Highly reproducible and reliable metal/graphene ...

Highly reproducible and reliable metal/graphene contact by ultraviolet-ozone treatment Wei Li, Christina A. Hacker, Guangjun Cheng, Yiran Liang, Boyuan Tian, A. R. Hight Walker, Curt A. Richter, David J. Gundlach, Xuelei Liang, and Lianmao Peng Citation: Journal of Applied Physics 115, 114304 (2014); doi: 10.1063/1.4868897 View online: http://dx.doi.org/10.1063/1.4868897 View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/115/11?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Anode modification of polymer light-emitting diode using graphene oxide interfacial layer: The role of ultravioletozone treatment Appl. Phys. Lett. 103, 073305 (2013); 10.1063/1.4818820 Thermal reversibility in electrical characteristics of ultraviolet/ozone-treated graphene Appl. Phys. Lett. 103, 063107 (2013); 10.1063/1.4818329 Ultraviolet/ozone treatment to reduce metal-graphene contact resistance Appl. Phys. Lett. 102, 183110 (2013); 10.1063/1.4804643 UV ozone treatment for improving contact resistance on graphene J. Vac. Sci. Technol. B 30, 060604 (2012); 10.1116/1.4754566 Contact resistivity and current flow path at metal/graphene contact Appl. Phys. Lett. 97, 143514 (2010); 10.1063/1.3491804

[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: 222.29.86.61 On: Sat, 10 May 2014 09:33:12

JOURNAL OF APPLIED PHYSICS 115, 114304 (2014)

Highly reproducible and reliable metal/graphene contact by ultraviolet-ozone treatment Wei Li,1,2 Christina A. Hacker,2 Guangjun Cheng,2 Yiran Liang,1 Boyuan Tian,1 A. R. Hight Walker,2 Curt A. Richter,2 David J. Gundlach,2,a) Xuelei Liang,1,a) and Lianmao Peng1

1 Key Laboratory for the Physics and Chemistry of Nanodevices and Department of Electronics, Peking University, Beijing 100871, China 2 Physical Measurement Laboratory, National Institute of Standards and Technology, Gaithersburg, MD 20899, USA

(Received 13 January 2014; accepted 5 March 2014; published online 17 March 2014) Resist residue from the device fabrication process is a significant source of contamination at the metal/graphene contact interface. Ultraviolet Ozone (UVO) treatment is proven here, by X-ray photoelectron spectroscopy and Raman measurement, to be an effective way of cleaning the metal/graphene interface. Electrical measurements of devices that were fabricated by using UVO treatment of the metal/graphene contact region show that stable and reproducible low resistance metal/graphene contacts are obtained and the electrical properties of the graphene channel remain C 2014 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4868897] unaffected. V

I. INTRODUCTION

Graphene is considered a promising material for the post-silicon era.1 Extraordinarily high mobilities (up to 200 000 cm2/Vs) have been obtained in graphene transistors,2 which make graphene an excellent candidate for radio frequency applications.3 However, due to the difficulties of obtaining low metal/graphene contact resistance (Rc), the maximum operating frequency reported for graphene transistors remains much lower than the III-V highelectron-mobility transistor (HEMT).1,3,4 The contact resistance is mainly determined by the unique interactions between the contact metal and the atomically thin, 2D graphene film. The weak metal/graphene interaction and the low available density of states around the Fermi energy in the graphene sheet limit the current injection efficiency from the metal to the graphene.5,6 The dipole layer formed at the metal/graphene interface, and the built-in electrostatic field near the contact area just in the channel due to charge transfer also play important roles in the high metal/graphene contact resistance.7,8 Various approaches have been advanced to reduce the metal/graphene contact resistance, including metal work function engineering and contact geometry.9–12 Besides the above difficulties, metal/graphene interface contamination introduced during the device fabrication process remains an important contributor to high metal/graphene contact resistances. Resist residue from the device fabrication process, e.g., the graphene transfer and subsequent lithography processes used for fabricating contacts to chemical vapor deposition (CVD)-grown graphene, is a general and significant source of the metal/ graphene interface contamination.13 It is equally problematic that this contamination affects device stability and reproducibility. Considering the different causes for the high Rc a)

