ref. 117.

is that the vdW gap between graphene and MoS2 is about 0.33 nm, resistance. Combining broken inversion symmetry and strong
as shown by DFT simulations104, which is larger than that in typical spin–orbit coupling in monolayer TMDCs leads to an interesting
metal–MoS2 contacts (such as 0.15 nm in Ti–MoS2; ref. 32). spin-split valence band, which enables the electrical manipulation
Graphene can also be used as a buffer layer between MoS2 and a of spins106–108 and makes semiconducting TMDCs an interesting
bulk metal. It has been demonstrated81 that using a layer of graphene material family for spintronics. Calculations have also predicted
between multilayer MoS2 and Ni contacts significantly reduces giant magnetoresistance in Fe/MoS2/Fe sandwich structures109. So
the SBH, resulting in a 20-fold improvement of the contact resist­ far, however, the realization of spin injection from ferromagnetic
ance and yielding one of the best-ever reported contact resistances contacts into TMDCs remains elusive. In this section we will briefly
with TMDCs. This promising result calls for more experiments, discuss the conditions that must be met, with regard to the contacts,
in particular with single- and few-layer TMDCs. Finally, note that to realize spin injection in TMDCs.
although excellent contact resistances were recently achieved by The simplest spintronics device is the ‘spin valve’, in which two
edge-contacting graphene31, so far the only reported attempts of ferromagnetic electrodes, bridged by a non-magnetic material, are
edge-contacting TMDCs have produced very resistive contacts105. polarized either parallel or antiparallel to each other. The magneto­
As we have seen, when scaling devices we must consider the resistance ratio ΔR/R between these two configurations is the fig­
existence of a finite ‘transfer length’ within which charges are ure of merit for this device. The current flow can be represented by
injected. Another limitation comes from the requirement of ‘ohmic’ a resistor network, with two parallel current paths corresponding
behaviour. For a FET to operate properly and without nonlinearities to the two spin channels110 (Fig. 8a). It can thus easily be seen that
in the transistor output, the contact resistance RC should be a small the magnetoresistance ratio drops when the resistance of the SC
fraction of the total resistance. For example, the ITRS 2012 require­ becomes large. The injection of a spin-polarized current into a SC is
ment for low-standby-power silicon-on-insulator FETs, a geometry therefore hampered by the conductivity mismatch between the SC
close to FETs based on 2D SCs, is that the total contact resistance and the (usually metallic) ferromagnetic electrodes111.
should be down to 20% of the total resistance by 2015. As a conse­ The most common way around this problem is to modulate
quence, RC should scale with channel length. The dashed black line the contact resistance by introducing tunnel barriers at the con­
in Fig.  7b depicts this criterion for a channel length of 22  nm. It tacts112,113. Tunnel barriers also have a spin-dependent resistance
is noteworthy that the performance of 1T contacts49 are compliant and can thus play the role of R and R in Fig.  8a, with arbitrar­
with the ITRS 22-nm-node requirement. ily high resistances. However, for very high resistances, the tran­
sit time of electrons becomes too long. Because of this, the contact
Spin injection resistance of the barriers must be well adjusted so that a significant
Not all applications require minimal contact resistance; a prime magneto­resistance signal can be observed.
example of this is the field of spintronics, in which the criterion The magnitude of the magnetoresistance signal as a function of
for success is impedance matching rather than minimizing contact contact resistance can be calculated using the approach proposed

1202 NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials

© 2015 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOI: 10.1038/NMAT4452 REVIEW ARTICLE
Table 1 | Material properties used in calculations of spin injection and contact resistance.
