Electronic structures and stability of Ni/Bi2Te3 and Co/Bi2Te3 interfaces

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2010 J. Phys. D: Appl. Phys. 43 115303

(http://iopscience.iop.org/0022-3727/43/11/115303)

Download details:

IP Address: 129.110.242.6

The article was downloaded on 05/03/2010 at 04:48

Please note that terms and conditions apply.

The Table of Contents and more related content is available

Home Search Collections Journals About Contact us My IOPscience

IOP PUBLISHING JOURNAL OF PHYSICS D: APPLIED PHYSICS

J. Phys. D: Appl. Phys. 43 (2010) 115303 (8pp) doi:10.1088/0022-3727/43/11/115303

Electronic structures and stability ofNi/Bi2Te3 and Co/Bi2Te3 interfacesKa Xiong1, Weichao Wang1, Husam N Alshareef1,2, Rahul P Gupta1,John B White3, Bruce E Gnade1 and Kyeongjae Cho1,4

1 Materials Science and Engineering Department, The University of Texas at Dallas, Richardson,TX 75080, USA2 Materials Science and Engineering, King Abdullha University of Science and Technology,Thuwal 23955-6900, Saudi Arabia3 Marlow Industries, 10451 Vista Park Road, Dallas, TX 75238, USA4 Physics Department, The University of Texas at Dallas, Richardson, TX 75080, USA

E-mail: ka.xiong@utdallas.edu and kjcho@utdallas.edu

Received 18 November 2009, in final form 4 February 2010Published 4 March 2010Online at stacks.iop.org/JPhysD/43/115303

AbstractWe investigate the electronic structures and stability for Ni/Bi2Te3, NiTe/Bi2Te3, Co/Bi2Te3

and CoTe2/Bi2Te3 interfaces by first-principles calculations. It is found that the surfacetermination strongly affects the band alignment. Ni and Co are found to form Ohmic contactsto Bi2Te3. The interface formation energy for Co/Bi2Te3 interfaces is much lower than that ofNi/Bi2Te3 interfaces. Furthermore, we found that NiTe on Bi2Te3 is more stable than Ni, whilethe formation energies for Co and CoTe2 on Bi2Te3 are comparable.

(Some figures in this article are in colour only in the electronic version)

1. Introduction

The solid-state thermoelectric (TE) cooler is of considerableinterest in many applications because of its advantages ofreliability, low-noise operation, miniaturization and highpower density [1–4]. Commercial TE cooling devices usedoped Bi2Te3 ((Bi,Sb)2Te3 for n-type and Bi2(Te,Se)3 forp-type) as the semiconductor material due to its high figureof merit (ZT ∼ 1) [5]. For TE coolers to meet the demandsof the industry in terms of high cooling power density, thesize of the device needs to be scaled down. Consequently,the electrical resistance between the contact metal and the TEmaterial plays an important role in the device performance [6]since the material ‘figure of merit’ (Z) is degraded by thecontact resistance and the relationship between Z and ZD

(device figure of merit) can be shown in the following equation:

ZD =(

L

L + 2rcσ

)Z, (1)

where L is the device length, rc is the contact resistance andσ is the bulk conductivity. For current bulk TE devices,electroless Ni is used as the contact metal, providing a contactresistance of ∼5 × 10−6 � cm2 [7]. The contact resistance

needs to be at least 10–100 times lower in order to maintainthe device scaling [8]. Thus, it is necessary to lower rc

by engineering the interface between the contact metal andTE materials. The contact resistance of an Ohmic contactdepends not only on the contact metal but also on the interfacialreactions. Furthermore, because Bi2Te3 has a small bandgap of ∼0.16 eV [9], theoretically speaking, it is possible toachieve a much lower contact resistance than is available withelectroless Ni.

