So far, all QCLs are purely unipolar n-type devices, so the model uses a single-band effective-mass Hamiltonian, which accounts for mixing with the valence bands through energy-dependent effective mass
\(m(E, z)~= m^*(z)\{1+[E-E_c(z)]/E_g(z)\}\), where
E is the total energy,
\(E_c(z)\) and
\(E_g(z)\) are the conduction band edge and band gap profiles, and
z is the growth direction. The in-plane dynamics are included by kinetic energy terms with the same energy-dependent effective mass. Such a choice, which is not ultimate, preserves the in-plane non-parabolicity, comparable to the results predicted by the 8-band k·p method [
15]. The full non-interacting Hamiltonian is expressed as
$$\begin{aligned} H=\frac{-\hbar ^2}{2}\frac{d}{dz}\frac{1}{m(E,z)}\frac{d}{dz}+V(z)+\frac{\hbar ^2k^2}{m(E,z)}, \end{aligned}$$
(1)
where the potential energy term
\(V(z) = E_c(z) + V_\mathrm{sc}(z)\) comprises the conduction band edge profile
\(E_c(z)\) and the electron mean field term
\(V_\mathrm{sc}(z)\), calculated self-consistently by the solution of the Poisson equation. As already mentioned, the calculations are made in real space, so the Hamiltonian of Eq. (
1) is discretized and the grid points (non-uniformly spaced on the
z-axis) define its base vectors; the discretized Hamiltonian
\({\mathbf {H}}\) can then be evaluated in the matrix form following [
16]
$$\begin{aligned} {\mathbf {H}}_{i,i}= & {} V(z_i)+\frac{\hbar ^2 k^2}{m(E,z_i)}-{\mathbf {H}}_{i,i+1}-{\mathbf {H}}_{i,i-1}, \nonumber \\ {\mathbf {H}}_{i,i\pm 1}= & {} -\frac{\hbar ^2}{m(E,z_{i\pm })z_{i,i\pm 1}(z_{i+1}-z_{i-1})}, \end{aligned}$$
(2)
where discretization site
i has coordinate
\(z_i\), sites
i,
j are at the distance
\(|z_j - z_i| \equiv z_{i,j}\),
\(z_{i\pm } = (z_{i\pm 1} + z_i)/2\) are halfway between
\(z_i\) and
\(z_{i+1}\) or
\(z_{i-1}\) and
\(z_i\), and
\(m(E,z_{i\pm })\equiv \frac{1}{2}(m(E,z_{i\pm 1})+m(E,z_i))\). It stems from Eq. (
2) that matrix
\({\mathbf {H}}\) is non-Hermitian. The Dyson and Keldysh equations take the forms [
17]
$$\begin{aligned} (E{\mathbf {I}} - {\mathbf {H}} - {\varvec{{\Sigma }}}^\mathbf{R})\mathbf{G}^\mathbf{R}= {\varvec{\lambda }}, \end{aligned}$$
(3)
$$\begin{aligned} (E{\mathbf {I}} - {\mathbf {H}} - {\varvec{\Sigma }}^\mathbf{R})\mathbf{G}^{<}= & {} {\varvec{\Sigma }}^<\mathbf{G}^\mathbf{A}, \mathbf {G^A=(G^R)^\dagger }, \end{aligned}$$
(4)
where
\(\mathbf{I }\) is the unity matrix, and
\({\varvec{\Sigma }}\)s and
\(\mathbf{G }\)s are the self-energy and Green’s function matrices with the elements (
i,
j) linking the discretization sites at
\(z_i\) and
\(z_j\). The matrix
\(\varvec{\lambda }\) in Eq. (
3) is a diagonal matrix with the elements
\(\varvec{\lambda }_{i,i} \equiv 2/(z_{i+1} - z_{i-1})\). For the Hamiltonian of Eq. (
1), all
\({\mathbf {G}}\)s are the functions of two parameters, i.e., total energy
