ARTICLESPUBLISHED ONLINE: 5 MAY 2013 | DOI: 10.1038/NMAT3633

Grains and grain boundaries in highly crystallinemonolayer molybdenum disulphideArend M. van der Zande1,2*†, Pinshane Y. Huang3†, Daniel A. Chenet2†, Timothy C. Berkelbach4,YuMeng You5, Gwan-Hyoung Lee2,6, Tony F. Heinz1,5, David R. Reichman1,4, David A. Muller3,7

and James C. Hone1,2

Recent progress in large-area synthesis of monolayer molybdenum disulphide, a new two-dimensional direct-bandgapsemiconductor, is paving the way for applications in atomically thin electronics. Little is known, however, about themicrostructure of this material. Here we have refined chemical vapour deposition synthesis to grow highly crystalline islandsof monolayer molybdenum disulphide up to 120µm in size with optical and electrical properties comparable or superior toexfoliated samples. Using transmission electron microscopy, we correlate lattice orientation, edge morphology and crystallinitywith island shape to demonstrate that triangular islands are single crystals. The crystals merge to form faceted tilt and mirrortwin boundaries that are stitched together by lines of 8- and 4-membered rings. Density functional theory reveals localizedmid-gap states arising from these 8–4 defects. We find that mirror twin boundaries cause strong photoluminescence quenchingwhereas tilt boundaries cause strong enhancement. Meanwhile, mirror twin boundaries slightly increase the measured in-planeelectrical conductivity, whereas tilt boundaries slightly decrease the conductivity.

Characterizing the structure and properties of grains andgrain boundaries is critical for understanding and control-ling material properties in the expanding array of two-

dimensional (2D) materials1. Studies have mapped the grainstructure of large-area graphene grown by chemical vapourdeposition (CVD), characterized defects such as grain boundariesat the atomic scale and demonstrated that these grain boundariescan strongly affect graphene’s electrical, optical and mechanicalproperties2–11. Much less is known about the grain structure andproperties of defects in other 2D materials, such as molybdenumdisulphide (MoS2). MoS2 has been widely explored in the formof nanotubes12, nanoparticles13 and thin films14,15 owing to itsexcellent tribological and catalytic properties. Recent work hasshown that exfoliated monolayer MoS2 is a 2D direct-bandgapsemiconductor16,17. Moreover, large-areamonolayerMoS2 can nowbe synthesized using CVD (refs 18–21), making it a promising can-didate for building atomically thin layered electrical22–24, optical16,25and photovoltaic26 devices.

Here, we grow high-quality crystals of monolayer MoS2exhibiting grain sizes up to 120 µm and possessing optical andelectronic properties comparable or superior to those of exfoliatedsamples. Our study applies a diverse set of characterizationmethods, combining transmission electron microscopy (TEM)of atomic and crystal structure with optical spectroscopy andelectrical transport to determine the influence of grain boundarieson the properties of the material. Using diffraction-filtered andatomic-resolution electron microscopy, we characterize single-crystal triangular islands and polycrystals containing tilt andmirrortwin grain boundaries. We findmirror twin boundaries are stitchedtogether predominantly through lines of 8- and 4-membered

1Energy Frontier Research Center, Columbia University, New York, New York 10027, USA, 2Department of Mechanical Engineering, Columbia University,New York, New York 10027, USA, 3School of Applied and Engineering Physics, Cornell University, Ithaca, New York 14853, USA, 4Department ofChemistry, Columbia University, New York, New York 10027, USA, 5Departments of Physics and Electrical Engineering, Columbia University, New York,New York 10027, USA, 6Samsung-SKKU Graphene Center (SSGC), Suwon, 440-746, Korea, 7Kavli Institute at Cornell for Nanoscale Science, Ithaca,New York 14853, USA. †These authors contributed equally to this work. *e-mail: av2466@columbia.edu.

rings, which produce mid-gap states in DFT calculations. Finally,we show that individual grain boundaries strongly affect thephotoluminescence observed inMoS2 monolayers and slightly altertheir in-plane electrical conductivity.

Growth and characterization of large-grain MoS2 crystalsFigure 1a,b shows optical images of monolayer MoS2 grown ona Si/SiO2 substrate. The samples are grown by CVD with solidMoO3 and S precursors18. In contrast to previous work18, we didnot use seeds to nucleate growth; instead, our best growths wereobtained with carefully cleaned substrates and by minimizing theexposure of the precursors to air during storage (see Methodsand Supplementary Fig. S1 for details). Figure 1a shows an opticalimage of a typical sample. Isolated islands (violet triangles) haveedge lengths ranging from 30 to 80 µm. On the right (nearest tothe solid-state precursors during growth), the islands merge intoa continuous film. Although non-uniformity in nucleation densityand crystal size is a limitation of our growth techniques, eachgrowth contains a ∼1×15mm region where hundreds of similar,isolated islands grow. Across different growths, the average sizeof islands varies between 20 and 100 µm, with individual trianglesup to 120 µm (Supplementary Fig. S2). Most islands are uniformmonolayers, except for small bilayer or multilayer patches (visiblein the centre of Fig. 1b).