Recently, we experimentally investigated the Ni/Bi2Te3

and Co/Bi2Te3 interfaces with the goal of decreasing thecontact resistance. Ni and Co films were sputtered ontopolycrystalline bulk Bi2Te3 (Se doped) with thicknesses of120 nm and 90 nm, respectively. After post-annealing thesamples at 200 ◦C in vacuum, we observed a 460 nm NiTeinterfacial region for the Ni/Bi2Te3 interface [10], while for theCo/Bi2Te3 we observed a very thin interfacial region (∼20 nm)and a small amount of CoTe2. These results are evident inthe XRD spectra (figure 1) and in the cross-sectional TEMimages (not shown). The formation of a thick NiTe interfacialregion is consistent with previous experimental work [11].More importantly, the experimentally determined I–V curvesof Ni on Bi2Te3 show a linear characteristic indicative of anOhmic contact. This has motivated us to assess the electronic

0022-3727/10/115303+08$30.00 1 © 2010 IOP Publishing Ltd Printed in the UK

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 1. XRD spectra comparing 90 nm sputtered Co film and120 nm sputtered Ni film on Bi2Te3 after annealing (in N2 ambient)at 200 ◦C for 60 min.

structures and stability of Ni/Bi2Te3, Co/Bi2Te3 and relatedinterfaces at an atomic level by first-principles calculations.In this study, we focus on investigating ideal abrupt interfacesthat are free of defects and interfacial layers, as we want todetermine what the intrinsic properties of these interfaces couldbe under ideal conditions and compare these with experimentalobservation.

2. The computational method and model

We performed the calculations using density functional theorywith the plane-wave basis code VASP [12]. The exchangecorrelation energy is approximated by the generalized gradientapproximation (GGA). The pseudopotential is described bythe projected augmented wave (PAW) method. A 6 × 6 × 1k-point Monkhorst–Pack grid was sufficient for calculatingtotal energies and a 10 × 10 × 1 grid for the density of states(DOS). We include the spin-polarized and spin–orbit coupling(SOC) effects in our calculations, as they strongly affect theelectronic structure of Ni (or Co) and Bi2Te3.

The interfaces are modelled by a superlattice containingone interface and 10 Å vacuum on top. To choosethe appropriate metal surface, we can either use theexperimental observation by constructing a large lattice-matched superlattice or find a suitable orientation that couldgive a reasonable cell size. The latter method has been adoptedas a compromise between the computational cost and accuracy.We consider Ni(1 1 1), Co(1 1 1), NiTe(0 0 1) and CoTe2(0 1 0)on top of Bi2Te3(0 0 0 1). The metal is compressed or expandedto match the Bi2Te3(0 0 0 1) surface and the correspondinglattice mismatches between these metals and Bi2Te3 are of1.5%, 1.5%, 10% and 4%, respectively. We found that thework functions of these metals are insensitive to the appliedstrain. The supercells contain 108–154 atoms, which aresufficient to converge the calculated band offsets. The Ni(1 1 1)(or Co(1 1 1)) slab is ∼10 Å thick, while the thicknesses ofthe NiTe(0 0 1) and CoTe2(0 1 0) slabs are ∼11 Å and ∼9 Å,respectively. Since Bi2Te3 has a layered structure containing

different blocks and each block has one Bi2Te3 unit, theBi2Te3 slab contains four blocks. In our model there is noneed to use hydrogen to passivate the Bi2Te3 surface, aseach block is charge neutralized so that it is a closed-shellsystem. For Ni/Bi2Te3 and Co/Bi2Te3 we consider two typesof interfaces: the Bi2Te3 terminated with either Bi or Te atoms.For NiTe/Bi2Te3 and CoTe2/Bi2Te3 we consider four types ofinterfaces: NiTe or CoTe2 is terminated with either Ni (Co) orTe atoms while Bi2Te3 is terminated with either Bi or Te atoms(denoted as Ni–Bi (or Co–Bi), Ni–Te (or Co–Te)), Te–Bi andTe–Te). In each case the atomic positions are relaxed and thein-plane lattice constants and all angles are kept fixed.

