Graphene

Graphene is a single layer of carbon chemical atoms arranged in a honeycomb lattice. A nearby gate can select predominantly Electron or electron hole conduction, as demonstrated by the graphene exfoliation and field-effect experiments.^[ref]

A useful question is how geometry changes motion. The nearest-neighbour model of graphene starts with three equivalent bonds from each site and two sites per primitive cell. Interference between those three hopping paths makes two energy bands meet; near their meeting, energy grows linearly with distance in wave vector. The resulting Dirac cones explain why an atomic sheet can imitate one aspect of a relativistic particle while remaining an ordinary material with contacts, defects and boundaries.

Take a planar honeycomb lattice with two sites $A,B$ per primitive cell and one localized atomic orbital per site. In this explicitly idealized tight-binding model, all three nearest-neighbour hopping matrix elements are $-t$, with $t>0$; there is no on-site offset or further-neighbour hopping. Ignore Electron interactions and treat physical spin as a separate degeneracy. Let the bond length be $b$, and choose the three vectors from an $A$ site to its neighbours asThe Bloch state amplitudes on $A,B$ form a two-component vector. A consistent phase convention gives the HamiltonianIts characteristic equation is $E^2=t^2|f{|}^2$, giving the two displayed energy bands. The two-component structure comes from the two sites, rather than two values of physical spin. This is a particularly small matrix model in which a crystal's geometry creates an unexpected low-energy law.

Historically, Philip Russell Wallace used a tight-binding model to investigate the electronic band structure of graphite in 1947. The material in that paper was layered graphite; the single-sheet model here isolates one simpler part of the problem.^[ref]

At either displayed wave vector, the three complex hopping amplitudes add to zero: $1+2\cos (2\pi /3)=0$. Therefore the energy bands touch. There are two inequivalent electronic valleys; other equivalent corners are related by reciprocal-lattice translations. Setting $\mathbf{k}={\mathbf{K}}_{\tau }+\mathbf{q}$ and retaining first-order terms giveswhere ${\sigma }_x,{\sigma }_y$ are Pauli matrices. For example, at ${\mathbf{K}}_{+}$, $f\simeq -3b(q_x-i{q}_y)/2$. The matrix has the form of a massless two-dimensional Dirac Hamiltonian. Here $v_F$ is a material-model velocity, rather than the vacuum speed of light; this description does not change an Electron's vacuum rest mass. It describes the dispersion relation of electronic excitations near a band crossing.

At ${\mathbf{K}}_{+}$ and positive energy, a normalized eigenvector for $\mathbf{q}=|\mathbf{q}|(\cos \varphi ,\sin \varphi )$ is $(1,e^{i\varphi })/\sqrt{2}$. Its Pauli matrix expectation values give the displayed direction. This pseudospin records the relative amplitudes and phase on the two sublattices. Real spin supplies another label; confusing these two labels would double-count states and misinterpret the two-band Hamiltonian.

Assume zero temperature, the linear Dirac cone, and both real-spin states and both electronic valleys. Define $n$ as excess Electron number per unit area, negative for an excess of electron holes, relative to neutrality. A filled circular region of radius $k_F$ in wave vector space containsThus the excess carrier density grows quadratically with the Fermi energy, unlike the constant density of states of a parabolic two-dimensional band. Differentiating gives the positive density of states per unit area, $D(E)=2|E|/[\pi (\hslash v_F{)}^2]$. Both the geometry of occupied states and the fourfold degeneracy matter to the coefficient.

A massless-cone description permits a nonzero cyclotron effective mass. For a circular constant-energy contour of area $S(E)=\pi E^2/(\hslash v_F{)}^2$, the semiclassical definition givesThe vanishing mass parameter in a Dirac Hamiltonian and the finite cyclotron response away from neutrality are different statements. This makes the temperature dependence of Shubnikov-de Haas oscillations a useful test of the dispersion rather than a measurement of an Electron's vacuum rest mass.

Andre Geim, Konstantin Novoselov and collaborators mechanically cleaved graphite, identified thin flakes, and fabricated multi-terminal Hall bars on oxidized silicon. A substrate gate changed the carrier density; longitudinal electrical resistance and the Hall coefficient supplied different checks. The 2004 report included one-, two- and three-layer devices and described a few-layer semimetal interpretation. Layer identification and electrical control were separate experimental achievements.^[ref]

The 2005 monolayer study used an optical microscope, electron microscopy and atomic force microscopy to select monolayers. It then measured Shubnikov-de Haas oscillations as temperature, gate voltage and magnetic field varied. Their damping inferred a density-dependent cyclotron effective mass; comparison with oscillation frequency tested a linear dispersion relation. A comparison with two-layer devices also distinguished their quantum Hall effect sequences. These are complementary tests of the monolayer model, rather than consequences of finding a thin-looking flake alone.^[ref]

Consider an ideal uniform graphene sheet connected to an Electron reservoir, with an insulated gate. Let $c_g>0$ be geometric capacitance per unit area, $e>0$ the elementary charge, and $V_0$ the neutral-point offset. Neglect temperature, trapped electric charge and spatial inhomogeneity. A change in gate voltage pays for both the electrostatic potential drop, $en/c_g$, and the change in the sheet's Fermi energy, $E_F/e$. Here $E_F$ is measured relative to the band crossing; it is negative for excess electron holes.

Combine this relation with the carrier density in a graphene cone. Write $u=|E_F|/e$, measured in volts, $V=|V_g-V_0|$, and $\alpha =e^3/[\pi (\hslash v_F{)}^2{c}_g]$. ThenThe positive root is unique because $u+\alpha u^2$ is strictly increasing for $u\ge 0$. The often useful geometric approximation $n\simeq c_g(V_g-V_0)/e$ neglects the Fermi energy term. It overestimates $|n|$ in this model, especially near neutrality where relatively few states are available.

For a deliberately chosen model with $c_g=0.010F/{\mathrm{m}}^2$, $v_F=10^{6}m/s$, $V_0=0$ and $V_g=1\mathrm{V}$, the formulas give $E_F\simeq 0.252\mathrm{eV}$ and $n\simeq 4.67\times 10^{12}\mathrm{c}{\mathrm{m}}^{-2}$. The purely geometric value would be $6.24\times 10^{12}\mathrm{c}{\mathrm{m}}^{-2}$. These are chosen-model predictions, not measurements of the historical devices. Numerical conversion uses the exact Planck constant and elementary charge.^[ref]

The quantum capacitance per unit area is $c_q=e^{2}D(E_F)$. Differentiating the gate equation givesThus the available electronic states act as a second contribution in series with geometric capacitance. At exact neutrality the ideal zero-temperature model has $c_q=c_{\mathrm{diff}}=0$; actual finite-temperature or spatially nonuniform samples need a broader description. A capacitance measurement can therefore probe the density of states, rather than geometry alone.