E and in-plane momentum modulus
k:
\({\mathbf {G}} = {\mathbf {G}}(E, k)\). The scattering self-energies that enter the formalism must be calculated with care. The usual formulation uses the Dyson equation in the form
\((E{\mathbf {I}} - {\mathbf {H}} - {\varvec{\Sigma }}^\mathbf{R})\mathbf{G}^\mathbf{R} = {\mathbf {I}}\), and so the Green’s functions have the units of energy
\(^{-1}\). In our case, the Green’s functions are per energy
\(\times\)length. The necessary integration over space must not be omitted, in order to preserve the consistency. For instance, for the quasi-elastic approximation of scattering with acoustic phonons, the formulation for the self-energy in [
18] is expressed as
$$\begin{aligned} {\varvec{\Sigma }}_{i,j}^{\mathbf {R,<}} = \delta _{i,j} \frac{k_BTD^2}{4\pi ^4\rho v_s^2 a}\int dk^2{\mathbf {G}}_{i,j}^{\mathbf {R,<}}, \end{aligned}$$
where
\(k_B\) is the Boltzmann constant,
\(v_s\),
\(\rho\),
D are the sound velocity, the density, and the deformation potential in the host material, and
a is the grid size. In our formulation, the integration over the spatial coordinate cancels term
a, so that the correct formula for this self-energy is
$$\begin{aligned} {\varvec{\Sigma }}_{i,j}^{\mathbf {R,<}} = \delta _{i,j} \frac{k_BTD^2}{4\pi ^4\rho v_s^2}\int dk^2{\mathbf {G}}_{i,j}^{\mathbf {R,<}}, \end{aligned}$$
(5)
where
\({\mathbf {G}}^{\mathbf {R,<}}\) are calculated from Eqs. (
3) and (
4). A similar, straightforward transformation can be applied to other self-energies with diagonal-only entries, such as alloy disorder or interface roughness. For the interactions, which give non-local self-energies, i.e., ionized impurities and polar optical phonons, the transformations are less intuitive. Self-energies for polar optical phonons should be calculated as [
18]
$$\begin{aligned}{\varvec{\Sigma }}_{i,j}^{{\mathbf {R}}}(E,k) &= \frac{\beta }{\pi }\int \frac{d^2q}{4\pi ^2} I_{lo}(|z_i-z_j| ,k,q) /\varvec{\lambda }_{j,j} \nonumber \\&\quad\times \left[ n_B{\mathbf {G}}_{i,j}^{{\mathbf {R}}}(E+\hbar \omega ,q) +(n_B+1){\mathbf {G}}_{i,j}^{{\mathbf {R}}}(E-\hbar \omega ,q) \right. \nonumber \\&\quad \left. +\frac{1}{2}\left( {\mathbf {G}}_{i,j}^{\mathbf {<}}(E-\hbar \omega ,q) -{\mathbf {G}}_{i,j}^{\mathbf {<}}(E+\hbar \omega ,q)\right) \right] , \end{aligned}$$
(6)
and
$$\begin{aligned}{\varvec{\Sigma }}_{i,j}^{\mathbf {<}}(E,k) &=\frac{\beta }{\pi }\int \frac{d^2q}{4\pi ^2} I_\mathrm{lo}(|z_i-z_j| ,k,q)/\varvec{\lambda }_{j,j}\nonumber \\&\quad\quad \times \left[ n_B{\mathbf {G}}_{i,j}^{\mathbf {<}}(E+\hbar \omega ,q) + (n_B+1){\mathbf {G}}_{i,j}^{\mathbf {<}}(E-\hbar \omega ,q)\right] , \end{aligned}$$
(7)
where
\(n_B\) is the Bose–Einstein factor for energy
\(\hbar \omega\),
\(\beta =e^2\hbar \omega (\epsilon _\infty ^{-1}-\epsilon _0^{-1})/2\), and
