correlation between the hysteresis and the initial defect density of graphene

5
Correlation between the hysteresis and the initial defect density of graphene Chunhum Cho, Young Gon Lee, Ukjin Jung, Chang Goo Kang, Sungkwan Lim, Hyeon Jun Hwang, Hojun Choi, and Byoung Hun Lee Citation: Applied Physics Letters 103, 083110 (2013); doi: 10.1063/1.4818770 View online: http://dx.doi.org/10.1063/1.4818770 View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/103/8?ver=pdfcov Published by the AIP Publishing This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 93.180.53.211 On: Mon, 17 Feb 2014 07:26:18

Upload: byoung

Post on 21-Dec-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Correlation between the hysteresis and the initial defect density of grapheneChunhum Cho, Young Gon Lee, Ukjin Jung, Chang Goo Kang, Sungkwan Lim, Hyeon Jun Hwang, Hojun Choi,

and Byoung Hun Lee Citation: Applied Physics Letters 103, 083110 (2013); doi: 10.1063/1.4818770 View online: http://dx.doi.org/10.1063/1.4818770 View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/103/8?ver=pdfcov Published by the AIP Publishing

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

93.180.53.211 On: Mon, 17 Feb 2014 07:26:18

Correlation between the hysteresis and the initial defect density of graphene

Chunhum Cho,1 Young Gon Lee,2 Ukjin Jung,2 Chang Goo Kang,2 Sungkwan Lim,1

Hyeon Jun Hwang,2 Hojun Choi,2 and Byoung Hun Lee1,2,a)

1Department of Nanobio Materials and Electronics, Gwangju Institute of Science and Technology,Oryong-dong 1, Gwangju 500-712, South Korea2School of Materials Science and Engineering, Gwangju Institute of Science and Technology, Oryong-dong 1,Gwangju 500-712, South Korea

(Received 20 May 2013; accepted 2 August 2013; published online 20 August 2013)

The role of the initial defects of graphene characterized by Raman spectroscopy is correlated with

the physical mechanisms causing the hysteretic device characteristics of graphene field effect

transistors (FETs). Fast charging related to the tunneling-induced charge exchange is found to be

closely correlated with the initial defect density, while slow charging related to environmental

influences such as the water redox reaction showed a weak correlation. It can be concluded that the

intrinsic quality of graphene should be improved to minimize the hysteresis of graphene FETs even

in an air-tight environment. VC 2013 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4818770]

Device instabilities such as a wide distribution of Dirac

points and hysteretic current-voltage (I-V) characteristics

have limited the applications of graphene in electronics. The

physical origins of the instability of graphene field effect

transistors (GFETs) have been actively studied to find a way

to minimize device scattering.1–10 Many factors affecting the

stability of GFETs have been identified including metallic

residues originating from the growth process, substrate mate-

rials, capping dielectrics, process-induced defects, imperfect

cleaning processes, initial defects, and operation ambient

and temperature.1–17

Especially, the time dependent device instability of

GFETs resulted in the hysteretic device characteristics. Two

main mechanisms, charge trapping through tunneling and an

interfacial redox reaction, have been suggested as the origin

of the hysteresis in GFETs by analyzing the time domain

response of the drain current.6,9,10 The two-trap model pro-

posed by Lee et al. could explain the time dependence of

hysteresis using two time constants.10 The slow time con-

stant is related to the chemical redox reaction at the gra-

phene/SiO2 interface, which can be modulated by limiting

the oxygen supply to the test environment.9 The fast time

constant is attributed to charge trapping through quantum

mechanical tunneling because of the temperature independ-

ence and short time constant. Yet, it is not clear how the tun-

neling component could be controlled and what the physical

origin of the tunneling component could be.

The primary candidates of the tunneling sites are either

graphene lattice defects or contaminants adsorbed in the sur-

face of the graphene. Structural defects or charged impurities

introduce a screening effect and impurity scattering in gra-

phene devices.18–23 In addition, lattice defects are known to

induce mid-gap states resulting in a charge transfer from per-

fect lattices to defects.18,19 Much research has reported a

strong correlation between electrical properties and lattice

defects or charged impurities.18–23 However, no studies have

addressed the relation between defects and hysteretic device

characteristics.

In this work, GFETs with various channel defect den-

sities were fabricated using laser-grown graphene, and their

I-V hysteresis characteristics were correlated with the initial

defect density of the graphene measured using Raman spec-

troscopy. The physical mechanism of the hysteresis, espe-

cially the fast tunneling components, was found to closely

correlate with the initial defect density measured by Raman

spectroscopy.

