Identifying the mechanisms of p-to-n conversion in unipolar graphene field-effect transistors

The GFETs were characterized at each step of the process (step 1: as-prepared p-type GFET, step 2: GFET after deposition of the Si₃N₄ passivation as shown in figure 1(a), step 3: GFET with the Si₃N₄ passivation after three weeks, step 4: GFET after the removal of the Si₃N₄ passivation) and the transfer characteristics of a typical device are illustrated in figure 1(b). A notable change of the threshold voltage and hole/electron mobility can be observed at various steps in the device fabrication. The threshold voltage of the as-fabricated GFET is positive (+60 V in step 1), and then changes to negative (−76 V in step 2). Subsequently in step 3, the threshold voltage becomes less negative and then positive again when the Si₃N₄ layer is finally removed in step 4. The shapes of the transfer curves in step 1 and step 4 are similar with a shift in threshold voltage, which indicates that the GFET after removal of the Si₃N₄ layer restores the p-type GFET characteristics. The hole and electron motilities can be extracted using the following formula, μ_FET = (1/V_DS) × (g_m/C_BG) × (L/W), where C_BG is the unit area back gate capacitance and W/L is the aspect ratio of the GFET [16]. Given V_DS = 100 mV,C_BG = 11.5 nF cm⁻², the maximum field-effect mobility of the featured GFET is tabulated in figure 1(c).

Zoom In Zoom Out Reset image size

During step 1, the GFET displayed p-type behavior due to the Ti layer serving as the hole dopant [14], as well as adsorption of oxygen and water molecules from air ambient [13]. The use of Ti is also necessary to prevent possible contaminations from the lithography process while defining the source/drain electrodes. In step 2, the negative threshold voltage as well as the increased electron conduction in the positive gate bias region in step 2 indicate that the unipolar p-type GFET had been successfully converted into a unipolar n-type GFET [28]. The change of GFET to n-type conduction is partially due to the removal of the oxygen/water molecules (p-type dopants) [20]. Romero et al reported that graphene on SiO₂ substrate possesses an n-type behavior GFET due to the surface states of SiO₂ donating electrons to the graphene. However, the GFETs have to undergo an annealing time of more than 20 h to achieve a negative threshold voltage [22]. In contrast, the n-type GFETs demonstrated in this work were achieved in less than 2 h during the Si₃N₄ process. Based on these electrical characteristics, the possibility of electrical doping can be ruled out as electrical doping can only shift the Dirac point of the transfer curves without any significant change in their shape [27]. Another possible explanation of the conduction change could be attributed to the exposure of NH₃ as n-type dopants [13]. It is worth noting that the GFET loses n-type properties slowly over a period of ∼3 weeks (step 3), and that the p-type characteristics can be completely revived after removing the Si₃N₄ layer (step 4). The cause of a GFET losing its n-type conductivity could also be the re-adsorption of the oxygen/water molecule related p-dopants after being in air for a long period of time. These trends of changing characteristics from p-type to n-type were repeatable across 11 out of 12 GFETs that were fabricated. The remaining one which displayed ambipolar characteristics is likely to be caused by a neighboring piece of graphite flake, resulting in the poor passivation of the GFET. In order to fully understand the mechanism behind the evolution of electrical behavior, a Raman study was systematically conducted on four GFETs.

The GFETs used in this study were all fabricated from monolayer graphene as verified by Raman spectroscopy (see figure 1(d) for a typical Raman spectrum). The intensive peaks observed at a Raman shift frequency of ∼1580 cm⁻¹ and ∼2680 cm⁻¹ correspond to the G peak and 2D peak, respectively. The G peak arises from the doubly degenerate zone center E_2g mode, while the 2D peak is the second order of zone boundary phonons [9]. The intensity of the 2D peak as compared to the G peak (I_2D/I_G = 1.25) as well as the full width at half maximum (FWHM) of the 2D peak (32.13 cm⁻¹) are indicators of monolayer graphene [9]. In order to eliminate the influence of Si₃N₄ in the Raman characterization results, the Raman spectrum was taken above the graphene passivated by Si₃N₄ and compared to another spectrum taken above the Si₃N₄ surface only (see figure 2(a)). The Si₃N₄ Raman profile demonstrates a slight hump around 1580 cm⁻¹, which coincides with the G peak of the graphene signal. The Si₃N₄ Raman profile was then subtracted away from the as-measured Raman scan to obtain the normalized Raman profile. These processes were repeated for all Raman spectra obtain in steps 2 and 3 of the experiments to eliminate the background signal generated from the Si₃N₄. From figure 2(b), it is worth noting that only the width of the D peak had been affected by the noise, due to the significantly low intensity of the D peak. Little shift or no influence to the D, G and 2D peaks were observed.

