Accepted Manuscript
Sharp switching behaviour in graphene nanoribbon p-n junction
Ahmed M.M. Hammam, Marek E. Schmidt, Manoharan Muruganathan, Hiroshi Mizuta
PII: S0008-6223(17)30563-8
DOI: 10.1016/j.carbon.2017.05.097
Reference: CARBON 12073
To appear in: Carbon
Received Date: 27 February 2017
Revised Date: 15 May 2017
Accepted Date: 30 May 2017
Please cite this article as: A.M.M. Hammam, M.E. Schmidt, M. Muruganathan, H. Mizuta,Sharp switching behaviour in graphene nanoribbon p-n junction, Carbon (2017), doi: 10.1016/j.carbon.2017.05.097.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service toour customers we are providing this early version of the manuscript. The manuscript will undergocopyediting, typesetting, and review of the resulting proof before it is published in its final form. Pleasenote that during the production process errors may be discovered which could affect the content, and alllegal disclaimers that apply to the journal pertain.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
1
Sharp switching behavior in graphene
nanoribbonp-njunction
Ahmed M. M. Hammam,a,b,* Marek E. Schmidt,a Manoharan Muruganathana and Hiroshi
Mizutaa
a School of Materials Science, Japan Advanced Institute of Science and Technology, 1-
1Asahidai, Nomi 923-1292, Japan
b Physics Department, Faculty of Science, Minia University, Main Road - Shalaby Land, Minia
11432, Egypt
Abstract 1
The experimental study of interband quantum mechanical tunneling in graphene nanoribbons
is a major step to realizing graphene tunneling-based field effect transistors (TFET). Here, we
report the sharp switching behavior observed in an electrostatically controlled graphene
nanoribbon p-n junction in pn and np biasing. We demonstrate current modulation with a slope
of 42 mV/dec over five order of magnitude in drain current at 5 K when the device is switched
from nn to np configuration. This slope is unaffected by temperature up to 50 K. The suppression
of carrier transmission in the OFF state is attributed to the finite bandgap of the ~15 nm wide
graphene nanoribbon channel. We show that the reported device characteristics can be explained
by band-to-band tunneling through the junction. This work is expected to offer valuable insight
into BTBT in GNRs and be a valuable contribution towards competitive graphene TFETs.
*Corresponding author. Tel: +81 761-51-1573. E-mail: [email protected] (Ahmed M.
M. Hammam)
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
2
Introduction 2
Complementary metal oxide semiconductor (CMOS) devices have continuously been
downscaled for decades, however, the stand-by power governed by the leakage current at their
OFF state is now exceeding the dynamic power at the ON state[1]. In order to overcome this
issue, various types of abrupt switches are currently examined, which are expected to realize a
subthreshold slope (SS) smaller than the theoretical limit of 60 mV/decade at room temperature
for Metal-oxide-semiconductor field effect transistors (MOSFET) and a lower threshold voltage,
leading to a remarkable reduction of the leakage floor. In this direction, tunnel field effect
transistors (TFETs) are currently under serious study as one of the promising candidates for
beyond-Moore electronics [2]. It is worth to mention that a conventional semiconductor TFET is
formed by a fixed p-doped source and n-doped drain region, where the gate is used to tune the
channel potential to realize ON and OFF states. In the ON-state, a p-n junction is formed
between source and channel. Over the past decade, various TFETs based on both group IV and
group III-V semiconductors have been studied. However, these TFETs commonly suffer from
seriously low ON-state current due to the large band gap of the employed semiconductors [2,3].
In contrast, TFET designs based on graphene, which features inherently zero band gap and
massless charge carriers, are expected to achieve a higher tunneling current at the ON-state.
Moreover, superior gate controllability due to the atomically thin two-dimensional (2D) nature of
graphene is expected. Unlike conventional semiconductors, in graphene, doping level and type
can be controlled electrostatically by a perpendicular electric field. Thus, the device performance
can be optimized by modulating the electrostatic gates. Despite its exceptional electrical
properties, graphene with its zero bandgap is handicapped in the field of digital electronics where
a finite bandgap is an essential requirement to control the carrier transport through a device.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
3
However, patterning 2D graphene sheets into very narrow ribbons has been reported to introduce
an energy gap[4–9]. Therefore, Graphene Nano-Ribbons (GNRs) have a significant potential to
enable graphene to be used in digital electronics.
The large number of theoretical studies about graphene TFETs[10–13]currently lack extensive
experimental verification to help identify the most crucial aspects and study the feasibility and
challenges. To the best of our knowledge, experimental demonstration of such kind of abrupt
switches based on band-to-band tunneling in GNRs have not yet been reported, although there
has been related research on graphene p-n junctions that can be divided into two categories:
Based on graphene (i) with energy gap, and (ii) without. In case of zero bandgap graphene p-n
junctions, carrier transport demonstrated a substantial evidence of the existence of the so-called
Klein tunneling (perfect transmission of the electrons in graphene with normal incidence onto a
potential barrier) phenomenon. Therefore it has been concluded that charge carriers cannot be
confined electrostatically in zero bandgap graphene p-n junctions[14–23].
