Effect of oxygen plasma etching on graphene studied with Raman spectroscopy and electronic transport

We report a study of graphene and graphene field effect devices after exposure to a series of short pulses of oxygen plasma. We present data from Raman spectroscopy, back-gated field-effect and magneto-transport measurements. The intensity ratio between Raman"D"and"G"peaks, I(D)/I(G) (commonly used to characterize disorder in graphene) is observed to increase approximately linearly with the number (N(e)) of plasma etching pulses initially, but then decreases at higher Ne. We also discuss implications of our data for extracting graphene crystalline domain sizes from I(D)/I(G). At the highest Ne measured, the"2D"peak is found to be nearly suppressed while the"D"peak is still prominent. Electronic transport measurements in plasma-etched graphene show an up-shifting of the Dirac point, indicating hole doping. We also characterize mobility, quantum Hall states, weak localization and various scattering lengths in a moderately etched sample. Our findings are valuable for understanding the effects of plasma etching on graphene and the physics of disordered graphene through artificially generated defects.


Introduction
Graphene has received much attention recently in the scientific community because of its distinct properties and potentials in nanoelectronic applications. Many reports have been made on graphene's very high electrical conductivity at room temperature [1,2] and its potential use as next-generation transistors [3], nano-sensors [4], transparent electrodes [5], and many other applications.
Plasma etching is a common tool used to pattern graphene nanostructures, such as Hall bars [2] and nanoribbons [6]. In addition, plasma etching is used to study how graphene's properties are affected by etching-induced disorder [7]- [9].
Previous reports have characterized oxygen plasma's effect on graphene's field effect conductivity [7,8], Raman spectra, and surface morphology as measured by atomic force microscopy (AFM) [7]. However, many aspects regarding the disorder generated by plasma etching in graphene remain to be better understood. For example, the initial stage of the disorder generation as reflected in Raman spectra (the so-called "graphite to nanocrystalline graphite" stage [23]) has not been fully characterized because of the corresponding narrow window for the amount of oxygen plasma exposure. The magneto-transport and carrier localization in plasma etched graphene have also not been studied before, to the best of our knowledge.
Here we present a study on graphene exposed to a various amount of oxygen plasma etching, focusing on its properties as characterized by Raman, field-effect measurements and magneto-transport measurements (including quantum Hall effect and weak localization).

Experimental procedure
Our graphene samples are fabricated by micromechanical exfoliation [1] of highly ordered pyrolytic graphite (HOPG, "ZYA" grade, Momentive Performance Materials) onto a p-doped Si wafer with 300 nm of SiO2. Single-layer graphene flakes, typically around 100 µm 2 in size, are identified using color contrast with an optical microscope [24] and then confirmed with Raman spectroscopy (using a 532 nm excitation laser) [25]. Graphene field-effect devices are fabricated using electron-beam lithography. The electrical contacts (5 nm-thick chromium and 35 nm-thick gold) are fabricated by electron-beam evaporation.
Our graphene devices are exposed cumulatively to short pulses (~ ½ seconds) of oxygen plasma in a microwave plasma system (Plasma-Preen II-382, figure 1) operating at 100 W. A constant flow of O2 is pumped through the sample space, and the gas is excited by microwave (manually pulsed on and off). The microwave generates an ionized oxygen plasma, which generates etched holes (observable by AFM similar as previously reported [7]) in graphene. The microwaveexcited plasma pulses are applied to the samples cumulatively, and field-effect and Raman measurements are performed as soon as possible (~5 min) in ambient atmosphere and temperature after each pulse. The magneto-transport data are taken using a He 3 superconducting magnet probe several days after plasma exposure.