Authors to whom correspondence should be addressed. Electronic addresses: [email protected] and [email protected]

0021-8979/2014/115(11)/114304/5/$30.00

discussed above, the interface contamination is the most straight forward challenge to overcome through device fabrication process optimization. Several methods such as plasma treatment and thermal annealing have been introduced as processing solutions to reduce the interface contamination and lower Rc.14–18 However, these methods are either difficult to control or restrict the choice of the substrate material (e.g., low cost glass or plastic) and potential application space. There is still lacking of a highly reproducible and reliable low resistance metal/graphene contact fabrication process. Most recently, ultraviolet ozone (UVO) treatment of the contact area prior to metal deposition was reported to be an effective way to remove the resist residue.19,20 By using this approach, we have improved the metal/graphene contact and have shown that Rc less than 200 X lm can be obtained for mechanically transferred, CVD-grown graphene.19 Though the UVO cleaning has been shown to be effective in reducing the metal/ graphene contact resistance, there are two issues to be understood more thoroughly which were not detailed in our previous paper.19 First, the UVO cleaning process should be quantified, so that it can be optimized to obtain better device perfomance. Second, more statistical data are needed to support that reliable low metal/ graphene contact resistance can be reproducibly obtained by the UVO treatement. However, such statistical data were seldomly supplied in previous reports.4,8,17 In this paper, we greatly advance our prior work19 with detailed quantitative X-ray photoelectron spectroscopy (XPS) and Raman spectroscopy studies and analytical analysis. The performance improvements we report in this study are presented with significantly more statistical analysis, and focus on the reliable electrical properties that can be reproducibly obtained by using UVO treatments. II. EXPERIMENTAL PROCEDURES

Large area monolayer graphene was synthesized on 25 lm thick copper foil by using CVD.21 Graphene films

115, 114304-1

C 2014 AIP Publishing LLC V


114304-2

Li et al.

J. Appl. Phys. 115, 114304 (2014)

were then transferred onto heavily doped silicon wafers with a thermally grown 300 nm SiO2 insulator using a copper etch and “modified RCA cleaning” process.22 Graphene transistors and test structures were fabricated at Peking University by using conventional contact photolithography to define openings in the photoresist, followed by metal deposition and lift-off. Just prior to the metallization step, the substrate was placed into a commercial UVO system to remove the resist residue remaining on the graphene surface. Only the contact regions in the openings defined in the resist layer were exposed to UVO, while the graphene channel areas were protected by the photoresist. After e-beam deposition of Ti (20 nm)/Au (80 nm) and lift-off, a second photolithographic process and oxygen plasma was used to pattern the graphene channel followed by photoresist removal in acetone. A transfer length method (TLM)19 test structure with pad spacing (L) of (4.5, 10.5, 16.5, 22.5, 28.5, 34.5, 40.5, and 46.5) lm was used to extract the contact resistance. The graphene channel width is 10 lm and the contact width is 6 lm for all the devices. The electrical characterization of the devices and the data analysis were performed at NIST. To better quantify the UVO cleaning process, we performed high-resolution XPS and Raman spectroscopy studies on control graphene samples that underwent the same photolithographic processing steps and UVO cleaning at NIST as were used in fabricating the electrical devices. In comparing preliminary studies on the films processed in the UVO reactor at Peking University and those from this detailed study using the UVO reactor at NIST, it was found that there are differences in the oxidative environments between the UVO reactors at the respective institutions. Specifically, the UVO exposure in the NIST reactor induces damage between 10 min and 16 min while no significant damage was detected after 22 min in Peking. However, the trends obtained from both systems are the same. III. RESULTS AND DISCUSSION

