You are on page 1of 95

ECE344

Semiconductor Devices and Materials


M. Fischetti
201 D Marcus Hall
Department of Electrical and Computer Engineering
University of Massachusetts
Amherst, MA 01003

Fall 2009
The crisis of Classical Mechanics

At the end of the XIX century classical physics consisted of two main areas:

1. Classical Mechanics: Developed by Galileo and Newton, it predicted the motion of celestial objects with
astonishing accuracy. It was also applicable (accounting for always present frictional effects) to the motion of
projectiles.
2. Classical Electromagnetism: Starting from the experiments of Coulomb, Ampère, Faraday, Lenz, etc., it
culminated with the mathematical formulation by Maxwell and the prediction that electromagnetic waves are
light (Hertz).

Both areas were highly successful in most cases in explaining experimental observations. Yet, at the threshold of
the XX century a few problems began to emerge, leading to the transition from ‘classical physics’ to ‘modern (or
quantum) physics’. The following were the major ‘puzzles’ which classical physics could not explain:

• Black Body spectrum: The ultraviolet catastrophe.

All objects emit ‘thermal radiation’, dependent on their temperature. Think of a wood log glowing red-to-white
in the fireplace, or of molten iron, etc. Roughly speaking, the composition of the object does not matter, only
its temperature: The hotter the body, the ‘bluer’ the body will appear.
Many objects do not reflect light (think of a chunk of coal) or reflect it poorly. Empirically it is found that the
light emitted by this class of bodies is almost completely independent of the details of their composition.
Thus, the concept of ‘black body’ idealizes this class of objects. It is defined as a body which absorbs light
(i.e., electromagnetic radiation), but does not reflect any of it, instead re-emitting it after having ‘thermalized’
it. The ideal model would be a black box in which radiation enters through a tiny hole. Electromagnetic wave
hitting the internal walls of the box reach equilibrium with the environment. Its spectrum (i.e., number of waves
– or ‘modes’ – n(ν)dν in an interval ν -ν + dν of frequencies ν ) is measured as light exits from the tiny hole.

ECE344 Fall 2009 1


– Classical electromagnetic theory and equipartition (i.e., the energy is equally distributed among all modes, no
matter what their frequency is) imply that the spectrum of a black-body should diverge at short wavelength.
That is: if n(ν) is density of modes with frequency in ν , ν + dν , then n(ν) ∝ ν 2 (Jeans’ law)
– Experiments show, instead, n(ν) ∼ e−aν/(kB T )
– Planck obtained the experimental spectrum assuming that:
1. the energy in each mode proportional to the ν of mode: Eν = hν with h a new (Planck) constant of
nature
2. energy is exchanged between waves and the walls of the black-body only via ‘chunks’ (‘quanta’) of size
∆Eν = hν
Planck’s assumption worked, but it lacked any justification!

classical theory (Jeans’ law)


ρ(ν) (10–17 joules/m3–Hz)

experimental result (T=1500K)

0
0 1 2 3 4
ν (1014 Hz)

ECE344 Fall 2009 2


• The photoelectric effects: Nothing works correctly...!

– Hitting a solid with light, electrons should be emitted with energy proportional to the intensity of the light
(since it’s proportional to |E|2 )...
– Instead, the electron energy grows with decreasing wavelength of the light and no electrons are emitted for
ν < ν0 , where ν0 is a function of the solid.
The light-intensity determines only the number of emitted electrons (Millikan, 1914)
– Einstein explained the experimental results with Planck’s ‘quanta’ (1905, Nobel prize in 1921). These ‘quanta’
stuff seems to be for real...
3
glass enclosure Millikan’s experiment (1914)

Stopping potential V0 (V)


grid light–intensity a =
I twice light–intensity b
metal 2

Ia a

LIGHT 1
electrons
ν0
Ib b

stopping potential V0 0
0 2 4 6 8 10 12
applied V V ν (1014/sec)

• Structure and spectra of atoms: Why don’t electrons fall into the nucleus?

– After Rutherford’s experiments (1913), atoms were viewed as electrons orbiting the nucleus. [Rutherford shot
radiation known as ‘α particles’ (now known to be He nuclei) at a very thin gold foil. Most α particles went through essentially
undeflected, but those who were deflected were scattered almost backwards. This showed that atoms are essentially ‘empty’ but
have a massive ‘core’ (or ‘nucleus’).] But Maxwell equations predicted that the electrons should radiate their energy
and fall into the nucleus.

ECE344 Fall 2009 3


– Planck’s quanta can explain this as well, although more is needed to explain the spectra: Experimentally it
is found that atoms somehow ‘excited’ (em e.g., by an electrical discharge, as in fluorescent light) emit light
only at some discrete wavelength (or frequencies). For example, hydrogen three major ‘series’ of lines obeying
the following empirical relationships (where R = 109,678/cm is the ‘Rydberg’ constant):
1. Lyman:  
1 1
ν = cR − 2 whith n = 2, 3, 4, ... (1)
12 n
2. Balmer:  
1 1
ν = cR − 2 whith n = 3, 4, 5, ... (2)
22 n
3. Paschen:  
1 1
ν = cR − 2 whith n = 4, 5, 6, ... (3)
32 n
• There’s more (Compton scattering, double-slit experiment,...). See any textbook on QM.
• Partial conclusion:
Light behaves sometimes as wave (after all, Maxwell equations are valid in most cases), sometimes as a particle
(photon)... Confusing...

ECE344 Fall 2009 4


Brief introduction to Quantum Mechanics

• de Broglie’s assumption to explain atomic spectra (among other things):


Why not assume that all forms of matter behave as particles or waves, depending on the situation? For a
particle of mass m the wavelength λ of the associated ‘pilot wave’ will be give by:

h
p = mv =
λ

This may explain discrete electron ‘orbits’ around the nucleus...


• How?. Consider how de Broglie’s suggestion might have explained some features of the hydrogen atom.
a. Show that the de Broglie assumption
h
p = mv = (4)
λ
and the ‘quantization condition’ that the circular orbit be an integer multiple of the length of the electron wavelength (that is
nλ = 2πr, where r is the radius of the orbit, n an integer) imply that only discrete orbits are allowed.
b. Calculate the total energy (kinetic plus potential) in each orbit characterized by an integer n.
c. Show how the results of part b explain the emission spectrum of hydrogen.
Hint: In part a find two equations, the first one describing the balance between the centrifugal and the Coulomb (centripetal) force.
Solve for the radius r and the angular velocity ω . b. Insert these expressions into the formulae for kinetic and potential energy.
Solution.
a. For uniform circular motion with angular velocity ω and radius r, the magnitude of the centrifugal force is mω 2 r. This must be
equal to the magnitude of the Coulomb force:
2 e2
mω r = , (5)
4π0 r2
having used MKS units (so 0 ≈ 8.85 × 1012 Coulomb/(Volt meter) is the dielectric constant of the vacuum).
The requirement that an integer number (say, n) of de Broglie wavelengths must ‘fit’ into an orbit implies, by Eq. (4):

nh
2πr = , (6)
mωr

ECE344 Fall 2009 5


since the velocity of the electron is v = ωr. From Eq. (5):

2 e2
ω = . (7)
4π0 mr3

From Eq. (6):


nh̄ 2 n2h̄2
ω = → ω = , (8)
mr2 m2 r 4
(where h̄ = h/(2π)). Equating Eqns. (7) and (8):

e2 n2h̄2 e2 n2h̄2
= → = ,
4π0 mr3 m2 r 4 4π0 mr

or:
4π0 n2h̄2
rn = , (9)
me2
where the subscript n has been added to show that only discrete values of the orbital radius are allowed. This was de Broglie’s main
goal.
b. Squaring this equation and inserting it into the first of the equations (8):

me4
ωn = , (10)
(4π0 )2 n3h̄3

where again the subscript n shows that only discrete angular velocities are allowed. The total energy will be:

1 2 r2 − e
En = mωn n .
2 4π0 rn

Multiplying Eq. (5) by r, the second term in the right-hand-side (rhs) of this equation becomes mω 2 r2 , so:

1 2 2
En = − mωn rn .
2
In words, the total energy is one half of the potential energy. This is a particular example of a general theorem of classical mechanics

ECE344 Fall 2009 6


called ‘virial theorem’. Now, using Eq. (8) to express ω 2 in terms of r in the equation above:

n2h̄2
En = − .
2mrn2

Finally, using Eq. (9) to eliminate rn :


me4
En = − .
2n2h̄2 (4π0 )2
So, de Broglie’s suggestion would have resulted in discrete energy levels for the electron in the hydrogen atom. Their magnitude is
surprisingly identical to the energy levels obtained from an exact solution of the Schrödinger equation. In particular, the energy of the
ground state (n = 1),
me4
En=1 = − ≈ − 13.1eV
2h̄2 (4π0 )2
is the ionization energy of hydrogen.
If we assume that H atoms emit light when the electron jumps from an orbit n to another orbit m at lower energy (i.e., En > Em ,
or n > m) the frequency of the emitted light will be given by:
    
me4 1 1 me4 1 1
hν = En − Em = − − → ν = c − . (11)
2h̄2 (4π0 )2 n2 m2 2h̄(4cπ0 )2 n2 m2

The factor in red is the Rydberg constant R, so, comparing Eq. (11) with Eqns. (1)-(3) above, we explain the Lyman, Balmer, and
Paschen series as the discrete wavelength of the light emitted by an electron falling from an orbit n to the orbit m=1 (Lyman), m=2
(Balmer), or m = 3 (Paschen).

• Schrödinger equation: How to explain structure of atom as a ‘boundary value’ problem.

Start from the classical energy (often called ‘Hamiltonian’ H ), kinetic plus potential:

p2 e2
H = − ,
2m 4π0 r

ECE344 Fall 2009 7


and ‘assume’ (or ‘guess’):
1. electrons (as well as all other particles) are associated to de Broglie’s waves
2. ‘discrete’ levels emerge as ‘eigenvalues’ of a wave equation (like standing waves on a bathtub, on the surface
of a drum, on a violin string or organ pipe...
3. the energy H must be replaced by ih̄∂/∂t,
4. the momentum p must be replaced by −ih̄∇ .
Then obtain the wave-equation for the ‘wavefunction’ Ψ(r, t):
 
2 2 2
h̄ ∇ e ∂Ψ(r, t)
HΨ(r, t) = − − Ψ(r, t) = i h̄ .
2m 4π0 r ∂t

If H is time-independent, set Ψ(r, t) = ψ(r)e−iE/h̄t with:

h̄2 2
− ∇ ψ(r) + V (r)ψ(r) = Eψ(r)
2m

(time independent Schrödinger equation)


• What’s the wavefunction? The Copenhagen interpretation (Bohr, Born)

After years of struggle and confusion, a group of physicists working at the Institute of Theoretical Physics in
Copenhagen (founded by Niels Bohr) proposed the following interpretation.

|Ψ(r, t)|2 dr = ρ(r, t) is the probability of finding an electron in a volume dr around r at time t. Of course:
 
ρ(r, t) dr = |Ψ(r, t)|2 dr = 1

ECE344 Fall 2009 8


and from Schrödinger equation one can show that:

∂ρ
+ ∇·S = 0,
∂t

where S = 2mih̄ [Ψ∇Ψ∗ − Ψ∗ ∇Ψ] is the ‘probability density current’

• A simple example: An 1D electron in a constant potential V (z) = V0 :

h̄2 d2 ψ(z)
− = (E − V0 )ψ(z) .
2m dz 2

Put ψ(z) = eikz and find k = ±[2m(E − V0 )]1/2 /h̄ or E = h̄2 k2 /(2m), so that h̄k is the z-component
of the electron momentum.
• Another simple example: Particle in a box.
Potential of a ‘quantum well’:

 ∞ for z ≤ 0
V (z) = 0 for 0 < z < L .

∞ for z ≥ L

Boundary conditions:

ψ(z) = 0 for z = 0 and z = L ,


since the electron cannot be in the ‘forbidden’ regions. Schrödinger equation:

h̄2 d2 ψ(z)
− = Eψ(z) for − L/2 < z < L/2 ,
2m dz 2

ECE344 Fall 2009 9


with solutions sin(kn z) only for discrete values of kn = πn/L:
   
2 1/2 nπz
ψ(z) = sin
L L

2 2 2
with En = n h̄ π2 .
2mL

The energy depends on the boundary conditions at L, like the note emitted by a vibrating violin string depends
on where one places the finger.
• A tunneling problem: Potential of a ’barrier’:

 0 for z ≤ 0
V (z) = V >0 for 0 < z < L .

0 for z ≥ L

General solution for electron of energy E < V coming from the left, partially reflected back and partially
transmitted: 

 Aeikz + Be−ikz for z ≤ 0
ψ(z) = Ceκz + De−κz for 0 < z < L ,

 F eikz for z ≥ L
with k = (2mE)1/2/h̄, κ = [2m(V − E)]1/2 /h̄.
Using continuity of ψ(z) and dψ(z)/dz at z = 0 and z = L one can find the coefficients A, B , C , D, and
F.
The ratio T = |F |2 /|A|2 is the ‘transmission coefficient’, the ratio R = |B|2 /|A|2 is the ‘reflection
coefficient’. Note that, classically, one would expect T = 0.
This is the first surprise of Quantum Mechanics: Particles can populate regions which are classically forbidden!

ECE344 Fall 2009 10


• More on the Copenhagen interpretation.

1. Quantum Mechanics has an intrinsic ‘probabilistic’ nature, different from all ‘classical’ physical theories. Just
consider the interpretation of the squared wavefunction |Ψ(r, t)|2 as the ‘probability’ of finding the particle
at position r at time t. So, we cannot say with precision where the particle will be at any given time, but we
can only define an ‘average’ position r:

r = Ψ∗(r, t) r Ψ(r, t) dr , (12)
V ol

(recall that the wavefunction is assumed to be normalized, that is, V ol Ψ∗(r, t) Ψ∗ (r, t) dr = 1), which
is called the ‘expectation value of the position r on the state Ψ. Similarly, we can define the expectation value
of the momentum,
 

p = Ψ (r, t) p Ψ(r, t) dr = −ih̄ Ψ∗ (r, t) ∇ Ψ(r, t) dr . (13)
V ol V ol

Note that for a free particle,



Ψ(r, t) = exp[i(k · r − ωt)], so that the expectation value of the momentum
is simply h̄k (since V ol Ψ∗(r, t) Ψ∗(r, t) dr = 1), while the expectation value of the position is not
defined: The particle is everywhere with equal probality, since |Ψ(r, t)|2 is constant, independent of r!
2. This is a general properties of many pairs expectation values: We can define precisely the value of one physical
quantity (the momentum, in our example above) at the expense of the other (the position). If we define the
‘uncertainty’ of a quantity as its root-mean-square variance,

2 ∗ 2
∆r = Ψ (r, t) [r − r] Ψ(r, t) dr , (14)
V ol

and similarly for the momentum,



2
∆p = Ψ∗ (r, t) [p − p]2 Ψ(r, t) dr , (15)
V ol

ECE344 Fall 2009 11


one can show that, in general,
∆r ∆p ≥ h̄ , (16)
which is the famous ‘Heisenberg uncertainty principle’.
3. The above example applies to the pair of ‘complementary variables’ position and momentum. Similar relations
apply to angular position and momentum, components of the angular momentum along different axes, and, in
a slightly different but well ‘publicized’ way, energy and time:

∆t ∆E ≥ h̄ , (17)

which states that we can violate energy-conservation for short-enough times.