On the other hand, when the graphene in the junction region has a finite energy gap,
suppression of Klein tunneling is theoretically reported [24,25]. Additionally, two important
considerations emerge in GNRs patterned by plasma etching techniques, namely, quantum
confinement (energy gap opening) and the effect of edge states that arises from unavoidable line
edge roughness (LER)[5,6,14,26–29]. Due to this edge disorder, the carrier transport in
semiconducting GNRs is dominated by the Coulomb blockade of multiple quantum dots[5,14]
and quantum confinement[30]. The former was observed by Liu et al.[14] who studied a p-n
junction in 30 nm wide GNR using one top gate and a global back gate for electrostatic
modulation. Müller et al.[31] fabricated a p-i-n junction using 30 nm wide GNR using buried
triple gates and attributed the electronic transport through the junction to band-to-band tunneling
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
4
(BTBT). However, electrical characterization at only one temperature was reported in their study
that focuses on the fabrication process, and the BTBT was not explained in detail.
In this paper, we present a study of the carrier transport through a low biased GNR p-n
junction in a wide range of temperatures extending from 5 to 300 K. Two individual top gates
with spatial separation of ~ 40 nm are used to define the PN junction electrostatically along the
GNR with a width of ~ 15 nm. By individually modulating the potential of the top gates, four
distinct configurations corresponding to nn, pp, pn and np are observed. Carrier transport shows
a remarkable step transition during switching the device configuration from nn to pn (forward
direction) or np (reverse direction), respectively, at lower temperature. Ion/Ioff ratios of 104 to 105
are realized at low temperature. Our detailed analysis shows that, although concurrent
mechanisms influencing the transport through the device, the sharp switching behavior can be
explained by considering BTBT through the p-n junction. The asymmetry between np and pn
biasing is explained based on 3D device simulations that show the effect of the top gate
alignment.
Experimental 3
3.1 Device Fabrication
The schematic of our device is shown in Fig. 1(a), and a scanning electron micrograph in 1(b). It
consists of a narrow GNR at the centre of the channel for p-n junction formation, with wider
graphene acting as the source and drain contacts. Here, a commercial chemical vapour deposition
(CVD) grown single-layer graphene is transferred on top of a SiO2 (90nm)/Si substrate. To
decrease the amount of PMMA residues after the transfer process that can degrade the carrier
mobility [32–34], we put the sample in hot Acetone for 30 min, followed by annealing in
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
5
forming gas for 3hrs at 250°C. Electron beam lithography (EBL) followed by e-beam
evaporation of 5/35 nm of Cr/Au and the lift-off technique are used to fabricate source drain
electrodes. High-resolution, negative tone resist, hydrogen silsesquioxane (HSQ), which is
converted into SiO2 after e-beam exposure, is used with a thickness of 35 nm to form the gate
dielectric and, at the same time, a hard mask. The narrow GNR is achieved by the reactive ion
etching (RIE) technique. It is worth mentioning that we did a second annealing (identical
conditions) just before coating the sample with HSQ to decrease the PMMA residues from the
previous fabrication steps. A part of the ~ 22 nm wide and 1µm long continuous HSQ hard mask
is shown in the inset of Fig. 1(b). Before the top gate fabrication, an additional passivation layer
of SiO2 was deposited by electron beam evaporation (10 nm), effectively reducing the gate-
channel leakage. Finally, a high-resolution positive tone resists (SML-100) was used to fabricate
the two Cr/Au (2/10 nm) top gates with spatial separation of 40 nm by the previously used EBL
lift-off technique. In the fabricated p-n junction device shown in Fig. 1 (b) the misalignment
between the top gate and the GNR of ∆X ≈ 325 nm is caused by the accuracy limit of the
lithography alignment process.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
6
Fig. 1 (a) Schematic illustration of our GNR p-n junction device. (b) Scanning electron
micrograph of the measured device with false colouring for better visibility. Insets show the
separation of ~ 40 nm between the top gates above the GNR, and the HSQ etching mask of ~22
nm width. ∆X is the misalignment of the top gates in respect to the GNR. (c) Ambipolar
characteristics at 300 K for Vd = 5 mV measured for individual gates. (d) Arrhenius plot used to
extract the energy gap Eg = 44 meV.
3.2 Characterization
A semiconductor device analyser (Agilent–B1500A) was used for all electrical measurements
performed under vacuum (~ 10-4 Pa) in a variable temperature helium closed cycle cryocooler
prober system. The base temperature of the system is T ≈ 4.8 K. To confirm the functionality of
each gate and the performance of the GNR, the drain current was measured as a function of the
top and back gate voltages. The room temperature measurements shown in Fig. 1 (c) demonstrate
the individual gate potential tunability that is important to achieve electrostatically controlled
individual p and n regions in the GNR device. In all gate modulations, it was observed that the
charge neutrality point (CNP) is located close to 0 V, demonstrating the nearly intrinsic
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
7
characteristics of the GNR. This characteristic is important as it minimizes the required gate
voltage offsets and offers symmetric tuning.
Results 4
As mentioned earlier, the width of the HSQ hard mask is ≈ 22 nm as shown in Fig. 1 (b),
however, the actual width of the patterned GNR is narrower due to the isotropic etching
component of plasma RIE [4,28,35]. From the linear high-temperature region of the Arrhenius
plot shown in Fig. 1 (d) , an energy gap (Eg) of ~ 44 meV was extracted for the fabricated GNR.
Using the empirical model W / ( is the reduced Plank constant, W is the GNR
width and = 106 m/s is the Fermi velocity)[8], the effective GNR width is thus estimated to be
~ 15 nm.