Material ρ2D (Ω ☐–1) T (K) μ (cm2 V–1 s–1) D (m2 s–1) τsf (ps) lSC
sf (nm) L (nm) References
Graphene 10 3
300 7,000 1.8 × 10–2
135 1,600 2,000 116
MoS2 104 74 200 1.3 × 10–3 10 115 100 63,125
MoSe2 104 300 30 7.8 × 10–5 9 26 100 126
WS2 104 74 100 6.4 × 10–4 88 237 100 99,124
WSe2 10 4
4 250 8.6 × 10 –6
1,000 92 100 100,127

in ref. 113 (Box 1). This gives us the magnitude of the ‘local’ signal. must be developed. Recent years have shown impressive progress
The nonlocal signal can be calculated using the approach described towards solving this problem. Several routes towards high-quality
in refs 114,115. Using experimentally determined values of resistiv­ electrical contacts have been identified, the most promising being
ity and valley lifetimes (since the valley and spin degrees of freedom the realization of seamless electrical contacts, in which native chem­
are locked in monolayer TMDCs106) (Table 1), with a spin injection ical bonds allow much easier charge transport and thereby lower
efficiency of γ = 0.5, we can estimate the magnetoresistance ratio for contact resistances. For example, metallic TMDCs can be used as
different TMDCs using equation (9) in Box 1. Note that these values covalently bonded electrical contacts to semiconducting TMDCs, or
apply only to out-of-plane spins, which are stabilized by the strong sp2 carbon–­carbon covalent bonding could be retained at graphene–
spin–orbit coupling. GNR junctions. However, most of the results so far have been
The results are shown in Fig. 8b,c. The much larger spin diffu­ obtained on graphene and MoS2. Material-specific properties such
sion length in graphene (Table  1) is related to its larger mobility. as constituent elements and atomic defects can strongly influence
Nevertheless, Fig.  8b shows that a measurable signal (comparable the electrical properties of the device. In this respect, our under­
with those in refs 115,116, in which Lchannel = 2 μm) can be expected standing of these contacts is still very limited, and more systematic
in reason­ably short junctions of WS2 and WSe2 (Lchannel = 100 nm). studies are needed, particularly on other TMDCs.
It is noteworthy that graphene and TMDCs lie on two differ­ Spintronics, another research area that depends critically on
ent sides of the conductivity-mismatch problem. In graphene, the controlling contact resistance, will gain prominence with the rising
contact resistance of direct ferromagnetic contacts is too low, and value of TMDC materials for spintronic and valleytronic applica­
decoherence occurs mainly through spin escaping into the elec­ tions. Initial estimates show that it should be possible to realize effi­
trodes. Contacts to TMDCs lie on the right side of the bell, where cient spin injection into this class of materials, giving hope that the
decoherence occurs inside the channel117. In graphene, introducing first experimental realizations of this will soon be reported.
a tunnel barrier increases the contact resistance and improves the
signal115. By a fortunate coincidence, the same tunnel barriers tend Received 30 March 2015; accepted 18 September 2015;
to decrease the contact resistance to TMDCs58,60. published online 20 November 2015
Figure 8c,d focuses on the more promising materials WSe2 and
WS2. Figure 8c shows the contact resistance as a function of sheet References
resistivity in order to compare with Fig. 7. The shaded area corre­ 1. Iijima, S. Helical microtubules of graphitic carbon. Nature 354, 56–58 (1991).
sponds to ΔR/R >2% for WSe2 at 4 K (note that the corresponding 2. Hayden, O., Agarwal, R. & Lu, W. Semiconductor nanowire devices.
Nano Today 3, 12–22 (2008).
region for WS2 at 74 K would be very similar). The data for back-
3. Novoselov, K. S. et al. Electric field effect in atomically thin carbon films.
gated single-layer WS2 (T = 74 K)99 lies far above the limit, whereas Science 306, 666–669 (2004).
electrolyte-gated single-layer WSe2 (T  =  4  K)100 displays a much 4. Radisavljevic, B., Radenovic, A., Brivio, J., Giacometti, V. & Kis, A. Single-layer
more appropriate contact resistance range. In addition to making MoS2 transistors. Nature Nanotech. 6, 147–150 (2011).
the contact resistance lower, this kind of doping also reduces the 5. Fang, H. et al. High-performance single layered WSe2 p-FETs with chemically
width of the depletion region, thus making the contact more suit­ doped contacts. Nano Lett. 12, 3788–3792 (2012).
6. Liu, W. et al. Role of metal contacts in designing high-performance monolayer
able for spin injection118. Chloride doping95 in 5–7 layers of WS2
n-type WSe2 field effect transistors. Nano Lett. 13, 1983–1990 (2013).
