Journal of the Korean Physical Society, Vol. 55, No. 1, July 2009, pp.

299303

Diameter Dependence of Charge Transport across Carbon Nanotube-Metal Contacts from First Principles
Yong-Hoon Kim
School of Computational Sciences, Korea Institute for Advanced Study, Seoul 130-722 and Department of Materials Science and Engineering, University of Seoul, Seoul 130-743

Young Moo Byun

School of Computational Sciences, Korea Institute for Advanced Study, Seoul 130-722 (Received 26 August 2008) By performing large-scale rst-principles density-functional theory and matrix Greens function calculations, we investigate the coherent charge transport properties of nite (8,0) and (10,0) singlewalled carbon nanotubes (CNTs) end-contacted to Au(111) electrodes. We nd that the device conductance varies signicantly with the CNT diameter, and analyze its origin in terms of the charge transfer at the CNT-metal interfaces and the decay of metal-induced gap states. We thus provide an atomistic picture that explains the diculty of constructing electronic devices based on small-diameter CNTs in a reliable and controllable way. Computational improvements we have made for an ecient large-scale matrix Greens function calculation will also be described.
PACS numbers: 73.63.-b, [Link], 71.15.-m Keywords: Carbon nanotubes, Nanoelectronic devices, Transport properties, Electronic structure

I. INTRODUCTION

Carbon nanotubes (CNTs) are promising building blocks for the next-generation nanoelectronics technology [1] due to their excellent mechanical and electrical properties [2,3]. In particular, CNT eld-eect transistors (CNTFETs) exhibiting an unprecedented high mobility, large current density, and a delicate response to the gate voltage have been successfully demonstrated [1, 3]. In the development of CNTFET applications, one critical yet still unresolved problem is our limited understanding of (and thus the lack of a method to control) the nature of the contacts between the semiconducting CNTs and the metal electrodes. One key advance here was the realization of the important role of the Schottky barriers that form at the CNT-metal contacts [4,5]. However, some research groups have more recently achieved Ohmic contacts with large-diameter (1.5 nm) CNTs [69]. It is now clear that the Schottky barriers resulting from charge redistribution throughout the metal-induced gap states (MIGS) critically depend on the CNT diameter, the electrode metal species, and the nature of CNT-metal contacts [8,9]. However, a microscopic and unifying picture that fully explains and predicts the large variations in the small-diameter-CNT charge transport properties
Corresponding

is still lacking [8,9]. In this work, we employ a rst-principles computational approach that combines density-functional theory (DFT) and matrix Greens function (MGF) calculations to investigate the coherent charge transport characteristics of nite 8-unit-cell (8,0) and (10,0) single-walled CNTs sandwiched between Au(111) electrodes in an endcontact geometry. Similar rst-principles studies have recently been reported, but they were restricted only to (8,0) CNTs, and the CNT diameter dependence was not considered [1012]. We will show that the device conductance varies signicantly according to the CNT diameter, in agreement with experiments. The microscopic origin of this behavior will also be analyzed by studying the decay of MIGS.

II. METHODS Based on the non-equilibrium Greens function theory [13], we used the matrix Greens function approach to calculate the device conductance. Depending on the nature of the device models (see section III.1), we performed three-dimensional (3D) [Fig. 1(a)] or two-dimensional (2D) (Fig. 2) DFT calculations within the Perdew-Burke-Enzhrof parameterization of the generalized-gradient approximation (GGA) [14]. We employed a modied version of the SeqQuest program
-299-

Author: [Link]@[Link]; Tel: +82-2-2210-5724; Fax: +82-2-2215-5863

-300-

Journal of the Korean Physical Society, Vol. 55, No. 1, July 2009

Fig. 1. (a) Device modeling of the innite perfect (10,0) CNT. We include 10 unit cells in the 3D main device model calculation and take two unit cells at both ends as contacts or the cental six unit cells as the core part. (b) Convergence of the surface Greens function at E = EF + 0.5 eV within the iteration method based on the simple linear mixing and the iteration scheme by Lpez-Sancho et al. [24] (c) The band structure of a (10,0) CNT calculated at the SZ level basis set, and (d) the corresponding DOS and (e) transmission.

Fig. 2. Device model based on (10,0) CNT a four-cell sandwiched between three-layer (3 3 3 3)R30 Au(111) slabs with d = 2.0 A.

