The Einstein Relation in Semiconductor Drift-Diffusion
1. What Are the Einstein Relations?
Inside a semiconductor, charge carriers are transported by two mechanisms: drift caused by gradients in the energy bands, and diffusion caused by gradients in carrier density. The Einstein relations link the corresponding mobilities and diffusion coefficients, so the two mechanisms balance correctly at thermal equilibrium. The drift–diffusion equations for electrons and holes are
\[ \begin{aligned} \boldsymbol{J}_n &= \underbrace{\color{red}{\mu_n n\nabla E_c}}_{\text{drift}} + \underbrace{\color{blue}{qD_n\nabla n}}_{\text{diffusion}},\\[5pt] \boldsymbol{J}_p &= \underbrace{\color{red}{\mu_p p\nabla E_v}}_{\text{drift}} - \underbrace{\color{blue}{qD_p\nabla p}}_{\text{diffusion}}. \end{aligned} \]
Here \(\boldsymbol{J}_n\) and \(\boldsymbol{J}_p\) are conventional electron and hole current densities, \(q\) is the positive elementary charge, \(n\) and \(p\) are free-carrier densities, \(\mu_n\) and \(\mu_p\) are mobilities, and \(D_n\) and \(D_p\) are diffusion coefficients. \(E_c\) and \(E_v\) are conduction- and valence-band edge energies in joules, so their gradients replace the explicit electric-field term. If band edges are stored in volts (energy divided by \(q\)), the equivalent drift term is \(q\mu_n n\nabla (E_c/q)\), and similarly for holes. The expressions here assume isothermal transport and locally fixed band structure; temperature gradients and spatially varying material properties need additional care. For background see Electrostatics, Charge carrier mobility and the full drift–diffusion derivation.
There are two terms in each equation. The red drift term describes transport driven by the band-edge slope; the blue diffusion term describes transport driven by density differences. In Figure ??, electrons move down the conduction-band slope and accumulate on the right. Diffusion pushes them back towards the lower-density region. Although electrons move opposite to conventional electron current, the current contributions are opposite and cancel at equilibrium.
In an unilluminated device at thermal equilibrium, including a p–n junction with a built-in field, the net electron and hole currents must each vanish locally:
\[ \boldsymbol{J}_{n,\mathrm{drift}}+\boldsymbol{J}_{n,\mathrm{diff}}=\boldsymbol{0}, \qquad \boldsymbol{J}_{p,\mathrm{drift}}+\boldsymbol{J}_{p,\mathrm{diff}}=\boldsymbol{0}. \]
If the Einstein relations are inconsistent with the carrier statistics, drift and diffusion need not cancel: a simulation may then predict a false dark current at zero applied voltage, or an offset in its dark current–voltage curve, as illustrated in Figure ??. This is particularly serious when simulating very small currents. Zero applied bias by itself does not imply equilibrium if the device is illuminated or otherwise driven; the zero-current statement here applies to the dark, thermally equilibrated device.
For the usual non-degenerate (Boltzmann) case, the Einstein relations are simply
\[ \boxed{D_n=\mu_n\frac{k_BT}{q},\qquad D_p=\mu_p\frac{k_BT}{q}.} \]
Here \(k_B\) is Boltzmann's constant and \(T\) is absolute temperature. The ratio \(k_BT/q\) is the thermal voltage (approximately 25.9 mV at 300 K). Higher mobility therefore requires a proportionally higher diffusion coefficient; the two are not independent choices in a consistent device model. The following sections derive this expression and explain what changes for degenerate or nonparabolic bands.
2. A Simple Derivation: The Classical Einstein Relation
Consider electrons in a uniform, isothermal, non-degenerate semiconductor. In thermal equilibrium the electron Fermi level \(E_F\) is spatially constant, and the carrier density follows the Boltzmann distribution. The distinction between Boltzmann and Fermi–Dirac occupations is discussed in Fermi–Dirac versus Maxwell–Boltzmann statistics:
\[ n=N_c\exp\!\left(\frac{E_F-E_c}{k_BT}\right). \]
Here, \(E_c\) is the conduction-band edge and \(N_c\) is the effective conduction-band density of states. With constant \(N_c\) and \(T\), taking the spatial gradient gives
\[ \nabla n=-\frac{n}{k_BT}\nabla E_c. \]
Substituting this density gradient directly into the electron current, using the conduction-band gradient notation, gives
\[ \boldsymbol{J}_n =\mu_n n\nabla E_c+qD_n\nabla n =\mu_n n\nabla E_c-qD_n\frac{n}{k_BT}\nabla E_c =n\nabla E_c\left(\mu_n-\frac{qD_n}{k_BT}\right). \]
Because the equilibrium current must vanish even where a built-in field exists, the expression in brackets must be zero. Therefore
\[ \boxed{D_n=\mu_n\frac{k_BT}{q}.} \]
Repeating the argument for holes gives \(D_p=\mu_p k_BT/q\). This is the familiar classical Einstein relation. It is often derived using parabolic bands, but the decisive approximation here is Boltzmann (non-degenerate) carrier statistics: the carriers are sufficiently dilute that occupation of available states is small.