Results and Analysis
Figure 2(a) shows representative Raman spectra for a single-layer graphene device (sample "1") taken after multiple cumulative exposures to oxygen plasma etching pulses. Of particular interest is the disorder-induced characteristic Raman "D" peak (~1350 cm -1 ) (other disorder-related peaks, such as the "D'" (~1620 cm -1 ) and "D+D'" (~2940 cm -1 ) peaks are also observed) [23]. The "D" peak initially rises with increasing exposure. After 14 pulses, the "D" peak has ~4 times the amplitude of the graphene's "G" peak (~1580 cm -1 ), but with additional exposure, the "D" peak begins to attenuate along with the "G" and "2D" (~2690 cm -1 ) peaks (both of which begin to attenuate since the first exposure). After 23 pulses, the "D" peak (while still significant) reduces to ~2 times the amplitude of the "G" peak and the "2D" peak is almost completely suppressed. Figure 2(b) shows the progression of the peak intensity ratios (ID/IG and I2D/IG) as functions of the number (Ne) of plasma-etching pulses. The dependence of the ratio of the intensities of the "D" and "G" peaks, ID/IG, on Ne shows 2 different behaviors in the regimes of "low" ( Ne < ~14) and "high" ( Ne > ~14) defect densities (referred to as "nanocrystalline graphite" and "mainly sp 2 amorphous carbon" phases respectively in Ref. 23). ID/IG begins at ~0 before the plasma exposure (the small "D" peak is likely due to the device fabrication process), increases with increasing Ne in the low-defect-density regime to ~4 after 14 plasma exposures, then decreases with further increasing Ne in the high-defectdensity regime to ~1.9 for Ne = 25. On the other hand, the ratio of the intensities of the "2D" and "G" peaks, I2D/IG, continuously decreases with increasing Ne from ~3 for Ne = 0 down to ~ 0.3 for Ne = 25. In figure 2(b), the data in the lowdefect-density regime are fitted to and those in the high-defect-density regime are fitted to The inset of figure 2(a) shows log(ID/IG) versus log(Ne). A line fit of the data in the low-defect-density regime gives a slope of ~1.1, confirming the approximate linear relationship between ID/IG and Ne in that regime, as found in equation (1).  A magneto-transport study is performed for a graphene field-effect device (sample "3") exposed to 2 oxygen plasma pulses. Prior to magneto-transport measurements, sample "3" is characterized by field-effect conductivity measurements and Raman spectroscopy before and after exposure, and the results are shown in figures 4(a) and 4(b) respectively. The CNP shifts from 2 V before exposure to 18 V after exposure, with the minimum conductivity (σmin, taken as the conductivity at the CNP) decreasing from ~275 µS to ~100 µS. We can extract the field-effect mobility (µFET) by examining the slope of the field-effect curve, conductivity (σ) versus back gate voltage (Vg), where Vg is sufficiently far away from the CNP and the curve is in the linear regime, using [26] where t = 300 nm is the thickness of the SiO2 and ε = 3.9×ε 0 = 3.45×10 −11 F/m is the permittivity of the SiO2. This gives room-temperature µFET ≈ 400 cm 2 /Vs after exposure, compared to µFET ≈ 9800 cm 2 /Vs before exposure.
The Raman spectra measured from sample "3" show the emergence of the characteristic "D" peak after exposure, We can also calculate mobility from the Hall effect measured in figure 4(d) using [26] ( ) This yields a post-exposure Hall mobility of ~600 cm 2 /Vs. We also note a pronounced peak in Rxx(B=0). This peak is due to weak localization, which is typically suppressed in low-disorder single-layer graphene devices [27].
Rxx(B) measurements in low B are also taken from sample "3" at various temperatures ranging from 0.5 K to 60 K and Vg = 10 V, shown in figure 5(a). These data characterize weak localization in the sample. Weak localization arises from the constructive interference between time-reversed multiple-scattering trajectories of phase coherent carriers, leading to coherent back-scattering of carriers and increasing the electrical resistance. When a perpendicular magnetic field is introduced to break time-reversal symmetry or the temperature is raised to destroy phase coherence, this interference is suppressed. From the low magnetic field data, we can extract the phase coherence length, L ϕ , as well as the intervalley and intravalley scattering lengths, Li and L* respectively, using [28][29][30] 1  2   1  ln   and   2 ,* , ,* , where The l and w denote the length and width of the graphene device and σ(B,Th) represents the magneto-conductivity at sufficiently high temperature (approximated using data at T = 60 K) for the weak localization feature to disappear. From the plots of these characteristic lengths as a function of temperature, seen in figure 5(b), we note Li and L* are relatively Tinsensitive, averaging ~14 nm and ~4 nm respectively. We also note L ϕ decreases with increasing T from ~23 nm at 0.5 K to ~10 nm at 60 K.