(square symbol) has also been reported as a representative example 7. Sarkar, D. et al. A subthermionic tunnel field-effect transistor with an
of molecular doping 94,119–123, which is another way of decreasing RC. atomically thin channel. Nature 526, 91–95 (2015).
In Fig. 8d we plot the optimal contact resistance needed to achieve 8. Splendiani, A. et al. Emerging photoluminescence in monolayer MoS2.
maximum magnetoresistance in WS2, along with its evolution as Nano Lett. 10, 1271–1275 (2010).
temperature is varied from 73 K to room temperature. These val­ 9. Mak, K. F., Lee, C., Hone, J., Shan, J. & Heinz, T. F. Atomically thin MoS2:
A new direct-gap semiconductor. Phys. Rev. Lett. 105, 136805 (2010).
ues were calculated using experimentally determined temperature-
10. Coleman, J. N. et al. Two-dimensional nanosheets produced by liquid
dependent figures of the mobility 99 and spin lifetime124. It is clear to exfoliation of layered materials. Science 331, 568–571 (2011).
see that as the temperature is increased the signal vanishes and the 11. Yoon, Y., Ganapathi, K. & Salahuddin, S. How good can monolayer MoS2
requirement on contact resistance becomes more stringent. transistors be? Nano Lett. 11, 3768–3773 (2011).
The use of tunnel barriers has proven to be efficient in reducing 12. Cao, W., Kang, J., Sarkar, D., Liu, W. & Banerjee, K. 2D semiconductor
the contact resistance and the SB between MoS2 and ferro­magnetic FETs—Projections and design for sub-10 nm VLSI. IEEE Trans. Electron Dev.
62, 3459–3469 (2015).
contacts58,60,61. Combined with molecular or electrolyte doping
13. Sarkar, D. et al. MoS2 field-effect transistor for next-generation label-free
strategies, this could lead to the demonstration of lateral spin valves biosensors. ACS Nano 8, 3992–4003 (2014).
based on WS2 or WSe2 transistors in the near future. 14. Lembke, D. & Kis, A. Breakdown of high-performance monolayer MoS2
transistors. ACS Nano 6, 10070–10075 (2012).
Summary and outlook 15. He, K., Poole, C., Mak, K. F. & Shan, J. Experimental demonstration of
Realizing good electrical contacts is a prerequisite for harnessing continuous electronic structure tuning via strain in atomically thin MoS2.
Nano Lett. 13, 2931–2936 (2013).
the full potential of 2D SCs. The atomic-scale thickness and pristine 16. Conley, H. J. et al. Bandgap engineering of strained monolayer and bilayer
surfaces of 2D materials make it difficult to reduce the contact resist­ MoS2. Nano Lett. 13, 3626–3630 (2013).
ance. New theoretical models and experimental approaches better 17. Castellanos-Gomez, A. et al. Local strain engineering in atomically thin MoS2.
suited to the low-dimensionality of the semiconducting material Nano Lett. 13, 5361–5366 (2013).

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1203

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLE NATURE MATERIALS DOI: 10.1038/NMAT4452
18. Bertolazzi, S., Brivio, J. & Kis, A. Stretching and breaking of ultrathin MoS2. 50. Kappera, R. et al. Metallic 1T phase source/drain electrodes for field effect
ACS Nano 5, 9703–9709 (2011). transistors from chemical vapor deposited MoS2. APL Mater. 2, 092516 (2014).
19. Geim, A. K. & Grigorieva, I. V. Van der Waals heterostructures. Nature 51. Lin, Y.‑C., Dumcenco, D. O., Huang, Y.‑S. & Suenaga, K. Atomic mechanism
499, 419–425 (2013). of the semiconducting‑to‑metallic phase transition in single-layered MoS2.
20. Li, L. et al. Black phosphorus field-effect transistors. Nature Nanotech. Nature Nanotech. 9, 391–396 (2014).
9, 372–377 (2014). 52. Sze, S. M. & Ng, K. K. Physics of Semiconductor Devices (Wiley, 2006).
21. Tao, L. et al. Silicene field-effect transistors operating at room temperature. 53. Leonard, F. & Talin, A. A. Electrical contacts to one- and two-dimensional
Nature Nanotech. 10, 227–231 (2015). nanomaterials. Nature Nanotech. 6, 773–783 (2011).
