The Semiconductor Equations
Nonuniform Band Structure and the Reference Energy
Most practical simulations deal with a band structure that varies in space—it is therefore necessary to define quantities such as the band energies with respect to a reference energy. In order to make the definition of boundary conditions simpler, the reference energy chosen is the Fermi energy in an equilibrium state when no potentials are applied to any of the boundaries in the system and when there are no thermal gradients in the system (note that in COMSOL Multiphysics, the reference temperature used to define the equilibrium Fermi energy can be changed in the Semiconductor interface’s Settings window under Reference Temperature). This is illustrated in Figure 3-7 for an isothermal, abrupt p-n junction (that is a boundary between a p- and n-doped region of a semiconductor with a constant doping level on the two sides of the device). In equilibrium the Fermi energy is well-defined and is constant throughout space (since there are no gradients in the Fermi level/chemical potential). As a result in the immediate vicinity of the junction, the bands bend to accommodate a constant Fermi energy. In this region of band bending the Fermi level lies close to the center of the gap and consequently the carriers are depleted — this is known as the depletion region. A space charge layer is associated with the depletion region since the charges of the ionized donors and acceptors are no longer compensated by free carriers. It is the space charge layer that generates an electric field and the corresponding potential gradient that results in the bending of the vacuum level, E0, and the conduction band and valence band edges (Ec and Ev, respectively). This results in a self-consistent picture of the band structure in the vicinity of the junction. As a result of the bending of the vacuum level and bands, a built-in potential, Vbi, develops across the junction. When an additional potential difference is applied to the p-side of the junction, Vp, the junction is in a condition known as reverse bias. The Fermi energy is no longer well-defined in the vicinity of the depletion region, but the equilibrium Fermi level, Ef0 is still used as a reference potential. Away from the depletion region the semiconductor is close to equilibrium and the concept of the Fermi level can still be applied. On the p-side of the junction, the Fermi energy shifts up to a value Ef1, which differs from Ef0 by qVp where q is the electron charge.
In the most general case the electron affinity, qχ = E0 Ec, and the band gap, Eg = Ec –Ev, can vary with position. However, the values can be considered to be material properties. This means that the conduction band edge cannot be parallel to the valence band edge or to the vacuum level (this is not shown in Figure 3-7).
Figure 3-7: Diagram showing the variation in the band structure in a p-n diode. The horizontal axis represents position in the device and the vertical axis indicates energy. The vacuum level E0, the bottom of the conduction band Ec, and the top of the valence band Ev, are shown. Top: The band structure when there is no applied potential. A built-in potential, Vbi, develops across the junction as a result of the space charge layers associated with the depletion region. The Fermi energy Ef0, is constant throughout the device. Bottom: A potential Vp is applied to the p-type side of the junction, resulting in a reverse bias. Away from the junction, where the material is close to equilibrium, a second Fermi level, Ef1, can be defined in the p-type region–shifted by the applied potential.
Quasi-Fermi Levels
Consider the motion of an individual electron moving in the curved conduction band (see Figure 3-8). In between collisions the energy of the electron remains approximately constant (although the force accelerates the electron, its velocity only increases slightly in between collisions). The energy of the electron measured with respect to the band edge, W, increases as a result of band bending. The total energy associated with an electron is given by:
Similarly the energy of a hole is defined with respect to the band edge in the following manner:
Figure 3-8: Motion of an individual electron within a curved band.
It is assumed that scattering processes within the band cause the electrons (or holes) to reach a collective equilibrium on timescales small compared to the simulation timescales (if highly nonequilibrium/hot electron effects are important within the band then it is necessary to account for the deviation of the Fermi function from a Fermi-Dirac distribution by solving an additional energy equation — currently this is not possible in the Semiconductor interface). Since the relaxation time for electrons within the conduction band is much less than the corresponding relaxation time for transitions across the band gap, the electron and hole distributions can have different associated Fermi levels (Efn for electrons and Efp for holes). In nonequilibrium circumstances (for example in the presence of an electric field or a thermal gradient) these quasi-Fermi levels are not coincident. The electron and hole quasi-Fermi levels typically lie within the band gap and consequently the electrons are scattered to states close to the band edge as they travel through space, as shown in Figure 3-8.
Simplifying the Boltzmann Equation
As discussed in The Semiclassical Model, the evolution of the distribution function for electrons and holes is governed by the Boltzmann equation (see Equation 3-40). Since this equation is difficult to solve, it is common to make simplifying approximations in its solution. Ultimately these approximations produce the drift-diffusion equations. The analysis presented here is based on Ref. 12 and Ref. 13.