\(I_\mathrm{lo}\) is the integral calculated as
$$\begin{aligned}&I_\mathrm{lo}(|z-z'| ,k,q)\nonumber \\&\quad =\int _0^{\pi /a} dq_z \frac{cos(q_z(z-z'))}{\sqrt{(q_z^2+q^2+k^2+q_0^2)^2-4k^2q^2)}}\nonumber \\&\quad\quad \times \left( 1-\frac{q_0^2(q_z^2+k^2+q^2+q_0^2)}{(q_z^2+q^2+k^2+q_0^2)^2-4k^2q^2)}\right) , \end{aligned}$$
(8)
For the impurity scattering, the self-energies are [
4]
$$\begin{aligned}&{\varvec{\Sigma }}_{i,j}^{\mathbf {R,<}}(E,k) =\frac{e^4}{16\pi ^2\epsilon ^2}\int qdq I_\mathrm{imp}(z_i,z_j,k,q)/ \varvec{\lambda }_{j,j}\nonumber \\&\quad\quad \times {\mathbf {G}}_{i,j}^{\mathbf {R,<}}(E,q),\nonumber \\&\quad I_\mathrm{imp}(z_i,z_j,k,q) =\sum \limits _k \frac{1}{\varvec{\lambda }_{k.k}} N_D(z_k)\nonumber \\&\quad\quad \times \int _{0}^{2\pi }d\theta \frac{e^{-\sqrt{q_0^2+k^2+q^2-2kq cos\theta }(|z_i-z_k|+|z_j-z_k|)}}{q_0^2+k^2+q^2-2kqcos\theta }, \end{aligned}$$
(9)
where
\(N_D(z)\) is the ionized impurity concentration profile,
\(\theta\) is the angle between vectors
k and
q,
\(q_0\) is the inverse Debye screening length, and
\(\epsilon\) is the static dielectric constant. In the case of non-uniform sampling, the calculation of integrals
\(I_\mathrm{lo}\) and
\(I_\mathrm{imp}\) is associated with a huge increase in computational effort and use of computer memory. However, as these integrals do not change when equations are iterated, they can be executed only once, tabulated and stored in computer memory. This is one more advantage of the chosen basis.
With the NEGF method, boundary conditions are applied through the contact self-energies. In our approach, the contact self-energy matrix
\(\mathbf {\Sigma _C^R}\) has only two nonzero elements, namely, for the device's left boundary
$$\begin{aligned} {\varvec{\Sigma }}_{{\mathbf {C}}~{1,1}}^{{\mathbf {R}}} =\tau _\mathrm{LD}\mathbf {g_{0,0}^R}\tau _\mathrm{DL}, \end{aligned}$$
(10)
where
\(\mathbf {g^R}\) is the Green’s function of the (uncoupled) lead [
19], and
\(\tau _\mathrm{LD}\),
\(\tau _\mathrm{DL}\) are coupling elements. The method for calculating the function
\(\mathbf {g^R}\), which imitates the boundary conditions appropriate for cascade structures, was described in [
5,
20]. With a non-uniform grid, the coupling elements are calculated according to Eq. (
2), i.e.,
$$\begin{aligned} \tau _\mathrm{LD}= & {} {\mathbf {H}}_{0,1}=\frac{\hbar ^2}{2}\frac{2}{(a_L+a_0)a_0}\frac{2}{m_L+m_1},\nonumber \\ \tau _\mathrm{DL}= & {} {\mathbf {H}}_{1,0}=\frac{\hbar ^2}{2}\frac{2}{(a_0+z_{1,2})a_0}\frac{2}{m_L+m_1}, \end{aligned}$$
(11)
where
\(a_L\equiv |z_0-z_{-1}|\),
\(a_0\equiv |z_1-z_0|\) (
\(z_0,~z_{-1}\) are the coordinates of the first and second site in the lead adjacent to the device), and
\(m_L=m(z_0)\),
\(m_1=m(z_1)\). The right contact self-energy
\({\varvec{\Sigma }}_{{\mathbf {C}}~{N,N}}^{{\mathbf {R}}}\) can be similarly calculated.