22. Lopez-Sanchez, O., Lembke, D., Kayci, M., Radenovic, A. & Kis, A. 54. Banerjee, K., Amerasekera, A., Dixit, G. & Hu, C. Temperature and current
Ultrasensitive photodetectors based on monolayer MoS2. Nature Nanotech. effects on small‑geometry‑contact resistance. IEEE Int. Electron Dev. Meet.
8, 497–501 (2013). 115–118 (1997).
23. Krasnozhon, D., Lembke, D., Nyffeler, C., Leblebici, Y. & Kis, A. 55. Heine, V. Theory of surface states. Phys. Rev. 138, A1689–A1696 (1965).
MoS2 transistors operating at gigahertz frequencies. Nano Lett. 56. Anwar, A., Nabet, B., Culp, J. & Castro, F. Effects of electron confinement
14, 5905–5911 (2014). on thermionic emission current in a modulation doped heterostructure.
24. Das, S., Chen, H.‑Y., Penumatcha, A. V. & Appenzeller, J. High performance J. Appl. Phys. 85, 2663–2666 (1999).
multilayer MoS2 transistors with scandium contacts. Nano Lett. 57. Mott, N. F. Impurity band conduction. Experiment and theory.
13, 100–105 (2013). The metal-insulator transition in an impurity band. J. Phys. Colloq.
25. Liu, H., Neal, A. T. & Ye, P. D. Channel length scaling of MoS2 MOSFETs. 37, C4‑301–C4-306 (1976).
ACS Nano 6, 8563–8569 (2012). 58. Chen, J.‑R. et al. Control of Schottky barriers in single layer MoS2 transistors
26. Liu, H. et al. Statistical study of deep submicron dual-gated field-effect with ferromagnetic contacts. Nano Lett. 13, 3106–3110 (2013).
transistors on monolayer chemical vapor deposition molybdenum disulfide 59. Qiu, H. et al. Electrical characterization of back-gated bi-layer MoS2
films. Nano Lett. 13, 2640–2646 (2013). field-effect transistors and the effect of ambient on their performances.
27. Landauer, R. Spatial variation of currents and fields due to localized scatterers Appl. Phys. Lett. 100, 123104 (2012).
in metallic conduction. IBM J. Res. Dev. 1, 223–231 (1957). 60. Dankert, A., Langouche, L., Kamalakar, M. V. & Dash, S. P. High-performance
28. Sharvin, Y. V. A possible method for studying Fermi surfaces. Sov. Phys. JETP molybdenum disulfide field-effect transistors with spin tunnel contacts.
21, 655–656 (1965). ACS Nano 8, 476–482 (2014).
29. Jena, D., Banerjee, K. & Xing, G. H. 2D crystal semiconductors: Intimate 61. Wang, W. et al. Controllable Schottky barriers between MoS2 and permalloy.
contacts. Nature Mater. 13, 1076–1078 (2014). Sci. Rep. 4, 6928 (2014).
30. Song, S. M., Park, J. K., Sul, O. J. & Cho, B. J. Determination of work function 62. Guo, Y. et al. Study on the resistance distribution at the contact between
of graphene under a metal electrode and its role in contact resistance. molybdenum disulfide and metals. ACS Nano 8, 7771–7779 (2014).
Nano Lett. 12, 3887–3892 (2012). 63. Radisavljevic, B. & Kis, A. Mobility engineering and a metal–insulator
31. Wang, L. et al. One-dimensional electrical contact to a two-dimensional transition in monolayer MoS2. Nature Mater. 12, 815–820 (2013).
material. Science 342, 614–617 (2013). 64. Chuang, H.‑J. et al. High mobility WSe2 p- and n-type field-effect transistors
32. Kang, J., Liu, W., Sarkar, D., Jena, D. & Banerjee, K. Computational study of contacted by highly doped graphene for low-resistance contacts. Nano Lett.
metal contacts to monolayer transition-metal dichalcogenide semiconductors. 14, 3594–3601 (2014).
