N-Doping of Graphene Through Electrothermal Reactions with Ammonia

See allHide authors and affiliations

Science  08 May 2009:
Vol. 324, Issue 5928, pp. 768-771
DOI: 10.1126/science.1170335

This article has a correction. Please see:

Negatively Doped Graphene Nanoribbons

The potential applications in electronic devices of graphene (single atom, thick layers of graphite) would be even greater if it can be accessed in both p- and n-doped forms. Graphene nanoribbons (long strips only tens of nanometers in width) are readily p-doped by adsorbates from the ambient atmosphere. Wang et al. (p. 768) show that when graphene nano-ribbons are electrically heated in an ammonia atmosphere, nitrogen is incorporated mainly at the edges of the ribbon and creates an n-type material. Field-effect transistors that operate at room temperature can be made from this material.


Graphene is readily p-doped by adsorbates, but for device applications, it would be useful to access the n-doped material. Individual graphene nanoribbons were covalently functionalized by nitrogen species through high-power electrical joule heating in ammonia gas, leading to n-type electronic doping consistent with theory. The formation of the carbon-nitrogen bond should occur mostly at the edges of graphene where chemical reactivity is high. X-ray photoelectron spectroscopy and nanometer-scale secondary ion mass spectroscopy confirm the carbon-nitrogen species in graphene thermally annealed in ammonia. We fabricated an n-type graphene field-effect transistor that operates at room temperature.

Recently, graphene has been made into semiconductors in the form of nanoribbons, leading to room temperature p-type graphene field-effect transistors (FETs) (1, 2). However, a fundamental problem has been that the edge structures and chemical terminations of graphene synthesized by various methods are unknown and uncontrolled, whereas their effects to the physical properties have been widely predicted (39). In particular, graphene nanoribbons (GNRs) edge-terminated by nitrogen species were shown to be electron-rich, leading to n-type transistor behavior (8). Therefore, it is essential to precisely control the edge structures and chemical terminations to obtain desirable device characteristics. Edge doping could present a new means of doping for nanoscale graphene.

We now report that GNRs can be functionalized by nitrogen species by high-power electrical annealing (e-annealing) in NH3 and exhibit n-type electronic doping. GNRs were synthesized chemically (1) or were lithographically patterned from pristine peel-off graphene (1012); the width ranged from below 10 nm up to ~150 nm. Chemically derived GNRs were dispersed on a 300-nm SiO2/Si chip, located and imaged by scanning electron microscopy (SEM) with 1-kV acceleration voltage (13) and by atomic force microscopy (AFM) [Fig. 1, B and C, and fig. S1; see also supporting online material (SOM) (14)]. We then fabricated FET-like devices on selected ribbons with palladium (Pd) metal source/drain (S-D) and highly doped Si backgate (Fig. 1, A and D).

Fig. 1

Electrothermal reaction of individual GNRs in NH3. (A) Schematics of a GNR device e-annealed under high current in NH3. The devices have metal S-D, highly doped Si backgate and 300-nm SiO2 gate dielectrics. The color gradient along the GNR represents the temperature profile: higher in the center region and lower near the contacts (18, 25). Atomic colors are as follows: H, green; O, red; N, purple; C, gray. The GNR shown here is assumed to have ideal zigzag edges, which may not correspond to the GNRs in experiments. D, drain; S, source; G, gate. (B) SEM (acceleration voltage = 1 kV) and (C) AFM images of a ~30-nm-wide GNR. (D) AFM image of the device fabricated on the GNR shown in (B) and (C).

We first focused on GNRs wider than ~20 nm. Under ambient conditions, the edges of these as-made GNRs were probably terminated by hydrogen, oxygen, hydroxyl groups, and carboxylic groups (15) and exhibited p-doping with the Dirac point at gate voltage Vgs > 40 V in current-gate voltage Ids-Vgs curves [fig. S3 (14)] (1, 2). The p-doping was partly attributed to the oxygen edge groups (6, 7), physisorbed oxygen molecules, and noncovalent poly(m-phenylenevinylene-co-2,5-dioctoxy-p-phenylenevinylene) (PmPV) coatings used in the synthesis process that are known to p-dope carbon nanotubes (16, 17). Pumping in a vacuum reduced the conductance of GNRs slightly [fig. S3 (14)], corresponding to a decrease in p-doping by partial desorption of oxygen, either from the GNRs or GNR-metal contacts (16).

