Transport Phenomenon

Texts: Ashcroft and Mermin (Chapter 13)

Ibach and Luth
Wolfe, Holonyak and Stillman – Semiconductors
 Semi-classical Conduction
For electrons in the nth band, under semi-classical conditions (with k defined in 1st BZ)
dr/dt = vn(k) = (1/ħ) ∂En/∂k (A)

ħ(dk/dt) = - e[E (r,t) + vn(k) x B (r,t)] (B)

give the time dependence of r and k in the presence external influences E and B

For a crystal of volume V, the number of electron states in RS volume element dk is

2dk/(8π3/V),
so, to the unit volume of crystal, the electronic contribution in DS from the element dk
in RS is dk/4π3 times f, the Fermi function

Thus, in thermal equilibrium, the nth band electron density (in DS) contributed by
the infinitesimal reciprocal space volume element dk at wave vector k is

f [En(k)] dk/4π3 = (dk/4π3) / [1 + exp (En(k) – μ )/kT] (C)

- A filled band (f =1) always remains filled because Pauli’s exclusion principle does
not allow more electrons in any state, if there is no overlapping band and inter-
band transitions are not allowed
 Semi-Classical Description in Phase Space
The No. of conduction electrons per unit volume contributed by the element dk in RS is
f [En(k)] dk/4π3 = (dk/4π3) / [1 + exp (En(k) – μ )/kT]

Phase space:
The No. of conduction electrons in real space volume element dr, is f[En(k)] drdk/4π3

which is also the No. of electrons in 6-dimensional phase (r-k) space element (drdk)

Note that for a completely filled band (all energies lie below Ef and so, f = 1 in ground
state), the density of electrons in phase (r-k) space is 1/4π3 at all times!

This is a direct consequence of the Liouville’s theorem which states that the phase
space density, 1/4π3 is preserved

That is, a filled band remains filled at all times, even in the presence of space and time
dependent electric and magnetic fields – hence is inert (also follows simply from PEP)
 Nonequilibrium Distribution Function
In the presence of space and time dependent perturbing applied fields and/or
temperature gradients, a nonequilibrium distribution function (NDF), gn(r,k,t) exists,
which is different from the equilibrium Fermi distribution (C), and is defined in
general, for the nth band electrons, such that:

gn(r,k,t) drdk/4π3 is the number of electrons from the nth band at time t in the
semi-classical phase space volume (drdk) about the point (r,k)

- In equilibrium, gn(r,k,t) reduces to the Fermi distribution function f(En(k))!

where, f(E) = [1 + exp (E – μ )/kT]-1

But in the presence of applied fields and /or temperature gradients, it will differ
from point to point and from its equilibrium form!

Collisions: As defined earlier, the probability of collisions in an infinitesimal time

interval dt is dt/τ,

which now, is more generally defined as , τ = τn (r, k) (D)

But, the primary role of collisions ‘remains’ (same as Drude assumption) to drive the
electronic system towards a ‘local’ thermodynamic equilibrium!
 Relaxation Time Approximation - Assumptions
1. The distribution of electrons emerging from collisions at any time does not depend
on the structure of the non-equilibrium distribution function, gn(r,k,t) just prior to
the collision
2. If the electrons in a region about r, have the equilibrium distribution appropriate to
a local temperature, T(r) (within domains of size >> a),

go(r,k) = [1 + exp {En(k) – μ(r)}/kT(r)]-1 (E)

then collisions will not alter the form of the distribution function!

Notice, go differs from ‘f’ primarily due to the local dependence on r and k and
implies local thermodynamic equilibrium under steady state condition

Thus:
(i) collisions do not provide any information about the non-equilibrium distribution
function, and
(ii) the main role of collisions is to maintain local thermodynamic equilibrium
 Relaxation Time Approximation – formulation
These assumptions determine the form of dgn(r,k,t), which describes just those
electrons that have ‘emerged’ from collision near (r,k) in the time dt

According to Assumption (1), ‘dg cannot depend on the form of NDF, i.e., gn(r,k,t),
so ‘dg’ can be determined for any particular form of ‘g’

The simplest case is when ‘g’ has the local equilibrium form, as it is unaltered by
collisions, according to Assumption (2)

But, we know that in time interval dt, a fraction dt/τ of electrons in band ‘n’
with wave vector k, near position r, will suffer a collision that does alter their band
index and/or wave vector

So, if the local equilibrium described by gno (r, k) is to remain unaltered, then the
distribution of those electrons that emerge from collisions, into the band
n with wave vector k, during the same interval must compensate for this loss

Thus: dgn(r, k, t) = [dt /τn(r, k) ] gno (r,k) (F)

This represents the mathematical formulation of Relaxation Time Approximation

 Calculation of Nonequilibrium Distribution Function
Assume, dN = g(r,k,t) drdk/4π3 (index ’n’ is dropped) (1)
is the number of nth band electrons at time t in the phase-space element drdk
Let’s count these electrons by grouping them as to when they had their last collision!