3. Results and discussion

The calculated lattice parameters for bulk Bi2Te3 are a =10.6 Å and θ = 24.13◦, in good agreement with theexperimental values [13]. The calculated band structure ofBi2Te3 is shown in figure 2. Without including the SOC, weobtained a band gap of 0.33 eV (figure 2(a)). Both conductionband minimum (CBM) and valence band maximum (VBM)are found to lie at �. The SOC effect scales down theband gap to 0.13 eV (figure 2(b)), less than the experimentalgap of 0.16 eV due to the well-known LDA error, in goodagreement with other theoretical results [9, 14–18]. The CBMand VBM now locate at a point between Z and �. Thecalculated partial DOS (figures 2(c) and (d)) show that theconduction band (CB) consists mostly of Bi 6p states whilethe valence band (VB) consists mostly of Te 5p states butthere is hybridization between these states, which indicatesthe covalent Bi–Te interaction. For the metals we used face-centred cubic (fcc) Ni and Co in our calculations. NiTe has ahexagonal structure (P 63/mmc) and CoTe2 has a cubic pyritestructure (Pmnn). Our DOS calculations reveal that theyboth are metallic, consistent with other work [19, 20]. Thecalculated lattice constants for these metals are also consistentwith the experimental data.

To determine the relative stability of these metal/Bi2Te3

interfaces, we calculated their interface formation energies.This is a function of chemical potentials and can be writtenas [21]

Emetal/Bi2Te3

form = Emetal/Bi2Te3

total − nEmetal − m

2EBi2Te3 − lµTe,

(2)

where Emetal/Bi2Te3

total is the total energy of the given supercell, n

and m are the number of atoms of metal and Bi, respectively,EBi2Te3 and EMetal are the total energies per formula unit inBi2Te3 and the metal, respectively, l is the number of excess (ordeficient) Te atoms. There is a single independent parameterdetermined by the growth conditions, the Te chemical potentialµTe. Its highest (the least negative) value is µbulk

Te , the energyof bulk Te chemical potential. Its lowest (the most negative)value is for thermodynamic equilibrium with Bi2Te3. Takingthe formation enthalpy of bulk Bi2Te3 as �H ∼ −1.3 eV [23],the limiting values of µTe are µTe = µbulk

Te + (�H/3) (Bi-rich,favouring Te substitutions) andµTe = µbulk

Te (Te-rich, favouringBi substitutions).

2

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 2. Calculated band structure and DOS of bulk Bi2Te3. (a) Band structure without considering SOC, (b) band structure with SOC,(c) partial DOS without SOC and (d) partial DOS with SOC.

Figure 3. Interface formation energies of various interfaces versusTe chemical potential relative to the Te-terminated Ni/Bi2Te3

interface. Solid lines correspond to the formation energies ofNi(Co)/Bi2Te3 interfaces. Dashed lines correspond to the formationenergies of NiTe/Bi2Te3 interfaces. Dotted lines correspond to theformation energies of CoTe2/Bi2Te3 interfaces.

Figure 3 shows the interface formation energy for thevarious interfaces as a function of the tellurium chemicalpotential µTe. For Ni (or Co) on Bi2Te3, the formation energyof the Bi-terminated interface increases with increasing µTe,while the energy of the Te-terminated interface is independentof µTe. The Bi-terminated interface is always more stablethan the Te-terminated interface. More importantly, theformation energies for Co/Bi2Te3 interfaces are much lowerthan Ni/Bi2Te3 interfaces, by ∼7 eV per formula unit. This

result may explain why Ni interacts more readily with Bi2Te3

to form a thick NiTe interfacial region while Co does not.On the other hand, the most stable NiTe/Bi2Te3 interface, theNi–Bi interface, has lower formation energy than the Ni/Bi2Te3

interfaces for all µTe. This finding is consistent with theexperimental fact that NiTe is more stable on Bi2Te3 thanNi. In contrast, the most stable CoTe2/Bi2Te3 interface, theCo–Bi interface, has lower formation energy than Co/Bi2Te3

interfaces only at µTe = −0.32 eV or below. Above µTe =−0.32 eV, the Bi-terminated Co/Bi2Te3 interfaced is the mostenergetically favourable.