The four possible configurations of the p-n junction (nn, pp, pn, np) are realized by
modulating the top gates independently while applying a drain bias Vd. For this, one of the top
gates is swept from -5V to +5V while the second gate is stepped in the same voltage range (-5V
to +5V). Note that, throughout this paper, the doping configuration is indicated by two letters
(such as np), where the first letter denotes the doping below Tg1 (positive voltage for n-doping,
negative voltage for p-doping), and the second letter denotes the doping below Tg2. Fig. 2 shows
the characteristic contour plot of drain current (Id) as a function of the two top gate voltages at 10
K (plots for other temperatures can be found in the supplementary information). Vd is fixed at 3
mV for all measurements at 25 K and above, and 12 mV below due to the high current
suppression at lower temperatures. As can be seen in Fig. 2, the low current region due to Tg2
modulation (transport gap) is wider than the low current region due to Tg1 modulation. This
difference is ascribed to the slight misalignment between the two top gates and the GNR in the
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
8
fabricated device as shown in Fig. 1(b). Tg1 covers a larger part of the wide graphene contact
region.
Fig. 2 Contour plot of drain current as function of the top gate voltages at T = 10 K. Drain
voltage Vd is 12 mV. The clearly defined doping regions induced by the electrostatically doping
are indicated by np, nn, pn, and pp, respectively. The three dashed lines indicate the location of
the profiles in Fig. 3(b-d).
The drain current (Id) at 10 K is shown for different gating configurations in Fig. 3. When
using the global back gate (Fig. 3 (a)), a symmetric modulation is observed with some random
oscillation around the CNP at Vg = -0.5 V. The symmetric modulation is consistent with the 300
K measurement shown in Fig. 1(c). When modulating by the synchronized top gates (along the
black dotted line in Fig. 2), the symmetric characteristics are observed, as well (Fig. 3 (b)),
however, some differences exist compared to the back gate modulation. Firstly, the ON current
(VTg = +/- 5 V) is higher. As the dielectric layer between the top gates and the graphene is
significantly thinner than between the substrate and graphene, the modulation at a given voltage
is stronger with a higher doping level. The higher OFF current is due to the charge neutrality
points (CNP) having a slight offset for the two top gates (see Fig.1 (c)), and the device cannot be
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
9
effectively turned off for synchronized top gate voltages (i.e. while the GNR underneath one gate
is completely turned off, the GNR under the other gate is still slightly ON). In contrast to these
two cases, when fixing one of the top gates and sweeping the other one, a remarkable step
change in the drain current is observed. For fixed VTg1 = 4.9 V (Fig. 3(c), corresponding to np
biasing), the step change with a rate of ~42 mV/dec over a range five order of magnitude in drain
current (5x105) is observed at VTg2 ≈ -1 V. The OFF state current is at least one order of
magnitude lower than for the back gate modulated current. In case of a fixed VTg2 = 4.9 V (Fig.
3(d), pn biasing), the rate of the step change at VTg1 ≈ -0.8 V is ~32 mV/dec but only over a
range of four order of magnitude in drain current (~1.3x104). Furthermore, the current is only
very slightly affected by the VTg1 outside of the suppression region.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
10
Fig. 3 Drain current as a function of gate voltage at T = 10K and Vd = 10 mV: (a) Back gate
voltage sweep. (b) Simultaneous Tg1 and Tg2 sweep. The GNR is modulated from nn
configuration to pp configuration. (c) Tg2 sweep at VTg1 = 4.9 V. The device is switched from nn
configuration to pn configuration by sweeping VTg2 from 5 V to -5 V (forward biasing). Inset
shows the transition region with sharp slope of 42 mV/decade over five orders of magnitude. (d)
VTg1 sweep at VTg2 = 4.9 V (reverse biasing) showing similar characteristics with 32 mV/decade
in the sharp switching region over four orders of magnitude.
Similar characteristics are observed at 100 K and slightly changed bias conditions, as well
(compare Fig. 4), however, two observations are made that require further elaboration. Firstly,
the slope of the abrupt current change decreases with increasing temperature. Secondly, there is a
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
11
clear difference between np and pn biasing, both in terms of the modulated gate voltage, but also
in terms of the fixed gate voltage. For the np biasing (Fig. 4(a)), the current is unaffected by VTg1
(from 2 to 3.9 V). In clear contrast, the current is strongly affected by VTg2 in the pn biasing, and
shows a constant current beyond the sharp switching (Fig. 4(b)). At 10 K this effect is observable
as well for the np biasing (see supplementary information) but strong current fluctuation and the
wide transport gap due to Tg2 masks this effect for the pn biasing. Although the source-drain
biasing direction is reversed in respect to the doping configuration for these two cases, such an
asymmetry is not expected and will be discussed in the next section.
Fig. 4 Device characteristics at 100 K for np and pn biasing conditions. (a) Drain current vs VTg2
(Vd = 3 mV and variable VTg1). (b) Drain current vs VTg1 (Vd = 3 mV and variable VTg2).
Discussion 5
5.1 Asymmetry of forward/reverse biasing
To understand the origin of the observed asymmetry between np and pn biasing, we have
performed 3D technology computer-aided design (TCAD) simulations. We have mentioned
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
12
earlier that the top gates have a misalignment of ∆X ≈ 325 nm as shown in Fig. 1(b). Thus, along
the GNR channel from source to drain, there are two areas not fully modulated by the top gates,
and a non-uniform modulation within the top gate regions are expected. In Fig. 5(a), the
perspective representation of the used 3D model is shown that reproduces the fabricated device.