Phys. Rev. X 4, 031005 (2014). 65. Liu, H. et al. Phosphorene: An unexplored 2D semiconductor with a high hole
33. Matsuda, Y., Deng, W.‑Q. & Goddard, W. A. Contact resistance for ‘end- mobility. ACS Nano 8, 4033–4041 (2014).
contacted’ metal−graphene and metal−nanotube interfaces from quantum 66. Lince, J., Carré, D. & Fleischauer, P. Schottky-barrier formation on a covalent
mechanics. J. Phys. Chem. C 114, 17845–17850 (2010). semiconductor without Fermi-level pinning: The metal-MoS2(0001) interface.
34. Kang, J., Sarkar, D., Liu, W., Jena, D. & Banerjee, K. A computational Phys. Rev. B 36, 1647–1656 (1987).
study of metal-contacts to beyond-graphene 2D semiconductor materials. 67. Jaegermann, W., Pettenkofer, C. & Parkinson, B. Cu and Ag deposition
IEEE Int. Electron Dev. Meet. 407–410 (2012). on layered p-type WSe2: Approaching the Schottky limit. Phys. Rev. B
35. Liu, W. et al. High-performance few‑layer‑MoS2 field‑effect‑transistor 42, 7487–7496 (1990).
with record low contact-resistance. IEEE Int. Electron Dev. Meet. 68. Schlaf, R., Klein, A., Pettenkofer, C. & Jaegermann, W. Laterally
19.4.1–19.4.4 (2013). inhomogeneous surface-potential distribution and photovoltage at clustered
36. Khatami, Y., Li, H., Xu, C. & Banerjee, K. Metal‑to‑multilayer-graphene In/WSe2(0001) interfaces. Phys. Rev. B 48, 14242–14252 (1993).
contact—Part II: analysis of contact resistance. IEEE Trans. Electron. Dev. 69. Klein, A., Pettenkofer, C., Jaegermann, W., Lux-Steiner, M. & Bucher, E.
59, 2453–2460 (2012). A photoemission study of barrier and transport properties of the interfaces of
37. Khatami, Y., Li, H., Xu, C. & Banerjee, K. Metal‑to‑multilayer-graphene Au and Cu with WSe2(0001) surfaces. Surf. Sci. 321, 19–31 (1994).
contact—Part I: contact resistance modeling. IEEE Trans. Electron. Dev. 70. Maurel, C., Coratger, R., Ajustron, F., Beauvillain, J. & Gerard, P. Electrical
59, 2444–2452 (2012). characteristics of metal/semiconductor nanocontacts using light emission in a
38. Chu, T. & Chen, Z. Understanding the electrical impact of edge contacts in scanning tunneling microscope. J. Appl. Phys. 94, 1979–1982 (2003).
few-layer graphene. ACS Nano 8, 3584–3589 (2014). 71. Maurel, C., Ajustron, F., Péchou, R., Seine, G. & Coratger, R. Electrical
39. Park, H.-Y. et al. Contact resistance reduction on graphene through n-type behavior of the Au/MoS2 interface studied by light emission induced by
doping and one-dimensional edge contact. MRS Spring Meeting T14.14 (2015). scanning tunneling microscopy. Surf. Sci. 600, 442–447 (2006).
40. Stokbro, K., Engelund, M. & Blom, A. Atomic-scale model for the contact 72. Dhanker, R., Chopra, N. & Giebink, N. C. Plasmonic internal photoemission
resistance of the nickel-graphene interface. Phys. Rev. B 85, 165442 (2012). for accurate device in situ measurement of metal–organic semiconductor
41. Popov, I., Seifert, G. & Tománek, D. Designing electrical contacts to MoS2 injection barriers. Adv. Funct. Mater. 24, 4775–4781 (2014).
monolayers: a computational study. Phys. Rev. Lett. 108, 156802 (2012). 73. Afanas’ev, V. V. Internal Photoemission Spectroscopy: Fundamentals and Recent
42. Kang, J., Liu, W. & Banerjee, K. High-performance MoS2 transistors with low- Advances (Elsevier, 2014).
resistance molybdenum contacts. Appl. Phys. Lett. 104, 093106 (2014). 74. Gourmelon, E., Bernède, J. C., Pouzet, J. & Marsillac, S. Textured MoS2 thin
43. Gan, L.‑Y., Zhao, Y.‑J., Huang, D. & Schwingenschlögl, U. First-principles films obtained on tungsten: Electrical properties of the W/MoS2 contact.