We first use photoluminescence to measure the quality andthickness of these triangles. Figure 1c shows photoluminescencespectra frommonolayer and bilayer MoS2. The photoluminescencepeaks at 1.84 and 1.95 eV respectively match the A and B direct-gap optical transitions16. The integrated intensity of the bilayerpeak is much weaker (∼7%) than the monolayer peak, reflecting

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 1© 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT3633

Energy (eV)

I / I Si

Monolayer

Bilayer

Mo

S

dc

100 µm

a

0.0

0.5

1.0

1.5

2.0

2.5

1.4 1.6 1.8 2.0 2.2

e

b

5 Å 10 µm

123 µm

Figure 1 | Large-grain MoS2 growth. a, Optical reflection image of a CVD growth of a typical large-grain MoS2 on a SiO2 (285 nm)/Si substrate. The imagecontrast has been increased for visibility; magenta is the bare substrate, and violet represents monolayer MoS2. b, Optical image of a monolayer MoS2

triangle. The triangle is 123 µm from tip to tip. c, Photoluminescence spectra from monolayer (red) and bilayer (blue) MoS2. Peak height is normalized tothe silicon Raman peak. The narrow spikes at high energy are the Raman transitions (see Supplementary Fig. S3a). d, High-resolution ADF-STEM image offreely suspended monolayer MoS2 on a TEM grid. The bright spots are molybdenum atoms; the grey spots are two stacked sulphur atoms. The lattice iscomposed of hexagonal rings alternating molybdenum and sulphur sites; top view and side views of the structure are overlaid. e, DF-TEM image of a largetriangle with the diffraction pattern inset. Together, the diffraction pattern and the DF-TEM image show that the triangle is a continuous single crystal. The∼2–4 µm brighter and darker areas are rotationally aligned bilayers of MoS2. Similar variations in contrast have been observed in bilayer graphene, wherethey reflect differences in stacking order29.

the expected evolution from a direct-gap to an indirect-gapsemiconductor. Surprisingly, the measured range of peak widths ofour CVDmonolayer MoS2 (50–60meV) is similar to that observedfor freely suspended samples of exfoliated MoS2 (also 50–60meV)and is much narrower than that of MoS2 exfoliated onto SiO2(100–150meV; ref. 16). These results suggest that the CVD-grownsamples are either of higher quality or are in a cleaner electrostaticenvironment than exfoliated samples.

To characterize the crystal structure of these islands, we used acombination of TEM, electron diffraction techniques and atomicresolution STEM. Figure 1d shows an image of the atomic structureof our CVD MoS2 monolayers taken by aberration-correctedannular dark-field scanning transmission electron microscopy(ADF-STEM). In ADF-STEM, a 60 keV, ångstrom-scale electronbeam scans over the sample, and an image is formed by collectingthe medium- to high-angle scattered electrons. The contrast ofADF-STEM images scales roughly as the square of the atomicnumber Z (ref. 27). Thus, the brightest spots in Fig. 1d arethe molybdenum atoms and the dimmer spots are the twostacked sulphur atoms. The hexagonal lattice is clearly visible, asindicated by the top- and side-view schematics in Fig. 1d. No pointdefects, atomic substitutions or voids were initially observed withinsingle crystals. However, defects readily form under extendedimaging (see Methods)28.

We used selected-area electron diffraction and dark-fieldTEM (DF-TEM) to characterize the large-scale crystal structureof the islands. Figure 1e, inset shows a selected-area electrondiffraction pattern of a triangular island roughly 45 µm across.The six-fold symmetry in the position of the diffraction spotsdemonstrates that the triangle contains no rotational boundaries,to within our measurement accuracy of 0.5◦. Figure 1e shows thecorresponding DF-TEM image: by using an aperture to selecta narrow range of diffracted beams, DF-TEM filters images

by crystallographic orientation2,11,29,30, and shows single crystalssuch as Fig. 1e in a single image. Similar analyses of dozens oftriangles show that triangular-shaped islands are predominantlysingle crystals. Under continued growth, these islands mergetogether to form aggregates or continuous sheets (Fig. 1a andSupplementary Fig. S2).

The large-grain, highly crystalline MoS2 monolayers, withoptical properties equivalent or superior to exfoliated samples,enable new experiments studying the growth, structure andproperties ofMoS2 crystals and grain boundaries.

Grain structure and orientation of CVD MoS2

We use electron diffraction techniques to uniquely identify thecrystal structure and edge orientation ofMoS2 triangles. Figure 2a,bshows the bright-field image and diffraction pattern of a single-crystal triangle. As seen in the schematic in Fig. 2a, the latticeof monolayer MoS2 is divided into molybdenum and sulphursublattices, which reduces the hexagonal lattice from six-fold tothree-fold symmetry. As a result, the six [1̄100] diffraction spotsare broken into two families: ka = {(1̄100), (101̄0), (01̄10)} andkb =−ka (Fig. 2b; ref. 31). Through both experiment and Blochwave simulations32 (see Supplementary Methods), we find the kaspots are higher in intensity (∼10% for an 80 kV electron beam)and point towards the molybdenum sublattice, as indicated bythe arrows in Fig. 2a,b.

We likewise observe this intensity asymmetry in DF-TEM.Figure 2d shows the bright-field image and diffraction pattern oftwo triangles. An aperture positioned at the magenta circle on theinset in Fig. 2d forms the dark-field image in Fig. 2e. As the trianglesare rotated 180◦ from one another, the aperture simultaneouslycollects the bright ka spot for the left triangle and the darkerkb spot for the right triangle. This produces a correspondingcontrast difference between the islands. We can thus use either

2 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

© 2013 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT3633 ARTICLES

2 µma

Rela

tive

inte

nsity

c

ed

Position

b

kb1 µm

ka

ka

kb

ka kb

[ ¬2110]

[ ¬1100

]

[1 ¬100

]

[ ¬2110]

0.0

0.5

1.0

1 µm

Experiment Simulation

Figure 2 | Diffraction imaging of crystal orientation and edge terminations. a, Bright-field image of a single-crystal triangle with a Mo-zigzag edgeorientation. b, Diffraction pattern from a. The asymmetry of the Mo and S sublattices separates the [̄1100] diffraction spots into two families:ka={(1̄100),(101̄0),(01̄10)} and kb=−ka. c, A line profile through experimentally measured diffraction spots (black) and Bloch-wave simulations (red).The higher intensity ka spots point towards the Mo sublattice, as indicated by the arrows in a,b. Both curves were normalized to the height of the [2̄110]peaks, and the red curve was offset horizontally for visibility. d, Bright-field TEM image of two triangles with S-zigzag edge orientations. The curvedappearance of the crystal edges contrasts with the sharper crystal edges of the Mo-zigzag edges in a. The inset diffraction pattern shows the location ofthe aperture used to form the image in e. e, Dark-field image of the region in d. As the triangles are rotated 180◦ from one another, the aperturesimultaneously collects the brighter ka spot for the left triangle and the darker kb spot for the right triangle. This produces a corresponding contrastdifference between the two islands that allows us to uniquely infer crystallographic orientation from dark-field images. To improve visibility, we haveclipped the bottom of the intensity range in this image.

the diffraction pattern or DF-TEM to uniquely identify the latticeorientation of MoS2 crystals.