We then e-annealed the devices in high vacuum (~10−6 torr) by double sweeping the S-D bias Vds (Fig. 2A). As Vds was increased to high biases, the slope of Ids-Vds curve decreased or even became negative, with a noticeable hysteresis between back-and-forth sweeps that indicated the removal of the p-doping sources (18). We recorded Ids-Vgs curves immediately after each e-annealing sweep and observed that the Dirac point gradually moved toward zero Vgs [fig. S2 (14)]. We continued to increase Vds during Ids-Vds sweeps until no hysteresis occurred, which indicated that most of the p-doping was removed (Fig. 2A and fig. S2). After e-annealing, the Ids-Vgs curve of GNR devices always showed the Dirac point at finite positive Vgs (typically 5 to 20 V) and slightly asymmetric hole and electron conduction (Fig. 2B). The residual p-doping was probably caused by oxygen species remaining on the edges (Fig. 3C) (6, 7), as well as the doping from the contact metal caused by the high work function of Pd (19).

Fig. 2

E-annealing of individual GNRs in vacuum and NH3. (A) Typical e-annealing process in vacuum for a w ~ 125-nm GNR. The process consisted of several double Ids-Vds sweeps (direction pointed by the arrows) with gradually increasing Vds. We stopped when no hysteresis existed between back and forth sweeps (blue curve). (Upper and lower insets) AFM images of the same device as-made and after e-annealing in vacuum, respectively. The height was reduced from ~1.5 to ~1.0 nm because of removal of PmPV coatings by e-annealing. (B) Ids-Vgs curves of the same GNR device as-made (red) and after e-annealing in vacuum (blue). The Dirac point moved from beyond 40 to ~8 V. (C) Ids-Vgs curves of the same GNR device in vacuum before (red) and after e-annealing in NH3 (blue). After e-annealing in NH3, we pumped the device to base pressure for overnight before taking the blue curve. The Dirac point moved from beyond 40 to ~–14 V. Vds = 1 mV in (B) and (C). (D) Calculated DOS of a w ~ 40-nm armchair GNR terminated partly by nitrogen species (14). The red dashed line denotes the Fermi level. (Left inset) The dependence of doping level (the position of the Fermi level from the Dirac point) on the density of substitutional N on the edges. (Right inset) Three unit cells of the simulated structure. There are two NH groups in the unit cell of the simulated structure. These edge groups are likely to coexist in real GNRs.

Fig. 3

Room temperature n-type graphene FETs. (A) Ids-Vgs curves of the as-made GNRFET in vacuum (p-type, red) and after e-annealing in NH3 (n-type, blue). Vds = 1 V for both curves. (B) Ids-Vds curves of the same device. Red curves were taken on an as-made device: Vgs = –40 V, –37 V, –34 V, –31 V, and –28 V from top to bottom. Blue curves were taken on e-annealed device: Vgs = 40 V, 35 V, 30 V, 25 V, and 20 V from top to bottom. The nonlinear characteristics for both p- and n-type transistors near zero bias were due to finite SB for both electrons and holes by Ti contact used for this part of the work (24). Higher performance devices can be made by using higher (lower) work-function metals for p- (n-) type transistors and by heavily doping the contact regions, respectively. (Insets) AFM images of the device before and after e-annealing. Height was reduced by ~0.4 nm after e-annealing due to removal of PmPV coatings. (C) Calculated DOS of a semiconducting 21-armchair GNR (w ~ 2.5 nm) terminated by oxygen-containing groups (14). The red dashed line is the projected DOS (PDOS) on the carbon backbone, which clearly shows a p-type doping effect. (Inset) Two unit cells of the simulated GNR. There are one C=O and two C–OH groups in each unit cell. These edge groups are likely to coexist in real GNRs. E, energy; EF, Fermi energy. (D) Calculated DOS of a 21-armchair GNR with nitrogen-containing groups on the edge sites, which is an n-type semiconductor (14). The DOS in the range of interest is mainly from the carbon atoms, and the PDOS on the carbon backbone has negligible difference with the total DOS within the energy range of interest. (Inset) Two unit cells of the edge structures of the simulated GNR. There are one NH and two C–NH2 groups in each unit cell. These edge groups are likely to coexist in real GNRs.

During e-annealing, hundreds of microwatts of power were injected into a single GNR, which caused electrothermal self-heating of the GNR. We probed the temperature of GNRs under high-power input by measuring the red shift of the Raman G band of a single GNR (14, 20). For a typical GNR under Vds = 2 V (356-μW input power), a G-band shift of ~4.4 cm−1 was observed [fig. S5 (14)], corresponding to an average temperature of ~300°C along the GNR (20). Thus, electrical annealing of GNRs to hundreds of degrees led to the removal of physisorbed oxygen and PmPV molecules and reduced p-doping, consistent with the GNR height decrease by ~0.3 to 0.6 nm after e-annealing removal of the coating on GNRs (Fig. 2A, insets).