All dielectrics (SiO2) are hidden for better visibility. Here we use Silvaco Atlas device
simulation framework [36], and the graphene is modelled as a 5 nm thin polysilicon layer with 1
µOhm.cm resistivity, dielectric relative permittivity of 3, and a balanced donor and acceptor
doping of 1019 cm-3 in order to reproduce the centered Fermi level of unmodulated graphene. The
energy gap of the narrow part is set to 44 meV, while this value is 0 meV for the remaining
graphene. The energy of the conduction and valence band of the graphene channel is affected by
the potential distribution between the top gates and the graphene. For the simulation, Vd is 10
mV. Fig. 5 (b) is the cross section through the simulated 3D model (cut parallel to yz plane
through the centre of the GNR at x=0) showing the potential distribution in the gate dielectric
with VTg1 -1 V and VTg2 4 V. In the p-n junction region, a continuous transition from negative to
positive potential is observed without intrinsic region. The potential gradient is directly affected
by the geometry (i.e. the top gate separation and the thickness of the HSQ and SiO2 layers).
Furthermore, in the regions inside of Tg1 and Tg2, it is clearly visible that the electrostatic
modulation of the GNR is reduced, but not eliminated. The band energies along the graphene
channel (x = 0 nm) are extracted as shown in Fig. 5(c) for the np, and in Fig. 5(d) for the pn
biasing condition for different variable gate voltages (the fixed gate is at 4 V).
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
13
Fig. 5 TCAD simulation results. (a) Perspective view of 3D model representation of device
shown in Figure 1b used for Silvaco TCAD 3D simulation. All SiO2 elements are hidden for
better visibility. (b) Cut plane parallel to the yz plane through the center of the GNR (x=0)
showing potential distribution induced by VTg1 = -1 V, VTg2 = 4 V and Id = 10 mV. The potential
varies with the position below the top gates due to the partially unmodulated GNR. (c+d)
Extracted band structure at center of GNR in the pn biasing configuration (VTg2 = 4 V) and np
biasing (VTg1 = 4 V). When sweeping the variable gate from 4 V to -4 V, initially a potential
barrier is present. Then, when the valence band below the variable gate is lifted above the
conduction band of the fixed gate, a relatively narrow tunnelling barrier is formed that allows the
rapid increase of current. The bias window is indicated by dashed lines. (e) Close-up of the PN
junction, highlighting the formation of the tunnelling barrier with length less than 40 nm.
For the nn configuration (4/4 V), a continuous doping along the channel is realized that allows
the high current. However, two potential bumps caused by the unmodulated GNR region are
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
14
present. As VTg1 is reduced to 0.4 V, a wide potential barrier is formed in the Tg1 region,
effectively blocking the current through the channel (compare Fig. 3(c)). At this point, the
thermally activated leakage current over the potential barrier is governing the Ioff level. Note that,
due to the edge roughness and the resulting potential variation in the band structure, multiple
quantum dot effect (Coulomb oscillations) is observed for all curves in Fig. 3 [27,37], and will
be discussed in more details later. Then, as VTg1 is further reduced to -0.8 V, the potential barrier
is further heightened and a decrease of current would be expected. However, as the valence band
edge on the Tg1-side of the junction is shifted above the energy of the conduction band edge on
the Tg2-side, a narrow barrier with the width WT is formed at the p-n junction inside the bias
window that is defined as the energy difference between the Fermi level in the source and drain
(EFS - EFD). This is highlighted in Fig. 5(e) where the junction region is shown for different pn
biasing conditions. For such a band structure with the narrow spacing (~ 40 nm) between Tg1 and
Tg2, the significant band-to-band tunnelling (BTBT) current through various mechanisms, such
as thermally assisted tunneling and hoping through trap states is expected [38]. The sharp
switching apparent in Fig. 3(c) and (d) occurs immediately after the current minimum is
observed. The 3D TCAD simulation results show that the valence band is shifted above the
conduction band at around 0.4 V of the variable gate, however, the current minima in Fig. 3(c)
and (d) are at gate voltages of ~ -1 V and -0.8 V, respectively. This, in fact, is consistent with the
simulation where the tunnelling window, which is defined as the energy difference between the
valence band edge in the p(n)-doped region and the conduction band edge in the n(p)-doped
region, is shifted into the bias window at those voltages.
When the gate voltages are further reduced, the previously described effect continues, namely
the potential barrier becomes higher and the tunnelling region is increased. The width WT of the
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
15
tunnel barrier is also decreased, but this change is relatively limited. Nevertheless, the current
level reached is comparable to the nn configuration in both cases, however, current saturation is
observed for pn biasing but not for np biasing. The 3D TCAD simulation shows that the
tunnelling barrier is identical for both configuration, thus a different reason has to exist.
The small potential bumps inside of the top gates show a clear, but weaker, modulation
compared to the fully covered areas (see red and blue dots in Fig. 5(c) and (d)). These bumps are
very similar, however, due to the source-drain biasing condition, the bump band energies in the
Tg1 region are up-shifted compared to Tg2. To help understand the consequences of this, the
different effects governing the transport of a charge carrier from source to drain in the pn
configuration will be described first. Most of the carrier at low temperature (depending on the
thermal distribution in the source region, charge carriers can overcome the source/Tg1 potential
bump and contribute to the thermally activated leakage current) travel along the p-doped channel
until they encounter the potential bump within Tg1. This bump will pose some resistance to the
carriers, but due to the thermal activation (we did not measure the device below 5 K) and edge-
roughness mediated transport [6] the barrier can be overcome. The charge carrier again enters the
p-doped region until it tunnels into the conduction band through the tunnelling barrier. To reach
the drain electrode, only the potential bump in the Tg2 region (which is fixed in this
configuration) is of importance. The charge carrier transport in the np configuration is very
similar. However, in the latter case, the fixed bump in the n-type region is in Tg1, and the BTBT
is from conduction to valence band.