Figure 2 also shows the dominant morphologies of MoS2triangles: molybdenum zigzag (Mo-zz) in Fig. 2a and sulphurzigzag (S-zz) in Fig. 2d (refs 33,34). These two edge terminationsare commonly observed in MoS2 nanocrystals33, and they are thetwo most energetically stable edge orientations34. We consistentlyobserve thatMo-zz triangles have sharper, straighter edges than S-zztriangles. This morphological difference allows us to easily identifycrystal orientation in optical images. High-resolution STEM imagesshow that despite their sharp appearance in DF-TEM, the edgesof both our S-zz and Mo-zz crystals are rough at the nanoscale(Supplementary Fig. S4).

In addition to single-crystal triangles, the CVD process producespolycrystalline aggregates that can be characterized by DF-TEM.Figure 3a shows two S-zz triangles intersecting at an angle of40◦± 0.5◦. Figure 3b shows a corresponding composite DF-TEMimage of the crystal structure. The composite image is constructedby overlaying colour-coded DF-TEM images2 acquired from thecrystallographic orientations shown in the diffraction pattern ofFig. 3a. The intersection of the grains forms an abrupt, faceted tiltboundary. Low-quality growths can produce the types of film inFig. 3c, which shows small, misshapen, crystals. The correspondingdiffraction pattern and dark-field overlay in Fig. 3d shows thatthe disordered crystals are randomly oriented polycrystallineaggregates. As indicated by the arrows in Fig. 3c, adjacent MoS2grains sometimes overlap to form rotationally misaligned bilayerregions. Similar combinations of abrupt and overlapped crystalinterfaces have been seen in CVD-grown graphene2,5,35. Unlikein graphene, the grain boundaries in MoS2 tend to be stronglyfaceted on the micrometre scale. These facets may originate fromthe relative surface energy of different edge configurations36,which will be more directionally dependent in MoS2 than in

homoelemental structures. Similar phenomena probably underliethe differences in edge roughness we see in Mo-zz versus S-zztriangles and the faceting at the edges of large crystals suchas in Fig. 1b.

In addition to tilt boundaries, we observed a second commonline defect in MoS2 crystals: mirror twin boundaries. In monolayerMoS2, mirror twins are the intersections of two MoS2 crystalswith a relative in-plane rotation of 180◦ that effectively swapsthe positions of Mo and S lattice sites. Figure 3e shows thebright-field image of two triangles meeting at 180◦, and Fig. 3fshows the corresponding DF-TEM image. The mirror twin appearsclearly as an intensity change in DF-TEM images, through thesame diffraction asymmetry seen in Fig. 2e. Unlike tilt boundaries,mirror twin boundaries such as in Fig. 3e,f grow within nucleationislands. Figure 3g,h and Supplementary Fig. S5 show anothercommonly observed shape that contains mirror twin boundaries:a symmetric, six-pointed star containing cyclic twins that resembleother inorganic and geological crystal growths37.

The atomic structure of MoS2 grain boundariesWe next examine the atomic structure of MoS2 grain boundariesusing ADF-STEM. Figure 4a,b show a mirror twin boundary ina 6-pointed star. Whereas the grain boundary follows the zigzagdirection on the micrometre scale, it exhibits faceting of ±20◦relative to this direction on the nanometre scale. This nanoscalefaceting is surprising because it avoids the highly symmetriczigzag boundary structure that would correspond to the microscalealignment seen in Fig. 3h. Figure 4b shows the atomic structure ofthe grain boundary. The overlaid purple and green polygons showthat the boundary is formed primarily from 4- and 8-memberedrings, with a recurring periodic 8-4-4 ringmotif.

STEM images in an ArXiv preprint posted during the reviewprocess of this work show a MoS2 tilt boundary formed from

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 3© 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT3633

c

d

e

f

g

h

a

b

10 µm 5 µm 1 µm 5 µm

Figure 3 | Tilt and mirror twin grain structures. a, Bright-field TEM image of two triangles that have grown together. Inset diffraction pattern shows the twocrystal orientations are 40◦±0.5◦ apart (measured between the red and cyan lines, which are at equivalent spots in the diffraction pattern).b, Colour-coded overlay of DF-TEM images corresponding with the two red- and cyan-circled spots in a shows a tilt grain boundary as a faceted lineconnecting the two triangles. c, Bright-field image of a region containing irregularly shaped MoS2 islands, with diffraction pattern inset. Red arrows indicateregions where adjacent grains overlap, forming rotationally misaligned bilayers. d, Coloured DF-TEM overlay shows that the irregular shapes arepolycrystalline aggregates. The crystals are connected both by faceted, abrupt grain boundaries, and∼1 µm overlapped bilayers. e, Bright-field image of amirror twin composed of 180◦-rotated triangles, with diffraction pattern inset. f, Dark-field image corresponding with the orange circle in e shows the twotriangles have different diffraction intensity, similar to the 180◦-rotated triangles in Fig. 2d,e. This intensity change indicates the presence of a mirror twinboundary between the darker and lighter regions. The small triangle in the centre is multilayer MoS2. g, Bright-field image of a 6-pointed star, with insetdiffraction pattern. h, DF-TEM image corresponding with the orange circle in g shows that the star contains several rotationally symmetric mirror twins,forming a cyclic twin.