4. Therefore, in general, according to the Copenhagen interpretation, we cannot talk of ‘values’ taken by an
observable A, but only of its ‘expectation value’,

 A = Ψ∗(r, t) A Ψ(r, t) dr . (18)
V ol

In some cases, the state Ψ of the particle will be such that one variable is precisely defined (such as the
momentum h̄k for a free particle), but the position will be undefined.
• Atoms: Building up the Periodic Table
We see in this wection how the Schrödinger equation – together with a few additional concepts – can allow us
to build (conceptually) the periodic table of the elements. In turn, this will give us some ideas about atomic
orbitals, which constitute the basic concepts underlying the formation of atomic bonds in molecules and crystals
(which is what we are after).
First, we need the concept of quantum number. The example of a particle in a three-dimensional (3D) box will
illustrate this.
– A little bit of math: A free 3D particle.
We have seen before how we can solve the Schrödinger equation in a couple of simple one-dimensional cases.

ECE344 Fall 2009 12


Let’s now consider a 3D case. The time-independent equation we wish to solve is:

h̄2 ∇2
− ψ(r) + V (r) ψ(r) = E ψ(r) . (19)
2m
Let’s assume that the potential is equal to a constant, V (r) = V0 , and let’s express the Laplacian and the
spatial coordinate r in terms of their components:
 
2 2 2 2
h̄ ∂ ∂ ∂
− ψ(x, y, z) + ψ(x, y, z) + ψ(x, y, z) + V0 ψ(x, y, z) = E ψ(x, y, z) .
2m ∂x2 ∂y 2 ∂z 2
(20)
A standard way to solve a partial differential equation of this type is to separate the variables, that is, to look
for a solution of the form:
ψ(x, y, z) = ψx (x) ψy (y) ψz (z) , (21)

where ψx , ψy , and ψz are three different functions in one variable only. Let’s now set V0 = 0 (which we can
always do, since the zero of the potential energy is arbitrary) and insert this into Eq. (20):
 
2 2 2 2
h̄ d ψx (x) d ψy (y) d ψz (z)
− ψy (y) ψz (z) + ψx (x) ψz (z) + ψx (x) ψy (y) = E ψx (x)ψy (y)ψz
2m dx2 dy 2 dz 2
(22)
Let’s now divide both sides by ψx (x)ψy (y)ψz (z):
 
2 2 2 2
h̄ 1 d ψx (x) 1 d ψy (y) 1 d ψz (z)
− + + = E. (23)
2m ψx (x) dx2 ψy (y) dy 2 ψz (z) dz 2

This equation tells us that the sum of 3 functions of different variables (x, y , and z ) must always be equal
to a constant (E ) for all possible values of x, y , and z . This is possible only if each function is equal to a

ECE344 Fall 2009 13


constant, that is:
h̄2 1 d2 ψx (x)
− = Ex , (24)
2m ψx (x) dx2
h̄2 1 d2 ψy (y)
− = Ey , (25)
2m ψy (y) dy 2
h̄2 1 d2 ψz (z)
− = Ez . (26)
2m ψz (z) dz 2
We have indicated with Ex , Ey , and Ez three arbitrary constants subject to the condition Ex +Ey +Ez = E .
We can solve each of these equations as we have done above (page 9) in the case of the free 1D particle.
Considering Eq. (24), for example, let’s write it as:

h̄2 d2 ψx (x)
− = Ex ψx (x) , (27)
2m dx2
which has a general solution
ik x
ψx (x) = Ax e x , (28)
where Ax is an arbitrary multiplicative constant (which we shall fix later by some ‘normalization’ procedure)
and
1
kx = 2mEx . (29)

Similarly for the y and z components, so that the full wavefunction has the form

ψ(r) = Ax Ay Az ei(kx x+ky y+kz z) = A eik·r , (30)

having defined the wavevector k as the vector with components (kx , ky , kz ) and A is an arbitrary constant.
Equation (30) describes the particle as a ‘plane wave’ with wavevector k. Note that its momentum p will be
h̄k (as we saw in 1D), and the energy of the particle is E = h̄2 k2 /(2m).
– A 3D particle in a box.
Let’s now go back to Eq. (19) or (20), but now let’s assume that the particle is confined in 3D cubic box of

ECE344 Fall 2009 14


sides L (e.g. the potential vanishes for −L/2 < x < L/2, −L/2 < y < L/2, −L/2 < z < L/2, but
it is infinite otherwise). Thus, we should look for solutions ψx (x), ψy (y), and ψz (z) similar to what we saw
above (page 10):
  1/2  
2 nx πx
ψx (x) = sin
L L
where nx is an integer, similar expressions being valid for y and z . The full wavefunction will now be:
  1/2      
8 nx πx ny πy nz πz
ψ(r) = sin sin sin , (31)
V L L L

where V = L3 is the volume of the box and the energy of the particle will be

(n2x + n2y + n2z )h̄2 π2


Enx ,ny ,nz = .
2mL2

Note that the normalization constant A = (8/V )1/2 in this case has been fixed by requiring that there is
unit probability of finding the electron inside the box:

dr |ψ(r)|2 = 1 .
V

Note that in 1D the particle has only one degree of freedom and we need only a single ‘quantum number’ n to
label the energy levels. In 3D the particle has 3 degrees of freedom and we now need 3 quantum numbers, nx ,
ny , and nz to label the possible states (that is, the possible ‘types’ of wavefunction) of the particle.
• The hydrogen atom.
The case of the H atom is conceptually similar to the case of the particle in a cubic box. Unfortunately, it is quite
more complicated from a mathematical perspective, because the potential confining the particle (the Coulomb
potential energy is −e2 /(4π0 r)) is not ‘flat’, but it depends on the distance r from the nucleus. Therefore,
we shall not go into a detailed solution of the Schrödinger equation, but we shall consider only its qualitative

ECE344 Fall 2009 15


features. In this case the Schrödinger equation is still separable in polar coordinates (r, θ, φ), thanks to the
fact that the potential does not depend on the angles θ and φ, but only on r . Thus, the full wavefunction can
be written as:
ψ(r) = ψ(r, θ, φ) = Rn,l (r) Θl,m(θ) Φm (φ) . (32)

The ‘radial function’ Rn,l (r) describes the way the wavefunction spreads away from the nucleus. The ‘angular
functions’ Θl,m (θ) and Φm (φ) describe how the wavefunction is distributed as a function of the polar and
azimuthal angles.
As we saw before in the case of the 3D particle in the cubic box, we need 3 quantum numbers to label the
possible energy levels: The ‘principal’ quantum number, n, can take the values:

n = 1, 2, 3, ... ∞ ,

and is related to the number we saw when dealing with the simple semiclassical model (pages 5-7). Usually
(that is, in the absence of magnetic fields and ignoring smaller effects) the electron energy depends only on this
number via an expression identical to what we saw on page 7. The angular quantum numbers l and m are
related to the angular momentum (h̄l is related to its magnitude and h̄m to its component along the polar
aixs) of the ‘orbiting’ electron and can take the values:

l = 0, 1, ... n − 1 , (n total number of values)

m = −l, −l + 1, ...0, 1, ... l − 1, l , (2l + 1 total number of values) .


For l = 0, we see that we must also have m = 0. In this case the wavefunction does not depend on the
angles θ and φ. These are fully spherically symmetric wavefunctions (for every n) and are called s-waves.
For l = 1 we can have three different wavefunctions for m=-1,0,+1. These are called p-waves and can be
expressed as functions shaped like ‘lobes’ pointing along the x, y , or z axis. For l = 2 we can have 5 states
(m=-2,-1,0,+1,+2), called d-waves. For l = 3 we have 7 f -waves, and so on.
• Spin and Pauli’s exclusion principle.
Before building atoms, we need to know two additional facts:

ECE344 Fall 2009 16


1. In addition to the 3 degrees of freedom that all particles have in 3D, most particles also have an additional
‘internal’ degree of freedom, s, which one can visualize as a ‘spin’ around a polar axis, much like to spinning
top or the planets. Thus, spin is an internal angular momentum of the particle. As we just saw for the orbital
angular momentum, also spin can take only discrete values in integer or semi-integer multiple of the reduced
Planck constant. Photons, some nuclei and some elementary particles have integer values of spin (s=0 or
h̄). For reasons we shall see later, these particles are called ‘Bosons’. Electrons, protons, and neutrons have
half-integer spin (s = h̄/2). They are called ‘Fermions’. Every electron can exists in 2 different spin states:
Spin ‘pointing up’ (s = h̄/2) or ‘pointing down’ (s = −h̄/2). Both are states with angular momentum of
magnitude h̄/2, but they differ in the direction of rotation.
2. In order to explain the periodic table and the structure of the atoms, the German physicist Wolfgang Pauli
had to invoke a new postulate (later demonstrated rigorously): Given an energy level, it cannot be occupied
by two or more Fermions (so, electrons) with identical quantum numbers. This is called ‘Pauli’s exclusion
principle’. In our case it tells us that a given atomic orbital labeled by the quantum numbers (n, l, m) can
be occupied at most by two electrons with opposite spin. Attempting to add a third electron would violate
Pauli’s principle.
• The periodic table.
We are now ready to build the elements. We shall do it as if we were building an onion, one ‘shell’ at the time, from
the inside out. We shall call ‘shell’ each layer defined by a common quantum number n, ‘subshells’ those with
common angular quantum number l. The word ‘orbital’ or ‘states’ denotes any combination of quantum number.

Let’s start with the simplest element, H. Its nucleus has a single positively charge proton and it is surrounded by
a single electron. The lowest possible orbital the electron can occupy is the ‘ground state’ (n=1, l=0, m=0).
We use the notation 1s1 , meaning that in the orbital 1 (n=1) which is an s-wave (l=0) we have one electron
(the superscript ‘1’). Note that the electron is bound rather weakly to the nucleus (we saw that the ionization
energy is about 13.6 eV). Therefore, H loses its electron quite easily (that is, it is easlily ionized). This trait is
common to all elements with only one electron in the outermost shell. These elements are very reactive (think
of hydrogen mixing with oxygen!)

ECE344 Fall 2009 17


The next simplest element is He. Its nucleus now has two protons (so that the ‘atomic number’ of He is
Z =2). However, two neutrons are required to avoid that the Coulomb repulsion between the protons causes a
disintegration of the nucleus itself. Two electrons orbit the nucleus and both are in the lowest energy ‘shell’,
provided they align their spins in opposite directions. We use the notation 1s2 . Note that we have fully
populated the first ‘shell’. The addition of another electron must cause the occupation of the next higher-energy
subshell n=2, because there is no more room in the n=1 shell. Thus, the third electron would have to go
farther away from the nucleus. So, He does not like to have another electron. On the other hand, stripping one
electron away from the He atom is very expensive in terms of energy required to do it, since both electrons sit
quite close to the nucleus. Therefore, helium does not ‘like’ to lose electrons (that is, becoming ionized). He,
like all elements which have a fully occupied shell, is chemically inert.

The third element, Li, has three protons in its nucleus (Z =3), so it must have 3 electrons orbiting it in order to
be charge neutral. The third electron, as we just saw, must go into the n=2 shell, so Li will have the electronic
configuration 1s2 2s1 . It is a very reactive metal, for the reasons we discussed dealing with H: It is very easy to
strip away the lone outer electrons, leaving behind a positive Li ion.

We now keep populating the n=2 shell, adding another electron in the s-orbital (berillium, Be, 1s2 2s2 with
Z =4), then 6 more into the 3 p orbitals, adding electrons with spin up and spin down in sequence. As we
move from the reactive metals (Li, Be) we build elements with 3 ‘outer’ electrons (boron, B, 1s2 2s2 2p1), 4
outer electrons (carbon, C, 1s2 2s22p2 ), 5 (nitrogen, N, 1s22s2 2p3 ), 6 (oxygen, O, 1s2 2s2 2p4 ), and 7 outer
electrons (fluorine, F, 1s2 2s2 2p5 ). As we do so, we move from elements which like to lose electrons, to those
who like to gain them. Eventually, with neon (Ne, Z =10, 1s2 2s2 2p6 ) we complete the n=2 shell and we hit
another inert gas, like He unlikely to react chemically. The 8 slots we have filled while populating the n=2 shell
are the 8 columns of the periodic table. The chemical properties of any two elements having the same number
of electrons in the outer shell are very similar.

So, as we fill the n=3 shell, we start from sodium (column I, Na, Z =11, 1s2 2s2 2p63s1 ) which, having
only one electron in the outer shell, behaves like Li. And we end up with argon (column VIII, Ar, Z =18,

ECE344 Fall 2009 18


1s22s2 2p6 3s2 3p6 ) which is another inert gas like He and Ne.

Things get a little more complicated now, because the energy of the 4s shell is lower than that of the 3d shell.
So, with potassium (K) and calcium (Ca) we fill the 4s shell, but then we go back filling the 3d shell. The fact
remains that – up until further complications caused by the f states of transiton metals – each element belongs
to its own column which determines the chemical properties vua the number of electrons in the outer shell.
Of interest to us are silicon (Si, Z =14, 1s22s2 2p6 3s2 3p2 ) and germanium (Ge, Z =32,
1s22s2 2p6 3s2 3p6 3d104s2 4p2 ) which, like C, belong to the IV column. Their 4 outer electrons are
s2 p2 and can easily ‘hybridize’ into 4 sp3 orbitals, forming the tetrahedal structure required to bond atoms in
their cubic crystal form. Elements of column III (B, Al, Ga, In) have outer electrons arranged as s2 p1 , that is,
one fewer electron than in Si and Ge. Elements of column V (N, P, As, Sb) have outer electrons arranged as
s2 p3 , that is, one more electron than in Si and Ge.

Finally, electrons in fully occupied shells are called ‘core electrons’. They do not contribute to the chemical
activity of the element. Electrons in the outer shell are called ‘valence electrons’.
• Molecules and Bonds.
If we bring several atoms together, several things may happen, depending on the electronic populations of the
outer shell.
1. Nothing. For example, bringing 2 Ne atoms together, both atoms retain their electronic configuration
unaltered, as they are both in a configuration of a completely filled outer shell with 8 electrons (2 in s orbitals,
6 in p orbitals). This is why elements of the VIII column are called ‘inert’ elements (or gases).
2. Formation of an ionic bond. Let’s consider Na and Cl. Na has 1 electrons in the outer s shell, Cl 7 valence
electrons (2 in the s shell, 5 in the p shell). Energetically it is advantageous for Na to release its electron
(weakly bound, as the core electrons shield the nuclear charge), while Cl ‘loves’ to acquire that electron and
complete the filling of its outer shell. Thus the sodium atom becomes positively ionized (Na+ ) and the clorine
atom becomes negatively ionized (Cl− ). Now we have a positive (Na+ ) and negative (Cl− ) particle which are
attracted by the strong Coulomb force. The two ions are said to have formed an ‘ionic bond’ and stick together
in the NaCl molecule (regular kitchen salt). Ionic bonds are usually quite strong with ‘binding energies’ of

ECE344 Fall 2009 19


several eV. Semiconducting crystals like GaAs (Ga in column III, As in column V, so this is a so-called III-V
compound semiconductor) are bound via ionic bonds. Water (H2 O) is similarly formed, although now the
charge-transfer takes place among 3 atoms: The 2 H atoms give their electrons to O which has 6 valence
electrons, and so it fills its outer shell. Thus we have two positively charged H+ bound to a doubly ionized
O2− ion. Ionic bonds are usually quite strong with ‘binding energies’ of several (of the order of 1-to-10) eV.
3. Formation of covalent bond. Let’s consider two oxygen atoms brought together. They are in the 2s2 2s4
configuration in the outer shell. They would both ‘love’ to fill their outer shells by adding two electrons. They
can do that by sticking together and sharing each two outer electrons, thus forming the O2 molecule. They
form a ‘covalent bond’. Si and Ge are a little more complex. They have 4 electrons in their outer shells
(s2 p2 ). They can ’hybridize’ to form 4 sp3 orbitals. While the Si2 molecule cannot exist, several Si atoms
can form a ‘network’ of tetragonal bonds and form a covalent crystal. Covalent bonds are the strongest bonds.
4. Metallic bond. This happens when we bring together a large number of metallic atoms (say, Li or Be). In this
case the single electrons belonging to each one of the many atoms become ‘shared’ among all ions. The ions
form a ‘lattice’ (as we shall see below dealing with crystals), the shared electrons keeping the lattice together.
The metallic bond has a strength comparable to that of ionic bonds.
5. Hydrogen bond. This is still a poorly understood bond present in water-ice, and, for example, binding the
bases to the chain of DNA molecules. An H atom between two molecules acts as a bridge binding (somewhat
weakly) the molecules together, as its lone electrons ‘resonates’ between them.
6. Van der Vaals force. Although not relevant to the structure of crystals, another type of interaction affects
solids and some relatively chemically inactive molecules. When brought together, rather than exchanging or
sharing electrons, atoms or molecules ‘polarize’ their electronic clouds. The resulting interaction is between
electric dipoles with opposite orientation. The force decays very quickly with distance. This is an interaction
typical of flat surfaces (think of two sheets of glass) and polymers.

