Semiconducting graphene from highly ordered substrate interactions
M. S. Nevius, M. Conrad, F. Wang, A. Celis, M. N. Nair, A. Taleb-Ibrahimi, A. Tejeda, E. H. Conrad
SSemiconducting graphene from highly ordered substrate interactions
M.S. Nevius, M. Conrad, F. Wang, A. Celis,
2, 3
M.N. Nair, A. Taleb-Ibrahimi, A. Tejeda,
2, 3 and E.H. Conrad ∗ The Georgia Institute of Technology, Atlanta, Georgia 30332-0430, USA Laboratoire de Physique des Solides, Universit Paris-Sud,CNRS, UMR 8502, F-91405 Orsay Cedex, France Synchrotron SOLEIL, L’Orme des Merisiers, Saint-Aubin, 91192 Gif sur Yvette, France UR1 CNRS/Synchrotron SOLEIL, Saint-Aubin, 91192 Gif sur Yvette, France
While numerous methods have been proposed to produce semiconducting graphene,a significant bandgap has never been demonstrated. The reason is that, regardless ofthe theoretical gap formation mechanism, disorder at the sub-nanometer scale preventsthe required chiral symmetry breaking necessary to open a bandgap in graphene. Inthis work, we show for the first time that a 2D semiconducting graphene film can bemade by epitaxial growth. Using improved growth methods, we show by direct bandmeasurements that a bandgap greater than 0.5 eV can be produced in the first graphenelayer grown on the SiC(0001) surface. This work demonstrates that order, a propertythat remains lacking in other graphene systems, is key to producing electronicallyviable semiconducting graphene.
It is well known that the first graphene layer grown onthe SiC(0001) surface is not electronic graphene. Thatis, the first “buffer” graphene layer does not show thelinear dispersing π -bands (Dirac cone) expected at the K -point of metallic graphene.[1–3] The lack of π -bandsin experimental band maps of the buffer layer[2] sup-ported the theoretical conclusion that sufficiently strongcovalent bonds between the buffer layer and the SiC in-terface would push the graphene π -bands below the SiCvalance band maximum.[4, 5] Aside from these very earlystudies, research on the SiC graphene buffer layer fadedand was subsequently eclipsed by a wide variety of otherunsuccessful ideas to open a band gap in exfoliated orchemical vapor deposition (CVD)-grown graphene.[8]One method to open a band gap in graphene is by pe-riodic bonding to either all A or all B sites, which breaksgraphene’s chiral symmetry (referred to as graphenefunctionalization). The buffer graphene, commensu-rately bonded to the SiC(0001) surface, should have beenan excellent example of a functionalized system that in-duces a bandgap. Despite the buffer graphene’s potentialto be functionalized by a commensurate and, most im-portantly, ordered array of Si or C atoms in the SiC, therewas a major research shift to functionalize CVD-growngraphene. Efforts to functionalize CVD graphene by anumber of other methods have been a major researcharea. As of this writing, no functionalized graphene, orgraphene modified by any other proposed method, hasbeen developed that produces a workable semiconductingform of graphene. The problem with these methods is theinherent disorder introduced in the functionalization[7]and growth process.[8] In fact, the lack of a graphenebandgap was the motivation to shift research to metaldichalcogenides despite the inability to grow them at thelevel of purity and order required for industrial scale elec-tronics.In this work, we use furnace-grown graphene to pro- FIG. 1. (a) An ARPES cut through the graphene K -point ofan under-grown 6 √ k y is perpendicular to Γ − K .The states g and g (dashed lines) observed by Emtsev et al.[2] are marked. (b) The same cut as in (a) for growth 20 ◦ Chigher. Circles mark the peak positions along part of the (cid:15) band. duce a structurally well ordered buffer graphene onthe SiC(0001) surface. Angle resolved photoemission(ARPES) measurements show new dispersing π -bandsthat are not observed in samples grown by previous meth-ods. These bands live above the SiC valance band max-imum near the Fermi Energy, E F . The new band struc-ture is a result of improved order caused by a highergrowth temperature which, for the first time, gives riseto a well ordered 6 × > . × a r X i v : . [ c ond - m a t . m t r l - s c i ] M a y FIG. 2. A constant E − E F = − × s , s ).Umklapp scattering of the Dirac cones from SiC 1 × × × the buffer graphene layer on SiC is a true semiconductor,the goal of the first graphene electronics research.