[15] which is based on norm-conserving pseudopotentials [16] and Gaussian basis sets. For the transmis-

sion calculations, we partitioned the Hamiltonian (H ) and the overlap (S ) matrices into the blocks belonging to the core region and the electrodes 1 and 2 and obtained the retarded Greens function in the scattering core region for each energy E around the Fermi 1 level EF : GC (E ) = (ESC HC + 1 + 2 ) , where + 1(2) = x1(2) gS 1(2) x1(2) is the self-energy describing the coupling between the core and the lead 1 (2), and where x1(2) = ESC 1(2) HC 1(2) . The surface Greens function gS 1(2) was extracted from the 3D bulk calculations for each electrode. The retarded Greens function was then used to compute the transmission: T (E ) = + T race 1 GC 2 G+ C , where 1(2) = i 1(2) 1(2) . We have previously described the computational advances we have made for the accurate and ecient largescale matrix Greens function calculations [17, 18]. In addition, we also parallelized the main block of our code where the energy-resolved transmissions, T (E ), are calculated with respect to the energy E within the Message Passing Interface library [19]. This led to an almost perfect linear scaling of the computational time with the number of processors and enabled us to easily treat large CNT-based device models [20, 21]. Finally, using a Landauer-B utticker-type formula, we calculated the two-terminal conductance in the device. The details of our computational approach are given elsewhere [17,

Diameter Dependence of Charge Transport across Carbon Yong-Hoon Kim and Young Moo Byun Table 1. Band gap values of an innite (10,0) CNT obtained using dierent basis set levels. C basis set Band gap [eV] single- 0.881 single- plus polarization 0.869 single(2s)/double(2p)- 0.815 double- 0.811

-301-

double- glus polarization 0.808

18,22,23] and will be further discussed in Sec. III.1.

III. RESULTS AND DISCUSSION

1. Models and Calculation Levels

In this work, we adopted simplied CNT-metal electrode interface geometries: namely, the clean-cut (10,0) and (8,0) CNTs with the diameter 7.85 A and 6.30 A, respectively, end-contacted to the perfectly at Au(111) electrodes with contact distance d = 2.0 A. Both CNTs were 8-cell lengths long (l 32.75 A), and each electrode was modeled by using a two-layer (3 33 3)R30 Au(111) slab. Thus, the numbers of atoms in the (10,0) and the (8,0) models were 428 and 364, respectively. To achieve high-accuracy DFT and transmission calculations, we carefully checked the numerical convergence with respect to the basis set size for two reference systems: (i) the innite (10,0) CNT [Fig. 1(a)] and (ii) small device models in which a 4-unit cell (10,0) is sand CNT wiched between three- and two-layer (3 3 3 3)R30 Au(111) slabs (Fig. 2). The former was treated by using 3D DFT calculations, while the latter by 2D DFT calculations (See section II). For the 3D (10,0) CNT model (i), we nd that the acceleration of time-consuming gS 1(2) calculations by adopting the iterative algorithm by LpezSancho et al. [18,24] is especially signicant due to the 1D nature of CNTs [20, 21] [Fig. 1(b)]. The extrema in the band structure [Fig. 1(c)] correspond to the vanHove singularities in the density of states (DOS) [Fig. 1(d)] and to the steps in the quantum conductance [Fig. 1(e)]. From the band gap values of the innite (10,0) CNT obtained from ve dierent basis sets, single- gSZ), single- plus polarization (SZP), single(2s)/double(2p) (SDZ), double- gDZ), double- glus polarization (DZP) are reported in Table 1. We conclude that the SDZ level basis set accurately reproduces the high-accuracy DZP level result. On the other hand, the SZ level basis set overestimates the band gap by 0.1 eV. For the 2D device models based on nite (10,0) CNTs (ii), by adopting the three-layer Au(111) slabs as electrodes and by using the DZP basis set for C and a single(5d)/double(6s) plus polarization (SDZP) level basis set for Au, we rst determined the optimal contact distance as d = 2.0 A. We also considered the (8,0) CNTbased device model to obtain an identical result. Next, as in the case of the innite CNT model (i), we checked the numerical convergence of various basis set levels with respect to the DZP/SDZP (C/Au) data. The results are

Fig. 3. Transmission functions of the four-cell (10,0) CNT device models (Fig. 2) calculated using several d = 2.0 A dierent basis set levels. The order of computational accuracy is (1) a DZP level for C atoms and a single(5d)/double(6s) plus polarization level for the Au atoms (three-layer Au slabs, solid lines), (2) a DZP polarization level for the two CNT edge layers of C atoms, a SDZ level for the remaining C atoms, and a SZP level for the Au atoms (two-layer Au slabs, dashed lines), (3) a DZP level for the two CNT edge layers of C atoms, a SZgevel for the remaining C atoms, and a SZP level for the Au atoms (two-layer Au slabs, long-dashed lines), and (4) a SZgevel for the C atoms and a SZP level for the Au atoms (two-layer Au slabs, dot-dashed lines). Although the full SZg basis set (4) is inadequate, adopting a DZP level basis set for the CNT edge layers (3) is satisfactory.