ECE344 Fall 2009 20


2 shared –
O
Na+ p electrons
H

Cl– O –

IONIC BOND COVALENT BOND VAN DER VAALS HYDROGEN


As we saw, the chemical properties of the element are fully determined by the valence electrons. The number of
electrons in the outer shell also determines the chemical valence of the atom. As a general rule, electrons with
fewer than 3 electrons in the outer (incomplete) shell have a chemical valence equal to the number of electrons.
For example, Li (one electron in the outer shell) has valence I, Ga (3 electrons in the outer shell ) has valence III.
If the outer shell contains 6 electrons or more, the valence of the element is 8 minus the number of electrons.
Elements with a more complicated electronic structure of the outer shell(s) can exhibit several different chemical
valence numbers.

Finally, it is important to realize that during the formation of a bond the electronic orbitals change their structure.
For example, during the formation of the O2 molecule the electrons ‘shared’ by the two oxygen atoms will ‘orbit’
both cores (nuclei+core electrons), forming new orbitals by a ‘linear superposition’ of the wavefunctions of the
2p orbitals of each atom. As shown in Fig. 3.2 of the text, out of these two ‘single atom’ orbitals we form two
molecular orbitals: A lower-energy ‘bonding orbital’ and a higher-energy ‘anti-bonding orbital’, respectively at a
slightly lower and higher energy then the original single-atom orbitals. The bonding orbital is occupied by the
two shared electrons and it exhibits a larger charge density (that is, a larger |ψ|2 ) between the atoms. In other
words, there is a large probability of finding the electrons between the atoms. Thus, this orbital contributes to
keeping the atoms together. The antibonding orbital is empty and it is associated with a charge density (again,
|ψ|2 ) which is larger away from the bond.

ECE344 Fall 2009 21


Crystals

The main purpose of this course is the study of how current (carried by electrons) flows in semiconductor
devices. Therefore, it is important to understand the properties of the electron motion in semiconductors.
Semiconductors, as we shall see later, are materials which have current-conduction properties intermediate between
insulators (glass, wood) and conductors (copper, aluminum). Their practical importance originates from two major
properties:

1. The addition of a minute amount of impurities in perfect semiconductors (at the level of a few parts per million)
can modify their current-conduction properties by huge amounts, transforming what is essentially an insulator
into a pretty good conductor. This property is extremely useful in designing ‘devices’ which use the changing
conductor/insulator properties of the material to make tiny electronic ‘switches’ (on/off, 0/1 in binary logic).
2. Their electronic structure (as we’ll see in the next section) includes the presence of an “energy gap” which
behaves like the energy step between two energy levels in an atom. This determines the ‘optical properties’ of the
material. That is, the wavelength (color) of the light which can be emitted and absorbed by the semiconductor.
This property allows the use of semiconductors to make devices interacting with (emitting/absorbing) light in
the visible and near-visible spectrum (lasers, light-emitting diodes, etc.)

Most semiconductors are made up of elements of the IV column of the periodic table (group IV elements,
such as Si and Ge, called elemental semiconductors) or are a combination of elements from the third and fifth
column (such as GaAs, InGaAs, GaN, InGaAsP called III-V compound semiconductors) or, finally, a combination
of group II and group VI elements (such as CdSe, ZnS, HgCdTe, called II-VI compound semiconductors).

The electrical conductivity of semiconductors depends on the arrangement of the atoms which constitute the
semiconductor itself. And this dependence is very strong, judging from the fact that a few impurities can modify
it dramatically. Thus, if we want to study electronic condution, we have to analyze how atoms are arranged in

ECE344 Fall 2009 22


semiconductors and how we can describe their electronic properties. First, let’s describe the ‘geometrical’ properties
of crystals. Their electronic properties will be the subject of the next section.

• Definitions
– Crystals are regular arrangements of atoms which look identical when viewed from 2 points

r and r = r + Rl ,

where
Rl = l1 a + l2 b + l3 c
is a ‘translation operation’ with integers l1 , l2 , and l3 and with ‘fundamental translation vectors’ a, b, and
c. The vector Rl is also called a lattice vector. The length of these vectors are called lattice constants a,
b, and c. One can define infinitely many translation vector. It is convenient to consider always the ’shortest’
ones, which we have called above ‘fundamental’ translation vectors..
– A cell is the parallelepiped formed by the fundamental translation vectors. Note that the repetition of a cell
obtained by applying a translation operation fills the entire space.
– As there are infinitely many possible choices for the translation vectors, so there are infinitely many possible
choices for the cells. It is often (not always, though!) convenient to work with the primitive or Wigner-Seitz
cell, which is the cell having the smallest possible volume, given by Vcell = |a × b · c|. It can generated
by connecting all lattice points with straight lines, bisecting these lines with planes normal to the lines, and
considering the volume enclosed by these planes.
– A lattice is the set of all points r generated by translation operations. The symmetry properties of a
three-dimensional lattice allow the existence of 14 types of Bravais lattices (those generated by Wigner-Seitz
cells).
– A basis is the set of the coordinates of all atoms within a primitive cell. For example, in the next figure we
see the structure of the GaAs (or) Si crystal: Ga atoms occupy the edges and the center of the faces of the
cube (the crystal is then said to be face-centered cubic, or fcc. As atoms are displaced from the edge at the
left by a(1/4, 1/4, 1/4). The basis of the cell is thus made up of the two coordinate vectors describing the

ECE344 Fall 2009 23


position of each atom in the cell: The vector (0, 0, 0) for Ga, the vector a(1/4, 1/4, 1/4) for As. Si (like C
in the diamond form, and Ge) is identical, but Si atoms occupy both sites in the cell. Note also that the cell
shown there is a cube and it is not the primitive cell!
We need to know both the lattice and the basis vectors to descrie fully the crystal. Thus, logically the crystal
structure is given by:
crystal structure = lattice + basis
– In order to describe surfaces of semiconductor crystals and directions along those surface, the following
conventions apply. Consider an arbitrary plane which intersects the axes a, b, and c at distances (A, B, C)
(in units of the lengths (a,b,c), the lattice constants). Take the reciprocal numbers (1/A, 1/B, 1/C) and
find the smallest integers (h, k, l) having the same ratios as the reciprocals. The integers h, k, l are the
Miller indices of the plane.
A particular plane characterized by the Miller indices h, k, l is denoted by (h, k, l). The set of all planes
equivalent (from a symmetry point of view) to the (h, k, l) plane (e.g., the (0, 2̄, 0, 0) and (0, 0, 3) planes
are equivalent to the (1, 0, 0) plane) is denoted by {h, k, l}.
– A direction in the crystal is denoted by the direction of the vector ha + kb + lc and it is labeled as [h, k, l].
The set of all equivalent directions is denoted by < l, m, n >.
– Note that in cubic crystals (and, in general, only in cubic crystals), the direction [h, k, l] is perpendicular to
the plane (h, k, l).

ECE344 Fall 2009 24


a c
b
a

ECE344 Fall 2009 25


• Cubic and zinc-blende lattices
– Most semiconductors of interest have a ‘tetragonal’ coordination (resulting from the sp3 hybridization of the
bonding orbitals we have mentioned for Si). This yields the face centered cubic (fcc) lattice we have seen
before. Note that in this case there is a single lattice constant, a. For the elemental group-IV semiconductors
its value is about 0.543 nm (Si), 0.565 nm (Ge). The figure on page 26 below illustrates the value of the
lattice constant for several III-V compound semiconductors.
Note that group-IV semiconductors are bound together via covalent bonds. They are ‘nonpolar’ materials,
since the bonds do not possess an electric dipole moment. On the contrary, III-V materials are bound via
bonds of a rather ‘mix’ character, a bit covalent but mostly ionic (e.g., group-III Ga has a smaller electronic
cloud than group-V As in GaAs, so the bond is between Ga+ and As−, although neither Ga nor As are fully
ionized, as in NaCl, for example.). These are ‘polar’ semiconductors.
– The WS cell of fcc lattices is generated by the fundamental vectors
a a a
a = (x̂ + ŷ) , b = (ŷ + ẑ) , c = (ẑ + x̂) .
2 2 2
– Point Group.
Symmetry is an important feature of crystals. Without it, we would unable to handle them as accurately as
we can. Indeed, ‘amorphous’ materials (such as glass) are way less understood than crystals.
The periodicity of the lattice is the first obvious symmetric property. This can be viewed as a symmetry under
translations: We can shift the lattice by a lattice vector and we recover exactly the same confuguration we have
started from. Other important symmetry operations are related to the structure of a crystal. For example, a
cube can be rotated, flipped, or ”inverted’ in many ways without changing its appearence. In general, there are
24 fundamental symmetry transformation which map the cubic lattice onto itself (the so-called Td sub-group):
All permutations of the 3 coordinates (6 operations) times sign-swapping two of them (4 operations).
Accounting for inversions, there are 48 symmetry operations in all (the so-called Oh group).
• Crystal growth, wafers, and doping.
Let’s refer mainly to the Streetman-Banerjee text (because of the nice figures) for a somewhat detailed description
of how Si is extracted chemically from sand, how large Si crystals are grown into ingots, and how wafers are cut,

ECE344 Fall 2009 26


lapped, and polished from the ingots.
In addition to large crystalline wafers, the substrates used for electronic applications, are often modified by
growning different semiconductors by ”epitaxy” (vapor phase – VPE – or molecular beam – MBE) onto different
substrates. For example, of the III-V compound semiconductors, InP allows the growth of the largest wafers
(up to 5”, compared to 30 cm – ≈ 12” – for Si wafers). Other III-V materials (GaAs, InAs, etc.) may be
grown on InP by epitaxy. However, it is best to grow materials which have the same lattice constant of InP.
The figure below shows the energy gap vs. the lattice constant of various ternary alloys. One can see that
Inx Ga1−x As matches the lattice constant of InP when the mole-fraction of In x is 53%. This is called InGaAs
’lattice-matched’ to InP and it is widely used in optoelectronics applications.

2.0
Ev (Γ)

BAND EDGE ALIGNMENT TO Au ( eV )


GaP
1.5 AlAs Ec (Γ)
AlSb Ec (X)
1.0 AlP Ec (L)
GaAs
Si
InP GaSb
0.5 Ge
InSb
0.0 GaSb
Ge InSb
Si InAs
–0.5 GaAs
GaP AlSb
InAs
–1.0 InP
AlAs
AlP
–1.5
5.4 5.6 5.8 6.0 6.2 6.4 6.6
LATTICE CONSTANT ( )
When a material with a lattice constant a larger than the lattice constant of substrate, a0 is grown epitaxially,
it will grow with lattice constant a0 . This will result in the material to be under biaxial ‘stress’ (tensile if

ECE344 Fall 2009 27


a < a0 , compressive otherwise). The energy stored in the stress will grow with growing volume (thickness) of
the material. Above a critical thickness, it will be energetically favorable to release the stress via the creation of
crystalline defects (stacking faults, dislocations, etc.). If these are electrically active, they will affect negatively
the electronic properties and, in general, should be avoided.

ECE344 Fall 2009 28


Energy bands in crystals

In this section we shall discuss the electronic properties of semiconductor crystals. These are characterized
mainly by the existence of allowed ‘energy bands’ separated by ‘energy gaps’. Electronic transport in semiconductors
depends crucially on both the characteristics of the energy bands and by the value of the gap between occupied
and empty states.

Understanding bands quantitatively requires solving the Schrödinger equation for many electrons in the
potential due to all ions in the crystal. For example, ignoring core electrons, a crystal composed of N Si atoms
can be viewed as a lattice of N Si+4 ions and 4N electrons embedded in the complicated Coulomb potential of
all of these ions.

The problem may appear daunting. However, the periodicity of the lattice allows major simplifications:
Roughly speaking, all we have to do (so to speak...) is to study the electronic properties of a single cell subject to
the periodic boundary conditions stemming from the symmetry of the lattice. Thus, we can reduce the problem to
one involving ‘only’ 8 electrons and 2 4-fold-ionized Si ions.

General theorems are available to compute the band structure of solids: Most notably, several schemes to
approximate the potential of the ions and all ’other electrons’ to a fixed pseudopotential independent of what the
other electrons do, so that we can use a single electron picture, and Bloch theorem, which tells us how to reduce
the complicated wavefunction as a product of a simple ‘free electron envelope’ wavefunction and a complicated
‘fast-wiggling’ wavefunction, the same in each cell. Thus each electronic state is labeled by something which
resembles the wavevector of a free electron, k, and can be written as:
ik·r
ψk (r) = e uk (r) , (33)

ECE344 Fall 2009 29


where uk ) (called a Bloch function) has the periodicity of the lattice. Notice that, similarly to the symmetries of
the crystal in ‘real space’ of all spatial coordinates r, so there are also corresponding symmetries in the space of all
wavevectors k: For eaxmple, the state of an electron traveling along the x-axis of cubic crystal will be described
by a wavefunction identical to that associated to an electron traveling along the z -axis, provided we replace kx
with kz . These symmetries allow us to consider only a Wigner-Seitz (primitive) cell in this ‘reciprocal space’. This
cell is called the first Brillouin Zone (BZ) of the crystal. It is enough to get wavefunction for values of k inside
the first BZ in order to get a full knowledge of the electronic properties of the lattice. These are tremendous
simplifications afforded by the symmetry of crystals.

Despite these powerfull simplifications and approximations, the problem is still mathematicaly somewhat
convoluted. So, we shall ignore quantitative aspects and discuss instead the band structure of solids only from a
qualitative point of view.

Let’s start by considering what happens when we bring 2 or more Si atoms together.

• Coupling atoms.
Consider two Si atoms, widely separated, as in the figure below. This shows the Coulomb potential around
each nucleus (∝ 1/r , where r is the distance from the nuclei), the energy of the 3 occupied levels, and their
electronic confuguration. Let’s recall here the important fact that there are 8 states available in the n=3 energy
level (2 3s-states and 6 3p-states), but Si has only 4 electrons in this shell, so that the outermost (valence)
shell is only half occupied.