analysis of MoS2/Ti2C and MoS2/Ti2CY2 (Y=F and OH) all-2D semiconductor/ J. Appl. Phys. 87, 1182–1186 (2000).
metal contacts. Phys. Rev. B 87, 245307 (2013). 75. McDonnell, S., Addou, R., Buie, C., Wallace, R. M. & Hinkle, C. L.
44. Leong, W. S., Nai, C. T. & Thong, J. T. L. What does annealing do to metal– Defect-dominated doping and contact resistance in MoS2. ACS Nano
graphene contacts? Nano Lett. 14, 3840–3847 (2014). 8, 2880–2888 (2014).
45. Gong, C., Colombo, L., Wallace, R. M. & Cho, K. The unusual mechanism 76. Werner, J. H. & Güttler, H. H. Barrier inhomogeneities at Schottky contacts.
of partial Fermi level pinning at metal–MoS2 interfaces. Nano Lett. J. Appl. Phys. 69, 1522–1533 (1991).
14, 1714–1720 (2014). 77. Lin, Y.‑F. et al. Barrier inhomogeneities at vertically stacked graphene-based
46. Ouyang, Y. & Guo, J. Carbon-based nanomaterials as contacts to graphene heterostructures. Nanoscale 6, 795–799 (2013).
nanoribbons. Appl. Phys. Lett. 97, 263115 (2010). 78. Mathai, A. J., Patel, K. D. & Srivastava, R. Studies on In–pWSe2
47. Kang, J., Sarkar, D., Khatami, Y. & Banerjee, K. Proposal for all-graphene Schottky diode by current–voltage–temperature method. Thin Sol. Film.
monolithic logic circuits. Appl. Phys. Lett. 103, 083113 (2013). 518, 4417–4424 (2010).
48. Nakanishi, T., Koshino, M. & Ando, T. Transmission through a boundary 79. McDonnell, S. et al. Hole contacts on transition metal dichalcogenides:
between monolayer and bilayer graphene. Phys. Rev. B 82, 125428 (2010). Interface chemistry and band alignments. ACS Nano 8, 6265–6272 (2014).
49. Kappera, R. et al. Phase-engineered low-resistance contacts for ultrathin MoS2 80. Park, W. et al. Contact resistance reduction using Fermi level de-pinning layer
transistors. Nature Mater. 13, 1128–1134 (2014). for MoS2 FETs. IEEE Int. Electron Dev. Meet. 5.1.1–5.1.4 (2014).

1204 NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials

© 2015 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOI: 10.1038/NMAT4452 REVIEW ARTICLE
81. Leong, W. S. et al. Low resistance metal contacts to MoS2 devices with 108. Mak, K. F., McGill, K. L., Park, J. & McEuen, P. L. The valley Hall effect in
nickel‑etched‑graphene electrodes. ACS Nano 9, 869–877 (2015). MoS2 transistors. Science 344, 1489–1492 (2014).
82. Chuang, S. et al. MoS2 p-type transistors and diodes enabled by high work 109. Dolui, K., Narayan, A., Rungger, I. & Sanvito, S. Efficient spin injection
function MoOx contacts. Nano Lett. 14, 1337–1342 (2014). and giant magnetoresistance in Fe/MoS2/Fe junctions. Phys. Rev. B
83. Baugher, B. W. H., Churchill, H. O. H., Yang, Y. & Jarillo-Herrero, P. 90, 041401 (2014).
Optoelectronic devices based on electrically tunable p–n diodes in a 110. Baibich, M. N. et al. Giant magnetoresistance of (001)Fe/(001)Cr magnetic
monolayer dichalcogenide. Nature Nanotech. 9, 262–267 (2014). superlattices. Phys. Rev. Lett. 61, 2472–2475 (1988).