The first step taken when simplifying the Boltzmann equation is usually to make the relaxation-time approximation, which assumes that the collision terms in the Boltzmann equations take the form:
Where f0 (E, Efn, T) is the Fermi function for electrons and f0h (E, Efp, T) is the Fermi function for holes. This is essentially the simplest form of the collision term that returns the electrons to the Fermi-Dirac distribution desired. In practice this assumption is a significant simplification; Ref. 1 and Ref. 2 provide more detailed discussions of collision mechanisms in real solids.
From Hamilton’s equations the spatial gradient of the electron total energy is equal to the rate of change of crystal momentum, which in turn is related to the applied force (Equation 3-35 in The Semiclassical Model), so that:
Using these results the Boltzmann equations become
For compactness, the explicit dependence of f0 and τ is dropped. Considering first the electron density, the assumption that the electron distribution function is close to its quasi-equilibrium state allows the distribution function to be written in the form:
where f1 << f0. The spatial gradient terms are dominated by terms involving the gradient of f0, so the f1 terms are neglected in comparison to these. Likewise the term df/dt is also small since the deviation from the quasi-equilibrium is small and the f0 term varies on timescales much slower than the collision time if the temperature is a function of time. The Boltzmann equation can be expressed in the approximate form:
(3-53)
Next the gradients in f0=1/(1+exp[(E-Efn)/(kbT)]) are expressed in terms of parameters related to the band structure, using the chain rule:
(3-54)
where the last step follows from the semiclassical definition of velocity (Equation 3-34 in The Semiclassical Model). From the chain rule the spatial gradient of f0 is given by:
Since f0=f0([E-Efn]/T):
therefore:
(3-55)
Substituting Equation 3-55 and Equation 3-54 into Equation 3-53, using the definition E = Ec + W and rearranging gives:
By definition the number of electrons is given by:
The final equality holds because f1 is an odd function when integrated over the region of k-space in the vicinity of a band edge and f0 is an even function (assuming, without loss of generality, that the origin of k-space at the band minimum). To see why this is the case, consider Equation 3-38 in The Semiclassical Model (shown below for convenience).
In equilibrium (f = f0) no current flows so the integrand must be an odd function (since it is not zero over all k-space). Since the origin is at a band minimum, E(k)/k is odd so f0 must be even. Deviations from equilibrium produce a current without changing the total number of electrons, so f1 must be an odd function. The electron current density can be written in the form:
Note that the definition E=Ec+W is used to split the energy terms.
The rigid band assumption is made next. This asserts that even when the bands bend, the functional form of W in k-space (measured with respect to the band edge) is unchanged. Thus W = W(k). Given this assumption the quantities inside the integrals are dependent only on the local band structure, except for the df0/dE term.
Next define the quantities:
(3-56)
(3-57)
To obtain:
(3-58)
Qn and μn are tensor quantities (although most semiconducting materials are cubic) so that the corresponding tensors are diagonal with identical elements and consequently can be represented by means of a scalar (cubic materials are assumed in the Semiconductor interface). Careful examination of Equation 3-56 and Equation 3-57 shows that these cannot straightforwardly be considered material constants because both depend on the quasi-Fermi level through the quantities df0/dE and 1/n. In the nondegenerate limit, the quasi-Fermi level dependence of these two quantities cancels out, since:
1/nexp[-Efn/(kBT)] and f0/Eexp[-Efn/(kBT)]
These quantities can therefore only strictly be considered material constants in the nondegenerate limit. In the degenerate limit it is, however, possible to relate Qn to the mobility for specific models for the relaxation time (see below). In principle a mobility model could be used, which is dependent on the electron quasi-Fermi level (or the electron density) in the degenerate limit.
Exactly the same arguments are applied for holes, leading to the equation for the hole current:
(3-59)
Here, the symbols introduced with the subscript change np have the same definitions as the electron quantities except that the relevant integrals are over the hole band.
Equation 3-58 and Equation 3-59 describe the evolution of the quasi-Fermi levels within a semiconductor. It is possible to formulate the whole equation system so that the hole and electron quasi-Fermi levels are the dependent variables, which is one of the discretization options available in the Semiconductor Module (the quasi-Fermi level formulation). Before deriving the more familiar drift-diffusion equations it is useful to derive the relationship between Qn and μn, and Qp and μp.