ECE344 Fall 2009 30


Si Si

3s23p2 3s23p2 = 3(sp3)4


= 3(sp3)4 (only half full)

2s22p6 2s22p6

1s2 1s2

Now let’s bring the nuclei much closer, so that the potential around each nucleus is affected significantly by
the other atom. The ‘core’ levels (n = 1 and n = 2) do not feel much the perturbation caused by the other
nucleus, but the highest-energy occupied level n=3 do feel a dramatic change. The orbitals now spread over
both nuclei. Clearly, their energy will change, because the potential itself has changed. Qualitatively, what
happens is that, starting from the original hybrid sp3 orbitals of a single atoms, we will have a ‘distortion’ of
each of them, caused by the new potential. In addition, we may form two superposition of the orbitals, much as
we had formed the hybrid sp3 . One superposition corresponds to orbitals with a large electron probablity-density
between the nuclei. This is a ‘bonding orbital’, since these ‘shared electrons’ constitute a covalent bond. Since
the electrons in this orbital sit close to the attractive nuclei, the energy of this orbital is low. The other linear
combination of orbitals we may form corresponds to orbitals having a small electron density between the nuclei.
These orbitals are ‘antibonding’, since they do not contribute to the covalent bond between the two Si atoms.
Their energy will be higher, since the electrons, on average, sit farther away from the nuclei.

ECE344 Fall 2009 31


Si Si
3(sp3)8 (empty)
3s23p2 = 3(sp3)4 3s23p2 = 3(sp3)4
3(sp3)8 (full)

2s22p6 2s22p6

1s2 1s2

The net effect is that the 4-fold degenerate energy levels corresponding to the 2 sp3 orbitals of the two separate
atoms, each containing one (of a possible maximum of 2) electrons, have split into 2 lower-energy bonding states
(each with a slightly diffrerent energy) and 2 higher-energy (also each at a slightly different energy) antibonding
states. The electrons will occupy the lowest-energy states to form the bond. The antibonding states will remain
unoccupied.

ECE344 Fall 2009 32


Ψ1 Ψ2

BONDING Ψ1+Ψ2

Ψ1–Ψ2

ANTIBONDING

Let’s now imagine what happens when we bring together 3 Si atoms: They will bond together forming a set of
3 occupied bonding states (at 3 different but very similar energies) and 3 antibonding unoccupied states.

• Energy bands.
Now let’s consider a Si crystal formed by, roughly, N = 1022 atoms. What happens now is very similar
to what we have seen in the case of 2 and 3 atoms. The N atoms will contribute a total of 8N possible
sp3 orbitals, originally all at the same energy. Only 4N of them are occupied. They will split into 4N
occupied orbitals contributing to the bond and 4N empty antibonding orbitals. The energy of each of the

ECE344 Fall 2009 33


4N bonding orbitals will be very close to that of any other bonding state. Indeed, they will spread over an
energy interval of a few tens of eV and we will have roughly 1022 energy levels in this small energy range.
They are so many and so close together that it makes sense to look at them as a ‘continuum’ of infinitely
many states populating an energy band. Since this is the band of bonding states made up using the valence
electrons, it is called the valence band of the crystal. Similarly, the set of 4N antibonding orbitals will
also form a band of unoccupied states. These states are spread away from the atoms throughout the whole
crystal. Assuming we found a way to occupy one of these states (we shall see how later on), the electron
in this state will not be ‘forced’ to bind the nuclei, but will be free to move around the crystal. If we
apply an electric field, the electron will be free to move according to the external force, thus contributing to
the conduction of current. Therefore, the band of antibonding states is called the conduction band of the crystal.

The valence and conduction bands are separated by an energy gap whose existence stems from considerations
similar to those which caused the presence of forbidden energies in the isolated Si atom.

• Metals, insulators, semiconductors.


We have just seen that the 4N valence electrons in a Si crystal occupy the entire valence band. On the other
hand, the conduction band, which could contain also 4N electrons is completely empty.

Suppose that we apply an electric field to a Si crystal (by placing in contact with two metallic plates, for
example, at the opposite ends of the crystal). In principle, the electrons would ‘like’ to respond to the field and
move away from the negatively-charged plate towards the positively-charged plate. But ‘motion’ requires gaining
kinetic energy. Therefore, the electrons in the valence band should acquire a bit of extra energy in order to move.
But since the valence band is full, they may gain energy only by ‘jumping’ to the next available empty states in
the conduction band. But this jump requires a large amount of energy (about 1.1 eV at room temperature).
Therefore, the electrons cannot move so that no current will flow in the crystal, despite the electric field we have
aplied... In other words, a perfect Si crystal (we’ll see below why we stress the word ‘perfect’) behaves like an
insulator.

ECE344 Fall 2009 34


Not all crystals are like Si. For example, if we take a crystal of N Na ions, the valence band will be constituted
by 2N states emerging from the s orbitals of each Na ion, but only N electrons populate the band, the ‘energy
gap’ lying at a much larger energy. Therefore, the valence band will be only partially (in this case, half) occupied
and electrons can acquire an arbitrarily small amount of energy from the applied electric field. Therefore, current
will flow in Na: Sodium is a conductor. Other metals, instead, have ‘overlapping bands’. The absence of a gap
makes good electrical conductors, much like Na.

Glass (silicon dioxide, SiO2 ) is another insulator, like Si, but its band-gap is about one order of magnitude larger
than that of Si (for Si we have seen that Eg ≈ 1.1 eV, while for SiO2 we have Eg ≈ 9.2 eV).

The figure below shows qualitatively the difference between semiconductors, insulators, and semiconductors. The
difference betweem semiconductors and insulators is only quantitative: If the forbidden energy gap is very large,
even the presence of impurities cannot modify the insulating poperties of the material. For semiconductors,
instead, the addition of a small number of selected foreign atoms (impurities) can alter dramatically their
properties. We shall see how this process (called ‘doping’) is intentionally done.

INSULATOR SEMICONDUCTOR METAL

EMPTY EMPTY EMPTY PARTIALLY


FULL
Eg
overlap
Eg
FULL FULL

FULL
FULL

ECE344 Fall 2009 35


• Parabolic approximation: Effective mass.
Here we introduce a fundamental property of the motion of electrons in crystals: Under some conditions
(restrictive, but still covering a useful range of applications), the motion of electrons in the conduction band
can be approximately described as the motion of a free electron, provided we modify its mass using instead as
effective mass.
To give an idea of how and why this comes about, let’s recall a simple fact: Dealing with the solution of the
Schrödinger equation for a free particle in one dimension (that is, zero potential energy everywhere), we saw that
the solution representing an electron moving along the x axis has the form Ae±ikx , where A is a normalization
constant. The energy of the electron as a function of the wavevector k is:

h̄2 k2
E(k) = , (34)
2m
where m is the mass of the electron. The figure below shows this ‘parabolic’ relation.
3
2 2
k
E(k) =
2m
KINETIC ENERGY E (eV)
2

0
–10 –5 0 5 10
WAVEVECTOR k (109 m–1)

ECE344 Fall 2009 36


The calculation of the structure of the energy bands in solids is quite complicated, as we said before. Typically
these calculations start by writing the complicated Schrödinger equation for wavefunctions of the Bloch form,
Eq. (33). For each ‘envelope’ of a give k-vector, one finds a set of energy levels En (k). Let’s consider Si and
ignore core electronic states (the 1s, 2s, and 2p orbitals). Then, these levels are labeled by a ‘band index’ n,
with n=1 through 4 corresponding to the valence bands (occupied eaqch by N electrons), lelels with n > 4
corresponding to the conduction bands. Plotting E(k) as the wavevector k scans the entire first BZ, we find
curves representing how the energy levels En (k) vary with k.

The figure below shows results for these types of calculations applied to the study of the band structure of Si.
The labels L, X , Γ, U , K , etc. label special highly simmetric points in the BZ. For example, the point Γ is
the center of all reflections, inversions, etc. for a cubic crystal. It is called the ‘center of the BZ’ (or, simply,
‘zone center’), shown in the figure above.

ECE344 Fall 2009 37


5

4
Γ2’
Γ2’
3
L3
Γ15 Γ15
2

ENERGY ( eV )
1 L1 X1

–1 Γ8

Γ7
–2 L4,5

L6
–3
X5
–4

–5

–6
X ZWQ L Λ Γ ∆ X U,K Σ Γ

Looking at the figure, we see that the valence bands reach a maximum energy at about 1.17 eV below the
minimum of the conduction band. Three valence bands reach this maximum value at the same point, the Γ
point. The lowest-energy conduction band, instead, reaches a minimum value (arbitrarily set to zero) near the

ECE344 Fall 2009 38


symmetry point X . Note that, by cubic symmetry, there are 6 equivalent X points: One along the positive kx
axis, one along the negative kx axis, 2 more on the y axis, and 2 final minima along the z -axis.

Let’s consider now the function the function En (k) for n = 5 (the first conduction band, so we also call it
ECB (k)) in the proximity of its minimum. Let’s consider, for simplicity, this function on a line going from the
BZ-center Γ to X : This line correcponds to wavevectors of the form k = (k, 0, 0). The function ECB (k))
reaches its minimum at the point k = (k0 , 0, 0), where k0 ≈ 0.852π/a. Let’s expand this function around
its minimum, k0 :

∂ECB (k0 , 0, 0) 1 ∂ 2 ECB (k0 , 0, 0) 2


ECB (k, 0, 0) ≈ ECB (k0 , 0, 0) + (k − k0 ) + 2
(k − k0 ) . (35)
∂kx 2 ∂kx

Now: Due to the fact that we have set the zero of the energy at ECB (k0 , 0, 0), the first term at the
right-hand-side (rhs) vanishes. Due to the fact that we have an extremum (a minimum) at k = (k, 0, 0), the
first derivative vanishes by definition of extremum, so that also the second term at the rhs vanishes. Thus,
Eq. (35) becomes:
1 ∂ 2 ECB (k0 , 0, 0) 2
ECB (k, 0, 0) ≈ 2
(k − k0 ) . (36)
2 ∂kx
Let’s now shift the kx by −k0 so that k → k − k0 and, finally, rewrite Eq. (36) as:

1 ∂ 2 ECB (k0 , 0, 0) 2
ECB (k, 0, 0) ≈ 2
k . (37)
2 ∂kx

Now let’s compare Eq. (37) with Eq. (34). Both are parabolic laws (i.e., the energy grows with the square of
the wavevector). We can make them look even more similar if we introduce the quantity:

1 1 ∂ 2 ECB (k0 , 0, 0)
= 2
. (38)
m∗ h̄ ∂k 2
x

ECE344 Fall 2009 39


With this definition, Eq. (37) can be rewritten as:

h̄2 k2
E(k) ≈ , (39)
2m∗
which is just the kinetic energy of a free electron, but with its mass m replaced by an effective mass m∗ .
The figure below shows a blow-up of the area covered by the red rectangle in previous plot, showing in more detail
the extrema of the conduction and valence bands, together with parabola (dotted red curves) approximating the
‘exact’ energy-vs-k relationship E(k) (also called ‘dispersion’) with the effective-mass parabolas.

1.0

0.5

0.0 X
ENERGY ( eV )

–0.5
Eg

–1.0
Γ

–1.5

–2.0
–0.5 0.0 0.5 1.0 1.5
k–VECTOR (2π/a)

ECE344 Fall 2009 40


Extending this analysis to three dimensions, we must replace the Taylor expansion in Eq. (35) with its
corresponding 3D expansion. Following the same procedure, we find that, rather than a simple scalar number,
the effective mass is a ‘tensor’. In other words, the ‘curvature’ of the parabola (and so the mass) changes
with direction. Thus, surface of constant energy (‘equienergy surfaces’) are not spherical but may have more
complicated geometries. For example, around the minimum of the Si CB we have just considered we should
write:


h̄2 2
kx ky2 + kz2
E(k) ≈ + . (40)
2 ml mt

We need two effective masses: A heavier ‘longitudinal’ effective mass (about 0.9 the free electron mass m0 ) as
we move along the line from Γ to X , and a much lighter ‘transverse’ mass (mt ≈ 0.19 m0 ) as we move away
from the minimum at (k0 , 0, 0) along a direction perpendicular to the kx -axis. Equi-energy surfaces are thus
‘ellipsoids of rotation’, solids generated by the rotation of an ellipse around the axis along its major (longitudinal)
axis.

The band structure of other cubic (fcc) semiconductor is quite similar. The two figures below show those of
Ge and GaAs. Note that in both semiconductors, as also in Si, the maximum of the valence bands is triply
degenerate and it occurs at the center of the BZ at the Γ-point. The minimum of the conductions band for
Ge, instead, occurs at the 8 symmetry points called L for Ge, with local minima at a slightly higher energy at
the 6 points X (and one more at Γ). For GaAs the minimum occurs at Γ. This makes GaAs a ‘direct-gap’
semiconductor, with both the VB maximum and the CB minimum at the same position in k-space. This has
important consequences on the optical properties of GaAs (and many other III-V compund semiconductors),
since electronic transitions involving photons are ‘vertical’, as we shall see later.

ECE344 Fall 2009 41


5 5

4 L4,5 4
L3 Γ15
L6 Γ8
3 3
Γ15
Γ8
2 Γ6 2

ENERGY ( eV )
X3
ENERGY ( eV )

1 Γ7 X5 1
L1
L6 Γ7
Γ1 X1
0 0
Ge Γ8
–1 –1 GaAs
Γ8
L4,5 Γ7
–2 –2 L4,5
L6 Γ7

–3 –3 L6
X5
X7
–4 –4
X6
–5 –5

–6 –6
X ZWQ L Λ Γ ∆ X U,K Σ Γ X ZWQ L Λ Γ ∆ X U,K Σ Γ

Despite some minor ‘clumsiness’ due to the need of an anisotropic mass, nevertheless we have obtained an
extremely powerful simplification: The whole set of complications caused by the lattice, its symmetries, the
nature of the ionic potentials, and, to some extent, the entire quantum-mechanical machinery underlying the

ECE344 Fall 2009 42


electronic properties of the crystal have been hidden and, to a certain exent, may be ‘ignored’. We are left with
the following compact description of electrons in the conduction band:

1. We can describe electrons as classical particles with a wavevector k. The physical momentum of electrons with
low energy (that is, with k close to the minimum of the band) is h̄k, as for free electrons. (As k moves away
from the minimum, h̄k, properly called crystal momentum, ceases to be simply related to the real momentum
of the electron.
2. The energy of the electron is simply given by the free-electron expression provided we use the effective mass
m∗ .
3. Sufficiently close to k0 , the electron velocity is given by the classical expression (for a simple isotropic mass):
dr p h̄k 1
= v = = = ∇kECB (k) . (41)
dt m∗ m∗ h̄
The second equality is obvious from the definition p = mv, having replaced m with m∗ . The third
equality comes from observation 1 above that p = h̄k. The final equality relies on the fact that this is the
expression for the group velocity exhibited by a packet of waves in a medium in which the frequency ω is a
function of wavevector, ω(k) = ECB (k)/h̄, similarly to electromagnetic waves in a medium in which the
dielectric constant depends on wavelength. These media are called ‘dispersive’. This is the origin of the name
‘dispersion’ usually given to ECB (k).
4. Suppose that, somehow, we manage to have electrons in the conduction band (either because the material is
a metal, or because we have given enough energy – with light of short-enough wavelength so that hν > Eg ,
for example – to electrons in the valence band so that some of them can ”jump” across the gap). If we apply
an electric field to the crystal, these electrons can move. It can be shown that the (crystal) momentum h̄k of
the electron obeys the equation of motion:
d(h̄k)
= −eF . (42)
dt
Replacing the (crystal) momentum h̄k with the momentum p, we recognize this equation as Newton’s second

ECE344 Fall 2009 43


law dp/dt = Force, so that it may seem trivial. However, we should recall that we are dealing with a very
complex system. In Eq. (42) the external force −eF is only the externally applied field and k is the wavevector
of the ‘envelope’ of the full wavefunction. From this perspective the result of Eq. (42) is actually amazing:
We can ignore all the complications due to the messy crystal potential giving raise to all sort of forces pushich
and pulling the electrons among the ions. We can ignore also all details about what the electron does near the
ionic cores – absorbed into the Bloch functions uk (r) – and treat the electrons as if only the external field
existed and acted only on its envelope wavefunction!
We can now couple Eq. (42) to Eq. (41). From the latter h̄k = m∗ dr/dt. Inserting this into the former we
have:  
d(h̄k) d d r
= m∗ = −eF , (43)
dt dt dt
which we can rewrite as:
∗ dr
m = −eF , (44)
dt2
which is just Newton’s second law.
The two equations (41) and (42) are called ‘acceleration theorems’. Despite their ‘familiar’ look, they are far
from trivial and constitute the foundations on which we shall build the theory of electronic transport in solids.