Figures 4(a) and (b) show the relaxed structures forBi-terminated and Te-terminated Ni/Bi2Te3 interfaces. Theinterfaces are formed by either Ni–Bi or Ni–Te bonds. Eachinterfacial Bi (or Te) atom is 6-fold coordinated and found toform three bonds with interfacial Ni atoms, as compared withthe coordination of their bulk atoms. Co/Bi2Te3 interfaceshave similar interfacial bonding, as Co has similar latticeconstants as compared with Ni. Figures 4(c) and (d) are therelaxed structures for the most stable NiTe/Bi2Te3 (Ni–Bi) andCoTe2/Bi2Te3 (Co–Bi) interfaces. For the Ni–Bi interface,the interface is formed by Ni–Bi bonds and the interfacial Biis 4-fold coordinated. The Co–Bi interface is terminated byCo–Bi and Te–Bi bonds, due to the more complicated interfacegeometry.

Figure 5 shows the spin-resolved and projected DOS ofvarious interfaces without including the SOC effect. Theenergy of the Fermi level is at zero. It shows that the DOSon atoms well away from the interface (see ‘Bi bulk’ and ‘Tebulk’) replicate those of bulk atoms, as there are no metalstates in the Bi2Te3 band gap (Eg ∼ 0.5 eV), while the DOS of

3

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 4. The interface region of the relaxed atomic structures for (a) Bi-terminated Ni/Bi2Te3, (b) Te-terminated Ni/Bi2Te3, (c) Ni–BiNiTe/Bi2Te3 and (d) Co–Bi CoTe2/Bi2Te3 interfaces. Bi : large spheres; Te : medium spheres; Ni : small gray spheres; Co : small blackspheres.

interfacial atoms (see ‘Bi interface’ and ‘Te interface’) showthe expected tailing of metal states into the Bi2Te3 band gap.Moreover, it reveals that Ni and Co prefer to be in a magneticstate, while Bi and Te away from the interface do not have anymagnetization effects. The position of the Fermi level dependson the interfacial bonding configuration. For Bi-terminatedinterfaces the Fermi level lies close to the VB edge, while forTe-terminated interfaces the Fermi level lies ∼0.3 eV abovethe VB edge, close to the CB.

Figure 6 shows the projected DOS of these interfaceswhen including the SOC effect. The SOC effect scales theband gap down to ∼0.1 eV. Since the SOC mainly lowers theenergies of the CB of Bi2Te3 (Bi p bands), for the Bi-terminatedinterfaces the Fermi level still locates at the VB edge, but forthe Te-terminated interfaces it now lies at the CB edge.

The p-type Schottky barrier heights, �p, of theseinterfaces are evaluated by the so-called ‘bulk plus lineup’method [23], which is expressed as

�p = �Ebulk + �Vinterface, (3)

where �Ebulk is the energy difference from the Fermi energy

of the metal and the Bi2Te3 VB top, which are obtained frombulk calculations and the energies are referenced with respectto their bulk electrostatic potentials. �Vinterface is the shift in theelectrostatic potential across the interface from metal (denotedas VNi) to Bi2Te3 (denoted as VBi2Te3), which is obtained fromthe supercell calculation by assuming that the potential farenough from the interface is analogous to the bulk potential.

Figure 7 shows the calculated planar-averaged potentialsalong the z-direction of Bi- and Te-terminated Ni/Bi2Te3

interfaces. To evaluate �Vinterface two regions away from theinterface were selected for extracting the reference potentialsof Ni (from 0 to 4 Å) and Bi2Te3 (from 27.26 to 34.70 Å andfrom 29.10 to 39.61 Å for Bi- and Te-terminated interfaces,respectively) [24]. In figure 7 the magnitudes of (EF − VNi)

and (EBi2Te3V − VBi2Te3) are almost the same as those in bulk

calculations, which are ∼4.7 eV and 6.45 eV, respectively,indicating that the supercell is large enough.