[10, 11]The substrates used in these studies were n-doped n = 2 × cm − CMP polished on-axis 4H-SiC(0001).The graphene was grown in a controlled silicon sublima-tion furnace.[12] Graphene growth is a function of tem-perature, time, and crucible geometry that sets the sil-icon vapor pressure. With the current crucible design,a monolayer (ML) graphene film will grow at 1520 ◦ C in20 min. Using the same crucible, the semiconductingbuffer layer discussed in this paper will grow at a tem-perature 160 ◦ C lower than the ML in the same amount oftime. Growing 20 ◦ C lower than the optimum buffer tem-perature will still give the same (6 √ × √ ◦ LEEDpattern (referred to as 6 √ π -bands discussed below.Early ARPES work on the UHV-grown 6 √ g and g at -0.5 and -1.6 eV were theonly band features between E F and the SiC valance bandmaximum.[2] These states were interpreted as localizedMott-Hubbard states hybridized from SiC surface dan-gling bonds. We can reproduce these states by heatingthe SiC 20 ◦ C cooler than the optimal buffer growth tem-perature. Figure 1(a) shows an ARPES cut through thegraphene K -point from this “sub-buffer” film. The twosurface states seen in previous work are clearly visible.However, by heating 20 ◦ C higher, a new dispersing band, (cid:15) ( k ) appears [see Fig. 1(b)]. The new surface state is ro-bust, being reproducible in multiple samples. Note thata faint linear Dirac cones appears at k y = 0. This is dueto a small amount of ML graphene ( < E F ).Another indication of the improved sample order isthe quality of the monolayer grown above the optimumbuffer. Figure 2 shows a constant energy cut throughpart of the Brillouin Zone (BZ) of a ML graphene film.In addition to the Dirac cone, replicas of the Dirac conefrom umklapp scattering processes are also visible. Allreplica cones from the K th K -point can be indexed us-ing reciprocal lattice vectors of the SiC 6 × G K ( m, n ) = m s + n s , where | s | = | s | = | a ∗ SiC | [seeFig. 2]. In the ordered ML films, replica cones are clearlyseen from both 1 st -order in the 6 × s , s )and from multiple scattering processes involving 1 st -order( s , s ) plus a SiC G vector (e.g., the G K (¯7 ,
0) and G K (¯7 , st -order replicas (i.e, n, m = 1).[14] The fact that so manyARPES replicas bands are observed in these films, alongwith the 6th order x-ray diffraction rods,[9] testifies tothe film’s improved order.Detailed ARPES measurements from these improvedsamples show that the (cid:15) ( k ) band [Fig. 1(b)] is a gappedgraphene π -band. Figure 3(a) shows a constant energycut though part of the BZ of a buffer layer graphene nearthe (cid:15) ( k ) band maximum. Three lobes are visible. Theselobes represent a second dispersing bands, (cid:15) ( k ), that ismarked in the Γ KM (cid:48) cut in Fig. 3(b) and (c). Again, aDirac cone from a small amount of ML graphene is visi-ble. The two bands are independent of the perpendicularmomentum k ⊥ ( E ) and therefore cannot be due to bulkbands. The tops of both bands lie ∆ E ∼ . E F , or 1.8 eV above the valance band maximum of SiCinterface, indicating that the buffer is a wide band gapsemiconducting form of graphene. A schematic of thetwo bands is shown in Fig. 4. The (cid:15) ( k ) band appears asa gapped π -band that disperses slower perpendicular toΓ K than along either Γ K or KM directions [see Table I].The linear part of the (cid:15) ( k ) band has a velocity, v , that issignificantly lower than the Fermi velocity, v F , reducingto nearly half v F perpendicular to Γ K [see Table I]. TABLE I. The band velocity ( v ) and effective mass ( m ∗ ) nearthe K -point near the π -band maximum. m ∗ is estimatedassuming parabolic bands near E F .Band v/v F m ∗ /m e ML Dirac cone 1.0 - (cid:15) ( ⊥ Γ K ) 0 . ± .
01 1 . ± . (cid:15) (Γ K ) 0 . ± . . ± . (cid:15) ( KM ) 0 . ± . . ± . (cid:15) ( ⊥ Γ K ) 0 . ± .