To chemically modify GNRs, we e-annealed the GNR devices in a ~1-torr NH3/Ar environment with carefully designed sequences and control experiments [fig. S3 (14)]. In NH3, we applied similar e-annealing sequences as in vacuum. After e-annealing, we pumped the chamber to base pressure for overnight (~8 hours) to fully remove physisorbed NH3 molecules (14). Comparing the Dirac point positions of the devices in vacuum before and after e-annealing in NH3, we observed a large ~−20-V shift (Fig. 2C and fig. S3B) that stayed stable and constant in vacuum. The shift caused by physisorbed NH3 molecules (21) was usually smaller in magnitude [typically –5 V; fig. S3B (14)], and control experiments showed that physisorbed NH3 molecules were unstable and removed in vacuum after overnight (~8 hours) pumping [fig. S3C (14)]. Considering the high temperature of GNRs during e-annealing in NH3, we propose chemical reactions between GNRs and NH3 leading to nitrogen functionalization, most likely at the more reactive edge carbon atoms (15). As a result, n-doping was introduced by the electron-rich nitrogen species (8). We cannot rule out the possibility of C–N bond formation at defect sites within the GNR plane; however, defect density in our chemically derived GNRs was low, based on their comparable electrical properties (including mobility, estimated to be a few hundred to ~1000 cm2/Vs for GNRs wider than 20 nm) to similar-width lithographically patterned ribbons from pristine graphene (2). Furthermore, we carried out e-annealing of lithographically patterned GNRs from pristine peel-off graphene in NH3 and qualitatively observed the same behavior as that of chemically derived GNRs [fig. S4 (14)]. These results suggest C–N formation most likely at the edge sites. Unlike the potassium doping approach (22), our doping approach introduced no appreciable charged impurities to degrade carrier mobility in GNR devices, as evidenced by similar p- and n-channel slopes in Ids-Vgs curves before and after e-annealing in NH3 (Fig. 2, B and C).

We theoretically investigated the effect of edge functionalization by calculating the band structure of GNRs terminated by oxygen- and nitrogen-containing species (14). Calculations showed that GNRs with edge functionalization by oxygen and nitrogen species were p- and n-doped, respectively (Figs. 2D and 3, C and D), which agrees with the Dirac point shifts observed in experiments. P-doping was originated from a sub-band introduced near the Fermi level by edge C=O double bonds (Fig. 3C). A nitrogen atom bonded to two C atoms at the edges (that is, a N atom substituting a C atom at the edges, Figs. 2D and 3D) was the most effective in n-doping (8), whereas NH2 groups terminating a GNR with perfect zigzag or armchair edges did not introduce n-doping (7). In real GNRs, however, the edges could be imperfect. N and NH substitution and NH2 termination could all be possible after e-annealing, which leads to an n-doping effect (Fig. 3D). The n-doping level was approximately proportional to the density of edge substitutional N atoms in GNRs (Fig. 2D, inset). Under the same density of edge substitutional N atoms, the calculated doping concentration per unit area was found to scale inversely with the GNR width, which results in the Fermi level closer to the Dirac point for wider ribbons. Although the calculated structures were not the same as GNRs in our experiments, our calculations did confirm the possibility of “bulk” p- and n-doping of GNRs (up to width w ~ 40 nm) by edge chemical groups.

Direct spectroscopy of individual GNRs is difficult given the small quantity of the material and the limited spatial resolution of most spectroscopic tools, so we carried out x-ray photoelectron spectroscopy (XPS) and nanometer-scale secondary ion mass spectroscopy (nanoSIMS) studies on graphene sheet (GS) films (23) thermally annealed in NH3 (14). Thermal reactions should give similar products as electrothermal reactions on graphene under similar reaction conditions. We thermally annealed a GS film under NH3 to 1100°C and, as control, a similar film sample under Ar to 800°C. XPS data (Fig. 4A) revealed that both samples showed similar C signals, indicating a similar amount of carbon material measured. We observed a clear N signal on the sample annealed in NH3, whereas we detected no N signal on the control sample (Fig. 4A, inset). From nanoSIMS (Fig. 4B), the sample annealed in NH3 showed similar O/C2 and C2H/C2 but greater CN/C2 ratio than the control sample, suggesting C–N bond formation in GSs by thermal annealing. Both XPS and nanoSIMS data provided spectroscopic evidence of N atoms incorporated into graphene during thermal annealing in NH3. Similar reactions were expected during electrothermal annealing of graphene. C–N bond formation should occur predominantly on the edges and defect sites in the plane, where the C atoms are much more chemically reactive than in the plane of perfect graphene (15). The observation of similar carrier mobilities after N-doping in our GNRs suggested no substantial bulk modification. However, the precise degree of bulk N substitution in graphene undergone reactions with NH3 at various temperatures should be investigated systematically.