Relating Qn to μn
For isotropic materials within the rigid band approximation, Equation 3-20 can be written as:
(3-60)
Since Ec (r) is independent of k the derivatives with respect to E can be replaced by derivatives with respect to W, enabling the (scalar) mobility to be written as:
where the semiclassical result v=(1/)(E(k)/k) is used and the xx element of the mobility tensor is evaluated to compute the scalar (the yy and zz elements produce the same result). From Equation 3-60:
The integral in the mobility expression can be transformed to spherical polar coordinates using the definitions:
and a functional form can be assumed for the relaxation time:
(3-61)
In Equation 3-61 r = −1/2 corresponds to acoustic phonon scattering and r = 3/2 is appropriate for ionized impurity scattering (Ref. 14). In general r can be considered a function of temperature. In COMSOL Multiphysics, r is assumed to be −1/2.
The mobility integral can now be written as:
To transform the integral over k to an integral over W note that W = h2 k2 /(2m∗) so that dk=(m*/22)1/2W -1/2dW. Writing each quantity in the integral as a function of W and rearranging gives:
Evaluating the angular integrals and integrating the energy integral by parts gives:
(3-62)
Where the result:
is used. Equation 3-62 can be rewritten in the form:
(3-63)
where the definition of the Fermi-Dirac function (Equation 3-46) and of f0 (Equation 3-5) is used.
Following the same procedure for Qn gives:
(3-64)
Equation 3-63 and Equation 3-64 show that Qn can be related to μn:
(3-65)
where the result Γ (j + 1) = jΓ(j) is used.
Similarly Qp is related to μp in exactly the same manner:
(3-66)
In COMSOL Qn and Qp are computed from μn and μp using Equation 3-65 and Equation 3-66 with r=−1/2. Note that in the nondegenerate limit:
The drift-Diffusion Equations
Having related Qn and Qp to the corresponding mobilities the next task is to derive equations relating the current to the carrier concentrations from Equation 3-58 and Equation 3-59. Once again the case of electrons is considered in detail and the results for holes are similar.
Inverting Equation 3-51 gives the following equation for the electron quasi-Fermi level:
(3-67)
where we have defined the inverse Fermi-Dirac integral (F-11/2(α)=η implies that α=F1/2(η)). Let α=n/Nc and η=F-11/2(α). To obtain the drift-diffusion equations Equation 3-67 is substituted into Equation 3-58. Note that Nc=Nc(T) so that η=F-11/2(n/Nc)=η(n,T). To compute the current the gradient of the quasi-Fermi level is required:
Since α=n/Nc and NcT3/2 (Equation 3-48):
(3-68)
In order to evaluate ∂η/∂α the result dFn(η)/dη=Fn-1(η) is required (see Ref. 15 for details). Given this result:
so that:
(3-69)
Substituting Equation 3-68 and Equation 3-69 into Equation 3-46 and then using α F1/2(η) gives:
(3-70)
Substituting Equation 3-70 into Equation 3-58 gives:
where the result η = (Efn − Ec)/(kBT) is used and follows from Equation 3-67.
Defining the function:
This result can be written in the form:
(3-71)
Following the same argument for the hole current gives:
(3-72)
The thermal diffusion coefficients Dn,th and Dp,th are defined as:
In the nondegenerate limit G(α)1 and the following equations are obtained:
For a relaxation time dominated by phonon scattering (currently assumed by the COMSOL software) r = −1/2.
Poisson’s Equation and the continuity equations
Equation 3-71 and Equation 3-72 define the hole and electron currents used by COMSOL Multiphysics in the Semiconductor interface. To solve a model the Semiconductor interface uses these definitions in combination with Poisson’s equation and the current continuity equations.
Poisson’s equation takes the form:
(3-73)
and the current continuity equations are given by:
(3-74)
where Un=ΣRn,i-ΣGn,i is the net electron recombination rate from all generation (Gn,i) and recombination mechanisms (Rn,i). Similarly, Up is the net hole recombination rate from all generation (Gp,i) and recombination mechanisms (Rp,i). Note that in most circumstances Un=Up. Both of these equations follow directly from Maxwell’s equations (see Ref. 6).
The Semiconductor interface solves Equations 3-73 and 3-74. The current densities Jn and Jp are given by Equation 3-71 and Equation 3-72, if the dependent variables are the carrier concentrations or the log of the concentrations. For the quasi-Fermi level formulation, Equation 3-58 and Equation 3-59 are used for the current densities.