In this experiment, forming a graphene sheet with a

wide range of defect densities was challenging. Such gra-

phene is necessary because it is best to fabricate GFETs in

the same batch to minimize the influence of fabrication vari-

ability. In this work, a pulsed KrF excimer laser was

employed to grow graphene with a non-uniform quality dis-

tribution. Other methods such as ion irradiation, plasma, and

impurity deposition were discarded because the influence of

such processes may further increase the variability.21–23

To form graphene with a wide range of defect density,

a Si-face 4H-SiC wafer with an on-axis orientation was

graphitized using a KrF laser (k¼ 248 nm) annealing pro-

cess.24 Since the surface of the SiC wafer was smoothed by

chemical mechanical polishing (CMP), an initial hydrogen

anneal to form a large area facet was not necessary. Before

the laser anneal, the sample underwent Piranha cleaning

followed by a dilute HF clean to remove organic and metal-

lic contaminants. The wafer was then loaded into a vacuum

chamber and exposed to multiple laser shots (600 shots,

repetition rate¼ 10 Hz, average fluence¼ 1.33 J/cm2, pulse

width¼ 25 ns) in high vacuum (<10�6 Torr) at room tem-

perature. The laser spot size was increased to 4 mm � 4 mm

using a laser beam homogenizer.

The graphene layer grown on the SiC substrate was

transferred to a SiO2 substrate as follows. A stack of Au thin

film/polymethyl methacrylate (PMMA)/thermal release tape

served as the transfer template. The graphene/Au/PMMA/

tape stack was detached from the SiC substrate, and the gra-

phene layer was transferred to the SiO2 substrate. Before the

transfer, the substrate was cleaned by sonication in methanol,

a)Author to whom correspondence should be addressed. Electronic mail:

[email protected]

0003-6951/2013/103(8)/083110/4/$30.00 VC 2013 AIP Publishing LLC103, 083110-1

APPLIED PHYSICS LETTERS 103, 083110 (2013)

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

93.180.53.211 On: Mon, 17 Feb 2014 07:26:18

acetone, and deionized (DI) water and treated with O2

plasma to remove organic contaminants. Finally, the tape

and Au were removed, leaving a graphene layer on the SiO2.

Using the transferred graphene, back-gated GFETs were fab-

ricated to examine their hysteretic characteristics. Since the

graphene region was large enough, contact photolithography

and O2 plasma were employed to pattern the graphene. Au

electrodes were used as the source/drain/gate. After pattern-

ing the graphene channel, Raman spectra of the graphene

channel were measured using a RENISHAW inVia Raman

microscope with an Arþ ion laser (k¼ 514.5 nm). Then, a

30 nm Al2O3 capping layer was deposited on the graphene

channel by atomic layer deposition (ALD) to minimize

p-type doping effects.9 In this process, trimethylaluminium

(TMA) precursor is used with a H2O/N2 gas at 130 �C.

Fig. 1(a) is a photograph of a transferred graphene layer

showing a few voids that were formed during the transfer pro-

cess. These voids are due to particles preventing the graphene

from bonding to the SiO2 substrate. The size of graphene is

� 2.5 mm � 2.5 mm, which is smaller than the focused laser

spot size (4 mm � 4 mm) because the graphene was not

formed at the edge of the exposed region because of rapid

thermal dissipation to the unexposed region. Fig. 1(b) is

AFM image of center region showing root-mean-square

(rms) roughness of �0.32 nm in 25 lm � 25 lm region.

Surface roughness is slightly higher than a clean graphene on

SiO2 (�0.16 nm)25 due to surface particles. Fig. 1(c) is the

AFM image of edge region and (d) is the line scan data show-

ing step height �1.1–1.2 nm. This is typical step height of

monolayer graphene which is higher than the layer-to layer

distance of bulk graphite (�0.34 nm) due to ambient species

or adsorbates between graphene and SiO2 substrate.26,27

The distribution of the quality of transferred graphene

was monitored by mapping the D/G intensity ratio as shown

in Fig. 2(a). Raman spectra of the transferred graphene were

mapped to check the defect density and thickness of gra-

phene.28 The D/G intensity ratio ranged from 0.25 to 1.1,

indicating a wide spectrum of defect density due to the dif-

ference in the local annealing condition originated from the

spatial non-uniformity laser beam. Since the defect density is

influenced by the non-uniformity of the laser beam, we

assumed the local quality variation is not significant within a

local device channel area (5.6 lm� 10 lm). A representative

Raman spectrum is shown in Fig. 2(b). G/2D intensity ratio

<0.65 and 2D peak well fitted by single Lorentzian curve

indicate the transferred graphene consists of a monolayer the

graphene.