3. The Generalised Einstein Relation
The simple result does not always apply. In heavily doped materials, or when the Fermi level approaches or enters a band, carriers obey Fermi–Dirac statistics rather than the Boltzmann approximation. The carrier density is then not a simple exponential of the Fermi level. The density of states may also be nonparabolic, disordered, or obtained numerically from a band-structure calculation. The carrier-density calculation itself is described in Charge carrier density.
Let the electron density depend on the energy separation \(\xi_n=E_{Fn}-E_c\), at fixed temperature and with the local density of states held fixed: \(n=n(\xi_n,T)\). The electron quasi-Fermi level \(E_{Fn}\) is constant at equilibrium. Taking a spatial gradient therefore gives
\[ \nabla n =\left.\frac{\partial n}{\partial\xi_n}\right|_T (\nabla E_{Fn}-\nabla E_c) =-\left.\frac{\partial n}{\partial\xi_n}\right|_T\nabla E_c. \]
Inserting this result into the current equation and requiring zero equilibrium current gives the generalised Einstein relation:
\[ \boxed{ \frac{D_n}{\mu_n} =\frac{n}{q\left(\partial n/\partial\xi_n\right)_T} =\frac{1}{q}\left(\frac{\partial\xi_n}{\partial\ln n}\right)_T. } \]
For holes, define \(\xi_p=E_v-E_{Fp}\), which increases with the hole population. Then the equivalent relation is
\[ \boxed{ \frac{D_p}{\mu_p} =\frac{p}{q\left(\partial p/\partial\xi_p\right)_T}. } \]
These equations have a useful physical interpretation: the derivative tells us how easily the carrier population changes when its chemical potential changes. A large change in density for a small change in quasi-Fermi energy means a large derivative and a smaller \(D/\mu\). If states fill up and the density responds less strongly, the ratio increases. Thus the Einstein relation is a property of the carrier statistics and density of states, not just a numerical factor multiplying the mobility.
For an arbitrary conduction-band density of states \(g_c(E)\), the electron density can be calculated as
\[ n=\int_{E_c}^{\infty}g_c(E) \frac{1}{1+\exp[(E-E_{Fn})/(k_BT)]}\,dE. \]
At a fixed band structure and temperature, its derivative can be evaluated from the same states:
\[ \left.\frac{\partial n}{\partial E_{Fn}}\right|_T =\frac{1}{k_BT}\int_{E_c}^{\infty}g_c(E)f(E)[1-f(E)]\,dE, \qquad f(E)=\frac{1}{1+\exp[(E-E_{Fn})/(k_BT)]}. \]
Since \(\partial\xi_n/\partial E_{Fn}=1\) at fixed \(E_c\), this derivative can be used directly in the generalised Einstein relation. This approach also works with a nonparabolic numerical density of states. For holes, use the corresponding valence-band density of states and the hole occupation \(1-f\), taking the derivative with respect to \(E_v-E_{Fp}\).
4. Parabolic Bands: Recovering the Classical Result
For a three-dimensional parabolic band, the electron density can be expressed in terms of a Fermi–Dirac integral:
\[ n=N_c F_{1/2}(\eta_n), \qquad \eta_n=\frac{E_{Fn}-E_c}{k_BT}. \]
We use the normalised definition \(F_j(\eta)=\Gamma(j+1)^{-1} \int_0^\infty x^j/[1+\exp(x-\eta)]\,dx\). Using \(dF_{1/2}/d\eta=F_{-1/2}\), the generalised Einstein relation becomes
\[ \boxed{ \frac{D_n}{\mu_n} =\frac{k_BT}{q} \frac{F_{1/2}(\eta_n)}{F_{-1/2}(\eta_n)}. } \]
An analogous equation holds for holes with \(\eta_p=(E_v-E_{Fp})/(k_BT)\). This result is valid for a three-dimensional parabolic band even when it is degenerate. The ratio of Fermi–Dirac integrals accounts for the effect of state filling.
In the non-degenerate limit, where \(\eta_n\ll-1\), both Fermi–Dirac integrals approach \(\exp(\eta_n)\). Consequently,
\[ \frac{F_{1/2}(\eta_n)}{F_{-1/2}(\eta_n)}\longrightarrow 1, \qquad \boxed{\frac{D_n}{\mu_n}\longrightarrow\frac{k_BT}{q}.} \]
This shows precisely where the familiar expression comes from. A parabolic band alone is not sufficient: to obtain \(D/\mu=k_BT/q\), the carriers must also be non-degenerate. In a degenerate parabolic band the Fermi–Dirac correction is needed; for a general nonparabolic band the density-derivative expression is usually the more useful starting point.