The curves in Fig. 4 for 100 K show that, in the np configuration, the current level is
unaffected by VTg1 and keeps increasing after the sharp switching. In contrast, for the pn
configuration, the current level after the sharp switching is approximately constant, and strongly
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
16
depends on VTg2. Based on the device design and 3D TCAD results, such a characteristic is only
possible if the transport from source to drain is governed by the Tg2 region, namely, the transport
through the Tg2 region has the highest resistance and this resistance changes with VTg2. The
potential bump inside Tg2 has a lower energy at a given voltage, thus, for p-doping, the potential
barrier is higher and thermal activation is required to overcome it. These 3D TCAD suggest that
BTBT occurs at the p-n junction, however, the saturation current after the sharp switching
behaviour is dominated by the potential bump and density of states (DOS) below Tg2 for pn and
np biasing configurations.
5.2 Origin of sharp switching
The sharp switching was observed in the np and pn biasing condition at low temperature (Fig.
3(c) and (d)), when one of the gates is biased so that the valence band edge in the variable gate
region is lifted into the bias window and at the same time above the conduction band edge in the
fixed gate region. This is evident from the 3D TCAD simulation results in the previous section.
We can thus sketch the simplified band structure of the p-n junction as shown in Fig. 6(a). In the
uniformly nn-doped state, a large drift current is enabled. As the energy of the band edges in the
Tg2 region are shifted up, the drift current is reduced and a thermally activated leakage current
dominates the transport. Finally, as the band overlap occurs, the tunnelling width WT is abruptly
reduced. As the tunnelling window is formed in the bias window (between EFD and EFS), the
large number of available charge carriers enables the increase of the drain current over five
orders of magnitude with the sharp slope. As the tunnelling energy window is further increased
(VTg2 is further decreased), the sharp switching saturates because the transport through the
remaining parts of the channel limit the allowed current.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
17
5.3 Quantum confinement and Coulomb oscillation
The Id/VTg data at low temperature shows significant peaks of comparable amplitude to the
sharp switching highlighted in Fig. (c) and (d). Due to the EBL-RIE GNR formation method
used in this work, we have to consider the effect of edge disorder that is unavoidable issue of it.
The carrier transport in GNR based devices is mainly affected by the multiple quantum dots,
Anderson localized states, and the disorder potential [5,6,14,26–28]. For short GNRs, the
multiple quantum dots effect has a stronger effect compared to the Anderson localized states.
The effect of multiple quantum dots is illustrated in Fig. 6(b), where the band structure of a GNR
with edge disorder is schematically sketched. Varying doping levels, (location of bands in
respect to the Fermi level) can cause the formation of isolated quantum dots along the GNR (Fig.
6 (b1)). Charge carriers have to overcome these potential barriers between the dots, greatly
influencing the current. The shape and number of the dots can change significantly with the
slight change of the Fermi level (such as change of gate voltage, compare Fig. 6(b2)), forming
continuous charge puddles. At higher doping levels (higher gate voltages), this effect is strongly
suppressed and the current at higher doping levels in Fig. 3 is smooth. Also, the effect of the
quantum confinement is reduced by thermal smearing, which is apparent when comparing Fig. 3
and Fig. 4. The current spikes in the low-current region sometimes exceed two orders of
magnitude, and thus we want to discuss if the reported sharp switching over five order or
magnitude might in fact be due to coincidental oscillation. For this purpose, the Id/Vd curves for
the p-n junction device at different top gate biasing conditions and temperatures is plotted in Fig.
6(c), showing a near-ohmic behaviour for the np configuration. For the intrinsic state (Tg1 and
Tg2 floating), the current level is more than two orders of magnitude lower and exhibit clear non-
linearity (see inset of Fig. 6(c)) which is evidence of a clear Coulomb blockade effect.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
18
Fig. 6 (a) Proposed band structure model of p-n junction. (a1) In the nn configuration (VTg2 = 4
V), continuous doping leads to high current. (a2) In the ni configuration, only low thermally
activated leakage current is observed together with coulomb oscillation. (a3) When the valence
band below Tg2 crosses above the conduction band below Tg1, band-to-band tunnelling leads to
sharp increase of current. (b) Schematic illustration of quantum dot formation in GNR with edge
irregularities. The size of charge puddles varies strongly at low doping levels. (c) Id/Vd at various
temperatures and doping configurations. In the intrinsic state, clear Coulomb blockade is
observed. For the np configuration, near-ohmic characteristics indicate existence of band-to-band
transport mechanism.
5.4 Temperature dependence of switching slope
The switching slope as function of temperature for the forward biasing condition has been
extracted for various temperatures and is shown in Fig. 7(a). A remarkable independence on
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
19
temperature is observed up to 25 K. It is worth to mention here that the SS degradation by
temperature is also observed in conventional semiconductor,[39–41] however, it follows a
continuous increase governed by the thermal activation energy, and the remarkable temperature
independence is not expected. By considering the BTBT phenomena, the transmission
probability can be expressed by the WKB approximation [2]:
, where WT is the screening tunnelling length, m* is the effective mass, Eg is the energy gap,
∆Ф is the tunnelling window (i.e. the energy difference between the valence and conduction
band right and left of the tunnelling barrier EVD-ECS) and ħ is the reduced Planck constant.