A schematic of the GFET and its scanning electron

microscope (SEM) image are shown in Fig. 3(a). Hysteresis

was measured using a pulsed current-voltage (I-V) measure-

ment system. The pulse Id-Vg curves were primarily meas-

ured at the electron conduction branch (right side of the

Dirac point) while applying a pulsed bias to the back gate.

The rise and fall time of the pulse was 300 ls and the pulse

width was 1 ms. Note that same gate bias (Vg-VDirac¼ 10 V)

was applied to restrain the effect of the voltage sweep range

on the hysteresis. A detailed explanation on the set-up of the

pulsed I-V measurement system can be found elsewhere.6,10

Fig. 3(b) shows representative Id versus (Vg-VDirac)

curves of GFETs. The drain current at the same Vg-VDirac

decreased as D/G ratio increased from 0.45 to 0.67. The hys-

teresis is defined as the difference in the Dirac points,

DVDirac, of two I-V curves generated by a gate pulse with a

negative Vg base bias and a positive Vg peak bias as shown

in Fig. 3(b). Since the charge traps gathered during a short

pulse bias cycle are easily dissipated during an off-cycle, the

hysteresis does not disappear or decreases even after multi-

ple cycles of I-V measurement. The drain current decreases

at þVg-VDirac are due to the time-dependent Dirac point shift

during the pulse width. Since DVDirac is due to the charge

generation during the gate pulse cycle, the equivalent num-

ber of trap charges, nt, generating DVDirac can be calculated

as follows:

FIG. 1. (a) Optical image of graphene transferred to SiO2. (b) AFM image

of center region showing rms roughness �0.32 nm in 25 lm � 25 lm region,

(c) AFM image of edge region showing the pinholes in the graphene due to

the insufficient graphitization of SiC, and (d) line scan data at the edge of

graphene show 1.1–1.2 nm of step height, which is close to the thickness of

monolayer graphene on SiO2 substrate.26,27

FIG. 2. (a) Raman mapping image

of D/G intensity ratio within 2 mm

� 2 mm region. (b) A representative

Raman spectrum of mono layer region.

083110-2 Cho et al. Appl. Phys. Lett. 103, 083110 (2013)

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

93.180.53.211 On: Mon, 17 Feb 2014 07:26:18

nt ¼ COX � DVDirac=q ; (1)

where COX is the capacitance of the SiO2. q is the charge of

electron. The charge density can be correlated with the initial

defect density, ndo, measured by Raman spectroscopy. ndo is

calculated using the D/G ratio as shown as28

ndo ¼1:8 � 1022

k4L

ID

IG

� �; (2)

where kL is the excitation laser wavelength. To illustrate the

statistical correlation between nt and ndo as shown in Fig. 4,

more than 15 GFETs were tested with various defect den-

sities ranging from 8.78 � 1010 to 1.84 � 1011 cm�2, which

corresponds to 0.34 and 0.72 of the D/G ratio. nt appears to

be proportional to the ndo of the graphene channel, although

the data distribution within the batch is still wide. The trend

line shown in Fig. 4 indicates that the electrically measured

trap density, nt, is �2.5 times the initial defect density, ndo,

measured by Raman spectroscopy. Since the electrically

measured nt extracted from DVDirac detects the charges from

the trap sites, which exist outside the graphene or on the sur-

face of the graphene, as well as the charges from the initial

defect sites, a one-to-one correspondence with the initial

defect density is not expected, but the rough correlation

between nt and ndo implies that the initial defects play an im-

portant role in the charge trapping process.

To further investigate the role of initial defects in the

graphene, the time dependence of the drain current decrease

at the peak bias of the pulse was analyzed using the two-trap

model as follows:6,10

Id ¼ I0 � A � exp�t

sA

� �þ B � exp

�t

sB

� �� �; (3)

where I0 is the initial drain current at the beginning of the

pulse peak, A and B, and sA and sB represent the magnitude

and speed of two different mechanisms. A and sA are respon-

sible for the fast time constant of the tunneling mechanism,

while B and sB are responsible for the slow time constant of

the chemical reaction. The procedure to extract the time con-

stants and their relative contributions from the transient char-

acteristics of the drain current is detailed elsewhere.10

Representative curves showing the time dependence of

the drain current under a pulsed gate bias are shown in the

left panel of Fig. 5(a). The transient portion of Id for GFETs

with three different initial defect densities is shown in the

right panel of Fig. 5(a). The fitting parameters for Eq. (3) are

calculated for the 30 ls to 2 ms range of transient drain cur-

rents. sA and sB obtained for the two-trap model are approxi-

mately fixed to average values (sA¼ 74 ls, sB¼ 628 ls),

assuming the time constants of the two-trap mechanism are

similar in devices from the same batch. Then, the amounts of

the contribution from fast and slow traps (parameter A and B

in Eq. (3)) are calculated as shown in Fig. 5(b). In general,

the contribution of A, i.e., the tunneling component, is much

higher than the chemical reaction component.