ECE344 Fall 2009 44


• The concept of “holes”.
We have already hinted at the possibilty of exciting electrons across the energy gap and bringing them into
the conduction band (CB) of a semiconductor. Indeed this happens spontaneously in all semiconductors at
room temperature. For example, in Si at 300 K there will be about 1.45 ×1010 electrons in the CB per cubic
centimeter. (This is called the ‘intrinsic electron concentration’, ni of Si at 300 K). This is simply the result of
the ‘thermal agitation’ of the ions and electrons in the crystal. Note that since we have about 5 × 1022 Si ions
in each cm2 , and each ion has 4 valence electrons, we have about 2 × 1023 electrons cm−3 in the valence band
(VB). Therefore, we see that ni is a very small number: We have one electron in the CB for every 1.4 × 1013
electrons in the VB!
We may also shine photons of energy hν > Eg ≈ 1.1 eV on Si and excited an electron from the VB to the
CB.
In either case, whether electrons are excited thermally or optically across the gap, we are left with an electron in
the CB and a ‘lack of one electron’ in the VB.
We have just seen that electrons in the CB can move: Each electron in the CB is surrounded by a very large
number of empty states. But what about the ‘lack of one electron’ in the VB?
When the CB is totally empty and the VB is totally full, we have observed that no current can flow. This is
because for every electron with wavevector k, and so velocity v = h̄k/m∗ , there will be somewhere else in the
full VB another electrons with wavevector −k, and so velocity −v. The total current density due to the NV B
electrons in the VB will then be:
NV B

J = −e vi = 0 . (45)
j=1
Let’s now take a single electron j and excite it to the CB. The total current in the VB will now be as before,
but we must substract the contribution to the total current due to electron j , which is −evj . Thus:

NV B

J = −e vi − (− e vj ) = e vj , (46)
j=1

ECE344 Fall 2009 45


since the first sum vanishes, as we have seen. This equation can be read as saying that the current in the VB is
carried by a single positive charge with velocity vj . We can interpret the single electron missing from the VB
as a hole added to the VB. The advantage of this interpretation is that we deal with the motion of a single
positively-charged particle, rather than having to deal with the motion of ∼ 2 × 1023 − 1 negatively-charged
particles.

The equations of motion (41) and (42) are also valid for holes, provided we replace the charge −e with +e and
the dispersion ECB (k) (or, more simply, the effective mass m∗ in the CB) with the dispersion in the valence
band, EV B (k) (or the effective mass in the VB).

The figure on page 48 illustrates qualitatively how electrons and holes move under the action of an electric field.
It shows a “band diagram”, in which the x-axis represents the position and the vertical axis represents the total
energy, which we shall measure in eV. The two lines represent the top of the valence band (lower line, labeld as
EV ) and the bottom of the CB, labeled as EC . At the far left and far right are regions of zero electric field.
Therefore the lines, representing potential energy, are ‘flat’. In the central region, instead, the slope of the lines
is proportional to the electric field, whose x component Ex = dEC /dx is positive (that is, the field points
to the right). [The relation Ex = dEC /dx comes from the definition of the electrostatic potential V (x) as Ex = −dV /dx.
The potential energy for electrons is −eV (x), since the electron charge is negative, −e. Since we have defined the line EC as the
potential energy – measured from the bottom of the CB –, we have −eV (x) = EC (x), so Ex = −dV /dx = (1/e)dEC /dx.
Finally, now note that when measuring energy in eV, we must divide the energy (measured in joules) by the magnitude of the electron
charge. So, numerically we have Ex = dEC /dx, when EC is expressed in eV].
Electrons can be thought as ‘entering from the right’ and move to the left, driven by the force −eEx . As
they do so, they move along horizontal lines, retaining the same total energy they had originally, but converting
potential energy into kinetic energy. This can be viewed as the vertical distance between the EC -line and the
location of the green dot (the electron) at a given position x. The parabolas superimposed to the EC -line shows
the dispersion E(k) at a given position. Accelerated electrons climb along the dispersion as their wavevector

ECE344 Fall 2009 46


changes according to Eq. (42) above:
e
kx (t) = kx (t0 ) − Ex (t − t0 ) . (47)

Since kx grows in magnitude (it becomes more negative, since electrons are accelerated in the direction opposite
to the field, being negatively charged), the kinetic energy of the electron grows. Occasionally, collisions ‘scatter’
electrons. We shall talk more about scattering processes in the following. For now, let’s just say that occasionally
electrons collide with thermally displaced ions in the lattice, or with other electrons, or with the Coulomb field
of the ionized dopants. Some of these collisions result in a re-direction of the electron velocity (or wavevector),
some also in a loss of energy (for example, to the lattice in collisions with thermal vibrations). The electron then
‘falls’ down along the dispersion. For simplicity, the figure shows a complete loss of kinetic energy, but this is
not necessarily always the case. Having lost energy and velocity, the electron starts again ‘downhill’, pushed by
the field. Eventually, they will enter the field-free region at left and leave the scene.
For holes, the situation is similar, but ‘flipped’ upside-down, because we have selected the y -axis as measuring
the electron (potential) energy. Thus, holes can be thought of as entering from the left and ‘float’ towards the
top at right, accelerated in the positive x direction by the electric field. Similar ‘quasi-parabolic’ dispersions
could also be shown for holes, but are not shown for simplicity... Actually: The top of the VB is somewhat
messier than the bottom of the CB...

ECE344 Fall 2009 47


ENERGY
E

e– EC
k

EV

electron
kinetic
energy

hole
kinetic
energy

h+

POSITION

ECE344 Fall 2009 48


Doping and Statistical Mechanics of Electrons and Holes
As we have already remarked before, the importance of semiconductors lies on the fact that adding small
amounts of impurities can make transform them from insulators to conductors. This is achieved via ‘doping’
(described below) and results in ‘excess carriers’ to be present: Electrons in the CB and/or holes in the VB. In this
section we see what is doping and how these excess carriers arrange themselves in the bands.

• Doping.
– So, once more: The technological importance of semiconductors stems from the fact that we can make them
decent insulators or excellent conductors by adding ‘free carriers’ (electrons in the CB, holes in VB).
– This is achieved by introducing ‘substitutional’ impurities. In Si (a group IV element, with 4 electrons in
the bonding sp3 valence orbitals) the substitution of a group III element (‘acceptors’ such B, Ga, In, having
only 3 valence electrons) causes the absence of one electron in the valence band for each acceptor atom. Si
so doped is said to be p-type, since current will be carried by holes, which we just saw that can be viewed
as positively charged ‘particles’. By ’accepting’ an extra electron (that is, giving away the ‘hole’), acceptors
become negatively charged ionic impurities. The substitution of a group V element (’donors’ such as P, As,
Sb, having 5 valence electrons) causes the presence of one extra electron in the CB for each donor impurity.
Si so doped is said to be n-type, since negatively charged electrons carry the electric current. By ’donating’
an extra electron acceptors become positively charged ionic impurities.
– Technologically doping can be performed in several ways:
1. By diffusion: Si is exposed to a gas rich of the dopant species. These diffuse into Si. Thus, we have
a depth-dependent distribution of dopants, dictated by the diffusion process: Higher concentration at the
surface, dropping rapidly with depth. This is called the ‘doping profile’. Diffusion proceeds typically (not
always, though) ‘interstitially’, that is, the dopant atoms ‘sneak in’ between Si atoms. As such, the dopants
are electrically inactive. Therefore, the wafer must be heated for long times (tens of minutes) at some high
temperature (typically 900 C to 1100 C) to allow dopant atoms to replace thermally dislodged Si atoms and
become ‘substitutional’ impurities. This is called an ’annealing step’ used for ‘dopant activation’. Historically
this was the first process adopted. However, as devices became smaller, the re-arrangement of the dopants

ECE344 Fall 2009 49


during the activation annealing step became intolerably large and duffusion is today almos always replaced
by ‘ion implanatation’ (see below).
2. By in situ doping: As the semiconductor is deposited or grown (by CVD for poly-crystalline Si or epitaxially,
for example), impurities are added during the process itself. This allows very accurate control of the
distribution of the dopant but, clearly, can be used only when growing the material and it is not an option
when dealing with wafers.
3. By ion implantation. The Si wafer is exposed to bombardment from a source of dopant ions (an ion
accelerator). The energy of the ions (of the order of hundreds of keV) allows them to penetrate the
crystal. The higher the energy, the deeper the ‘range’. An annealing step is now necessary not only
to activate dopants, but also to repair the damage created by the bombardment. Very fast anneals are
now used (‘flash’ anneals, laser anneals, etc.), usually a few tens of seconds at 1050 C or so. The short
annealing time and the ability to accurately tailor the profile have made ion-implantation the method of choice.

– It is easy to see why impurities like dopants are easily ionized. Thus, considering for example donors like
P in Si, the fifth ‘extra’ valence electron of P is easily excited into the conduction band, without remaining
attached to the dopant. We can view the 5-th-electron+impurity system as resembling the hydrogen atom:
The five-fold ionized P core is surrounded by 4 electrons in the sp3 bonding orbitals, exactly as if it were the
original Si atom, but it is singly positively charged. The extra electron orbits the core much like an electron in
the H atom. We can use exactly the same theory, but with two major modification: First, the electron has a
different (smaller, in Si and most – but not all – semiconductors) effective mass and. Second, the dielectric
constant of Si is larger (as it is in all semiconductors). Let’s see how this affects the energy levels.

The external potential caused by the substitutional ionized impurity is a ‘screened’ Coulomb potential: It is
screened by:
1. the valence electrons and by some lattice rearrangement, which is accounted for by the macroscopic dielectric
constant s of the semiconductor
2. the free carrier (holes in the VB, electrons in the CB)

ECE344 Fall 2009 50


Ignoring for now the second screening effect, the potential will be

e2
Vdop (r) = −
4πs r
Thanks to effective-mass approximation, we can treat this potential as an hydrogen-like problem, as we said
above. Assuming a scalar effective mass (that is: the same in all directions, which we saw is not true for Si,
but we shall ignore this complication for now), the energy levels (measured from the band extremum) will be:

m∗ e4
En = − .
32π3 n2 h̄2 2s

Note how the magnitude of these energies is reduced by a factor m∗ 20 /(m0 2s ) with respect to the hydrogen
atom. Since for most semiconductors m∗ /m0 ∼ 0.1 and 0 /s ∼ 0.1, the ionization energy about 104
times smaller than that of H, so it is of the order of 1-to-10 meV. This is smaller than the the thermal energy
kB T (≈ 25 meV at 300 K). Thus, we see that dopants are almost always ionized. ‘Freezout’ (that is, the
condition in which free carriers are trapped into impurity states and the semiconductor actually becomes an
insulator) occurs at very low temperatures. Note also that the Bohr radius of the n-th excited state,

4πn2 h̄2 s
an = − ,
m∗ e2
is much larger than that of the H atom, being typically of the order of tens of nanometers. This justifies the
use of the macroscopic dielectric constant, since the orbit of the electrons extends over so many lattice cells
that we can approximate the crystal as a homogeneous macroscopic medium.
• Density of States.
We now deal with the important concept of ‘density of states’, which is of utmost importance in studying the
way electrons (holes) populate the CB (VB). First we must consider the density of states in momentum-space
(or k-space). Then, we translate this into density of available states per unit energy. We shall then deal with

ECE344 Fall 2009 51


a basic property of the statistics of quantum particles and, finally, we will be able to calculate the density (in
space and energy) of electrons in CB and holes in the VB as a function of the concentration of dopants and
temperature.
– A free particle in a 3D box.
Let’s recall once more the derivation we followed before to solve the Schrödinger equation for a free electron
in a cubic box of side L (see pages 12-15):

∂ 2ψ ∂ 2ψ ∂ 2ψ 2
+ + + k ψ = 0,
∂x2 ∂y 2 ∂z 2

with k2 = 2mE/h̄2 .
Let’s separate the problem: We look for a solution of the form ψ(x, y, z) = ψx (x)ψy (y)ψz (z). Substituting
this into the Schrödinger equation and diving by ψ , one gets:

1 d2 ψx 1 d2ψy 1 d2 ψz 2
+ + + k = 0.
ψx dx2 ψy dy 2 ψz dz 2

Since this is true for any x, y , and z , we must have

1 d2 ψx 2
= some constant = − k .
ψx dx2 x

Thus, accounting for the boundary condition (wavefunction must vanish at the wall of the cube):

ψ(x, y, z) = A sin(kx x) sin(ky y) sin(kz z)

2 2 2 2
kx + ky + kz = k
nx π ny π nz π
kx = , ky = , kz = ,
L L L

ECE344 Fall 2009 52


with nx , ny , nz = ±1, ±2, ±3, ....
The number of solutions per unit volume of k-space will be (obviously?) L3 /π 3 . But since solutions differing
only for an overall sign are identical and there are 8 possible permutations of this type, the number of allowed
k-states per unit volume in k-space will be 8 times smaller. Finally, accounting for the electronic spin (ignored
so far) the number of states will be twice as large. Thus, the density of k-states will be L3 /4π 3 .
– Density of states at a given energy.
The density of k-states with energy between E and E + dE can be obtained by multiplying the density in
k-space by the volume of the spherical shell with wavevector between k and k + dk, that is

2 L3
4πk dk .
4π3

(Recall Jean’s law n(ν) ∝ ν 2 ? Since for photons 2πν = ck, where c is the speed of light, the expression
above is just Jean’s law for the black-body spectrum).
We now interested in expressing the number of states with magnitude of k in the shell of magnitudes
between k and k + dk in terms of states with energy between E and E + De. To do that, we use
E(k) = h̄2 k2 /(2m∗ ) and change variable from k to E : Since k = (2m∗ E)1/2/h̄ and

h̄2 k m∗
dE = dk → kdk = dE
m∗ h̄2
we have
2 (2m∗ E)1/2 m∗ m∗ (2m∗ )1/2 E 1/2
k dk = k kdk = dE = dE ,
h̄ h̄2 h̄3
so that the number of states per unit volume (recall that L3 is the volume of the crystal) with energy between
E and E + dE , ρ(E), will be:

m∗ (2m∗ E)1/2
ρ(E) dE = dE ,
h̄3 π2

ECE344 Fall 2009 53


where E is measured from the band extremum.
– A useful trick.
Suppose we must perform a sum over all k-states in k-space. Since the states are very dense, it makes sense
to replace the sum with an integral (and it also makes our life easier, since integrals are in general easier
to do that sums). This can be done by noticing that summing over k-points in a given volume of k-space
is equivalent to integrating over the same volume and dividing by the volume of each k-point, which is the
inverse of the density. Thus
 
 L3 1  2
→ 2 dk or → dk
(2π)3 V (2π)3
k k

– DOS for the Si CB:


Consider the ellipsoid  
h̄2 kx2 ky2 + kz2
E100 (k) = + .
2 mL mT
with the longitudinal axis along the kx axis. Using the so-called “Herring-Vogt transformation”,
    1/2   1/2
∗ m0 1/2 ∗ m0 ∗ m0
kx = kx , ky = ky , kz = kz ,
mL mT mT

one can show that the DOS in the CB of Si is

md (2mdE)1/2
ρ(E) = .
h̄3 π2

where the ‘density of states effective mass’ is md = (mL m2T )1/3 .