Hence, the SS becomes[42]:
As can be seen from Eq. (2), the SS modelled by the WKB approximation does not depend on
temperature. By using the experimentally determined SS at low temperature, the energy gap of
44 meV and m* = 0.05 m0 [9], we calculate a tunnelling length of 8.3 nm. This value is lower
than the estimated 20 nm from the 3D TCAD simulation results shown in Fig. 5(e). This
disagreement between Eq. (2) and the 3D TCAD simulation might be due to the LER. If a
variation of the tunnelling length occurs at the junction between the p-side and n-side, the
effective tunnelling length at the junction is reduced [43]. For temperatures above 25 K, the
experimental data shows a noticeable deviation. It should be noted that the SS due to BTBT
≈ − √∗ !"ΔФ% &'
(( ≈ )*'+ !"ΔФ%√∗ , &
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
20
cannot be observed directly, instead, it is derived from the transport behaviour of the complete
device including irregularities, source/drain leads, OFF-state current and thermally activated
leakage current. The suppression of the latter is greatly reduced at higher temperature [7,11,44]
when the thermal energy becomes comparable to the energy gap. In fact, in our p-n junction we
observe a rapid increase of the off state current by temperature. Thus, when a different effect
than the increase in tunnelling current becomes more dominant, or the change of current due to
BTBT is relatively small compared to other leakage paths, the SS degradation is predicted and
the measured characteristics do not follow the model.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
21
Fig. 7 p-n (a) Temperature dependence of switching slope and modelling results. A good
agreement is found for WT = 8.3 nm. (b+c) Device resistance as function of temperature in (b) np
configuration (VTg1 = 4 V, VTg2 = -4 V) and (c) nn configuration (VTg1 = VTg2 = 4 V) showing
distinctively different characteristics. (d) Sketch schematically illustrating influence of device
temperature on current contributions. At higher temperature, the thermally activated leakage
contribution increases (above ECD).
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
22
The device resistance as function of temperature is shown in Fig. 7(b) for the np and in Fig.
7(c) for the nn configuration. The drain voltage is 12 mV (3 mV) at low (high) temperature as
mentioned earlier. The resistance increases in both cases at low temperature, which is explained
by the multiple quantum dots effect. For the nn configuration, the curve follows the general trend
of semiconductors, where the increase of temperature is associated with an increase of the charge
carrier concentration, which leads to a decrease in the resistance before phonon scattering leads
to an increase of resistance at increasing temperature. In stark contrast, the resistance in the np
configuration has a minimum at around 100 K, before increasing significantly with temperature.
The origin of this increase is illustrated in Fig. 7(d), where the band structure of the junction in
the np configuration is schematically sketched together with the source Fermi-Dirac distribution
at low and high temperature, respectively. At low temperature, a larger number of carriers is
available in the BTBT energy window, and the thermally activated leakage current is suppressed
(due to the energy gap and low concentration above ECD). With increasing temperature, however,
the FD distribution smears out in a way that the total charge contribution for transport is reduced
(suppression due to energy gap below Tg2 is enhanced). The FD distribution of the carrier within
the bias window changes considerably with temperature. Therefore, the tunnelling current
decreases by increasing the temperature [7]. Finally, above 250 K, the device resistance
decreases again (see Fig. 7(b)). This is due to the enhanced thermally activated leakage current.
Note that the absolute resistance at 300 K of ~300 kΩ in the np configuration is still considerably
higher than in the nn configuration (~160 kΩ), signifying the presence of the potential barrier.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
23
Conclusion 6
The sharp switching behaviour observed in a graphene p-n junction with a finite bandgap of
~44 meV shows good agreement with a device model based on band-to-band tunnelling. When
the channel is modulated continuously by the back or synchronized top gates, the sharp
switching is not observed due to the lack of BTBT. The asymmetry between np and pn
configuration is explained as a consequence of the top gate misalignment. Device simulations
shows that potential bumps are formed in the unmodulated channel regions, which affect the bias
window differently. The current noise in the OFF-state is caused by the edge roughness induced
by the reactive ion etching technique to pattern the GNR. Finally, we show the temperature
dependence of the switching slope and the device resistance, highlighting that the observed
device characteristics can only be explained by BTBT.
Acknowledgements 7
This work is supported by the Center Of Innovation (COI) program of the Japan Science
Technology Agency. The authors are very grateful to Shunri Oda and Michiharu Tabe for the
fruitful technical discussions. Ahmed M. M. Hammam acknowledges the financial support from
the Egyptian Ministry of Higher Education (Culture Affairs and Missions Sector).
References 8
[1] R. Puri, L. Stok, S. Bhattacharya, Keeping hot chips cool, in: Proc. 42nd Des. Autom. Conf.
2005, 2005: pp. 285–288. doi:10.1145/1065579.1065653.
[2] A.M. Ionescu, H. Riel, Tunnel field-effect transistors as energy-efficient electronic
switches, Nature. 479 (2011) 329–337. doi:10.1038/nature10679.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
24
[3] H. Lu, A. Seabaugh, Tunnel Field-Effect Transistors: State-of-the-Art, IEEE J. Electron
Devices Soc. 2 (2014) 44–49. doi:10.1109/JEDS.2014.2326622.