a line of 5–7 rings38. This finding complements our atomic-resolution images of mirror twin structures: both tilt and mirrortwin boundaries are commonly found in our films. These resultsare consistent with a recent theory paper that predicts that inMoS2,either odd- or even-membered rings can form grain boundariesdepending on the tilt angle and stoichiometry, among severalfactors39. For comparison, graphene’s tilt boundaries are mostcommonly formed by 5- and 7-membered rings2,40, whereas twinboundaries have been observed with 8-5-5 motifs4. Our data alsosuggest that a priori simulation of boundary structures may bechallenging: for example, our 8-4-4 motif does not match thetheoretically predicted structure in ref. 39, mostly owing to itssurprising faceted structure.

In our observed 8-4-4 motif, neighbouring 4-rings meet at asulphur site, which appears to have four nearest neighbours ratherthan the usual three. This change in coordination requires either theS atoms to have dangling bonds, or the Mo atoms to change theiroxidation state. Using first-principles density functional theory41(DFT), we find that after relaxation, the 8-4-4 boundary structureis at a local energetic minimum (Fig. 4c).

Figure 4d shows the DFT-calculated density of states (DOS) ofdefect-free monolayer MoS2 (black curve). The light green bandindicates the bandgap of pristine MoS2. When a grain boundaryis implanted into the MoS2, the DOS develops mid-gap states(pristine + GB, red dashed curve) that appear mainly in theprojected DOS of the atoms in the boundary (GB alone, filledblue curve). Figure 4e shows a spatial plot of the local DOS ofpristine+ GB in the plane of the Mo atoms, integrated in theenergy range of the pristine bandgap; themid-gap states are spatiallylocalized at the boundary. Analogous calculations on alternativemirror twin boundary structures (Supplementary Fig. S6) confirmthat this behaviour is generic. These localizedmid-gap states, typicalof defects in semiconductors, are important because they can affectthe optical and transport properties of thematerial4,42.

Optical properties of MoS2 grain boundariesUsing the understanding of the grain structure developed above, wecan systematically explore the effects of individual grain boundarieson the optical and electronic properties ofMoS2.

We measure the optical properties of MoS2 usingphotoluminescence mapping, where we step a focused excitationlaser over the sample and record photoluminescence spectra ateach point. Figure 5a,e shows optical images of a mirror twin island(Fig. 5a) and a tilt boundary island (Fig. 5e). Figure 5b–d,f–h showscorresponding maps of integrated photoluminescence intensity,peak position and peak width. Away from the boundaries, bothislands show similar emission intensities, peak positions (1.84 eV)and peak widths (57meV in the mirror sample and 60meV inthe tilt sample, reflecting typical sample-to-sample variation). Atthe grain boundaries, the islands show strong photoluminescencemodifications: the mirror boundary is a clearly visible straight linewith 50% quenching of intensity (Fig. 5b), an 8meV upshift in peakenergy (Fig. 5c) and a 5meV peak-broadening (Fig. 5d). The tiltboundary, which has a faceted structure similar to Fig. 3b, shows asurprising 100% enhancement in emission strength, 26meVupshiftand 5meV broadening. The width of the signal change aroundthe tilt boundary is also larger than around mirror boundaries.Supplementary Fig. S3b,c shows corresponding Raman mapsof the tilt boundary.

The data demonstrate that photoluminescence is stronglyaffected by the presence of grain boundaries and that thesechanges depend on the grain boundary angle. Several factors maycontribute. In the simplest theory, photoluminescence quenchingarises from defects in semiconductors, such as mid-gap statesat the boundaries, which can act as centres for non-radiativerecombination43. Our calculations indicate that these effects,even after accounting for exciton diffusion, are too small toexplain the photoluminescence results (Supplementary Methods).Two further aspects of the grain boundaries probably play a

4 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

© 2013 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT3633 ARTICLES

zig-zag20°

20°

5 nm

1 nm

c

¬2 ¬1 0 1 2Energy (eV)

0.0

0.5

1.0

1.5

2.0

Pristine

GB alone

2.5 e

Pristine + GB

DO

S (e

V¬

1 /at

om)

d

Local midgap DOS

b

a

Figure 4 | Grain boundary atomic structure. a, High-resolution ADF-STEM image of a mirror twin boundary. The boundary is visible just below theannotated line. The annotation indicates the nanoscale faceting of the boundary at±20◦ off of the zigzag direction. b, Zoomed-in image of the grainboundary shows a periodic line of 8-4-4 ring defects. c, An atomistic model of the experimental structure shown in b. Energy minimization with DFTconfirms that this boundary is locally stable. d, The total DOS of defect-free, that is pristine, MoS2 (black), the total DOS of MoS2 with the grain boundary(red dashed) and the projected DOS of the atoms along the grain boundary (blue filled). The dashed grey line denotes the Fermi energy of pristine MoS2,and the light green shaded area indicates the pristine bandgap. All states have been given a Gaussian broadening of 0.07 eV. e, A 2D spatial plot of thelocal mid-gap DOS (integrated in the plane of the Mo over a 1.7 eV range about the Fermi energy of the pristine MoS2). The colour scale of the density is0–0.05 states per bohr3.

role in controlling photoluminescence: doping and strain. First,local changes in doping may occur at grain boundaries. Forexample, our observed 8-4-4 defects in the mirror boundaryare molybdenum rich, which would n-dope the boundary,whereas 5–7 defects observed at a tilt boundary are sulphur-rich38, which would p-dope the boundary. Previous investigationshave shown that photoluminescence is strongly affected bycharge density, with increased/decreased electron density causingphotoluminescence quenching/enhancement44. The influence ofcharge is thus consistent with the observed trends in the strengthof the photoluminescence at these different types of boundary.Other sources of doping can include spatial differences in growthkinetics or the preferential attachment of adsorbates at boundaries.Second, strain around the boundaries maymodify the bandgap45 orcause the boundary region to lift off the electrically disordered SiO2surface, enhancing photoluminescence emission16.