ECE344 Fall 2009 54


• A quick review of Statistical Physics.

Consider a large (that is, at the human, macroscopic scale) volume of gas. This volume will contain a very large
number of molecules, N ≈ 1023 . Clearly, we cannot even think of using Newton’s laws to describe the gas:
We would have to know the position and velocity of each molecule of the gas at some initial time, and solve the
huge number N of equations of motion, accounting for the forces among the particles. Even if we could solve
the all of these equations, it would be next to impossible to know all of the initial conditions required to get the
full solutions. Yet, we can describe the gas if we are willing to be content with information only about some
global, averaged quantities: We do not require the knowledge of all positions and velocities of all molecules
at all times... we would not know what to do with this knowledge anyway... Instead, we ask for information
about the volume and pressure of the gas given its temperature. This is the information provided by the laws of
Thermodynamics.

Statistical Mechanics tackles the problem differently in a different, intermediate way in an atempt to bridge
the gap between Newton’s laws and Thermodynamics: We shall still use Newton’s laws, but in an averaged,
statistical fashion.

In our example, we do not require to have the precise knowledge of all coordinates and velocities. Rather, we
shall be satisfied with some ‘statistical’ knowledge, like the average kinetic energy of a particle, or the probability
that it will have a particular velocity.

Let’s consider again our gas of N molecules. Let’s assume that is contained in a volume V in thermal contact
with a much larger system, so large that its temperature can be assumed not to change appreciably when it
exchanges energy with our gas. Let’s also assume that our gas is at some very high temperature, having been
so prepared initially. We ask how the gas will be described when it reaches thermal equilibrium with the large
reservoir. Simple (?) considerations tell us that the gas will evolve in time towards equilibrium by moving
towards the most probable distribution of molecules among all possible energy states under the constraints that
1. the number of molecules in the gas does not change, and 2. the total energy (gas + reservoir) remains

ECE344 Fall 2009 55


constant (while exchange of energy between gas and reservoir is allowed to occur). Thermodynamics tells us that
energy will flow from the ‘hotter’ gas to the ‘colder’ reservoir, untill they will be both at the same temperature.
Microscopically, as we just said, we can view the initial situation as highly ‘unlikely’, the equilibrium situation
being one of the most likely. One can do this ‘configuration counting’ and come to a solution. If we assume
that the molecules in the gas are distinguishable (that is, we can think of labeling them and that we are able
to ‘recognize’ them), then, under these assumptions, some simple (but algebraically cumbersome) bookkeeping
gives us the following result: The probability fM B (E)dE that a molecule will have kinetic energy in the
interval (E , E + dE ) will be given by the probability distribution:
 
E
PM B (E)dE = A ρ(E) exp − dE = A ρ(E) fM B (E) , (48)
kB T

where A is some normalization (so that the total probability adds to unity, we shall not worry here about its
value), T is the (absolute) temperature of the reservoir, and kB = 1.38 × 10−23 J/K is called the Boltzmann
constant. The function ρ(E) is the density of states we found before and we shall discuss its appearance shortly.
We have given the exponential function its own symbol, fM B = exp[−E/(kB T )], for later convenience.
Now note that the temperature is a measure of how likely it will be for a molecule to have a particular value
of energy E : The higher the temperature, the more likely it will be for the molecule to be found at energy
E . Clearly, PM B (E) can also be interpreted as the ‘energy distribution’ of the molecules in the gas: When
properly normalized, N PM B (E)dE will tell us how many molecules, on average, will be found in the energy
interval between E and E + dE . At zero temperature all molecules will be at the lowest possible energy state
E = 0. As T grows, the distribution becomes ‘flatter’, more and more energy states becoming significantly
populated. Equation (48) is known as the Maxwell-Boltzmann distribution and it applies to classical particles
only, as we shall see below.

Why does the DOS ρ(E) appear in Eq. (48)? The best to way to see it is to re-express E in terms of the
velocity v and then count all particles per unit volume in the gas. We can simply look at all velocities, multiply
each entry in our tally by the probability that a particle will be there, and add them all up to obtain the particle
density n = N/V . Since we shall deal mostly with wavevectors and quantum mechanical particles in the

ECE344 Fall 2009 56


following, let’s carry this exercise dealing with k instead (which, since v = h̄k/m∗ , amounts to doing the same
‘counting’), pretending that Eq. (48) is valid in this case. Thus by the trick for summing over k-vectors we saw
before (top equation on page 54),
  
2 E
n = A dk exp − . (49)
(2π)3 kB T

Let’s go to polar coordinates in k space as we did above, change integration variable from k to E , and we get:
     
21/2 m∗3/2 E E
n = A dE E 1/2 exp − = A dE ρ(E) exp − . (50)
π2 h̄3 kB T kB T
This shows that the probability distribution must indeed contain the DOS.

Finally, let’s compute the average energy of each particle in the gas: By definition, the average energy will be
the average of E weighted by the probability function ρ(E)fM B (E):



const 0 dE E ρ(E) fM B (E) 0 dE E 3/2 exp[−E/(kB T )]
E =
∞ =
∞ , (51)
const 0 dE ρ(E) fM B (E) dE E 1/2 exp[−E/(k T )]
0 B

where ‘const’ is some normalization constant which cancels out from the calculation. Let’s change integration
variable by setting x2 = E/(kB T ), so that dE = 2kB T xdx and E = x2 kB T :

∞ 4 −x2
0 dx x e (3/8)π1/2 3
E = kB T
∞ = kB T = kB T . (52)
dx x2 e−x2 (1/4)π 1/2 2
0

(The integrals can be calculated with some trickery... Ask if you want to know... don’t ask what you do not want
to know...) This is a particular instance of a general result known as the principle of ‘equipartition’: At thermal

ECE344 Fall 2009 57


equilibrium any system will have an amount (1/2)kB T of energy associated to each degree of freedom. In our
case, each particle has 3 degrees of freedom (it can move in 3 directions), so its energy will be 3 × (1/2)kB T .
Finally, that the total energy of the gas will be (3N/2)kB T .

• The Fermi-Dirac distribution function.


The arguments given above apply to ‘classical particles’, that is, particles which can be viewed as distinguishable.
Electrons and other microscopic particles do not obey the classical Maxwell-Boltzmann statistics because they
cannot be consider distinguishable. This apparently trivial distintion actually bears profound consequences. It
can be shown through very complicated arguments that indistinguishable particles with half-integer spin (h̄/2
like electrons, protons and neutrons, 3h̄/2 like weird heavy elementary particles and some nuclei, etc.) obey
Pauli’s principle. On the other hand, indistinguishable particles with zero or integer spin (0 like some nuclei,
h̄ like photons, 2h̄ like gravitons, etc.) can populate any quantum state in arbitrarily large numbers. Because
of their properties of being indistinguishable and of obeying these additional constraints, it can be shown that
half-interger-spin particle at thermal equilibrium will occupy an energy state E according not to Eq. (48) above,
but according to the same probability distribution but replacing fM B (E) with:
1
fF D (E) =  E−E  . (53)
1 + exp kB T
F

This is called Fermi function and the resulting distribution,

PF D (E)dE = A ρ(E) fF D (E) , (54)

is called ‘Fermi-Dirac’ distribution. The energy EF is called Fermi energy (or ‘Fermi level’, or often ’chemical
potential’). It is a measure of the density of particles: Suppose we are at zero T and we must fill our ‘quantum
box’ with 4 particles. Two particles will go into the first energy state, E1 , two more into the second, E2 , by
Pauli’s principle. The Fermi energy will by slightly abobe E2 , since at T = 0 FF D is a step function equal to
1 when E < EF , equal to zero otherwise.

ECE344 Fall 2009 58


Particles of zero or integer spin obey a mathematical similar (but physically profoundly different) statistics. They
will be distributed according to the same probability distributions abve, but replacing fM B (E) or fF D (E)
with
1
fBE (E) =   . (55)
E
exp k T − 1
B
This is called the ‘Bose-Einstein’ distribution. Its peculiar feature is that at very low T all particles ‘condense’
into the lowest possible energy state, since there is no Pauli’s exclusion principle to prevent this from happening.
This phenomenon, called ‘Bose condensation’, has been observed experimentally in several quantum systems.

For those interested, below we see how one can use Pauli’s principle and the indistinguishability of electrons to
derive the Fermi-Dirac distribution.
– Let’s consider how a system of Ne ‘indistinguishable’ electrons at thermal equilibrium with the environment will be distributed over
N energy levels, each with energy Ei and containing gi (i = 1, ..., N ) available states labeled by different quantum numbers, by
N
placing ni electrons in each energy level i. The total energy of the system is Etot = i=1 Ei ni .
The number gi is called the ‘degeneracy’ of the level i.
Pauli’s principle tells us that we cannot put more than one electron in each state, so ni ≤ gi .
– The number Wi of different ways in which ni electrons can be placed into the gi available states in the ith level is:

gi !
Wi = .
(gi − ni )!ni !

(Think of gi bins in which we can put a ball or leave it empty, having ni balls in total. We can put the first ball in any of the gi
bins, the second ball in any of the remaining gi − 1 bins, etc., for a total of gi ! possible ways. But, since neither the balls nor the
’voids’ can be distinguished, we must divide this number by the ni ! equivalent ways we can place the ni balls into the bins and by
the (gi − ni )! equivalent ways we can arrange the gi − ni voids.)
– Thus, the total number of possible arrangements is just
  gi !
W = Wi = .
(gi − ni )!ni !
i i

– The most likely distribution should be determined by maximizing W under the constraints of energy and particle conservation. It’s

ECE344 Fall 2009 59


convenient to maximize ln W instead. Let’s write

ln W = [ln gi ! − ln(gi − ni )! − ln ni !] .
i
Let’s use Stirling’s approximation, valid for large x:

ln x! ≈ x ln x − x ,

and write

ln W ≈ [ gi ln gi − gi − (gi − ni ) ln(gi − ni ) + (gi − ni ) − ni ln ni + ni ] =
i

= [ gi ln gi − (gi − ni ) ln(gi − ni ) − ni ln ni ] .
i
Then
 ∂ ln W  
d(ln W ) = dni ≈ [ ln(gi − ni ) + 1 − ln ni − 1 ] dni = ln(gi /ni − 1) dni .
∂ni
i i i

now comes from solving d(ln W ) = 0 subject to the conditions
Our solution  i dni = 0 (the total number of particles cannot
change) and E dn
i i i = 0 (the total energy should not change). This requires the use of the ‘method of Lagrange multipliers’:
We multiply each equation describing a constraint by a constant (a ‘multiplier’) and add these equations to the original equation:

[ ln(gi /ni − 1) − α − βEi ] dni = 0 .
i
Here the Lagrange multiplier α refers to particle conservation, the multiplier β to energy conservation. Since this must be valid for
each i, a solution will be given by:
ln(gi /ni − 1) − α − βEi = 0 for all i.
Therefore, the density of electrons in the ith energy level Ei will be:

ni 1
f (Ei ) = = .
gi 1 + eα+βEi

ECE344 Fall 2009 60


For closely spaced levels we can go to the continuum. Moreover, more complex arguments of statistical mechanics imply
α = −EF /(kB T ) and β = 1/(kB T ), where T is the temperature, kB is Boltzmann constant, and EF the Fermi energy.
Thus:
1
f (E) = .
1 + e(E−EF )/(kB T )

Here we should recall that the Fermi energy (or ‘Fermi level’, or the ‘chemical potential’ of the grand-canonical
ensemble), the total electron density, and the temperature are related, as we’ll see below.
– Some useful approximations to the Fermi-Dirac distribution are the following:

 θ(E − EF ) for kB T << EF
fF D (E) ≈ ,

e(E−EF )/(kB T ) for kB T >> EF (Maxwell-Boltzmann distribution)

where θ(x) is the Heavyside step function, θ(x) = 1 for x < 0, θ(x) = 0 otherwise.
• The electron (hole) density in the CB (VB).
Let’s now use the Fermi-Dirac distribution to calculate how electrons and holes are distributed energetically in
the CB and VB of semiconductors.
– The total density of free electrons in the lowest-energy CB of a semiconductor will be given by:

1  2
n = fF D [E(k)] = fF D [E(k)] dk .
V (2π)3
k

Going to spherical coordinates in k-space:


 2π  π  ∞  ∞
1 2 1
n = dφ dθ sin θ dk k fF D [E(k)] = 2 dk k2 f [E(k)] .
4π3 0 0 0 π 0

For an isotropic effective mass m∗ , E(k) = h̄2 k2 /(2m∗ ), so that k = (2m∗E)1/2 /h̄ and dE =

ECE344 Fall 2009 61


(h̄2 /m∗ )kdk. So:
  3/2  ∞  ∞
1 2m∗ 1/2
n = dE E f (E) = dE ρ(E) f (E) ,
2π2 h̄2 0 0

recalling the definition of the DOS at energy E we have obtained earlier.