84. Cheng, R. et al. Electroluminescence and photocurrent generation from 111. Schmidt, G., Ferrand, D., Molenkamp, L. W., Filip, A. T. & van Wees, B. J.
atomically sharp WSe2/MoS2 heterojunction p–n diodes. Nano Lett. Fundamental obstacle for electrical spin injection from a ferromagnetic metal
14, 5590–5597 (2014). into a diffusive semiconductor. Phys. Rev. B 62, R4790–R4793 (2000).
85. Lee, C.‑H. et al. Atomically thin p–n junctions with van der Waals 112. Rashba, E. I. Theory of electrical spin injection: Tunnel contacts
heterointerfaces. Nature Nanotech. 9, 676–681 (2014). as a solution of the conductivity mismatch problem. Phys. Rev. B
86. Podzorov, V., Gershenson, M. E., Kloc, C., Zeis, R. & Bucher, E. High- 62, R16267–R16270 (2000).
mobility field-effect transistors based on transition metal dichalcogenides. 113. Fert, A. & Jaffrès, H. Conditions for efficient spin injection from a
Appl. Phys. Lett. 84, 3301–3303 (2004). ferromagnetic metal into a semiconductor. Phys. Rev. B 64, 184420 (2001).
87. Cohen, S. S. Contact resistance and methods for its determination. 114. Takahashi, S. & Maekawa, S. Spin injection and detection in magnetic
Thin Sol. Film. 104, 361–379 (1983). nanostructures. Phys. Rev. B 67, 052409 (2003).
88. Berger, H. H. Models for contacts to planar devices. Solid State Electron. 115. Han, W. et al. Tunneling spin injection into single layer graphene.
15, 145–158 (1972). Phys. Rev. Lett. 105, 167202 (2010).
89. Scott, D. B., Hunter, W. R. & Shichijo, H. A transmission line model 116. Tombros, N., Jozsa, C., Popinciuc, M., Jonkman, H. T. & van Wees, B. J.
for silicided diffusions: Impact on the performance of VLSI circuits. Electronic spin transport and spin precession in single graphene layers at
IEEE J. Solid-State Circuits 17, 281–291 (1982). room temperature. Nature 448, 571–574 (2007).
90. Nagashio, K., Nishimura, T., Kita, K. & Toriumi, A. Metal/graphene contact 117. Seneor, P. et al. Spintronics with graphene. MRS Bull. 37, 1245–1254 (2012).
as a performance killer of ultra-high mobility graphene analysis of intrinsic 118. Albrecht, J. D. & Smith, D. L. Electron spin injection at a Schottky contact.
mobility and contact resistance. IEEE Int. Electron Dev. Meet. 1–4 (2009). Phys. Rev. B 66, 113303 (2002).
91. Xia, F., Perebeinos, V., Lin, Y., Wu, Y. & Avouris, P. The origins and limits of 119. Zhao, P. et al. Air stable p-doping of WSe2 by covalent functionalization.
metal–graphene junction resistance. Nature Nanotech. 6, 179–184 (2011). ACS Nano 8, 10808–10814 (2014).
92. Liu, H. et al. Switching mechanism in single-layer molybdenum disulfide 120. Chen, K. et al. Air stable n-doping of WSe2 by silicon nitride thin films with
transistors: An insight into current flow across Schottky barriers. ACS Nano tunable fixed charge density. APL Mater. 2, 092504 (2014).
8, 1031–1038 (2014). 121. Fang, H. et al. Degenerate n-doping of few-layer transition metal
93. Baugher, B. W. H., Churchill, H. O. H., Yang, Y. & Jarillo-Herrero, P. Intrinsic dichalcogenides by potassium. Nano Lett. 13, 1991–1995 (2013).
electronic transport properties of high-quality monolayer and bilayer MoS2. 122. Farmer, D. B. et al. Chemical doping and electron−hole conduction
Nano Lett. 13, 4212–4216 (2013). asymmetry in graphene devices. Nano Lett. 9, 388–392 (2009).
94. Kiriya, D., Tosun, M., Zhao, P., Kang, J. S. & Javey, A. Air-stable surface 123. Sarkar, D. et al. Functionalization of transition metal dichalcogenides with
charge transfer doping of MoS2 by benzyl viologen. J. Am. Chem. Soc. metallic nanoparticles: Implications for doping and gas-sensing. Nano Lett.
