Properties of Self-Aligned Short-Channel

Graphene Field-Effect Transistors Based on

Boron-Nitride-Dielectric Encapsulation and Edge
Contacts
EE23BTECH11006 - Ameen Aazam
EE23BTECH11021 - Ganne Gopi Chandu
EE23BTECH11035 - Kunal Vaibhav Thorawade
EE23BTECH11041 - Md Ayaan Ashraf

1 Introduction
Graphene is a single layer of carbon atoms arranged in a two-dimensional honeycomb lattice.
Known for its remarkable properties, it is extremely strong, lightweight, and highly conductive
for both electricity and heat. Its unique structure gives it exceptional mechanical flexibility and
a high surface area, making it a revolutionary material in fields like electronics, energy storage,
and nanotechnology. Discovered in 2004, graphene has since been at the forefront of materials
research due to its potential to transform industries with applications ranging from flexible displays
to advanced batteries and high-speed transistors.

Figure 1: The atomic structure of graphene. Carbon atoms are arranged in a 2D honey-comb
lattice.

— EE23BTECH11041 : Md Ayaan Ashraf

1
 It’s a carbon allotrope which exhibits a regular hexagonal, honeycomb like crystal structure. The
carbon atoms are sp2 hybridized with the pz orbital participating in the π bond. Which forms the
basis of highly conducting two dimensional electronic gas (2DEG) of graphene on both of its sides.
By controlling the electron cloud we can form the basis of GFET. But there is a significant trade off
of using graphene due to it’s different dispersion relation in contrast to conventional semiconducting
materials. It follows a linear E-K relation, rather than the common parabolic ones,

E(k) = ±(h̄)(vf )|k| (1)

Figure 2: Band structure of graphene. (a) Dispersion relation of graphene schematically depicted
in the k-space. (b) The linear energy dispersion around the K-point.

which rises the Dirac cones in the momentum space (The real lattice space can be transformed
into a momentum space via Fourier transform, which will also be hexagonal periodic like in the
real space). And in the Dirac cones, the valence band and conduction band touches each other
causing the zero band gap property of graphene. As it lacks band gap, it allows both electrons
and holes to conduct in the channel under specific bias conditions (ambipolar conduction) and we
can not technically ”turn off” the device and this fact makes graphene a poor candidate for digital
applications (where we look for the best Ion /Iof f ratio, but they can exhibit at best a ratio of 100).

— EE23BTECH11006 : Ameen Aazam

So instead we try exploit its high mobility and conductance properties in analog and high
frequency RF applications, where their transconductance (gm ) and output conductance (gd ) are
advantageous for power amplification. GFETs face challenges with poor dielectric interfaces and
ohmic contacts. Graphene’s electrical properties are highly sensitive to anything in contact with
its surface, and some of the best mobility values have been achieved with graphene suspended in
a vacuum. The I-V characteristics of GFETs, especially those with short channel lengths, often
lack strong saturation. This is primarily due to high contact resistances combined with the weak

2
electrostatic control caused by the thick oxide layers (tox ).

Using hexagonal boron nitride (h-BN) as a dielectric material to encapsulate graphene enables
low-field mobilities of over 1,000,000 cm2 /Vs at room temperature, comparable to the best results
achieved with suspended graphene. Like graphene, h-BN has a hexagonal crystal structure with
strong in-plane bonds and no dangling bonds out of the plane, making it an excellent substrate and
gate dielectric for graphene. Its layered structure also allows for very thin layers to be used as the
gate dielectric, enabling a small effective oxide thickness.

In this review, we present the first short-channel self-aligned GFETs (<100 nm) based on full
h-BN encapsulation of the graphene sample. Source and drain contacts in these devices are made
by forming edge contacts with the graphene, a technique being used for the first time in short-
channel FET structures. Interestingly, this method shows an opposite trend in contact resistance
compared to conventional top contacts, where holes typically have lower resistance than electrons.
As a result, most GFETs are usually set up as p-type devices, but the devices dicussed below are
biased as n-type transistors.