Note that the upper integration limit extends to ∞. Of course, this is incorrect, strictly speaking, since we are
limiting ourselves to the low-energy region of the lowest-energy CB. Yet, the Fermi-Dirac distribution decays
so quickly (exponentially) at high energies, that the error we make is negligible.
Three cases are noteworthy:
1. Low-temperature, degenerate limit. At low T , recalling that f (E) is unity for E < EF but it vanishes
for E > EF , we have what’s called the degenerate limit:
  3/2  E   3/2 3
1 2m∗ F 1/2 1 2m∗EF kF
nlow T ≈ dE E = = ,
2π2 h̄2 0 3π2 h̄2 3π2

where kF = (2m∗ EF )1/2 /h̄ is the Fermi wavevector. Note that the Fermi energy is positive. This means
that the Fermi level is above the minimum of the CB for electrons (or below the maximum of the VB for
holes).
2. High-temperature, non-degenerate limit. On the contrary, at high temperature we have the so-called
non-degenerate limit in which the FD-distribution approaches the form of the MB distribution (so that this
is also called the ‘semiclassical limit’): fF D (E) ≈ e−(E−EF )/(kB T ) and
  3/2  ∞
1 2m∗n
nhigh T ≈ eEF /(kB T ) dE E 1/2 e−E/(kB T ) =
2π2 h̄2 0
  3/2  ∞
1 2m∗nkB T E /(kB T ) 1/2 −x
= e F dx x e ,
2π2 h̄2 0

ECE344 Fall 2009 62


having set x = E/(kB T ) in the last step and having indicated by m∗n the effective mass at the bottom
of the CB to distinguish it from the effective mass at the top of the VB which we shall use later. Now we
should recall the definition of the ‘Gamma’ function (which is an analytic extension of the factorial function
(n − 1)! to all real numbers):
 ∞
Γ(s) = dx xs−1 e−x .
0
Knowing that

Γ(1/2) = π, Γ(s + 1) = s Γ(s) ,
we get:
  3/2
1 2m∗nkB T
nhigh T ≈ eEF /(kB T ) = Nc e−(EC −EF )/(kB T ) ,
4 πh̄2

where in the last step we have used explicitly the fact that we have always measured EF from the
bottom of the CB. In the case of the ‘ellipsoidal’ CB of Si we should replace the isotropic mass m∗n with
mn,d = 62/3 (mL m2T )1/3 , the factor of 6 stemming from the 6-fold degeneracy of the CB minima. For
holes we get, similarly,

3/2
1 2m∗p kB T (−EG −EF )/(kB T ) −(EF −Ev )/(kB T )
phigh T ≈ e = Nv e ,
4 πh̄2

so that the product pn at equilibrium in the non-degenerate limit is

 3
1 kB T ∗ 3/2 −EG /(kB T ) 2
pn = (mn,d mp ) e = ni ,
2 πh̄2

ECE344 Fall 2009 63


the square of the ‘intrinsic carrier concentration’, ni ≈ 1.4 × 1010 cm−3 in Si. The quantities

  3/2 3/2
1 2m∗n kB T 1 2m∗p kB T
NC = and NV = ,
4 πh̄2 4 πh̄2

are called effective density of states for electrons and holes, respectively. Other useful forms are:

n ≈ ni e(EF −Ei )/(kB T ) , p ≈ ni e(Ei −EF )/(kB T ) ,

where Ei is the Fermi level in undoped (intrinsic) Si, which we shall derive shortly.
3. General case.
The general case of temperature not too high or not too low is mathematically a little more complicated,
because there is no analytical solution to the integral. However, the result is widely available in the literature
and it is expressed in terms of what are called ‘Fermi integrals’. In general we have:
  3/2  ∞   3/2  ∞
1 2m∗ 1/2 1 2m∗kB T 2 x1/2
n = dE E f [E(k)] = √ dx ,
2π2 h̄2 0 4 πh̄2 π 0 1 + ex−η

where, as before, we have changed the integration variable to x = E/(kB T ), we have set η = EF /(kB T ),

and we have multiplied and divided by 2/ π to recover the prefactor Nc introduced above. As just stated,
the integral cannot be solved in closed form, but it is a well-know integral (appropriately called Fermi-Dirac
integral of order 1/2 and labeled by F1/2 (η)). So, the electron density is usually written in terms of this
integral as:
  3/2
1 2m∗ kB T
n = F1/2 (η) = Nc F1/2 (η) .
4 πh̄2
– Determining the Fermi level.
Two more cases are extremely important: How to determine the Fermi level in an intrinsic (i.e., undoped)
semiconductor and in a doped material.

ECE344 Fall 2009 64


1. Fermi level in intrinsic semiconductors: Since the semiconductor is not electrically charged, we use the
so-called ‘charge neutrality condition’: There must be an equal number of electrons and holes. In formulae:
−(Ec −EF )/(kB T ) −(EF −Ev )/(kB T )
n = p → Nc e = Nv e ,

so  
kB T Nc
EF = Ei = Emg − ln ,
2 Nv
where we have defined the ‘mid-gap’ Emg = (Ev + Ec )/2. Since NC /NV = (m∗n /m∗p )3/2 is not
too different from unity, the log term above is usually negligible and we can almost always approximate

Ei ≈ Emg .

2. Fermi level in extrinsic (doped) Si: Use again the charge neutrality condition which now reads:
+
n = p + ND ,
+
for an n-type semiconductor. Here ND is the density of ionized donors. The probability that an electron
will occupy the energy level ED associated to the H-like impurity potential will be
1
f (ED ) = ,
1 + gD e(ED −EF )/(kB T )
where gD is a degeneracy factor dependent on the microscopic structure of the impurity, typically taken as
1/2 since only one electron can occupy the impurity state while we have derived the Fermi function assuming
+
2 electrons with opposite spins per state. Therefore: ND = ND [1 − f (ED )] ≈ ND (in words: dopants
are almost always ionized, except at very low temperatures) and:
   
NC ND
n ≈ p + ND → EF ≈ EC − kB T ln = Ei + kB T ln .
ND ni

ECE344 Fall 2009 65


An analogous expression holds for p-type semiconductors:
   
NV NA
p ≈ n + NA → EF ≈ EV + kB T ln = Ei − kB T ln .
NA ni

Let’s summarize these results. In an intrinsic semiconductor the Fermi level is essentially at midgap,
EF ≈ Ei ≈ Emg . If we dope it n-type, EF moves up towardes the minimum of the CB and its distance
from Ei – usually denoted with φB – will be:
 
ND
EF − Ei = φB = kB T ln .
ni

As ND grows we reach a point at which EF enters the CB. The semiconductor at this point becomes
degenerate and the MB-approximation we have employed for fF D (E) ceases to be valid. A correct analysis
should make use of the Fermi-Dirac integrals. Similarly, if we dope the semiconductor p-type, the Fermi level
moves ‘down’ towards the top of the VB. Its distance from the intrinsic Fermi level Ei will be
 
NA
EF − Ei = −φB = − kB T ln .
ni

The figures in the following page show the position of the Fermi level for both electrons and holes at two
temperatures in Si.

ECE344 Fall 2009 66


ECE344 Fall 2009 67
– Compensation.
In many cases both donors and acceptors are present in the same region of semiconductor. This may happen
either unintentionally (e.g., donor-like unwanted defects are present in a region intentionally doped p-type)
or intentionally (em e.g., n-type dopants are implanted in a p-type wafer). In this case the species with the
higher concentration will determine the type of the doping: If ND > NA the material will be n-type and
vice versa. However, in this case, the concentration of free electrons in the CB will be ND − NA , not ND .
This is due to ‘compensation’: The NA holes which we would have if the material were doped only p-type,
in our case recombine with NA of the ND electrons in the CB, leaving ND − NA free electrons. Thus,
everything we have discussed so far applies with the simple modification of replacing the doping NA or ND
with the ‘net doping’ |NA − ND |.
In the general case of a compensated semiconductor, local charge neutrality can be written as:
− +
n + NA = p + ND .

– Mobility and Ohm’s law.


In the next section we shall discuss in detail how free carriers move under the action of an external field. Here,
however, it is worth giving a first introduction to the subject, in order to discuss how the concentration of
dopants in a semiconductor determines its ‘resistivity’.
Consider an n-type semiconductors, just to fix the idea, and consider again the distribution of electrons in the
CB:  ∞
n = ρ(E) f (E) dE ,
EC
having indicated explicitly the lower integration limit as the CB-bottom and having dropped the subscript ‘FD’
in the Fermi-function f (E). We may rewrite this expression in terms of the electron distribution in k-space:

2
n = dk f [E(k)] . (56)
(2π)3
This way of re-expressing the carrier density shows clearly that electrons are distributed over a volume of
k-space which becomes significantly large at high density. They can populate states of significantly large

ECE344 Fall 2009 68


energy and velocity even at equilibrium. However, the net velocity is clearly zero, because for every electron
occupying a state k (and so with velocity h̄k/(2m∗)), there will be another electron occupying a state −k
(and so with velocity −h̄k/(2m∗)). One can picture the ‘electrin gas’ as a swarm of bees hovering a fixed
position: Lots of movement, high instantaneous veolcity of each bee, but no net ‘drift’: The swarm goes
nowhere fast.
Let’s now apply an electric field to the ‘swarm’ of electrons. Let’s assume that the field is positive along
the x-axis. Since for each electron we have dpx /dt = −eFx , the total momentum Px of the swarm of n
electrons will change according to

dPx  dpx
= n = − e n Fx . (57)
dt  f ield dt

This would imply an endless acceleration. However, electrons in a crystal will suffer collisions (with the charged
ionized dopants, with lattice vibrations, with other electrons, with interfaces of the device, etc.). Without
going into details of what happens during these collisions, let’s take a statistical view and assume that on the
average the electrons are going to collide once every τ seconds. Let’s also assume that at each collision each
electron loses all of its momentum. Then, the rate at which momentum is lost by the ‘swarm’ of electrons will
be: 
dPx  Px
= − . (58)
dt  collisions τ
[It is not really necessary to assume total loss of momentum at each collision. Rather, the equation above is valid provided 1/τ
is re-interpreted as the ‘momentum relaxation rate’ (that is, the rate at which momentum is lost by the ‘swarm’) rather than the
collision rate]
At steady state with the field, the momentum lost via collisions must equal and oppostite to the momentum
gained via the acceleration from the field. Therefore, equating Eq. (57) to Eq. (58):
 
dPx  dPx 
− = , (59)
dt  f ield dt  collisions

ECE344 Fall 2009 69


or
Px
− e n Fx = . (60)
τ
Since Px = nm∗ vx , where vx  is the x-component of the velocity each electron in the swarm, averaged
over all electrons, we have from the equation above:

vx  = − Fx . (61)
m∗
This should be viewed as a net ‘drift’ of the swarm along the x-direction, under the action of the field balanced
by the ‘drag’ of the collisions.
The x component of the current density, Jx , carried by the swarm of electrons will be:

e2 τ n
Jx = −envx  = Fx . (62)
m∗
The quantity µn = eτ /m∗ is called ’electron mobility’. It is measured in cm2 /Vs and it is the constant of
proportionality between the electric field and the average electron drift velocity:

vx = µn Fx . (63)

Note that Eq. (62) is simply Ohm’s law: The current is proportional to the applied bias: If we have a chunk
of semiconductor with cross-sectional area A and length L and we apply a voltage V between two ends of
the ‘chunk’, as shown in the figure below, the current I will be I = Jx A and the field along the x-direction
will be Fx = V /L. Thus, Eq. (62) can be rewritten as:

Ae2 τ n V V
I = Jx A = = A e n µ n , (64)
m∗ L L
or, defining the resistance R = L/(enµn A),

V = IR , (65)

ECE344 Fall 2009 70


which should be a familiar relation between voltage and current.

L
d
y
W
x
– Resistance, resistivity, conductivity, etc.
Looking again at the figure illustrating our ‘chunk’ of semiconductor, we should define several quantities which
we shall need in the following.

∗ Mobility. We have already seen that the mobility is the constant of proportionality between electric field
and electron velocity. In vector form, vd = µn F, where the subscript d refers to the fact that we mean
the ‘drift’ velocity of the ‘swarm’ and we have dropped the ‘average symbols‘ ... This is an ‘intensive’
property of the material; that is, it does not depend on the shape or dimensions of the sample, but only on
its intrinsic properties.
∗ Conductivity. The conductivity σ is another intensive property and it is the proportionality constant between
electric field and current:
J = σF. (66)

ECE344 Fall 2009 71


From Eq. (62) we have σ = enµn . It is measured in 1/(ohm·cm). Note that the unit of resistance ‘ohm’
(indicated by the capital Greek letter Omega, Ω) is equivalent to volt·seconds/Coulomb (Vs/C). Its inverse,
1/ohm, is often known as ’mho’ (‘ohm’ spelled backwards).
∗ Resistivity. The resistivity ρ is the final intensive quantity we need to define. It is simply the inverse of the
conductivity:
1 1
ρ = = . (67)
σ eµn n
It is measured in ohm.cm. For example, for n-type Si with 1016 donors per cubic centimeter, using µn ≈
1,500 cm2 /Vs at 300 K,

1
ρ = ≈ 0.4 Ωcm . (68)
1.602 × 1019 × 1, 500 × 1016

∗ Resistance. As we have seen before, after Eq. (64), the resistance is an ‘extensive’ quantity. This means
that is depends on the shape and dimension of the sample. For the geometry of the figure,

L
R = , (69)
enµn A

so that the resistance drops as the cross-sectional area increases (as in wires of different diameters or ‘gauge’)
and increases as the length of the sample increases.
∗ Sheet resistance. Finally, the sheet resistance is the defined as the resistance of a square of sample of
thickness d. From Eq. (69), setting W = L, so that A = Ld,

L 1 ρ
Rs = = = (70)
enµn Ld enµn d d

The sheet resistance is measured in ohm/square (Ω/✷) and it depends on the thickness of the sample.
– DOS and Conductivity Effective Masses.
The effective mass m∗ which appears in the expression for the mobility, µn = eτ /m∗ , is called the

ECE344 Fall 2009 72


‘conductivity mass’ and should be denoted as mc to distinguish it from the other effective mass we have
defined, the density-of-states (DOS) effective mass, md . We have introduced the effective mass as the inverse
of the curvature of the bottom of the CB along the direction of the applied field. For a material with an
isotropic CB, such as GaAs, this is the same in all directions. For Si (and also Ge) the situation is a little more
complicated. Recall that in Si the C has six minima close to the six symmetry points X . The energy of the
electrons in terms of their k-vectors is given by:

2 2 2
h̄2 kL kT 1 + kT 2
E(k) = + , (71)
2 mL mT

where kL is the component of the wavevector along the longitudinal principal axis of the ‘football’, and
kT 1 and kT the 2 components perpendicular to it. So, considering the field along the x direction, for the
‘footballs’ with the longitudinal axis along the x-direction the mass relevant to the mobility will be mL , but
for the other 4 footballs it will be the transverse mass mT . Since at equilibrium 1/6 of the electron occupy
each football, the inverse mass 1/m∗ averaged over all footballs (we shall call them more properly ‘ellipsoids’
from now on!) will be:  
1 1 1 2
= + . (72)
mc 3 mL mT
In a way, the conductivity mass is obtained by taking an average of the inverse masses along the three principal
directions and taking the inverse of the results. From the way we have defined it, the conductivity mass
describes the inertia of the electron, and so its acceleration and velocity under an external field. On the
contrary, the DOS mass describes the density of available states at a given energy, as we saw on page 54. In
Si the two masses differ numerically: Using mL = 0.91 (in units of the free electron mass) and mT = 0.19,
we have mc ≈ 0.26 while md = (mLm2T )1/3 ≈ 0.32.
In more general situations, the conductivity mass can be related directly to the electron group velocity, and
so to the inverse of the slope of the dispersion E(k) along the direction of the field. It also often called
the ‘optical mass’, since it is the mass which appears in optical transitions (excitation of electrons to higher
energies due to absorption of light).

ECE344 Fall 2009 73


– Temperature and doping dependence of the mobility.
Let’s consider once more the expression for the electron mobility,

µn = . (73)
mc
While mc is simply a constant, function of the material, the scattering time (or, more properly, the ’momentum
relaxation time’) τ can vary in the same material. For exemple, if the electrons collide frequently with ionized
donors (which is the case at low temperatures) , then if we consider a sample of Si with a higher doping
concentration collisions will become more frequent, the time τ will become shorter and the mobility will
decrease as the doping increases. Also, since the ‘scattering potential’ of each impurity will appear weaker to
electrons of a higher kinetic energy, increasing the temperature of the sample will reduce the frequency of the
collisions, and thus the mobility will increase.
On the contrary, if electrons collide most frequently with ions of the crystal displaced from their equilibrium
position by thermal agitation (which is the case at higher T ), then increasing the doping will clearly have no
effect, but increasing the temperature will now reduce the mobility, since collisions with ‘agitated’ ions will
become more frequent.
As a rule of thumb, at room temperature ‘phonon scattering’ (that is, collisions with thermally-agitated ions)
dominates for doping concentrations up to 1017 cm−3 or so, impurity scattering dominates above 1019 cm−3 ,
the grey area in-between being affected by both. When phonon scattering dominates, the mobility drops with
temperature as roughly µn,phon ∝ T −3/2 , while when impurity scattering dominates, µn,imp ∝ T 3/2 .
A well-known rule to ‘add’ mobilities due to different scattering processes is ‘Matthiessen’s rule’: Mobilities
are added like capacitances, by adding the inverse and taking the inverse of the results. In our example, the
total mobility due to scattering with phonons and impurities, µn,tot can be obtained from:
1 1 1
= + . (74)
µn,tot µn,phon µn,imp

ECE344 Fall 2009 74


– High-field effects and ‘hot electrons’.
Let’s recall that the mobility is the constant of proportionality between applied electric field and average
electron drift velocity,

vd = µn F .