136, 7853–7856 (2014). 15, 2852–2862 (2015).
95. Yang, L. et al. Chloride molecular doping technique on 2D materials: WS2 and 124. Mai, C. et al. Exciton valley relaxation in a single layer of WS2 measured by
MoS2. Nano Lett. 14, 6275–6280 (2014). ultrafast spectroscopy. Phys. Rev. B 90, 041414 (2014).
96. Cui, X. et al. Multi-terminal transport measurements of MoS2 using a van der 125. Mai, C. et al. Many-body effects in valleytronics: Direct measurement of valley
Waals heterostructure device platform. Nature Nanotech. 10, 534–540 (2015). lifetimes in single-layer MoS2. Nano Lett. 14, 202–206 (2014).
97. Li, S.‑L. et al. Thickness scaling effect on interfacial barrier and electrical 126. Kumar, N., He, J., He, D., Wang, Y. & Zhao, H. Valley and spin dynamics in
contact to two-dimensional MoS2 layers. ACS Nano 8, 12836–12842 (2014). MoSe2 two-dimensional crystals. Nanoscale 6, 12690–12695 (2014).
98. Liu, W., Sarkar, D., Kang, J., Cao, W. & Banerjee, K. Impact of contact 127. Wang, G. et al. Valley dynamics probed through charged and neutral exciton
on the operation and performance of back-gated monolayer MoS2 emission in monolayer WSe2. Phys. Rev. B 90, 075413 (2014).
field‑effect‑transistors. ACS Nano 9, 7904–7912 (2015). 128. Du, Y., Liu, H., Deng, Y. & Ye, P. D. Device perspective for black phosphorus
99. Ovchinnikov, D., Allain, A., Huang, Y.‑S., Dumcenco, D. & Kis, A. Electrical field-effect transistors: Contact resistance, ambipolar behavior, and scaling.
transport properties of single-layer WS2. ACS Nano 8, 8174–8181 (2014). ACS Nano 8, 10035–10042 (2014).
100. Allain, A. & Kis, A. Electron and hole mobilities in single-layer WSe2.
ACS Nano 8, 7180–7185 (2014).
101. Yoon, J. et al. Highly flexible and transparent multilayer MoS2 transistors with Acknowledgements
graphene electrodes. Small 9, 3295–3300 (2013). This work was performed in frames of the Marie Curie ITN network MoWSeS
102. Bertolazzi, S., Krasnozhon, D. & Kis, A. Nonvolatile memory cells based on (grant no. 317451). We acknowledge funding by the European Commission under the
MoS2/graphene heterostructures. ACS Nano 7, 3246–3252 (2013). Graphene Flagship (grant agreement no. 604391), the US National Science Foundation
103. Yu, L. et al. Graphene/MoS2 hybrid technology for large-scale two- under Grant CCF-1162633 and the US Air Force Office of Scientific Research under
dimensional electronics. Nano Lett. 14, 3055–3063 (2014). Grant FA9550-14-1-0268 (R18641).
104. Ma, Y., Dai, Y., Guo, M., Niu, C. & Huang, B. Graphene adhesion on MoS2
monolayer: An ab initio study. Nanoscale 3, 3883–3887 (2011).
105. Xu, S. et al. High-quality BN/WSe2/BN heterostructure and its quantum Additional information
oscillations. Preprint at http://arxiv.org/abs/1503.08427 (2015). Reprints and permissions information is available online at www.nature.com/reprints.
106. Xiao, D., Liu, G.‑B., Feng, W., Xu, X. & Yao, W. Coupled spin and valley Correspondence should be addressed to K.B. and A.K.
physics in monolayers of MoS2 and other group-VI dichalcogenides.
Phys. Rev. Lett. 108, 196802 (2012).
107. Gong, K. et al. Electric control of spin in monolayer WSe2 field effect Competing financial interests
transistors. Nanotechnology 25, 435201 (2014). The authors declare no competing financial interests.

NATURE MATERIALS | VOL 14 | DECEMBER 2015 | www.nature.com/naturematerials 1205

© 2015 Macmillan Publishers Limited. All rights reserved