High-resolution XPS and Raman spectroscopy were used to monitor the UVO cleaning process and to investigate the chemical and disorder changes during the UVO contamination removal process. The carbon (C) and silicon (Si) atomic percent obtained from the XPS measurements on the CVD-grown graphene on Si/SiO2 substrates, which went through the same device fabrication steps and the UVO treatment process are shown in Fig. 1(a). The C signal arises from both the graphene and the resist residue contamination. The Si signal originates from the substrate. A sharp increase of C atomic percent after photolithography can be ascribed to the photoresist residue introduced onto the graphene surface in the device fabrication process. This is consistent with the fluorine (F) content observed in the XPS spectra shown in Figure 1(b) that can only be attributed to the photoresist residue after photolithography. This residue is a general and significant source of instability, irreproducibility, and high contact resistance for graphene devices. After a 5 min UVO cleaning, the C atomic percent was reduced to roughly the same level of the as-transferred sample. This indicates the removal of photoresist residue, and is confirmed by the

FIG. 1. (a) The C and Si atomic percent (left axis) and Raman D-to-G peak area ratio (right axis) changes during the UVO cleaning process. (b) The Fluorine content change during the UVO cleaning process. The curves were vertically shifted for clarity. (c) Raman spectrum of the graphene under different UVO treatment time. Laser wavelength: 514 nm, spot size: 2 lm.

disappearance of the F signal shown in Fig. 1(b). Additional UVO treatment from 5 min to 16 min further reduces the C atomic percent. We ascribe this reduction to the removal of the poly (methyl methacrylate) (PMMA) residue, which was introduced during the transfer process of the CVD-grown graphene.22 The C atomic percent was observed to decrease dramatically if the UVO treatment time was extended even further. This drastic change is attributed to the onset of the UVO attacking the graphene. XPS spectra shows that even the silicon oxide below the graphene is affected after 35 min UVO treatment. The Si atomic percent is found to change inversely with the C signal. This can be understood initially as the removal of PMMA and eventual removal of some of the carbon atoms in graphene, which leads to a gradually thinner film on the SiO2 substrate yielding an increased intensity in the photoemitted electrons.


114304-3

Li et al.

Raman spectroscopy was also used to monitor changes in the graphene quality during the residue removal process (see Fig. 1(c)). The defect (D) peak around 1350 cm1 correlates to the degree of disorder in graphene.23–25 The Lorentzian fitted D-to-G peak area ratio is shown in Fig. 1(a) right axis. The D-to-G peak area ratio remains relatively unchanged for the first 10 min UVO treatment. However, a pronounced D-to-G peak area ratio increase is observed after 16 min UVO treatment, which is due to the increased disorder in the graphene due to damage induced by the UVO exposure, and is consistent with the XPS results. In contrast, our preliminary studies on graphene films processed in the UVO reactor at PKU did not reveal such obvious decomposition of graphene, even for films exposed for 40 min to UVO treatment. These results indicate that the interface contamination was cleaned already by the UVO treatment before defects were introduced into the graphene significantly. The clean graphene surface is important for good metal/graphene contact, which is confirmed by the following device performance measurements. We find the UVO cleaning process to be instrument dependent because we observed two distinctly different times when defects were introduced into graphene, i.e. 16 min for the NIST system and 22 min for the PKU system, respectively. This should be expected, because the cleaning rate of a particular UVO system depends on its specific configuration and use in individual laboratories (e.g. exhaust rate, feed gas, exposure time, sample-to-grid lamp distance, sample temperature, and lamp intensity). However, the working principle of all UVO systems are the same, and the cleaning trends should be the same, just as we mentioned before. Actually, our result in Fig. 1 provided a method, using Raman or XPS monitoring, to obtain the optimized treatment time for a specific UVO system. To investigate the effects of the UVO treatment of the metal/graphene contact region on the performance of the graphene device, we carefully measured and analyzed the electrical characteristics of a large number of samples. Fig. 2 shows output and transfer characteristics of a typical