Interestingly, A and B showed significantly different

trends with ndo. Parameter A appears to be correlated with

the ndo, while parameter B appeared nearly independent of

it. This result means that the initial defects in the graphene

influence the charge trapping caused by the tunneling mecha-

nism, but do not affect charge generation caused by the re-

dox reaction. The transient Id curves showing a large

difference in a short time scale support the result of this

modeling. The correlation between the initial defect density

and tunneling-induced charging is intuitively understandable

because the preferred charge transfer from the defect sites

has been observed using scanning tunneling microscopy

(STM).19,29 Such charge transfer is promoted by the change

in the energy level of the defect states, which usually shifts

to a midgap state. On the other hand, the weak correlation

between the slow charging mechanism, i.e., the chemical

reaction component, and the initial defect density is more

difficult to explain because the chemical reaction at the

FIG. 3. (a) Schematic image and SEM

image of graphene FETs. Scale bar is

40 lm. (b) Representative pulsed Id-Vg

curves of graphene FETs with various

D/G ratios. Because of different con-

ductivity, drain currents are normal-

ized by a value of the drain current at

the Dirac point. Vd¼ 0.5 V, rise and

fall time¼ 300 ls, pulse width¼ 1 ms.

FIG. 4. Defect density dependence of hysteresis represented by charge

density.

083110-3 Cho et al. Appl. Phys. Lett. 103, 083110 (2013)

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

93.180.53.211 On: Mon, 17 Feb 2014 07:26:18

surface of the graphene can be catalyzed by impurity mole-

cules that bond at defect sites as a result of the low formation

energy of the chemisorption compared to perfect lattices.30

One possible explanation is the supply limited reaction due

to the Al2O3 passivation of the graphene channel. Lee et al.reported that the limited oxygen supply can suppress the sur-

face reaction component.10

The implication of this finding is that the initial quality

of graphene is critically important in stabilizing graphene

devices even after optimizing other factors, including the

substrate material, integration process, and device structure,

to minimize the oxygen exposure as suggested by Lee

et al.10 If the quality of the graphene cannot be improved to

minimize the hysteresis, an alternative way to passivate the

influence of the initial defect sites at the graphene, such as a

forming gas anneal for silicon devices, will be necessary.

In summary, the charge trap density of GFETs and the

initial defect density of graphene are found to correlate to

each other. In particular, the tunneling component of the

charge trapping has shown a strong correlation with initial

defect density. While the chemical reaction component of

the hysteresis could be minimized by limiting the oxygen

supply to the graphene, the majority of hysteresis can be

eliminated only by reducing the initial defect density of the

graphene.

This work was supported by the Pioneer Research

Center Program (2012-0009462) and by the Inter-ER

Cooperation Projects funded by the MKE and KIAT

(R0000499).

1P. Joshi, H. E. Romero, A. T. Neal, V. K. Toutam, and S. A. Tadigadapa,

J. Phys.: Condens. Matter 22, 334214 (2010).2M. Lafkioti, B. Krauss, T. Lohmann, U. Zschieschang, H. Klauk, K. v.

Klitzing, and J. H. Smet, Nano Lett. 10, 1149 (2010).3Z.-M. Liao, B.-H. Han, Y.-B. Zhou, and D.-P. Yu, J. Chem. Phys. 133,

044703 (2010).4H. Wang, Y. Wu, C. Cong, J. Shang, and T. Yu, ACS Nano 4, 7221

(2010).5G. Kalon, Y. Jun Shin, V. Giang Truong, A. Kalitsov, and H. Yang, Appl.

Phys. Lett. 99, 083109 (2011).

6Y. G. Lee, C. G. Kang, U. J. Jung, J. J. Kim, H. J. Hwang, H.-J. Chung, S.

Seo, R. Choi, and B. H. Lee, Appl. Phys. Lett. 98, 183508 (2011).7A. Veligura, P. J. Zomer, I. J. Vera-Marun, C. J�ozsa, P. I. Gordiichuk, and

B. J. van Wees, J. Appl. Phys. 110, 113708 (2011).8H. Xu, Y. Chen, J. Zhang, and H. Zhang, Small 8, 2833 (2012).9C. G. Kang, Y. G. Lee, S. K. Lee, E. Park, C. Cho, S. K. Lim, H. J.