07 0 . ± . (cid:15) (Γ K ) - 1 . ± . The (cid:15) ( k ) band is 3-fold symmetric, extending towardsΓ and dispersing perpendicular to Γ K . Figure 3(c) shows FIG. 3. (a) A constant energy cut through the graphene BZ near the K-point ( E − E F = − . hν = 70eV). Dashed linesmark the boundary of the BZ. (b) A cut through the surface bands in the Γ KM (cid:48) direction. Circles mark the peak positionsalong part of the (cid:15) and (cid:15) band along with a few higher binding energy bands. A weak Dirac cone from a partial ML is shown.(c) (b) A cut perpendicular to Γ K through the (cid:15) band [vertical black dashed line in (a)]. Circles mark the peak positions ofthe (cid:15) band. a cut perpendicular through the lobe in Fig. 3(a). Theband velocity of (cid:15) ( k ) perpendicular to Γ K is nearly thesame as monolayer graphene [see Table I]. The (cid:15) bandhas an effective mass ( m ∗ ) that ranges between 0.55 to1.5 m e , while (cid:15) is a light band perpendicular to Γ K .In a broad sense, the gapped band structure near the K -point strongly suggests chiral symmetry breaking thatmixes the π bands from the K and K (cid:48) points.[15] Anyperiodic potentials that break the AB symmetry in thegraphene through bond formation, chemical or strainfields, or finite size effects can open a gap in graphene.Weak interactions like those in bilayer graphene onlyproduce small gaps.[16] The strain necessary to induceKekule distortions,[17] which produce band gaps of theorder observed in these samples, are large enough totear the graphene.[18] That strain level is inconsistentwith the 0.4% strain measured by X-ray scattering.[9, 19]Strong bonding using Aryl functionalization[7, 20] hasgiven large gaps, but the disorder inherent during thefunctional group’s incorporation into the graphene lat-tice leads to poorly resolved band structure and lowmobilities.[21]A theoretical understanding of the buffer layer, andtherefore an understanding of the origin of the observedgap, is difficult because of the excessive calculation timeassociated with exploring different models for the large6 √ √ √ × √ bonds between 2/3 of the interfacial Si atoms FIG. 4. A schematic representation of the (cid:15) ( k ) and (cid:15) ( k )buffer layer bands near the top of the π -bands around the K -point. and the buffer buffer graphene caused the π -bands toshift above and below the conduction band minimumand valance band maximum, respectively. The calcula-tions also predicted a metallic, slightly delocalized, sur-face state near E F due to the remaining unbounded SiCSi atoms in the interface, similar to the states observedexperimentally in the earlier, less ordered samples likein Fig. 1(a).[2] These approximate models are clearly in-sufficient to explain the observed bands. Only one abinito calculation by Kim et al.[22] has calculated theband structure for the buffer using a full 6 √ FIG. 5. (a) The calculated graphene buffer layer (6 √ × √ ◦ cell on a relaxed bulk terminated SiC(0001)surface.[22] Green open circles are buffer carbon atoms thatare bonded to the SiC surface. Black circles are carbon notbonded to substrate. Chains of atoms define superhexgon re-gions. (b) An outline of the calculated bands (red dashedlines) from a bulk terminated SiC plus buffer layer fromRef. [22] overlaid onto the experimental bands. The theo-retical bands have been shifted 0.13 eV lower to better matchthe (cid:15) band. interface,[22] it does give some insight into the origin ofthe observed gap when compared to the ARPES results.Kim et al.[22] find that about 25% of the carbonatoms in the buffer graphene are covalently bonded toSi atoms on the SiC interface. The resulting structureis a hexagonal network of graphene ribbon-like struc-tures with the remaining buffer carbon atoms covalentlybonded to the SiC surface [see Fig. 5(a)]. Similar hexag-onal networks, either structural or electronic, have beenobserved in Scanning Tunneling Microscopy or producedtheoretically.