J. Appl. Phys. 115, 114304 (2014)

10.5 lm channel length device for different UVO treatment times measured in air. There is a marked increase in drain current for the UVO treated devices. Because the channel region was protected by the photoresist during the UVO treatment and the resistivity of the graphene channel proved to remain unchanged,19 the increased current for the UVO treated device results from the large decrease of the contact resistance. The measurements for TLM contact resistance extraction were taken at room temperature in air with the back-gate electrode grounded. For our Ti/Au contacted test structures, the neutrality point is shifted positive by many 10s of volts in air (Fig. 2(b)). It is known that Rc measured far from the Dirac point is almost independent of the gate bias.8 The reproducibility of the decreased contact resistance by UVO treatment is investigated for a statistically relevant sample size of devices for different UVO treatment times. Fig. 3 shows evolution of the total resistance with the UVO exposure time for 130 devices with different channel lengths (nearly 16 devices for each channel length). The devices fabricated without UVO treatment have higher total resistance and exhibit a wider range of resistance values. This variability likely arises from different amounts of photoresist contamination on the graphene surface from device to device. In stark contrast, the resistance values are significantly lower for the devices fabricated with the UVO treatment of the contact region, as shown in Fig. 3(a). Most importantly, the scatter in the resistance values is significantly reduced. This provides a measure of the repeatability one can obtain by interface conditioning, which is of particular importance for large scale device fabrication. We note that a few of the devices that underwent UVO treatment still show relatively high resistance. We believe that this is due to the microstructural imperfections in the channel region (cracks), which has been discussed in previous work.22 Fig. 3(b) shows the corresponding average resistance of the devices in Fig. 3(a). Because all devices have the same channel width and contact width, it is appropriate to extract Rc in the TLM framework. The extracted Rc and Rs for the sets of devices is given in

FIG. 2. Typical width normalized output (a) and transfer (b) characteristics of 10.5 lm channel length devices during the UVO treatment measured in air.


114304-4

Li et al.

J. Appl. Phys. 115, 114304 (2014) TABLE I. Extracted contact resistance (Rc) and sheet resistance (Rs) of the devices in Fig. 3(b).

Rc (X lm) Rs (X/ⵧ)

FIG. 3. (a) The total resistance statistics of 130 devices with different channel lengths that underwent different UVO cleaning process measured at Vgs ¼ 0 V in air. (b) Average total resistance corresponding to those in (a) and the TLM fitting respectively, the error bar represents the standard deviation. (c) Channel mobility statistics of devices treated by different UVO exposure time, the box contains 50% data. The line and little square in the box represent the median and average values.

No UVO

10 min UVO

16 min UVO

25 min UVO

42326 6 2675 494 6 184

1429 6 508 511 6 35

573 6 147 450 6 10

568 6 242 445 6 19

Table I. Devices that underwent a 25 min UVO treatment have a contact resistance of 568 X 6 242 X lm, which compares favorably with most of the reported results.26 Moreover, this value is extracted from the average device resistances. When one considers the variation in the quality of the graphene channel between devices, then it is reasonable to expect even lower contact resistance for optimized devices. Our best obtained values for Rc are less than 200 X lm.19 We believe Rc can be further decreased when combining the UVO cleaning with contact metal and contact geometry engineering. It is important to reiterate that on the metal/graphene interface was affected by the UVO exposure, while the graphene channel itself was protected by the photoresist. Thus, the effect of the UVO on the graphene channel is expected to be negligible and the mobility of devices exposed to different UVO cleaning conditions should not be degraded. A first indication that the graphene channel is unaffected by the UVO treatment is gleaned from the extracted sheet resistance (RS) for the sets of devices. The average values for RS are given in Table I and are found to remain relatively unchanged. To confirm this assumption, the transfer characteristics of the devices were measured in vacuum chamber (