Hwang, and B. H. Lee, Carbon 53, 182 (2013).10Y. G. Lee, C. G. Kang, C. Cho, Y. Kim, H. J. Hwang, and B. H. Lee,

Carbon 60, 453 (2013).11S. S. Sabri, P. L. L�evesque, C. M. Aguirre, J. Guillemette, R. Martel, and

T. Szkopek, Appl. Phys. Lett. 95, 242104 (2009).12S. M. Song and B. J. Cho, Nanotechnology 21, 335706 (2010).13W. H. Lee, J. Park, Y. Kim, K. S. Kim, B. H. Hong, and K. Cho, Adv.

Mater. 23, 3460 (2011).14X. Liang, B. A. Sperling, I. Calizo, G. Cheng, C. A. Hacker, Q. Zhang, Y.

Obeng, K. Yan, H. Peng, Q. Li, X. Zhu, H. Yuan, A. R. Hight Walker, Z.

Liu, L. Peng, and C. A. Richter, ACS Nano 5, 9144 (2011).15J. A. Robinson, M. LaBella, M. Zhu, M. Hollander, R. Kasarda, Z.

Hughes, K. Trumbull, R. Cavalero, and D. Snyder, Appl. Phys. Lett. 98,

053103 (2011).16Y.-F. Lin, S.-T. Wang, C.-C. Pao, Y.-C. Li, C.-C. Lai, C.-K. Lin, S.-Y.

Hsu, and W.-B. Jian, Appl. Phys. Lett. 102, 033107 (2013).17J. Yamaguchi, K. Hayashi, S. Sato, and N. Yokoyama, Appl. Phys. Lett.

102, 143505 (2013).18N. M. R. Peres, F. Guinea, and A. H. Castro Neto, Phys. Rev. B 73,

125411 (2006).19L. Tapaszt�o, P. Nemes-Incze, G. Dobrik, K. Jae Yoo, C. Hwang, and L. P.

Bir�o, Appl. Phys. Lett. 100, 053114 (2012).20D. Van Tuan, J. Kotakoski, T. Louvet, F. Ortmann, J. C. Meyer, and S.

Roche, Nano Lett. 13, 1730 (2013).21J.-H. Chen, C. Jang, S. Adam, M. S. Fuhrer, E. D. Williams, and M.

Ishigami, Nat. Phys. 4, 377 (2008).22K. Kim, H. J. Park, B.-C. Woo, K. J. Kim, G. T. Kim, and W. S. Yun,

Nano Lett. 8, 3092 (2008).23J.-Y. Hwang, C.-C. Kuo, L.-C. Chen, and K.-H. Chen, Nanotechnology

21, 465705 (2010).24H. J. Hwang, C. Cho, S. K. Lim, S. Y. Lee, C. G. Kang, H. Hwang, and

B. H. Lee, Appl. Phys. Lett. 99, 082111 (2011).25Z. Cheng, Q. Zhou, C. Wang, Q. Li, C. Wang, and Y. Fang, Nano Lett. 11,

767 (2011).26M. Ishigami, J. H. Chen, W. G. Cullen, M. S. Fuhrer, and E. D. Williams,

Nano Lett. 7, 1643 (2007).27B. Krauss, T. Lohmann, D.-H. Chae, M. Haluska, K. von Klitzing, and

J. H. Smet, Phys. Rev. B 79, 165428 (2009).28L. G. Cancado, A. Jorio, E. H. M. Ferreira, F. Stavale, C. A. Achete, R. B.

Capaz, M. V. O. Moutinho, A. Lombardo, T. S. Kulmala, and A. C.

Ferrari, Nano Lett. 11, 3190 (2011).29J. R. Hahn, H. Kang, S. Song, and I. C. Jeon, Phys. Rev. B 53, R1725

(1996).30D. W. Boukhvalov and M. I. Katsnelson, Nano Lett. 8, 4373 (2008).

FIG. 5. (a) In the left panel, drain cur-

rent reduction as function of time dur-

ing the pulse width. Drain current is

fitted by the two-trap model formu-

lized by Eq. (3). In the right panel, rep-

resentative fitting lines with various

defect densities. To compare each

other, fitting lines are aligned by the

value of the current at saturation.

(b) Defect density dependence of A

and B extracted from fitting lines by

Eq. (3).

083110-4 Cho et al. Appl. Phys. Lett. 103, 083110 (2013)

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

93.180.53.211 On: Mon, 17 Feb 2014 07:26:18