[4] M.Y. Han, B. Özyilmaz, Y. Zhang, P. Kim, Energy Band-Gap Engineering of Graphene
Nanoribbons, Phys. Rev. Lett. 98 (2007) 206805. doi:10.1103/PhysRevLett.98.206805.
[5] F. Sols, F. Guinea, A.H.C. Neto, Coulomb Blockade in Graphene Nanoribbons, Phys. Rev.
Lett. 99 (2007) 166803. doi:10.1103/PhysRevLett.99.166803.
[6] M. Evaldsson, I.V. Zozoulenko, H. Xu, T. Heinzel, Edge-disorder-induced Anderson
localization and conduction gap in graphene nanoribbons, Phys. Rev. B. 78 (2008) 161407.
doi:10.1103/PhysRevB.78.161407.
[7] D. Jena, T. Fang, Q. Zhang, H. Xing, Zener tunneling in semiconducting nanotube and
graphene nanoribbon p−n junctions, Appl. Phys. Lett. 93 (2008) 112106.
doi:10.1063/1.2983744.
[8] C. Stampfer, E. Schurtenberger, F. Molitor, J. Güttinger, T. Ihn, K. Ensslin, Tunable
Graphene Single Electron Transistor, Nano Lett. 8 (2008) 2378–2383.
doi:10.1021/nl801225h.
[9] G. Liang, N. Neophytou, D.E. Nikonov, M.S. Lundstrom, Performance Projections for
Ballistic Graphene Nanoribbon Field-Effect Transistors, IEEE Trans. Electron Devices. 54
(2007) 677–682. doi:10.1109/TED.2007.891872.
[10] J. Knoch, S. Mantl, J. Appenzeller, Impact of the dimensionality on the performance of
tunneling FETs: Bulk versus one-dimensional devices, Solid-State Electron. 51 (2007) 572–
578. doi:10.1016/j.sse.2007.02.001.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
25
[11] Q. Zhang, T. Fang, H. Xing, A. Seabaugh, D. Jena, Graphene Nanoribbon Tunnel
Transistors, IEEE Electron Device Lett. 29 (2008) 1344–1346.
doi:10.1109/LED.2008.2005650.
[12] Y. Yoon, S. Salahuddin, Inverse temperature dependence of subthreshold slope in graphene
nanoribbon tunneling transistors, Appl. Phys. Lett. 96 (2010) 013510.
doi:10.1063/1.3280379.
[13] N. Ma, D. Jena, Interband tunneling in two-dimensional crystal semiconductors, Appl.
Phys. Lett. 102 (2013) 132102. doi:10.1063/1.4799498.
[14] X. Liu, Electrostatic confinement of electrons in graphene nanoribbons, Phys. Rev. B. 80
(2009). doi:10.1103/PhysRevB.80.121407.
[15] R.V. Gorbachev, A.S. Mayorov, A.K. Savchenko, D.W. Horsell, F. Guinea, Conductance
of p-n-p Graphene Structures with “Air-Bridge” Top Gates, Nano Lett. 8 (2008) 1995–
1999. doi:10.1021/nl801059v.
[16] A. Laitinen, G.S. Paraoanu, M. Oksanen, M.F. Craciun, S. Russo, E. Sonin, P. Hakonen,
Contact doping, Klein tunneling, and asymmetry of shot noise in suspended graphene,
ArXiv150204330 Cond-Mat. (2015). http://arxiv.org/abs/1502.04330 (accessed February
22, 2016).
[17] J. Liu, S. Safavi-Naeini, D. Ban, Fabrication and measurement of graphene p #8211;n
junction with two top gates, Electron. Lett. 50 (2014) 1724–1726.
doi:10.1049/el.2014.3061.
[18] N. Stander, B. Huard, D. Goldhaber-Gordon, Evidence for Klein Tunneling in Graphene
$p\mathrm\text-n$ Junctions, Phys. Rev. Lett. 102 (2009) 026807.
doi:10.1103/PhysRevLett.102.026807.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
26
[19] T. Low, S. Hong, J. Appenzeller, S. Datta, M.S. Lundstrom, Conductance Asymmetry of
Graphene p-n Junction, IEEE Trans. Electron Devices. 56 (2009) 1292–1299.
doi:10.1109/TED.2009.2017646.
[20] S. Sutar, E.S. Comfort, J. Liu, T. Taniguchi, K. Watanabe, J.U. Lee, Angle-Dependent
Carrier Transmission in Graphene p–n Junctions, Nano Lett. 12 (2012) 4460–4464.
doi:10.1021/nl3011897.
[21] J. Baringhaus, A. Stöhr, S. Forti, U. Starke, C. Tegenkamp, Ballistic bipolar junctions in
chemically gated graphene ribbons, Sci. Rep. 5 (2015) 9955. doi:10.1038/srep09955.
[22] M.I. Katsnelson, K.S. Novoselov, A.K. Geim, Chiral tunnelling and the Klein paradox
in graphene, Nat. Phys. 2 (2006) 620–625. doi:10.1038/nphys384.
[23] B. Huard, J.A. Sulpizio, N. Stander, K. Todd, B. Yang, D. Goldhaber-Gordon, Transport
Measurements Across a Tunable Potential Barrier in Graphene, Phys. Rev. Lett. 98 (2007)
236803. doi:10.1103/PhysRevLett.98.236803.
[24] X. Xu-Guang, Z. Chao, X. Gong-Jie, C. Jun-Cheng, Electron tunneling in single layer
graphene with an energy gap, Chin. Phys. B. 20 (2011) 027201. doi:10.1088/1674-
1056/20/2/027201.