— EE23BTECH11041 : Md Ayaan Ashraf

2 Overview of hBN encapsulated GFET

A relatively thick hexagonal Boron Nitride layer is used as the device substrate. The channel
consists of a monolayer graphene, encapsulated between two hBN layers (hBN is chosen to provide
a smooth surface and similar hexagonal lattice causing less impurities), the bottom one acting as
the gate dielectric, while the top one maintains graphene’s pristine quality. The top gate is is a
metal gate stack deposited above the encapsulated graphene. For the source and drain self aligned
contacts are used.
Self-aligned contacts refer to a fabrication technique where the metal contacts to the exposed
graphene channel are precisely positioned relative to the graphene lattice without the need for
separate alignment steps. This process ensures better carrier injection efficiency between the metal
surface to the graphene channel, lesser scattering and minimal contact resistance and parasitic
capacitance between the metal and the channel interface.

Figure 3: Cross section of GFET in different phases of fabrication

3
 — EE23BTECH11021 : Ganne Gopi Chandu

3 Analysis
3.1 Fitting the IV characteristic : Virtual Source Model
In a typical short channel FET IV characteristic, we have two regions, the linear regime, where the
drift velocity increases linearly with the electric field in the channel and the saturation mode, where
the velocity hits the saturation. But in between them there is a smooth transitioning, which can not
be accounted by merely two piecewise models. We can see that the IV curve can be characterized
by the drift velocity. So if we somehow we have an estimate of the drift velocity as a function of
VDS which sweeps gradually from being proportional to electric field to being constant, we should
be closer to the actual IV characteristic, without the need of piecewise descriptions. As a first
approximation we can go like,
1 1 1
= + (2)
< vx (VDS ) > µn VDS /L vsat
 
VDS /VDSAT
< vx (VDS ) > = vsat (3)
1 + VDS /VDSAT
= FSAT (VDS )vsat (4)

Where VDSAT = vsat L/µn

We can see in the linear regime µn VDS /L is quite lower than vsat , so it dominates and this is
the other way around in the saturation regime. And there is a gradual sweep between these two
regimes. But turns out that this modeling is not sufficient to fit the actual IV characteristic, so we
need an extra parameter , which is plugged in as following,
" #
VDS /VDSAT
FSAT = 1/β
(5)
[1 + (VDS /VDSAT )β ]

The value of β is empirically adjusted to fit the IV characteristic and the value is typically 2±0.1.

So the current (rather current per unit width) can be modeled as,

ID
= (Qx0e + Qx0h )vsat FSAT (6)
W
Where Qx0e and Qx0h are the net charges contributed by electrons and holes respectively, given as,
Z ∞
8π E 1
Qx0e = ·
2 1 + e(E+Ecs )/kB T
dE (7)
(hvF )
Z0 ∞
8π E 1
Qx0h = ·
2 1 + e(E+Ecd )/kB T
dE (8)
0 (hvF )

Ecs and Ecd are the local energy potentials at the virtual source and drain.

— EE23BTECH11006 : Ameen Aazam

4
Figure 4: (a) Scatter plot of the linear relationship between the inverse of the extracted mobility
with the inverse of the gate length for the devices in Fig. 2. This gives a linearly extrapolated
effective mobility and mean-free-path of 13 734 cm2/Vs and 848 nm, respectively. (b) Linear
relationship between the inverse of the injection velocity and gate length for the same devices in
(a). The extrapolation indicates a ballistic velocity of 9.3 × 107 cm/s or 93% of the Fermi velocity.

3.2 Ballistic Transport & Mobility

by ballistic transport, we mean that the electrons in the graphene channel move almost without
any scattering and collision, from the source to the drain. Basically, the mean free path of electrons
in graphene is exceptionally long (850nm here), so for short channel GFETs, the electrons move
streamline without scattering.