Electrical properties of MoS2 grain boundariesAn important question for device applications is whether grainboundaries disrupt or modify electrical transport. We fabricatedfield-effect transistors (FETs) from islands containing a single grainboundary (Fig. 5i, inset). The FET channels followed three differentconfigurations: within a grain (pristine), across the grain boundary(perpendicular) and along the grain boundary (parallel). Figure 5i,jshows plots of conductance versus gate voltage (transfer curves) forfour representative devices fabricated on mirror twin islands: twopristine devices (black and magenta), one parallel device (orangecurve) and one perpendicular device (cyan curve).

First, we compare the variability of pristine, single-crystaldevices. In devices from different growths, room-temperature,field-effect mobilities ranged from 1 to 8 cm2 V−1 s−1, and typicalon/off ratios ranged between 105 and 107. Multiple devicesfabricated within each single crystal yielded mobilities that wereidentical to within our measurement error. Specifically, the pristinedevices in Fig. 5i,j (black and magenta curves) show n-typebehaviourwithmobilities of 3–4 cm2 V−1 s−1. These ranges of valuesare comparable to those reported for back-gated FETs fabricatedwith mechanically exfoliated MoS2 on SiO2 (refs 22,24,46,47;in the absence of high-K dielectrics) and equivalent to the bestreported values for monolayer CVDMoS2 (refs 18,19). Our presentmobilities are probably limited by substrate effects and the localelectrostatic environment48,49.

Next we compare transport across grain boundaries. In themirror twin device shown in Fig. 5i,j, the perpendicular device (cyancurve) shows nearly identical performance to the pristine devices(magenta and black curves), indicating that themirror twin bound-ary has little effect on channel conductivity. However, the electricalcharacteristics of the parallel devices differed measurably from thepristine devices. Although the transfer curves of the parallel devicein Fig. 5i,j (orange curves) are qualitatively similar to that of thepristine devices, they have a 25% larger on-state conductivity and60% larger off-state conductivity. This change is only slightly largerthan the device-to-device variation, but was consistent across fourparallel devices measured on mirror twins (particularly in the offstate). The data impose a rough upper limit for conductivity alongthe one-dimensional grain boundary of 1.5 µS-µm in the on-state

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 5© 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT3633

Mirror boundary

a dc

e

i j

10 µm

g

b

f h

10 µm

5 µm

2.5

2.0

1.5

1.0

0.5

0.0

100

10¬1

10¬2

10¬3

10¬4

10¬5

G

(µS)

G

(µS)

¬70 ¬35 0

Vg (V)

35 70 ¬70 ¬35 0Vg (V)

35 70

Tilt

bou

ndar

yM

irro

r bo

unda

ry

Optical imagePhotoluminescence

quantum yieldPhotoluminescence

peak positionPhotoluminescence

peak width

Parallel to GBPerpendicular to GBPristinePristine

Figure 5 | Optical and electronic properties of mirror and tilt boundaries. a–d, Optical measurements of an island containing a mirror twin boundary.e–h, Corresponding measurements for an island containing a tilt boundary. a,e are optical images; b–d and f–h are colour plots of photoluminescence. Inb,f, red is the relative quantum yield, with colour scale 0–1100 a.u. We see 50% quenching at the mirror twin boundary and a 100% enhancement at the tiltboundary. In c,g, green is the peak position, with colour scale 1.82–1.87 eV. There is an upshift of 8 meV at the mirror twin boundary, and a much stronger26 meV upshift in the tilt boundary. In d,h, cyan is the peak width with colour scale of 53–65 meV. The peak broadens from 55 to 62 meV at the boundaryin both samples. i,j, Linear and logarithmic electrical transport transfer curves of 4 FETs fabricated from the mirror twin MoS2 island shown in the inset of i.The curves correspond with pristine regions (magenta and black), and regions containing a grain boundary running perpendicular (cyan) and parallel(orange) to the flow of electrons. All data were measured at room temperature, using the Si growth substrate as a back-gate and a source–drain bias of500 mV.

(Fig. 5j at Vg =+70V) and 120 pS-µm in the off state (Fig. 5j atVg =−70V). These values can be compared to overall sheet con-ductivity for pristine devices of 2.2 µS/� in the on state and 56 pS/�in the off state. This analysis indicates that the few-atom-wide twinboundaries, although still semiconducting, have similar conductiv-ity of up to a 1-µm-wide strip of pristinematerial. This result is con-sistent with the mid-gap states we predict in the material. The fulleffect ofmid-gap states, however, is probably limited by the disorderin the grain boundary, whichwill prevent a continuumof states.

In contrast to our measurements of the mirror twin, FETdevices of tilt boundaries show a decrease in conductance inboth the parallel and perpendicular configurations. On one tiltboundary, shown in Supplementary Fig. S8, the on-state boundaryconductance of both parallel and perpendicular devices decreases by

30% compared with pristine crystals. However, measuring six angledevices yielded no consistent change in conductance, decreasingby 5–80% compared with pristine. This variability is unsurprising:in graphene, electrical measurements are highly dependent onthe structure of the boundary5. We expect these variations to bewider and more complex in semiconducting MoS2, where theelectronic structure of the boundary should depend on the tiltangle and atomic structure39. Significantly, in all mirror twin andtilt boundaries we measured, the change in conductance from anysingle grain boundary is only slightly more than the device-to-device variation—an important conclusion for the applicability ofMoS2-based electronics.