Discussion
The Raman spectrum taken before exposure indicates single-layer graphene with ID/IG > 2 and a symmetric Lorentzian 2D peak [25]. The evolution of ID/IG as a function of induced plasma disorder can be attributed to a gradual evolution from the sp 2 -bonded carbon found in graphene into amorphous carbon with appreciable sp 3 bonding [23]. This evolution is classified into 2 main regimes -a so-called "graphite to nano-crystalline graphite" phase characterized by increasing ID/IG with increasing disorder (or "low-defect-density regime"), and a so-called "nanocrystalline graphite to manly sp 2 amorphous carbon" phase characterized by decreasing ID/IG with increasing disorder (or "high-defect-density" regime) [23].
In the low-defect-density regime, an empirical formula known as the Tuinstra-Koenig relation [23,31,32] has been developed to extract the crystalline domain size Ld (which also characterizes the average separation between defects): (8) and λ is the Raman excitation wavelength (532 nm in our case). In the high-defect-density regime, it has been proposed that ID/IG versus Ld can be fitted to the equation [23,32] where the constant D(λ) is obtained by imposing continuity between the two regimes.
A recent work [13], however, has suggested a new relationship between ID/IG and Ld, in the low-defect-density regime as (10) In our experiment, we assume the total exposure time to be proportional to the defect concentration, 1/Ld 2 , therefore We find that our data agrees with equation (9) in the high-defect-density regime (dot-dashed line fit in figure 2(b)).
The gradual decrease of the "2D" peak is also consistent with previous work [7,9,11]. The decreasing I2D/IG versus Ne is likely due to a combination of the positive doping of the graphene and the defect-induced suppression of the lattice vibration mode corresponding to the 2D peak.
The field-effect measurements show a strong positive shift of the CNP after plasma etching, most likely caused by p-doping molecules, e.g. water, attaching to the edges of etched holes [33]. If we use data in figure 3 as a guideline, 4 plasma pulses (which caused the CNP to up-shift by ~80 V, indicating a carrier density increase of ~6×10 12 cm 2 ) gives ID/IG ≈ 1.4, from which we extract (using equation (10)) Ld ≈ 9.1 nm, which corresponds to a defect concentration of ~8.4×10 11 cm 2 . We therefore estimate an average of ~7 holes doped per defect. Such hole doping could also cause a small blue shift in the Raman "G" peak position [34], which, however, is not resolved within the resolution of our Raman measurements.
The electrical transport measurements show the oxygen plasma exposure significantly decreases the sample's σmin (at Dirac point) and mobility. Interestingly, even with this high level of disorder and low mobility (accompanied by pronounced weak localization), we still observe the half-integer quantum Hall effect in the form of well-developed plateaus in Rxy corresponding to Landau level filling factors at 2, 6 and 10. We also note in figure 5(d) that although the quantum Hall plateaus in Rxy are reasonably well-developed for ν = ±6 and ±10, the corresponding Rxx (which normally approaches 0 for quantum Hall effect states) still take substantial values (> ~0.8 h/e 2 ). More studies are needed to better understand this behavior and whether it could be related to, for example a proposed "dissipative" quantum Hall effect [35] or charge inhomogeneity (puddles) [33] in graphene.
Comparing the measured Raman spectrum from sample "3" after exposure with the spectra in figure 2 indicates the sample is in the low-defect-density regime. For ID/IG ≈ 3 ( figure 4(b)) we can calculate Ld ≈ 6 nm from equation (10). This is on the similar order of magnitude as Li (~14 nm) and L* (~4 nm) extracted from weak localization fitting. Though it is still difficult to pinpoint exactly what kinds of disorder in graphene give rise to these scattering lengths, it has been proposed that atomically sharp defects largely contribute to Li while larger-length-scale disorder such as charged impurities can contribute to L* [10,28].

Summary
In summary, we have studied the disorder in graphene caused by oxygen plasma etching. Raman spectra show a characteristic progression of ID/IG with an increasing level of disorder, indicating an evolution from a graphene lattice to a more amorphous carbon phase in 2 distinct regimes, in which we can empirically extract defect length scales from ID/IG. The    (a) Conductivity as a function of back gate voltage measured in single-layer graphene sample "3" before and after 2 oxygen plasma pulses. The inset of (a) is an optical image of sample "3". For all electrical measurements, the current was supplied from lead 1 to lead 5, with Rxx measured from 3 to 4 and Rxy from 6 to 4. (b) Raman spectra of sample "3" taken before and after plasma exposure. The laser excitation wavelength is λ = 532 nm. The spectra are offset vertically for clarity. (c) Measured resistance as a function of back gate voltage of post-exposure sample "3" at 18 T and 0.5 K.  Extracted characteristic lengths from weak localization fittings as functions of temperature. The inset shows magnetic-fielddependent ∆σ (solid line) and the result of fitting (dashed line) using equation (4) to extract L ϕ , Li and L*. ∆σ has been symmetrized between two opposite magnetic field directions.