The study uses a model to analyze experimental data and determine how mobility and carrier
injection velocity change with the channel length (LG ). It shows that as the channel length shrinks,
mobility decreases due to the onset of ballistic transport, where electrons move without scattering.
The mobility is calculated using an equation that takes into account the mean-free path (λmf p )
and effective mobility (µef f ). From the results, we found a mean-free path of 848 nm and an
effective mobility of 13,734 cm²/Vs. The carrier velocity was also found to be close to graphene’s
Fermi velocity. With the exception of one device with a 1.24 µm channel, all the tested devices
are ballistic, meaning electrons travel without scattering. These graphene devices are able to
demonstrate ballistic transport at channel lengths greater than 0.5 µm, and their mobility is among
the highest reported for graphene-based devices. Notably, the achieved ballistic velocity is more
than five times faster than the velocities observed in traditional CMOS devices.

— EE23BTECH11041 : Md Ayaan Ashraf

3.3 Ceq
We know that the drain current is directly proportional to carrier mobility and carrier velocity
in both long and short-channel MOSFETs. Since graphene has both higher carrier mobility and
velocity than silicon, we would expect its transconductance and output conductance to be higher

5
than those of Si-MOSFETs. However, we observe that the values of transconductance and output
conductance for graphene devices are significantly lower than those of silicon devices at comparable
channel lengths. The reason for this is the effect of Quantum Capacitance and ambipolar
conduction.
Quantum capacitance reflects the ability of materials to accommodate additional carrier charges
in its bands when external voltage is applied. So naturally it depends upon the density of states
which can be given as,

CQ = q 2 D(E) (9)

where:
• q: Electron charge
• D(E): Density of states as a function of energy
• E: Energy level
Now for graphene, the DOS varies linearly with energy going to 0 at the fermi level, which
gives rise to a very small quantum capacitance. So in series with the gate capacitance, the over
all capacitance also becomes very low (lower than conventional silicon based FETs), causing lesser
control over the channel. It can be interpreted in a different way as well. As the energy increases
from the fermi level, the DOS increases from 0, so, changing a little bit of electron dnesity requires
a large gate voltage as it will required required numbers of states are more squed toward higher
energy, referring a lesser control of the gate. And this directly translates to the transconductance
of GFETs, as we’ve seen, for a classical MOSFET, the drain current can be given as,
µn Cox w
ID = (Vov )2 (10)
2 L
here due to the low Cq , the Cox,eq , also lowers,
Cox CQ
Cox,eq = (11)
Cox + CQ
causing an overall lowering of ID and hence gm for the same applied gate overdrive voltage.
However, since CQ (134 micro F/cm2 ) is very large in conventional materials, we can approxi-
mate Cox,eq ≈ Cox when CQ ≫ Cox for almost any oxide thickness. In contrast, for graphene, the
value of CQ is much smaller (on the order of 1 micro F/cm2 ). By observing the following graph:
We can see that the value of Ceq is very low despite the thin oxide thickness (which results in a
greater Cox ). This is due to the low CQ . Since Ceq is directly proportional to gm , we observe that
the value of gm is lower for graphene. This low value of gm affects the performance and usability
of GFETs.

— EE23BTECH11035 : Kunal Vaibhav Thorawade

4 Conclsuion
Short-channel GFETs with an equivalent oxide thickness (EOT) of less than 3.5 nm exhibit some
of the highest mobilities and carrier velocities achieved so far. Ballistic behavior is observed with

6
mean-free paths greater than 800 nm, even under high biases, opening up exciting possibilities for
devices based on ballistic transport over longer distances. For example, devices utilizing electron fo-
cusing can be realized at room temperature in graphene. However, despite the fact that graphene’s
carrier velocities are more than four times greater than those in silicon, the fundamental limita-
tions of graphene—such as its ambipolar conduction and quantum capacitance—result in poorer
transconductance (gm) and output conductance (gds) characteristics compared to silicon CMOS
devices.

— EE23BTECH11021 : Ganne Gopi Channu