[23–25] Kim et al.’s DFT calculations showthat the π -orbitals of carbon atoms on the superhexago-nal boundaries (or ribbons) give rise to several bands nearthe K -point above and below E F . These bands are over-laid on our measured bands in Fig. 5(b). We have shiftedthe calculated bands by -0.13eV to match the (cid:15) ( k ) bandmaximum. Like the experimental (cid:15) ( k ) and (cid:15) ( k ) bands,the theoretical model shows that the covalent bondingto the SiC does not completely destroy the π -bands asearlier calculations predicted. Nonetheless, the calcula- tions do not reproduce several important features of theexperimental bands. The calculations to not predict theformation of a band gap. They also do not correctly re-produce the dispersion of the (cid:15) band, especially from Γto K .We suggest that while the ribbon structure producedin the model of Kim et al.[22] may be correct, the largeamount of covalent bonds associated with a bulk ter-minated SiC likely over estimates the graphene-SiC in-teraction. It is more likely that the buffer graphene isbonded to the SiC through a smaller number of sites.A lower number of graphene-Si bonds is more consis-tent with STM measurements that suggest the bufferlies above a small set of Si-trimers.[26] A reduced buffer-SiC bonding geometry is also consistent with both x-rayscattering[27] and x-ray standing wave experiments,[28]which find a reduced Si-concentration and an increasedC-concentration in the SiC layer below the buffer. Wesuggest that reduced substrate bonding would still besufficient to strain the buffer to produce the observed rib-bon network. The ribbon network, which is now isolatedfrom the substrate, would result in a finite size inducedband gap, as observed.[29]In this work we show that, despite claims to the con-trary, a semiconducting form of graphene can be manu-factured. Using improved growth methods, we have beenable to produce a well ordered single layer of graphene onthe SiC(0001) surface. The first graphene layer, known asthe buffer layer, is a semiconducting form of graphene.Using ARPES, we show that the improved sample or-der leads to new bands with a band maximum that lies∆ E ∼ . > . ∗ email: [email protected][1] T. Ohta, et al. , Phys. Rev. Lett. , 206802 (2007).[2] K.V. Emtsev, et al. , Phys. Rev.
B 77 , 155303 1-10 (2008).[3] C. Riedl, et al. , Phys. Rev. Lett. , 246804 (2009).[4] F. Varchon, et al. , Phys. Rev. Lett. , 126805 (2007).[5] A. Mattausch and O.Pankratov, Phys. Rev. Lett. ,076802 (2007). [6] K. S. Novoselov, et al. , Nature , 192 (2012).[7] E Bekyarova, S Sarkar, S Niyogi, M E Itkis, and R CHaddon,
J. Phys. D: Appl. Phys. , 154009 (2012).[8] K.S. Novoselov, V.I. Fal’ko, L. Colombo, P.R. Gellert,M.G. Schwab, and K. Kim, A roadmap for graphene, Nature , 192 (2012).[9] M. Conrad, F. Wang, M.S. Nevius, A. Coati, P.F. Miceli,and E.H. Conrad (to be published).[10] C. Berger, et al. , J. Phys. Chem.
B 108 , 19912 (2004).[11] C. Berger, et al. , Science , 1191 (2006).[12] W.A. de Heer, et al. , Proc. Nat. Acad. Sci. , 16900-16905 (2011).[13] K.V. Emtsev, et al. , Nature Mat. , 203 (2009).[14] A. Bostwick, et al. , Nature Phys. , 36 (2007).[15] E. McCann, V. I. Fal’ko, Phys. Rev. Lett. , 086805(2006).[16] T. Ohta, et al. , Science , 951-954 (2006). [17] K. Okahara, et al. , Chem. Phys. Lett. , 462468 (1994).[18] S-H. Lee, et al. , ACS Nano , 29642969 (2011).[19] T. Schumann, et al. , Phys. Rev.
B 90 , 041403(R) (2014).[20] E. Bekyarova, et al. , it J. Am. Chem. Soc. , 13361337(2009).[21] S. Niyogi, et al. , Nano Lett. , 40614066 (2010).[22] S. Kim, J. Ihm, H.J. Choi, and Y.-W. Son, Phys. Rev.Lett. , 176802 (2008).[23] F. Owman and P. M˚artensson,
Surf. Sci. , 126 (1996).[24] C. Riedl, et al. , Phys. Rev.
B 76 , 245406 (2007).[25] F. Varchon, P. Mallet, J.-Y. Veuillen, and L. Magaud,
Phys. Rev.
B 77 , 235412 (2008).[26] G.M. Rutter, et al. ,[27] J. Hass, J. E. Mill´an-Otoya, P. N. First, and E. H. Con-rad,
Phys. Rev.
B 78 , 205424 (2008).[28] J.D. Emery, et al. , Phys. Rev. Lett. , 215501 (2013).[29] M. Fujita, K. Wakabayashi, K. Nakada and K. Kusakabe,
J. Phys. Soc. Japan65