shown in Fig. 3. We conclude that it is important to accurately treat the C atoms in the CNT edges in contact with the metal slabs because the device conductance is dominated by the CNT-metal contact resistance, as will be discussed in Sec. III.2. We nd that the SZ basis can still provide a satisfactory description of the device transmission properties as long as we adopt the DZP level basis set for the CNT edge layers in contact with metals. In this work, because the CNTs under consideration are rather short (8 unit cells long), we adopted a DZP level basis set for the two edge layers of C atoms, an SDZ level basis set for the remaining bulk-part C atoms, and the SZP polarization level basis set for the Au atoms.

2. Charge Transport Characteristics

The charge transport properties of the 8-cell (10,0) and (8,0) CNTs end-contacted to Au(111) electrodes with d = 2.0 A are presented in Fig. 4. Comparing these with

-302-

Journal of the Korean Physical Society, Vol. 55, No. 1, July 2009

Fig. 4. Transmissions (solid lines) of the nite (a) (10,0) and (b) (8,0) CNTs. Transmissions of the innite (10,0) and (8,0) CNTs are shown together for comparison (dotted lines). The valence- and the conduction-band parts of the innite CNT transmissions have been separately shifted to approximately t the band edges of nite CNTs.

Fig. 5. DOS projected on the unit cells of (a) (10,0) and (b) (8,0) CNTs. In general, upon moving away from the CNT edge cell (cell 1) into the central cell (cell 4), we observe a rapid decay of MIGS. However, because of the dierent contact strength in the two devices, the behavior of the PDOS peak decay is dierent in the two cases.

the data obtained for innite (10,0) and (8,0) CNTs, we nd that the transmission gaps are changed from their bulk values because of the band gap oscillations in short CNTs [25,26], and the transmission values are, in general, noticeably suppressed. The step-like transmission characteristics of the innite CNTs are destroyed and only spiky transmission peaks are observed. For the (10,0) CNT [Fig. 4(a)], we observe a bipolar-type transmission. Compared with the (10,0) CNT counterpart, the (8,0) CNT junction shows transmission peaks that are shifted to the lower energy range by about 0.2 eV to exhibit an n-type transmission [Fig. 4(b)]. In addition, the (8,0) CNT device shows an extra transmission contribution within the band gap on the n-type energy side. To understand the dierences between the two device models, we rst refer to the recent work by Shan and Cho [27]. They showed that, while single-walled CNTs with diameters larger than 1 nm have rather uniform work functions that are close to that of a graphene sheet (4.66 eV), smaller-diameter (<1 nm) single-walled CNTs have work functions that deviate from this value and signicantly depend on the CNT chirality [27]. More specically, the work functions of

(n,0) zigzag CNTs increase and those of (n,n) armchair CNTs decrease with decreasing CNT diameters. Therefore, the work function of the (8,0) CNT is 0.1 0.2 eV larger than that of the (10,0) CNT, which explains the energetically-downshifted transmission spectrum of the (8,0) CNT device compared with that of the (10,0) CNT counterpart. To further analyze the dierences between the two device models and especially the nature of the extra nregion transmission peaks in the (8,0) junction, we now consider the nature of the interface charge transfer and the MIGS by using the projected DOS (PDOS). The DOS projected onto the four CNT unit cells from the edge cells (cell 1) to the central cells (cell 4) in the two device models are shown in Fig. 5. Oscillations in the DOS with length scales that are smaller than one CNT unit cell have been averaged out. In both device models, we observe a signicant MIGS PDOS with a strong PDOS peak at E = EF + 0.2 eV for the edge cells. This implies a strong charge transfer and dipole formation at the CNT-Au interfaces. In general, the MIGS rapidly decay away from the boundary cells (cell 1) to the central cells (cell 4) for both devices. However, the

Diameter Dependence of Charge Transport across Carbon Yong-Hoon Kim and Young Moo Byun

-303-

nature of the E = EF + 0.2 eV PDOS peaks is rather dierent in the two cases. In the (10,0) CNT device, the EF + 0.2 eV PDOS peak at the edge cells also rapidly decays and does not contribute to the nal transmission [Fig. 5(a)]. By contrast, the decay of the EF + 0.2 eV PDOS peak is much slower in the case of the (8,0) CNT junction, and the MIGS resulting from the interface dipoles are noticeable even at the fourth cell [Fig. 5(b)]. This is precisely the origin of the small, but nite, n-type transmission within the band gap for the (8,0) CNT device [Fig. 4(b)]. While this transmission contribution would eventually disappear for longer (8,0) CNT junctions, we can conclude that the nature of the CNT-metal contacts sensitively depends on the CNT diameter and especially that the interface charge transfer is stronger for smaller-diameter CNTs.