Let r(t’), k(t’) be the solutions of semi-classical equations (A) and (B) that pass
through the point (r,k) at time t’ = t
Then the number of electrons that emerge from collisions into the volume
element dr’dk’ at r(t’), k(t’) during the time interval dt’ about t’ and were ‘aimed’ to
reach the point (r,k) at time t (if there was no more collision), according to RTA (F) is

[dt’/ τ(r(t’), k(t’))] go(r(t’), k(t’)) dr’dk’/4π3 (2)

Now if P(r, k, t; t’) is the fraction of these electrons, which actually survive without
collision from time t’ to t and reach the point (r,k) at time t, then multiplying (2)
with this probability and summing over all possible times for the last collision, i.e.,
t
dr’dk’/4π3
-
∫ [dt’/τ(r(t’), k (t’))] go(r(t’),k(t’)) P(r, k, t; t’) (3)
∞
gives dN (taking, dr’dk’ = drdk, as phase space volumes are preserved) at time ‘t’
 Calculation of NDF (cont’d)
t

g(r,k,t) = ∫-∞dt’ g (r(t’),k(t’)) P(r,k, t; t’) / τ (r(t’), k (t’))

o (4)

t
or, g(t) = ∫
-
dt’ go(t’) P(t; t’) / τ(t’) (4 is re-written, with the other variables hidden) (4a)
∞

Computation of P(t, t’):

The fraction of electrons that survive from t’ to t is lesser than the fraction that survive from
t’+dt’ to t by a factor [1 - dt’/ τ(t’)], which is the probability of not colliding during the interval dt’

i.e., P(t; t’) = P(t; t’+dt’) [1 - dt’/ τ(t’)]

which, in the limit, as dt’ → 0, leads to the differential equation

∂P(t; t’)/∂t’ = P(t; t’) /τ(t’) (5)

and which, under the boundary condition P(t; t) =1, leads to

t
P(t; t’) = exp [- ∫ dt”/τ(t”) ] (6)
t’
 Calculation of NDF (cont’d)
t
(4a) g(t) = ∫ dt’ go(t’) P(t; t’) / τ(t’)
-∞

(5) ∂P(t; t’)/∂t’ = P(t; t’) /τ(t’)

t
Substituting (5) in (4a): g(t) = ∫ dt’ go(t’) ∂P(t; t’)/∂t’
-∞

Integrating by parts and taking P(t; - ∞) = 0, since infinitely long survival is not possible!

t
g(t) = go(t) - ∫-∞dt’ P(t; t’) dgo(t’)/dt’ (7)

Replacing go(t) by go(r,k) and g(t) by g(r,k,t)

t
g(r,k,t) = go(r,k) - ∫ dt’ P(r,k, t; t’)
-∞
dgo(r(t’),k(t’))/dt’ (7a)

which expresses the non-equilibrium distribution function as the local equilibrium

distribution plus a time dependent correction
 Evaluation of dgo(t’)/dt’
Since go depends on time only through E(k(t’)), T(r (t’)) and μ(r (t’)), so

dgo(t’)/dt’ = (∂go/∂E)(∂E/ ∂k).(dk /dt’)

+ (∂go/∂T)(∂T/ ∂r).(dr /dt’)
+ (∂go/∂μ)(∂μ/ ∂r).(dr /dt’)

As v = dr/dt’ = (∂E/ ∂k)/ħ and go(r,k) = [1+ exp {(E(k) – μ(r)/kT(r)}]-1 (≡ f)

So, (∂go/∂E) = ∂f/∂E

∂go/∂T = ∂f/∂T = - (E – μ )/T (∂f/∂E)

and ∂go/∂μ = ∂f/∂μ = - ∂f/∂E

So from (7a): g(r,k,t) = go(r,k) - ∫- dt’ P(t; t’) dgo(t’) /dt’

∞

Or, g(r,k,t) = go(r,k) - ∫- dt’ P(t; t’) (∂f/∂E) v. [( F - ÑT (E – μ )/T ) - Ñμ ] (8)

∞
 The NDF expression
t

g(r,k,t) = go(r,k) + ∫- dt’ P(t; t’) [(- ∂f/∂E) v. (-eE - Ñμ – ÑT(E – μ )/T )] (8a)
∞

Bringing back all the variables/symbols,

gn(r,k,t) = gno(r,k) +
t

∫ dt’ Pn(r,k,t; t’)[(- ∂f/∂E) vn(k(t’)).(-eE (r,t’) - Ñμ(r,t’) – ÑT(r,t’)(En(k(t’)) – μ(r,t’))/T(r,t’)]

-∞ (9)

This is the general expression for gn(r,k,t), in which, the Fermi function is evaluated at
local temperature and chemical potential and the quantities in the brackets depend on
t’ through their arguments r(t’) and k(t’)