In this work, we produced monolayers of large-area, highlycrystallineMoS2 on insulating surfaces and imaged the crystal edges,

6 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

© 2013 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT3633 ARTICLESgrain structure and grain boundaries with electron microscopy.We identified the location of individual grain boundaries on thesurface and systematically compared the optical and electronicproperties to electronic structure calculations. In particular, weused the repeatability of the mirror twin structure to correlatemeasurements of the structure, photoluminescence and electronictransport of a single grain boundary structure. In tilt boundaries,photoluminescence and electrical transport measurements varymarkedly for different boundaries. Controlling grain boundarieswill be essential for optimizing MoS2electronic and photovoltaicdevices, where defects in large-area heterostructures can lead tointerlayer leakage or unwanted exciton recombination.

MoS2 is just the first member of a large family of layeredtransitionmetal dichalcogenides to be synthesized as amonolayer50.The above techniques are crucial for exploring synthesis strategiesto optimize the grain properties, and provide a template for studiesof the microscopic and macroscopic impact of grain structure onother monolayer membranes. Our results are thus a significant stepforward in realizing the ultimate promise of atomic membranes inelectronic, mechanical and energy-harvesting devices.

MethodsGrowth procedure. We grew monolayer MoS2 by CVD using a method similarto ref. 18, but without seeds to nucleate growth. Growth substrates were Si with285 nm of SiO2. Substrates were cleaned in acetone and isopropanol, followedby 2 h in H2SO4/H2O2 (3:1) and 5min of O2 plasma. They were then loadedinto a 2-inch CVD furnace and placed face-down above a crucible containing14mg of MoO3 (≥99.5% Sigma Aldrich #100932642) with another cruciblecontaining 120mg of sulphur (≥99.5% Sigma Aldrich #101144903) locatedupstream. The CVD growth is performed at atmospheric pressure while flowingultrahigh-purity nitrogen. The growth recipe is: sit 4 h at 105 ◦C with 500 s.c.c.m.,ramp to 700 ◦C at 15 ◦Cmin−1 with 10 s.c.c.m., sit 5min at 700 ◦C, cool to570 ◦C without feedback (∼20min) with 10 s.c.c.m., open furnace and flow 500s.c.c.m. for rapid cooling.

High yield is achieved by using ultraclean substrates and with fresh precursors.The yield is significantly lowered by dirty substrates and old precursors. In thesecases, any monolayers will be smaller and misshapen, as outlined in SupplementaryFig. S1. Closest to the source, the film becomes a mixture of multilayer stacksand other crystal allotropes, whereas furthest from the source, nothing grows (seeFig. 1a and Supplementary Fig. S2).

TEM sample preparation. For the TEM grids, we use 10-nm-thick amorphousSiN windows (TEMwindows.com) for DF-TEM samples and holey Quantifoilgrids (Ted Pella) for ADF-STEM. Polymer transfer layers were prepared byspinning poly(methyl methacrylate) (PMMA) A2 onto finished MoS2 on siliconoxide/silicon chips at 4000 r.p.m. for 60 s to form 50-nm-thick layers. The chipswere floated on 1M KOH. The KOH etched the silicon oxide epi-layer, causingthe chips to fall off, and leaving the polymer and MoS2-coated polymer membranefloating on the surface. The membrane was transferred to deionized water severaltimes, and then scooped onto a TEM grid and dried. The PMMA was removed bybaking the TEM grids in an ultrahigh-vacuum heater at 300 ◦C for 10min or anatmospheric pressure Ar/H2 gas flow for 4 h. Before atomic-resolution imaging, webaked the samples overnight in an ultrahigh vacuum at 130 ◦C.

ADF-STEM. ADF-STEM imaging was conducted using a NION UltraSTEM100.Although our 60 kV beam is below threshold for damage in MoS2 (ref. 28),defects readily form under extended imaging, possibly catalysed by surfacecontaminants from the transfer process. Imaging conditions were similarto those in ref. 27. Using a 25–30mrad convergence angle, our probe sizewas close to 1.3 Å. STEM images were acquired with the medium-angleannular dark-field detector with acquisition times of between 16 and40 µs per pixel. The atomic resolution images in Fig. 4 were smoothed toimprove contrast.

DF-TEM. TEM imaging and diffraction were conducted using an FEI TechnaiT12 operated at 80 kV, which did not cause any apparent damage to the MoS2membranes at these lower dose densities. Acquisition times for DF-TEMimages (both displaced-aperture and centred DF-TEM) were 5–50 s perframe. The intensity ranges of DF-TEM images were generally clipped toimprove contrast.

DFT modelling. DFT calculations were performed with the PW91 generalizedgradient approximation for the exchange-correlation functional and ultrasoftpseudopotentials, as implemented in the Quantum Espresso electronic structure

package41. Structural relaxations were carried out at the gamma point, andsingle-point calculations were performed with appropriately convergedMonkhorst-type k-point grids. Supercells were built with about 10 Å separationbetween replicas to achieve negligible interaction.

Photoluminescence. The photoluminescence measurements were performed ina Renishaw InVia Raman Microscope using a 532 nm laser and a 1,800 lmm−1grating. The curves in Fig. 1c were taken at 100 µW laser power for 10 s. Althoughthe indirect bandgap transition is not visible in the bilayer, this is not unusual ona SiO2 substrate. Photoluminescence maps were taken at 100 µW laser power for1 s with a 300 nm step size.

Electrical measurements. We fabricated back-gated FETs directly on the285-nm-thick SiO2 dielectric on degenerately doped silicon growth substrates.Grain boundaries were identified by photoluminescence as shown in Fig. 5. TheMoS2 was shaped by patterning a PMMAmask with electron-beam lithography andetching with O2 plasma for 20 s at 50 W with 20 s.c.c.m. flow rate. The electrodeswere then patterned by electron beam lithography with a bilayer PMMA stack anda subsequent evaporation of Al/Cr/Au (50 nm/5 nm/50 nm). We did not anneal thesamples after fabrication.