IV. SUMMARY The limited understanding of the nature of smalldiameter semiconducting carbon nanotube (CNT)-metal electrode contacts is a major obstacle for the realization of high-performance CNT-based nanoelectronics applications. To resolve this problem, here we examined the charge-transport properties of nite 3.3 nm-long (10,0) and (8.0) single-walled CNTs end-contacted to Au(111) electrodes with d = 2.0 A. We found that the charge transport properties of the two device models were signicantly dierent because the CNT work function, the level alignment and charge transfer at the CNT-metal contacts, and the nature of the MIGS decay from the CNT boundary into the CNT core critically depend on the CNT diameter. We thus provided an atomistic picture that explicitly demonstrates the diculty of constructing reliable devices based on small-diameter CNTs.

ACKNOWLEDGMENTS This work was supported by the University of Seoul 2007 Academic Research Fund.

REFERENCES
[1] P. Avouris, Z. Chen and V. Perebeinos, Nature Nanotechnology 2, 605 (2007).

[2] J.-C. Charlier, X. Blase and S. Roche, Rev. Mod. Phys. 79, 677 (2007). [3] M. P. Anantram and F. Leonard, Rep. Prog. Phys. 69, 507 (2006). [4] S. Heinze, J. Terso, R. Martel, V. Derycke, J. Appenzeller and P. Avouris, Phys. Rev. Lett. 89, 106801 (2002). [5] J. Appenzeller, J. Knoch, V. Derycke, R. Martel, S. Wind and P. Avouris, Phys. Rev. Lett. 89, 126801 (2002). [6] A. Javey, J. Guo, Q. Wang, M. Lundstrom and H. Dai, Nature 424, 654 (2003). [7] Y. Yaish, J.-Y. Park, S. Rosenblatt, V. Sazonova, M. Brink and P. L. McEuen, Phys. Rev. Lett. 92, 464011 (2004). [8] W. Kim, A. Javey, R. Tu, J. Cao, Q. Wang and H. Dai, Appl. Phys. Lett. 87, 173101 (2005). [9] Z. Chen, J. Appenzeller, J. Knoch, Y.-M. Lin and P. Avouris, Nano Lett. 5, 1497 (2005). [10] P. Pomorski, C. Roland and H. Guo, Phys. Rev. B 70, 115408 (2004). [11] K. Odbadrakh, P. Pomorski and C. Roland, Phys. Rev. B 73, 233402 (2006). [12] J. J. Palacios, P. Tarakeshwar and D. M. Kim, Phys. Rev. B 77, 113403 (2008). [13] S. Datta, Quantum Transport: Atom to Transistor (Cambridge University Press, Cambridge, 2005). [14] J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). [15] P. Schultz, SeqQuest Code Project (Sandia National Laboratories, [Link] [16] M. Fuchs and M. Scheer, Comput. Phys. Commun. 119, 67 (1999). [17] Y.-H. Kim, J. Korean Phys. Soc. 52, 1181 (2008). [18] Y.-H. Kim, J. Nanosci. Nanotechnol. 8, 4593 (2008). [19] W. Gropp, E. Lusk and A. Skjellum, Using MPI: Portable Parallel Programming with the Message Passing Interface, 2nd ed. (The MIT Press, Cambridge, 1999). [20] Y.-J. Kang, Y.-H. Kim and K. J. Chang, Phys. Rev. B 76, 205441 (2007). [21] Y.-J. Kang, Y.-H. Kim and K. J. Chang, Curr. Appl. Phys. 9, S7 (2009). [22] Y.-H. Kim, S. S. Jang, Y. H. Jang and W. A. Goddard III, Phys. Rev. Lett. 94, 156801 (2005). [23] Y.-H. Kim, J. Tahir-Kheli, P. A. Schultz and W. A. Goddard III, Phys. Rev. B 73, 235419 (2006). [24] M. P. Lpez Sancho, J. M. Lpez Sancho and J. Rubio, J. Phys. F: Met. Phys. 14, 1205 (1984). [25] A. Rochefort, D. R. Salahub and P. Avouris, J. Phys. Chem. B 103, 641 (1999). [26] B. Gao, J. Jiang, Z. Wu and Y. Luo, J. Chem. Phys. 128, 084707 (2008). [27] B. Shan and K. Cho, Phys. Rev. Lett. 94, 236602 (2005).