B does not appear explicitly, since the Lorentz force is perpendicular to v, but is
implicitly present through the time dependences of r(t’) and k(t’)
 Simplifications in NDF & Assumptions
(9) g(r,k,t) = go(r,k) +
t
∫dt P (r,k,t; t’) [(-∂f/∂E) v (k(t’)).(-eE (r,t’) - Ñμ(r,t’) - ÑT(r,t’) (E (k(t’)) - μ(r,t’))/T(r,t’)]
n n n
-∞
1. Weak electric fields and temperature gradients
Note that the contribution to integral from the time interval (t – t’) larger than τ is
negligibly small!
Within this time ~ τ, the electric field causes a ∆k, which is negligibly small compared
to the zone dimensions.
Similarly, the temperature gradients can be normally considered to cause a small
change of temperature within a mean free path
As the electric field and temperature gradients commonly applied in solids are weak,
so approximation to their linear order is justified in (9), which comes essentially
from the term containing E and ÑT explicitly
Hence, the t’-dependence of the integrand in (9) can be calculated at zero field and
constant T, i.e., k(t’), r(t’) and hence, v(k (t’)) can be treated as independent of E or ÑT

However, B cannot be similarly linearized, since, it can be strong enough to move

electrons through ∆k, comparable to zone dimensions (without change of E(k), within a
relaxation time, so, B is implicitly present through the time dependences of r(t’) and k(t’)
 Simplifications and Assumptions (cont’d)
(9) g(r,k,t) = go(r,k) +
t
∫dt’ Pn(r,k,t; t’) [(-∂f/∂E) vn(k(t’)).(-eE (r,t’) - Ñμ(r,t’) - ÑT(r,t’) (En(k(t’)) - μ(r,t’))/T(r,t’)]
-∞
2. If spatially uniform electromagnetic fields and temperature gradients as well as
position independent relaxation times are assumed, then:
The entire integrand becomes independent of r(t’) and the t’ dependence is only
through v(k(t’)) if B ≠ 0, P(t,t’) and E & T (in case, these are time dependent)

3. Energy dependent relaxation time: If τ depends on k only through En(k), which is

conserved in a magnetic field, then τ(t’) will also not depend on t’ through k(t’)
t
So in (6): P(t; t’) = exp [- ∫ dt”/τ(t”) ] - the integration becomes trivial
t’
and we get: P(t; t’) = exp [-(t – t’) / τ(E(k)]

Under these assumptions, in the presence of a magnetic field (B ≠ 0):

g(k,t) = go(k) +
t
∫ dt’ [exp {-(t – t’)/τ(E(k)}] (- ∂f/∂E) v(k(t’)). [-eE(t’) - Ñμ(t’) -ÑT(t’)(E(k) – μ(t’)/T(t’))]
-∞ (10)
 NDF under RTA (B = 0)
(10) g(k,t) = go(k) +
t

∫ dt’ [exp {-(t – t’)/τ(E(k)}] (- ∂f/∂E) v(k(t’)). [-eE(t’) - Ñμ(t’) -ÑT(t’)(E(k(t’)) – μ(t’))/T(t’)]
-∞

However, in the absence of B, v(k(t’)) = v(k)

And for time-independent E and ÑT, the time integration in (10) is simple, leading to

g(k) = go(k) + τ(E(k) (- ∂f/∂E) v(k) . [- eE - Ñμ – ÑT (E(k) – μ)/T] (11)

This is the expression for the non-equilibrium distribution function, under relaxation time
approximation, in the presence of time-independent applied electric field and
temperature gradient, but no magnetic field

DC electrical conductivity:
For constant T (Ñ T= 0, Ñμ = 0), (11) leads to only the effect of electric field
g(k) = go(k) - eE . v(k) τ(E(k)) (- ∂f/∂E) (12)
defines dc electrical conductivity!
 DC Electrical Conductivity
(12) g(k) = go(k) - eE . v(k) τ(E(k)) (- ∂f/∂E)

As the number of electrons per unit volume in the volume element dk is g(k) dk/4π3,
the current density is

j = - e ∫v(k) g(k) dk/4π3 = (-e2/4π3) ∫ v(k) [v(k) . E)] τ(E(k)) (∂f/∂E) dk

- The integration is over a primitive cell and represents the contribution to current
density from the (nth) partially filled band under consideration
- Notice that go does not contribute, since it represents the equilibrium Fermi
distribution, under which no current flows
If the volume element dk is replaced by dS dkn = dS dE/|ÑkE|,
then the integration over k-space can be replaced by surface integration over a
constant energy surface, followed by integration over the entire energy range, i.e.,

∞
j = (- e2/4π3) ∫o ∫S E v(k) [ v(k) . E ] τ(E(k)) (∂f/∂E) dS dE/ |ÑkE|
Or,
∞
j = (- e2/4π3) ∫o { ∫S E v(k) [ v(k) . E ] τ(E(k)) dSE } (∂f/∂E) dE/ |Ñ E|
k (13)