All devices were measured at room temperature under 10−5 torr insidea Lakeshore probe station, using the silicon substrates as a global backgate.We used an Agilent 4155C semiconductor parameter analyser to perform2-point conductance measurements. FET transfer curves shown in Fig. 5 weretaken at 500mV drain bias. The devices showed ohmic contacts from −500 to500mV source–drain bias.

Received 2 November 2012; accepted 15 March 2013;published online 5 May 2013

References1. Novoselov, K. S. et al. Two-dimensional atomic crystals. Proc. Natl Acad. Sci.

USA 102, 10451–10453 (2005).2. Huang, P. Y. et al. Grains and grain boundaries in single-layer graphene atomic

patchwork quilts. Nature 469, 389–392 (2011).3. Li, X. et al. Large-area synthesis of high-quality and uniform graphene films on

copper foils. Science 324, 1312–1314 (2009).4. Lahiri, J., Lin, Y., Bozkurt, P., Oleynik, I. I. & Batzill, M. An extended defect in

graphene as a metallic wire. Nature Nanotech. 5, 326–329 (2010).5. Tsen, A. W. et al. Tailoring electrical transport across grain boundaries in

polycrystalline graphene. Science 336, 1143–1146 (2012).6. Grantab, R., Shenoy, V. B. & Ruoff, R. S. Anomalous strength characteristics of

tilt grain boundaries in graphene. Science 330, 946–948 (2010).7. Ruiz-Vargas, C. S. et al. Softened elastic response and unzipping in

chemical vapor deposition graphene membranes. Nano Lett. 11,2259–2263 (2011).

8. Liu, Y. & Yakobson, B. I. Cones, pringles, and grain boundary landscapes ingraphene topology. Nano Lett. 10, 2178–2183 (2010).

9. Yazyev, O. V. & Louie, S. G. Topological defects in graphene: Dislocations andgrain boundaries. Phys. Rev. B 81, 195420 (2010).

10. Yu, Q. et al. Control and characterization of individual grains and grainboundaries in graphene grown by chemical vapour deposition. Nature Mater.10, 443–449 (2011).

11. Huang, P. Y., Meyer, J. C. & Muller, D. A. From atoms to grains: Transmissionelectron microscopy of graphene.Mater. Res. Soc. Bull. 37, 1214–1221 (2012).

12. Brontvein, O. et al. New high-temperature Pb-catalyzed synthesis of inorganicnanotubes. J. Am. Chem. Soc. 134, 16379–16386 (2012).

13. Li, Y. G. et al. MoS2 nanoparticles grown on graphene: An advancedcatalyst for the hydrogen evolution reaction. J. Am. Chem. Soc. 133,7296–7299 (2011).

14. Eda, G. et al. Photoluminescence from chemically exfoliated MoS2. Nano Lett.11, 5111–5116 (2011).

15. Botello-Mendez, A. R., Lopez-Urias, F., Terrones, M. & Terrones, H.Metallic and ferromagnetic edges in molybdenum disulphide nanoribbons.Nanotechnology 20, 325703 (2009).

16. Mak, K. F., Lee, C., Hone, J., Shan, J. & Heinz, T. F. Atomically thin MoS2: Anew direct-gap semiconductor. Phys. Rev. Lett. 105, 136805 (2010).

17. Splendiani, A. et al. Emerging photoluminescence in monolayer MoS2.Nano Lett. 10, 1271–1275 (2010).

18. Lee, Y. H. et al. Synthesis of large-area MoS2 atomic layers with chemicalvapour deposition. Adv. Mater. 24, 2320–2325 (2012).

19. Zhan, Y. J., Liu, Z., Najmaei, S., Ajayan, P. M. & Lou, J. Large-area vapor-phasegrowth and characterization of MoS2 atomic layers on a SiO2 substrate. Small8, 966–971 (2012).

20. Liu, K-K. et al. Growth of large-area and highly crystalline MoS2 thin layers oninsulating substrates. Nano Lett. 12, 1538–1544 (2012).

21. Shi, Y. M. et al. van der Waals epitaxy of MoS2 layers using graphene As growthtemplates. Nano Lett. 12, 2784–2791 (2012).

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 7© 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT3633

22. Radisavljevic, B., Radenovic, A., Brivio, J., Giacometti, V. & Kis, A. Single-layerMoS2 transistors. Nature Nanotech. 6, 147–150 (2011).

23. Dean, C. R. et al. Boron nitride substrates for high-quality graphene electronics.Nature Nanotech. 5, 722–726 (2010).

24. Britnell, L. et al. Field-effect tunneling transistor based on vertical grapheneheterostructures. Science 335, 947–950 (2012).

25. Yin, Z. et al. Single-layer MoS2 phototransistors. ACS nano 6, 74–80 (2012).26. Fontana, M. et al. Photovoltaic effect in gated MoS2 Schottky junctions.

Preprint at http://arxiv.org/arXiv:1206.6125 (2012).27. Krivanek, O. L. et al. Atom-by-atom structural and chemical analysis by

annular dark-field electron microscopy. Nature 464, 571–574 (2010).28. Komsa, H-P. et al. Two-dimensional transition metal dichalcogenides under

electron irradiation: Defect production and doping. Phys. Rev. Lett. 109,035503 (2012).

29. Brown, L. et al. Twinning and twisting of tri-and bi-layer graphene. Nano Lett.12, 1609–1615 (2012).

30. Hirsch, P., Howie, A., Nicholson, R., Pashley, D. W. & Whelan, M. J. ElectronMicroscopy of Thin Crystals (Kreiger, 1965).