Fig. 4

Spectroscopy of GS films thermally annealed in NH3 and Ar. (A) XPS on GS film thermally annealed in NH3 (blue) and Ar (red, control sample). The peaks around 285 and 399 eV are assigned as C and N peaks, respectively. (Inset) Zoom-in view of the XPS data near the N peak. The sample annealed in NH3 shows a clear N signal, whereas the control sample does not. (Sub-inset) Schematics of GSs thermally annealed in NH3. a.u., arbitrary units. (B) Relative ionic O/C2, CH2/C2, and CN/C2 ratios detected by nanoSIMS (14). The sample annealed in NH3 has a much higher CN/C2 ratio than the control sample, indicating C–N bonds formed during thermal annealing in NH3. The intensities and error bars are means and SEs from different regions of interest in the collection area. The finite N signal of control sample came from residual intercalant (tributyl ammonium hydroxide) used for making GSs (23). (Inset) Schematics of nanoSIMS experiment.

Tuning the electronic properties of graphene by chemistry of the edges and/or defects will have a large impact on graphene properties and applications. Previously, we demonstrated p-type sub–10-nm GNRFETs with as-made GNRs (1, 2). In this experiment, we used the e-annealing approach in NH3 to demonstrate n-type sub–10-nm GNRFETs operating at room temperature (Fig. 3). We used lower work-function metal Ti as the contact metal with a 5-nm Pd buffer layer to enhance electron conduction in the devices. As-made W < ~5 nm GNRFETs were p-type with Ion/Ioff ratio ~ 105 (Fig. 3A), because of the aforementioned p-doping sources. We first e-annealed the GNRFETs in vacuum, after which the devices became ambipolar [fig. S6 (14)]. We then exposed the devices in NH3, followed by e-annealing in NH3. After e-annealing in NH3, the GNRFET turned largely to n-type, with Ion ~ 1 μA, Ion/Ioff ratio ~ 105, and similar subthreshold slope as the as-made p-type GNRFETs. Physisorption of NH3 alone had only weak effect on transistor characteristics with small difference between Ids-Vgs curves in vacuum and NH3 [fig. S6 (14)]. We never succeeded in making n-GNRFETs by simple NH3 physisorption. The nonlinearity near zero bias in Ids-Vds curves for both n- and p-type transistors was due to finite Schottky barriers (SB) for both electrons and holes at Ti contacts (24). Ti contacts were used to observe a clear p- to n-FET evolution through the electrothermal reaction. For optimal device performances, one should use the highest and lowest possible work-function metal contacts for p- and n-type transistors respectively. Heavy doping at the contacts could also be invoked to improve performance. Ultra thin high-κ dielectrics could greatly improve the switching characteristics of p- and n-GNRFETs.

We calculated the density of states (DOS) of a semiconducting 21-armchair GNR (~2.5 nm wide) terminated by oxygen- and nitrogen-containing groups and observed p- and n-doping, respectively (Fig. 3, C and D), again confirming the edge chemical effects to the bulk properties of GNRs in the narrow-width regime. Taken together, the ability to control graphene chemistry is an important step toward controlled graphene electronics. Our results suggest that edge doping represents a new approach to dope graphene ribbons and affect its bulk properties, an interesting feature not likely in edge-free seamless carbon nanotubes. This technique, combined with precise control of edge shape (i.e., zigzag or armchair), may lead to precise determination of GNR device characteristics in the future. The current work also opens new possibilities of doing further chemistry on graphene.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S6


References and Notes

  1. Experimental details and supporting data are available as supporting material on Science Online.
  2. This work was supported in part by the Microelectronics Advanced Research Corporation Materials, Structures, and Devices Focus Center; Intel; and Office of Naval Research (ONR). The work done at the University of Florida was supported in part by NSF and ONR. The work at LLNL was performed under the auspices of the U.S. Department of Energy, contract DE-AC52-07NA27344.
View Abstract

Stay Connected to Science

Navigate This Article