The calculated �p for Bi-terminated Ni/Bi2Te3 andCo/Bi2Te3 interfaces are −0.1 eV and −0.17 eV, respectively.This means that for Bi-terminated Ni/Bi2Te3 and Co/Bi2Te3

interfaces, the Fermi level lies below the Bi2Te3 VB edge so

4

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 5. The spin-resolved and projected partial DOS of (a) Bi-terminated and (b) Te-terminated Ni/Bi2Te3 interfaces, (c) Bi-terminatedand (d) Te-terminated Co/Bi2Te3 interfaces, (e) Ni–Bi NiTe/Bi2Te3 and (f ) Co-Bi CoTe2/Bi2Te3 interfaces. The DOS are calculatedwithout including the SOC effect.

that this interface forms a p-type Ohmic contact. In contrast,for Te-terminated Ni/Bi2Te3 and Co/Bi2Te3 interfaces, thecalculated �p are 0.2 and 0.28 eV. This means that in both casesthe Fermi lies above the Bi2Te3 CB edge and gives an n-typeOhmic contact if we use the experimental band gap of 0.16 eV.The calculated band offsets of these interfaces are consistentwith the experimental observation. Thus, the band offsetstrongly depends on the interfacial bonding. Similarly, we

calculated �p for NiTe/Bi2Te3 and CoTe2/Bi2Te3 interfacesand found that they all give an Ohmic contact, as summarized intable 1.

The formation of different types of Ohmic contactwith respect to the termination of the interfaces can beexplained in terms of local bonding. The formation ofthe metal/semiconductor interface causes charge transferfrom the metal to the semiconductor, due to their different

5

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 6. Projected DOS of (a) Bi-terminated and (b) Te-terminated Ni/Bi2Te3 interfaces, (c) Bi-terminated and (d) Te-terminatedCo/Bi2Te3 interfaces, (e) Ni–Bi NiTe/Bi2Te3 and (f ) Co–Bi CoTe2/Bi2Te3 interfaces. The DOS are calculated with the SOC effect.

electronegativities (ENs). Despite the small differences ofEN values (Pauling criterion) of the elements in our system(Ni: 1.9, Co: 1.88, Bi: 2.02, Te: 2.1), the formation of dipoles atdifferent interface configurations makes the Fermi level easilysweep across the small band gap of Bi2Te3. The larger �p

of Te-terminated Ni(Co)/Bi2Te3 interfaces as compared with

that of Bi-terminated interfaces is similar to what was foundpreviously for metal/III–V (or II–VI) semiconductor interfaces[25]. It has been reported that the interface terminated by theanion of semiconductor has larger �p than that terminated bythe cation, due to the dipole formation between the charge atthe semiconductor surface and its image charge at the metal

6

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

Figure 7. Average potentials of the (a) Bi-terminated and(b) Te-terminated Ni/Bi2Te3 interfaces. The average potentials ofNi and Bi2Te3 along the z-direction (horizontal lines) are taken inregions away from the interface.

Table 1. Calculated types of Ohmic contacts for variousmetal/Bi2Te3 interfaces.

Interface Ohmic contact

Ni Bi-terminated p-typeTe-terminated n-type

Co Bi-terminated p-ypteTe-terminated n-type

NiTe Ni–Bi p-typeNi–Te n-typeTe–Bi n-typeTe–Te n-type

CoTe2 Co–Bi p-typeCo–Te p-typeTe–Bi n-typeTe–Te p-type

side. As a result, this lowers or raises the average potentialin the semiconductor with respect to its value in the metal,depending on the sign of the dipole (e.g. see �Vinterface infigures 7(a) and (b)).

Our calculated results show that the ideal Ni/Bi2Te3

and Co/Bi2Te3 interfaces intrinsically form Ohmic contacts.

Moreover, it is interesting that NiTe and CoTe2 also formOhmic contacts on Bi2Te3. This indicates that Co and Ni aresuitable contact materials. Co is a better choice because ofits low diffusivity. However, it should be noted that althoughour results are in agreement with the experimental data, theeffects of interface defects (e.g. dopant, vacancy, etc) andinterfacial layers (e.g. oxidation) should not be neglected. Inparticular, for addressing the contact resistance issues, theseextrinsic effects would change the interface chemistry andhence modify the entire electronic structure of the interface.We have studied the impact of oxygen at these interfaces, whichwill be published elsewhere, and found that oxygen indeedaffects the electronic structure of the interfaces. Therefore, toachieve low contact resistance, we should consider the overalleffect by including all these aspects.