31. Brivio, J., Alexander, D. T. L. & Kis, A. Ripples and layers in ultrathin MoS2membranes. Nano Lett. 11, 5148–5153 (2011).

32. Spence, J. C. H. & Zuo, J. M. Electron Microdiffraction (Plenum, 1992).33. Lauritsen, J. V. et al. Size-dependent structure of MoS2 nanocrystals. Nature

Nanotech. 2, 53–58 (2007).34. Byskov, L., Nørskov, J., Clausen, B. & Topsøe, H. Edge termination of MoS2

and CoMoS catalyst particles. Catal Lett. 64, 95–99 (2000).35. Robertson, A. W. et al. Atomic structure of interconnected few-layer graphene

domains. ACS Nano 5, 6610–6618 (2011).36. Howe, J. M. Interfaces in Materials: Atomic Structure, Thermodynamics and

Kinetics of Solid-vapour, Solid–Liquid and Solid–Solid Interfaces (Wiley, 1997).37. Zhang, H., Ha, D-H., Hovden, R., Kourkoutis, L. F. & Robinson, R. D.

Controlled synthesis of uniform cobalt phosphide hyperbranched nanocrystalsusing tri-n-octylphosphine oxide as a phosphorus source. Nano Lett. 11,188–197 (2011).

38. Najmaei, S. et al. Vapor phase growth and grain boundary structure ofmolybdenum disulphide atomic layers. Nature Mater. Preprint at http://arxiv.org/arXiv:1301.2812 (2013).

39. Zou, X., Liu, Y. & Yakobson, B. I. Predicting dislocations and grain boundariesin two-dimensional metal-disulphides from the first principles. Nano Lett. 13,253–258 (2012).

40. Kim, K. et al. Grain boundary mapping in polycrystalline graphene. ACS Nano5, 2142–2146 (2011).

41. Paolo, G. et al. QUANTUM ESPRESSO: A modular and open-source softwareproject for quantum simulations of materials. J. Phys. Condens. Matter 21,395502 (2009).

42. Pantelides, S. T. The electronic structure of impurities and other point defectsin semiconductors. Rev. Mod. Phys. 50, 797–858 (1978).

43. Pankove, J. I. Optical Processes in Semiconductors (Prentice-Hall, 1971).44. Mak, K. F. et al. Tightly bound trions in monolayer MoS2. Nature Mater. 12,

207–211 (2013).45. Scalise, E., Houssa, M., Pourtois, G., Afanas’ev, V. V. & Stesmans, A.

Strain-induced semiconductor to metal transition in the two-dimensionalhoneycomb structure of MoS2. Nano Res. 5, 43–48 (2012).

46. Late, D. J., Liu, B., Matte, H. S., Dravid, V. P. & Rao, C. N.Hysteresis in single-layer MoS2 field effect transistors. ACS Nano 6,5635–5641 (2012).

47. Qiu, H. et al. Electrical characterization of back-gated bi-layer MoS2 field-effecttransistors and the effect of ambient on their performances. Appl. Phys. Lett.100, 123104 (2012).

48. Das, S., Chen, H-Y., Penumatcha, A. V. & Appenzeller, J. High performancemultilayer MoS2 transistors with scandium contacts. Nano Lett. 13,100–105 (2012).

49. Yoon, Y., Ganapathi, K. & Salahuddin, S. How good can monolayer MoS2transistors be? Nano Lett. 11, 3768–3773 (2011).

50. Gutiérrez, H. R. et al. Extraordinary room-temperature photoluminescencein triangular WS2 monolayers. Nano Lett. http://dx.doi.org/10.1021/nl3026357 (2012).

AcknowledgementsOverall project coordination, sample growth, and electrical and optical characterizationwere supported as part of the Center for Re-Defining Photovoltaic EfficiencyThrough Molecular-Scale Control, an Energy Frontier Research Center fundedby the US Department of Energy (DOE), Office of Science, Office of BasicEnergy Sciences under Award DE-SC0001085. A.M.v.d.Z was supported bythe EFRC as a research fellow. Electron microscopy was performed at andsupported by the Cornell Center for Materials Research, an NSF MRSEC (NSFDMR-1120296). P.Y.H. was supported under the National Science FoundationGraduate Research Fellowship Grant No. DGE-0707428. D.A.C. was supportedby a Columbia University Presidential fellowship and a GEM PhD Fellowshipsponsored by the Center for Functional Nanomaterials at Brookhaven NationalLab. T.C.B. was supported under the Department of Energy Office of ScienceGraduate Fellowship Program (DOE SCGF), administered by ORISE-ORAUunder Contract No. DE-AC05-06OR23100. The authors thank S. Gondarenko,R. Hovden, I. Meric, J. Ravichandran, L. Wang, J. Richmond-Decker and P. Kimfor helpful discussions.

Author contributionsA.M.v.d.Z. supervised and coordinated all aspects of the project. MoS2 growth was carriedout by D.A.C. and A.M.v.d.Z. Electrical characterization and analysis was carried out byA.M.v.d.Z., D.A.C. and G-H.L. under the supervision of J.C.H. Electron microscopy anddata analysis were carried out by P.Y.H. with D.A.M.’s supervision. Optical spectroscopyand data analysis were carried out by A.M.v.d.Z. and Y.Y. under T.F.H.’s supervision.DFT simulations were carried out by T.C.B. under D.R.R.’s supervision. A.M.v.d.Z.,P.Y.H., D.A.C., T.C.B., Y.Y., T.F.H., D.A.M. and J.C.H. wrote the paper.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints. Correspondenceand requests for materials should be addressed to A.M.v.d.Z.

Competing financial interestsThe authors declare no competing financial interests.

8 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

© 2013 Macmillan Publishers Limited. All rights reserved.