5. Carrier Pressure: Another Way to Understand Diffusion
There is another useful way to understand where the diffusion term comes from. In the momentum-balance equation, derived from the Boltzmann Transport Equation, carriers exert a pressure because they have a distribution of velocities and momenta. A gradient in this carrier pressure produces a force that drives carriers from regions of higher pressure towards regions of lower pressure. This is the microscopic origin of diffusion. See the Boltzmann Transport Equation derivation for the connection to momentum balance.
For an isotropic three-dimensional parabolic conduction band, define the electron pressure \(P_n\) by
\[ P_n=\frac{2}{3}\int_{E_c}^{\infty}(E-E_c)g_c(E)f(E)\,dE. \]
Here \(g_c(E)\) is the conduction-band density of states and \(f(E)\) is the occupation probability. The factor \(2/3\) follows from the parabolic dispersion in three dimensions. In the classical, non-degenerate limit this becomes the familiar ideal-gas expression
\[ P_n=nk_BT. \]
For a uniform, isothermal material, the electron current can therefore be expressed using a pressure gradient instead of a density gradient:
\[ \boldsymbol{J}_n=\mu_n n\nabla E_c+\mu_n\nabla P_n =\mu_n n\nabla E_c+qD_n\nabla n. \]
Comparing the two diffusion terms shows that the relevant quantity is how rapidly pressure changes with carrier density, not simply pressure divided by density:
\[ \boxed{\frac{D_n}{\mu_n}=\frac{1}{q}\left(\frac{\partial P_n}{\partial n}\right)_T.} \]
Since \(P_n=nk_BT\) in the Boltzmann limit, \(\partial P_n/\partial n=k_BT\), and the classical Einstein relation follows immediately. When the electron gas becomes degenerate, pressure no longer increases linearly with density. For a three-dimensional parabolic band, using the same normalised Fermi–Dirac integrals as above gives
\[ P_n=N_c k_BT F_{3/2}(\eta_n),\qquad \frac{\partial P_n}{\partial n} =k_BT\frac{F_{1/2}(\eta_n)}{F_{-1/2}(\eta_n)}. \]
This recovers exactly the generalised Einstein relation derived in Section 4. It is important not to confuse \(P_n/n=k_BT F_{3/2}(\eta_n)/F_{1/2}(\eta_n)\) with \(\partial P_n/\partial n\): they are equal only in the classical limit. For nonparabolic bands, pressure requires a more general velocity- and band-structure-dependent treatment; the carrier-density derivative in Section 3 is usually a more convenient way to calculate the Einstein relation within a consistent drift–diffusion model.
6. Numerical Accuracy: Avoiding False Equilibrium Currents
In a drift–diffusion simulator, the generalised Einstein relation is straightforward to write down but can be difficult to evaluate accurately. The main challenge is the derivative of carrier density with respect to quasi-Fermi energy. A common approximation is a finite difference:
\[ \frac{\partial n}{\partial E_{Fn}} \approx\frac{n(E_{Fn}+\Delta E)-n(E_{Fn}-\Delta E)}{2\Delta E}. \]
If \(\Delta E\) is too large, the derivative is inaccurate because of truncation error. If it is too small, two nearly equal carrier densities are subtracted, which can amplify floating-point round-off error or numerical noise in the band-structure and density-of-states calculations. The resulting error is then amplified when forming \(D/\mu=n/[q(\partial n/\partial E_{Fn})]\), particularly where the derivative is small.
Even a small relative error can matter because, in equilibrium, the drift and diffusion currents may each be substantial while their sum should be exactly zero. A small mismatch can leave a false residual current. This can be especially noticeable in simulations involving nonparabolic bands, numerical densities of states, low carrier densities, or strongly varying heterostructures.
Where possible, it is preferable to differentiate the occupation function analytically under the density-of-states integral, as in Section 3, rather than finite-differencing two separately calculated densities. For very dilute carriers, use a stable Boltzmann-limit expression when appropriate. It is also important to use consistent numerical integration, carrier statistics, band-edge conventions and discretisation of drift and diffusion.
Finally, a correct Einstein relation is necessary but not by itself sufficient to eliminate spurious current: the spatial discretisation must preserve equilibrium too. In particular, flux discretisations should respect constant equilibrium Fermi levels; an equilibrium-preserving scheme such as Scharfetter–Gummel is commonly used for classical drift–diffusion. A practical simulator check is to run a dark, zero-bias equilibrium calculation and confirm that the electron and hole currents vanish throughout the device to the required numerical tolerance, including across material interfaces. For the treatment of junctions and band offsets, see Interfaces and The built-in potential.
The central lesson is simple: the Einstein relation is what makes the drift and diffusion descriptions thermodynamically consistent. Its familiar classical form is easy to use, but more general carrier statistics require a correspondingly general and numerically accurate relation. The broader transport framework is developed in the Boltzmann-to-drift–diffusion derivation.
👉 Next step: Now continue to Calculating charge carrier density