4. Conclusion

In summary, first-principles calculations of band offsetsand stability of Ni/Bi2Te3, NiTe/Bi2Te3, Co/Bi2Te3 andCoTe2/Bi2Te3 interfaces are presented. The formationenergies of Co/Bi2Te3 interfaces are much lower thanNi/Bi2Te3 interfaces, making the Ni/Bi2Te3 interface moreprone to reaction as temperature is increased. Moreover,NiTe is found to be more stable on Bi2Te3 than Ni, while theformation energy for CoTe2 on Bi2Te3 is comparable to that ofCo. Our results are in agreement with the experimental data.All the interfaces give Ohmic contacts, indicating Ni and Coare suitable contact materials for Bi2Te3.

Acknowledgments

This work is supported by the II–VI Foundation, a privatefoundation. The authors thank Dr Jeff Sharp for usefulsuggestions.

References

[1] Wood C 1988 Rep. Prog. Phys. 51 459[2] Dresselhaus M S, Chen G, Tang M Y, Yang R and Ren Z 2007

Adv. Mater. 19 1043[3] Snyder G J and Toberer E S 2008 Nature Mater. 7 105[4] Semenyuk V 2003 Proc 22nd Int. Conf. on Thermoelectrics

(Montpellier, France) 631pp[5] Nolas G S, Sharp J and Goldsmid H J 2001

Thermoelectrics—Basic Principles and New MaterialsDevelopments (Berlin: Springer)

[6] Fleurial J-P 1999 Proc 18th Int. Conf. on Thermoelectrics(Baltimore, MD) 294pp

[7] Gupta R P, White J B, Iyore O D, Chakrabarti U,Alshareef H N and Gnade B E 2009 Electrochem.Solid-State Lett. 12 H302–4

[8] Semenyuk V 2001 Proc 20th Int. Conf. on Thermoelectrics(Beijing, China) 391pp

[9] Goldsmid H J 1964 Thermoelectric Refrigeration (New York:Plenum)

[10] Iyore O D, Lee T H, Gupta R P, White J B, Alshareef H N,Kim M J and Gnade B E 2009 Surf. Interface Anal. 41 440

[11] Lan Y C, Wang D Z, Chen G and Ren Z F 2008 Appl. Phys.Lett. 92 101910

7

J. Phys. D: Appl. Phys. 43 (2010) 115303 K Xiong et al

[12] Kresse G and Furthmuller J 1996 Comput. Mater. Sci. 6 15Kresse G and Furthmuller J 1996 Phys. Rev. B 54 11169

[13] Francombe M H 1958 Br. J. Appl. Phys. 9 415[14] Huang B-L and Kaviany M 2008 Phys. Rev. B 77 125209[15] Scheidemantel T J, Ambrosch-Draxl C, Thonhauser T,

Badding J V and Sofo J O 2003 Phys. Rev. B68 125210

[16] Youn S J and Freeman A J 2001 Phys. Rev. B 63 085112[17] Larson P 2003 Phys. Rev. B 68 155121[18] Kim M, Freeman A J and Geller C B 2005 Phys. Rev. B

72 035205

[19] Yadava Y P and Singh R A 1985 J. Mater. Sci. Lett. 4 1421[20] Terieff P and Schicketanz H 1996 J. Alloys Compounds 232 26[21] Xiong K, Delugas P, Hooker J, Fiorentini V and Robertson J

2008 Appl. Phys. Lett. 92 113504[22] Gorbachuk N P and Sidorko V R 2004 Powder Metall. Met.

Ceram. 43 284[23] Van de Walle C G and Martin R M 1987 Phys. Rev. B 35 8154[24] Fonseca L R C and Knizhnik A A 2006 Phys. Rev. B

74 195304[25] Berthod C, Binggeli N and Baldereschi A 2003 Phys. Rev. B

68 085323

8