[25] M.R. Setare, D. Jahani, Electronic transmission through p–n and n–p–n junctions of
graphene, J. Phys. Condens. Matter. 22 (2010) 245503. doi:10.1088/0953-
8984/22/24/245503.
[26] J. Güttinger, F. Molitor, C. Stampfer, S. Schnez, A. Jacobsen, S. Dröscher, T. Ihn, K.
Ensslin, Transport through graphene quantum dots, Rep. Prog. Phys. 75 (2012) 126502.
doi:10.1088/0034-4885/75/12/126502.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
27
[27] P. Gallagher, K. Todd, D. Goldhaber-Gordon, Disorder-induced gap behavior in graphene
nanoribbons, Phys. Rev. B. 81 (2010) 115409. doi:10.1103/PhysRevB.81.115409.
[28] M.Y. Han, J.C. Brant, P. Kim, Electron Transport in Disordered Graphene Nanoribbons,
Phys. Rev. Lett. 104 (2010) 056801. doi:10.1103/PhysRevLett.104.056801.
[29] M. Manoharan, H. Mizuta, Edge irregularities in extremely down-scaled graphene
nanoribbon devices: role of channel width, Mater. Res. Express. 1 (2014) 045605.
doi:10.1088/2053-1591/1/4/045605.
[30] Y.-M. Lin, V. Perebeinos, Z. Chen, P. Avouris, Electrical observation of subband formation
in graphene nanoribbons, Phys. Rev. B. 78 (2008) 161409.
doi:10.1103/PhysRevB.78.161409.
[31] M.R. Müller, A. Gumprich, F. Schütte, K. Kallis, U. Künzelmann, S. Engels, C. Stampfer,
N. Wilck, J. Knoch, Buried triple-gate structures for advanced field-effect transistor
devices, Microelectron. Eng. 119 (2014) 95–99. doi:10.1016/j.mee.2014.02.001.
[32] A. Pirkle, J. Chan, A. Venugopal, D. Hinojos, C.W. Magnuson, S. McDonnell, L. Colombo,
E.M. Vogel, R.S. Ruoff, R.M. Wallace, The effect of chemical residues on the physical and
electrical properties of chemical vapor deposited graphene transferred to SiO2, Appl. Phys.
Lett. 99 (2011) 122108. doi:10.1063/1.3643444.
[33] Y.-C. Lin, C.-C. Lu, C.-H. Yeh, C. Jin, K. Suenaga, P.-W. Chiu, Graphene Annealing: How
Clean Can It Be?, Nano Lett. 12 (2012) 414–419. doi:10.1021/nl203733r.
[34] T. Iwasaki, M. Muruganathan, M.E. Schmidt, H. Mizuta, Partial hydrogenation induced
interaction in a graphene–SiO2 interface: irreversible modulation of device characteristics,
Nanoscale. 9 (2017) 1662–1669. doi:10.1039/C6NR08117G.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
28
[35] J. Sun, T. Iwasaki, M. Muruganathan, H. Mizuta, Lateral plasma etching enhanced on/off
ratio in graphene nanoribbon field-effect transistor, Appl. Phys. Lett. 106 (2015) 033509.
doi:10.1063/1.4906609.
[36] Silvaco Atlas—Device Simulation Framework, (n.d.).
http://www.silvaco.com/products/tcad/device_simulation/atlas/atlas.html (accessed
February 8, 2017).
[37] F. Molitor, C. Stampfer, J. Güttinger, A. Jacobsen, T. Ihn, K. Ensslin, Energy and transport
gaps in etched graphene nanoribbons, Semicond. Sci. Technol. 25 (2010) 034002.
doi:10.1088/0268-1242/25/3/034002.
[38] R. Nouchi, Contact resistance at planar metal contacts on bilayer graphene and effects of
molecular insertion layers, Nanotechnology. 28 (2017) 134003. doi:10.1088/1361-
6528/aa5ec2.
[39] R.J. Van Overstraeten, R.P. Mertens, Heavy doping effects in silicon, Solid-State Electron.
30 (1987) 1077–1087. doi:10.1016/0038-1101(87)90070-0.
[40] A. Schenk, A model for the field and temperature dependence of Shockley-Read-Hall
lifetimes in silicon, Solid-State Electron. 35 (1992) 1585–1596. doi:10.1016/0038-
1101(92)90184-E.
[41] M.J. Kerr, A. Cuevas, General parameterization of Auger recombination in crystalline
silicon, J. Appl. Phys. 91 (2002) 2473–2480. doi:10.1063/1.1432476.
[42] J. Knoch, S. Mantl, J. Appenzeller, Impact of the dimensionality on the performance of
tunneling FETs: Bulk versus one-dimensional devices, Solid-State Electron. 51 (2007) 572–
578. doi:10.1016/j.sse.2007.02.001.
MANUSCRIP
T
ACCEPTED
ACCEPTED MANUSCRIPT
29
[43] M. Luisier, G. Klimeck, Performance analysis of statistical samples of graphene nanoribbon
tunneling transistors with line edge roughness, Appl. Phys. Lett. 94 (2009) 223505.
doi:10.1063/1.3140505.
[44] Z. Chen, Y.-M. Lin, M.J. Rooks, P. Avouris, Graphene nano-ribbon electronics, Phys. E
Low-Dimens. Syst. Nanostructures. 40 (2007) 228–232. doi:10.1016/j.physe.2007.06.020.