Zoom In Zoom Out Reset image size

The pristine graphene had a negligible D peak at ∼1350 cm⁻¹ as previously seen in figure 1(d). After the deposition of the Au/Ti electrodes, the defect peak started to surface as seen in figure 3(a), step 1. The D peak arises from the Ti adatoms that were attached to the graphene surface [14]. A significant decrease of the D peak intensity was seen after the deposition of Si₃N₄ at step 2 (n-type). Conversely, Geng and co-workers observed an increase in the D peak when the graphene was heated in NH₃ ambient at 900 °C [10]. The increase in D peak was due to disorder induced within the graphitic plane in their case. We proposed that the differences between these two experiments were due to the different chemical doping mechanisms [18]. The substitution transfer doping usually requires high temperature and/or existing vacancies in the lattice structure to occur [11, 21]. The increase of the D peak intensity is due to the disruption of the lattice structure by the addition of a foreign atom [21]. From our results, we eliminate the influence of the substitution transfer doping mechanism based on two facts: (1) our experiment was carried out at a maximum temperature of 200 °C, (2) the decrease of I_D/I_G from 0.5 in step 1 to 0.079 in step 2. We suggest that the doping caused by NH₃ below 200 °C was only weakly adsorbed onto the graphene surface and resembled that of the surface transfer doping mechanism [15]. Another interesting observation of the D peak is that even after the removal of the Si₃N₄ passivation (step 4), the I_D/I_G ratio remains below 0.1 as compared to the initial GFET (step 1) (see figure 3(b)). The recovery of the D peak could show that defects to the lattice such as Ti adatoms had been partially removed [12]. This recovery is due to the annealing process at 200 °C in nitrogen ambient [11].

Zoom In Zoom Out Reset image size

For the doping scenario, the shifts of the G and 2D peaks also provide additional information on the type of doping [18]. In order to eliminate the influence of spatially dependent Raman shifts, at least two Raman spots were collected on each investigated sample [12]. In figure 4, a downshift of 3.46 cm⁻¹ and 7.4 cm⁻¹ from step 1 to 2 was observed on the G and 2D peak respectively. The reason for the downshifting of both the G and 2D peaks could be twofold: (1) reduction of the p-type dopants and/or (2) increase of the n-type dopants [7, 8, 14]. Downshifting of the G and 2D peaks at the same time has been experimental observed when the p-dopants in graphene are reduced [6, 14]. Similarly, this downshifting agrees well with the interaction with n-dopants such as nitrobenzene [6, 7]. From step 2 to step 3, the G and 2D peaks upshift by 3.21 cm⁻¹ and 7.84 cm⁻¹, respectively. This is consistent with the increase of p-type dopants (oxygen/water molecules) as suggested by the ambipolar behavior in figure 1(b). By comparing step 1 and step 4, both the G and 2D peaks have downshifted, which is in agreement with the reduction of Ti adatoms (p-dopants) [14]. From the trends of the G and 2D peaks, the influence of p-dopants is likely to dominate over the n-dopants, which is of a lower doping level as will be discussed later.

Zoom In Zoom Out Reset image size

The intensity ratio between the 2D peak and the G peak (I_2D/I_G) is known to be sensitive to the doping concentration. An increase in the doping concentration for both n- and p-dopants will result in a decrease of I_2D/I_G [5, 6, 18]. From figure 5, the ratio of I_2D/I_G at step 2 is the highest (lowest doping concentration) and step 3 is the lowest (highest doping concentration). These results reaffirm that the n-dopant concentration was lower than the p-dopant concentration, and that the G and 2D peak shifts were dominated by the p-dopants. From step 2 (n-type) to step 3 (ambipolar), the decreases of the I_2D/I_G to 0.80 indicates that the doping concentration increases. As the GFET was left in the atmosphere ambient for this interval, we propose that the increase in doping concentration is largely due to the re-adsorption of the oxygen/water molecules onto the GFET surface because of the diffusion of gaseous species through the Si₃N₄ film. The transition from step 3 (ambipolar) to step 4 (p-type) also results in a decrease of the doping concentration, which could indicate that most of the NH₃ dopants had been removed by the wet etching process of Si₃N₄ passivation. By comparing the GFET at step 1 and step 4, the doping concentration at step 4 is lower, which is in agreement with our previous observation of the D peak. Now we come to address the mobility variation shown in figure 1(c). One may see that the defect levels in steps 2, 3, and 4 are comparable (see D peaks in figure 3), which rules out lattice disorder as the reason for varying mobility. As shown in figure 5, the Raman peak intensity ratio I_2D/I_G is a minimum at step 3 in comparison with those at steps 2 and 4. Increasing doping concentration level will reduce the field-effect mobility because of dopant-induced scattering [3], leading to the higher electron and hole mobilities with ambipolar conduction (step 3).

Zoom In Zoom Out Reset image size

In order to further confirm that the increase in threshold voltage was due to re-adsorption of oxygen/water molecules, 750 nm-thick Si₃N₄ was deposited onto the GFET and the thickness of the Si₃N₄ was reduced step by step while monitoring the shift in threshold voltage (see figure 6). As discussed previously, the GFETs display p-type characteristics after removing the Si₃N₄ passivation. These devices undergo the same Si₃N₄ passivation process, and n-type behavior was restored again when passivated by 750 nm-thick Si₃N₄; demonstrating that the n-type modulation is repeatable on the same GFET. When the Si₃N₄ passivation layer is reduced by less than 200 nm, the GFET displayed p-type conduction with positive threshold voltages. This clearly demonstrates that if the passivation Si₃N₄ is too thin or porous, exposure to atmospheric ambient will results in the re-adsorption of oxygen/water molecules.

Zoom In Zoom Out Reset image size

In order to verify our hypothesis about the role of ammonia, another set of experiments has been conducted as follows. The Si₃N₄ passivation was removed and then subsequently replaced by thick SiO₂ passivation (see supplementary information available at stacks.iop.org/Nano/24/195202/mmedia). However, in the case of SiO₂, no n-type conduction can be observed. The only difference between the two deposition processes (Si₃N₄ and SiO₂) is the distinct reaction gases. That is NH₃ was changed to N₂O for SiO₂ deposition, resulting in the obvious contrast between the conduction behaviors with Si₃N₄ and SiO₂ passivation on the same device. This clearly demonstrates the importance of the NH₃ for the modulation of the GFET from p- to n-type conversion.