The physical interpretation is obvious: If the scattering time τ is constant, then applying a field twice as
strong will result in the ‘swarm’ of elecrons to drift at a velocity twice as high.
However, as the field gets larger and larger some additional effects come into play. To get an idea of what
happens, we should make two observations:
1. First, note that the scattering time is an average of the scattering times, τ (E), of all electrons at all energies
in the swarm: By the usual expression of ‘weighted averages’,

1 dE 1/τ (E)ρ(E) f (E)


=
. (75)
τ dE ρ(E) f (E)

In general, the scattering ‘frequency’ (usually called the ‘scattering rate’) 1/τ (E) is a function of the
electron kinetic energy. For example, scattering with ionized impurities weakens as E grows because the
depth of the Coulomb potential appears to decrease as the electron energy increases. On the contrary, the
frequency for scattering with phonons (and many other processes) increases like ρ(E). This is a general
principle: The more ‘final states’ are available after a collision, the easier it is to scatter and so the larger is
the scattering rate.
(Think of parallel-parking your car: It takes significantly longer to park – so, a lower ‘parking frequency’ – than to exit the parking
space – and so higher ‘deparking’ frequency . This is because when parallel-parking there are a small number of ‘final spaces’ (and
so, ‘states’) available (not too close to the curb, not too close to the car in front, not too far away from the curb, not to close
to the car behind, etc.) Instead, when exiting the parking space, one van afford to be sloppier, since there are many more ‘final
spaces’ available.
So, if electrons for some reason begin to be distributed with a probability f (E) different from the equilibrium

ECE344 Fall 2009 75


Fermi-Dirac distribution, then the average τ in Eq. (75) will change.
2. As electrons are accelerated by the field, they gain an average energy eF vd τ over the distance L = vd τ
covered between collisions. Since the velocity grows as E 1/2 and τ (E) decreases as E −1/2 as E grows
(as we have just seen in the case of phonon scattering), the distance L (called the ‘mean free path’) remains
roughly constant. Therefore, the average electron energy grows roughly linearly with growing electric field.
For small field, this ‘extra’ energy, eLF , remains smaller than the average energy each electron has at
equilibrium, 3kB T . Therefore we can consider that the swarm of electrons remains at equilibrium (‘cold
electrons’) as long as
eF vdτ = eLF < kB T , (76)
or
kB T
F < . (77)
evd τ
But as soon as the field grows above the critical value Fc = kB T /(evd τ ), the collisions are not able to
keep the electrons at equilibrium. In this case the distribution f (E) begins to ‘heat up’, more and more
electrons populating higher energy states and the average electron energy increases almost linearly with
increasing field.
Putting these two observations together, we see that when scattering with phonons dominate, as soon as
electrons get hot (that is, for F > Fc ), the average scattering rate 1/τ increases. Thus, τ decreases and
so does the mobility. The linear relation vd vs. F begins to ‘droop’ (or ‘saturate’). This is illustrated
in the following figure in Si at 300 K. This figure shows calculations of the electron average energy (left)
and drift velocity (right) as a function of electric field, as obtained from several research groups worldwide.
Experimental data (as indicated in the right frame) are also available for the drift velocity. Unfortunately,
measuring the average energy is extremely hard, so direct experimental data do not exist. The left frame
shows that electrons have an average energy of (3/2)kB T ≈ 40 meV for fields smaller than about 104
V/cm. For larger fields, the electron energy increases almost linearly with the field, as expected on the grounds
of out simple arguments above. On the right frame we see that for fields below about 104 V/cm the velocity
increases linearly with the field, as implied by concept of mobility, vd = µn F . But as soon as the electrons
become hot, the velocity increases slowly and will eventually reach the saturated value of about 107 cm/s at

ECE344 Fall 2009 76


fields of the order of 105 V/cm. This is the ‘hot electron’ regime, while the regime of small fields is called the
‘ohmic regime’ (since Ohm’s law applies, as we have seen above).

– What about holes?


Everything we’ve said so far applies to holes with the obvious modifications. We shall use a hole mobility µp ,
we can define in an analogous way the resistivity of p-type semiconductors and, finally, as shown in the last
figure, we can encounter a hot-hole regime at large applied fields. Note, however, that in Si the hole velocity
is smaller than the electron velocity in the entire range, from the ohmic to the hot-hole regime. In general, the
total current flowing in a device due to the drift of carriers will be the sum of the electron and hole currents:

Jx = (enµn + epµp )Fx . (78)

Note that both electrons and holes contribute with the same sign: Holes are moving in the direction of positive
field and contribute with a plus-sign to the current, as they are positively charged. Electrons move in the
opposite direction of negative field, but contribute to the current with a negative sign.

ECE344 Fall 2009 77


– Fermi levels at equilibrium.
So far we have considered ‘homogeneous’ systems. These are systems in which the type of crystal, the doping
concentration and type (p or n), the carrier concentration and type (holes or electrons) and the applied electric
field are constant everywhere. Electronic devices, on the contrary, are intentionally highly inhomogeneous
structures. What happens to the Fermi level in such a case?
Consider the case of two different semiconductors, possibly also differently doped, brought ideally together,
forming a ‘junction’ with material 1 on the left, material 2 on the right. In general, there will be a different
doping and electron concentration in the two regions. So we might also expect different Fermi levels at
different position. However, this does not happen: As soon as the two materials are brought together and
electrons are allowed to move, they will redistribute themselves: If there are more electrons on the left, so that
the Fermi level of material 1, EF 1 , is larger than that of material 2, EF 2 , electrons will move to the right,
thus lowering EF 1 and raising EF 2 , until the two Fermi levels will be equal.
To see more precisely how this happens, let’s consider the rate at which electrons flow from left to right: At
the left we have ρ1 (E)f1 (E) electrons at energy E (having indicated with ρ1 (E) the DOS in semiconductor
1 and with f1 (E) the Fermi function {1 + exp[(E − EF 1 )/(kB T )]}−1 in material 1). These electrons
can move to the right and occupy available empty states. Since 1 − f2 (E) is the probability that a state at
energy E in material 2 will be empty, the total density of empty states on the right will be ρ2 (E)[1 − f2(E)].
Therefore, the total rate of electrons at energy E moving left-to-right will be equal to the product of the
concentration of electrons on the left and the concentration of available empty states on the right:

rate left-to-right = ρ1 (E) f1 (E) ρ2 (E) [1 − f2 (E)] . (79)

Similarly,
rate right-to-left = ρ2 (E) f2 (E) ρ1 (E) [1 − f1 (E)] . (80)
At equilibrium, since there can be no transfer of energy across the discontinuity, these two rates of fluxes at a
given energy must be equal (if they were not, electrons and energy would keep flowing one direction or other,
and we would not be at equilibrium). So:

ρ1 (E) f1 (E) ρ2 (E) [1 − f2 (E)] = ρ2 (E) f2 (E) ρ1 (E) [1 − f1 (E)] , (81)

ECE344 Fall 2009 78


or

ρ1 (E) f1 (E) ρ2 (E) − ρ1 (E) f1 (E) ρ2 (E) f2 (E)] = ρ2 (E) f2 (E) ρ1 (E) − ρ2 (E) f2 (E) ρ1 (E) f1 (E) .
(82)
The second term on the rhs cancels with the second term on lhs and we have the final result at equilibrium
f1 (E) = f2 (E), which is what we had ‘guess’ before. We conclude with the most important result that at
equilibrium the Fermi level is constant everywhere. In formula in a one-dimensional system inhomogeneous in
the x direction:
dEF (x)
=0. (83)
dx
Note that the balance of fluxes at a given energy we have used in Eq. (81) is called ‘detailed balance’.

ECE344 Fall 2009 79


Generation, Recombination, and Diffusion
• Generation of excess carriers
We have seen that at a temperature of absolute zero semiconductors are insulators: No free electrons are present
in the CB, no free holes in the VB. However, at finite temperatures the ‘thermal agitation’ of the ions of the
lattice occasionally ‘kicks’ electrons into the CB, leaving holes in the VB. The density of these free carriers is
what we have defined ‘intrinsic carrier concentration’ ni .
This is but one of many ways in which free carriers can be excited. Thermal processes such as these can
also be assisted by defects located energetically inside the gap: An electron in the VB can ‘jump’ into some
empty defect/impurity level in the gap, leaving a hole in VB; subsequently, the electron can jump even higher
from the defect-level to the CB. These two-step processes pobviously involve jumps of smaller energy and
can actually dominate the total thermal-generation rate when the density of defects is sufficiently high. In
addition, generation of free carriers can also be triggered by light, by radiation (such as other fast electrons or
ions bombarding the crystal and ‘knocking’ electrons from the VB to CB). All of these processes result in the
generation of ‘electron-hole (e-h) pairs’.
Generation of e-h pairs by light is a very important process which is at the foundation of all opto-electronic
devices, such as light-emitting diodes (LED) and lasers. When we shine light on a crystal, if the light has a
sufficiently long wavelength λ, then its energy E = hν = hc/λ will be too small to excite an electron across
the gap. Light then travels unperturbed in the material, which then appears to be transparent for light of that
wavelength (think of glass for wavelength in the visible spectrum). But if E > Eg , that is λ < hc/Eg ,
valence electrons can absorb the photon and jump to the CB. The number of such processes which occur per
unit length when a photon travels in the crystal is called the absorption coefficient, α(λ), and its reciprocal,
1/α is called the ‘absorption length’. If the light hits the sample with intensity (or ‘photon flux’) I0 , then the
intensity will decay with distance x inside the sample according to:
dI(x)
= − αI(x) ,
dx
so that I(x) = I0 e−αx. This shows that the light intensity decays by a factor of e for every absorption length

ECE344 Fall 2009 80


1/α of propagation.
Plotting α as a function of wavelength λ (or, equivalently, photon energy E = hν = hc/λ) gives important
information about the electronic structure of the crystal, mainly about the band gap (above which energy light
is absorbed, as seen above) and its DOS.

• Recombination processes
Having created e-h pairs (thermally, with light, electron or ion bombardment, etc.), they will tend to ‘recombine’:
Whenever an electron and a hole paths cross, electron can fall into the attractive potential of the hole and
’annihilate’ itself and the hole in the VB. These ‘recombination’ processes can be broadly divided into two
classes: ’Radiative’ recombination, in which the e-h energy (typically, the bandgap Eg ) is released in the form
of light, or ‘non-radiative’, in which the excess energy is transferred to the lattice in the form of extra ionic
agitation. Recombination processes can also be assisted by traps or defect: The electron first falls into a trap
with energy-level inside the gap. A passing hole can then recombine with the trapped electron.
Regardless of the nature of the recombination processes, we can describe them with a ‘recombination rate’ r :
Since at equilibrium the electron and hole concentrations n0 and p0 (the subscript ‘0‘ refers to equilibrium
values) are such that n0 p0 = n2i , recombination processes will attempt to return the ‘perturbed’ concentration
n and p to their equilibrium values, so that the recombination rate must have the form:

r = αr (np − n2i ) .

Indeed at equilibrium np = n2i so that r = 0. The ‘recombination constant’ αr depends on the microscopic
nature of process (whether recombination is radiative or not, on the nature of the trap, if trap-assisted, etc.).
Let’s assume that we have generated an excess concentration of electrons and holes, by shining light on the
sample. Let n(t) and p(t) be the concentrations of electrons and holes as function of time. If at time t = 0
we turn off the light, the excess carriers will recombine and their concentration will decrease according to the
equation:
dn(t) 2
= −r = αr ( ni − np) .
dt
Let’s assume that n(t) = n0 + δn(t), p(t) = p0 + δp(t). Since we have generated pairs, δn = δp.

ECE344 Fall 2009 81


Also, if the sample was initially doped p-type, and if the light intensity we have used was not too strong, we can
assume that we have generated a small number of e-h pairs compared to p0 . Thus:

dn(t) dδn 2 2
= = αr [ ni − (n0 + δn)(p0 + δp)] ≈ αr [ ni − n0 p0 − (n0 + p0 )δn] ,
dt dt

having ignored terms of the order of (δn)2 . Since n0 p0 = n2i , the first and second terms inside the square
bracket cancel. Since the sample was assumed to be p-type, p0 >> n0 and we can ignore n0 . Thus:

dδn(t) δn(t)
≈ − αr p0 δn(t) = − , (84)
dt τn

where τn = 1/(αr p0 ) is the ‘electron lifetime’. This equation has the solution δn(t) = δn(0) exp(−t/τn ):
The concentration of excess electrons decays exponentially from its initial value δn(0) with the time constant
τn .
Suppose that now we keep shining light on the sample and that gop is the rate at which we generate e-h pairs
(measured in pairs per unit volume and unit time, so in m−3 s−1 ). Then recombination process will attempt to
reduce the concentration of excess electrons with a term similar to Eq. (84). However at the same time light
will continue to generate excess with a rate gop . Thus the rate at which the electron density changes will be
determined by the competition of these two processes:

dδn(t) δn(t)
≈ − + gop . (85)
dt τn

At steady state we must reach a constant concentration of electrons, so that δn/dt = 0 and δn = gop τn .
Similarly, we will have a steady-state concentration of excess holes, δp = gop τp .
A way to describe the concentration of excess carriers using the same mathematical machinery used at equilibrium
is based on the introduction of the concept of ‘quasi-Fermi level’. We saw that at equilibrium we can use

ECE344 Fall 2009 82


interchangeably the variables n, p, and EF , since:
   
EF − Ei Ei − EF
n0 = ni exp p0 = ni exp . (86)
kB T kB T

Off-equilibrium we now define the ‘quasi-Fermi levels’ for electrons, EF n , and holes, EF p , such that:
   
EF n − Ei Ei − EF p
n = ni exp p = ni exp . (87)
kB T kB T

Occasionally, with noticeable lack of good taste, EF n and EF p are called ‘imref’ (‘fermi’ spelled backwards).
We can view this as a simple mathematical change of variable, but it is better to view it as a convenient way
to keep using equilibrium concepts also off equilibrium. As long as we recall that EF n and EF p are NOT real
Fermi levels and are NOT well-defined from the point of view of Statistical Mechanics, these concepts are useful.
• Diffusion
When excess carriers are excited across the gap, they are often generated in a spatially inhomogeneous way
(that is, their concentration varies depending on position). For example, when we generate carriers with light,
they will be generated in high concentration only at the spot where the light beam strikes the sample. Clearly,
such an inhomogeneous spatial distribution cannot last: The random motion of the the electrons – hitting each
other, hit by thermally agitated ions, etc. – will randomly distribute them throughtout the entire volume. Such
a process, which occurs very frequently in nature, is called ‘diffusion’. The textbook draws the nice analogy with
molecules of perfume in a room. At first very localized within the small volume around the source of perfume
(the bottle or the area were we spray the perfume), with time their scent will ‘diffuse’ thorughout the entire
room. This is due to the random nature of the collision-dominated motion of the molecules: At the boundary
between the perfume-containing volume and the rest of the room, at a given time each molecule will have, on
average – a 50% probability of being scattered outside the volume, a 50% chance of being scattered back into
the perfume-containing volume. So, on average, half of the molecules will leave the original volume. With time,
the perfume-containing volume expands while, at the same time, the concentration of perfume molecules in the
original volume decreases. The process continues untill the entire volume is filled with a uniform concentration of

ECE344 Fall 2009 83


molecules. Therefore, note that the diffusion process is driven by the ‘gradient of the concentration’: Wherever
the concentration of molecules is larger, molecules will diffuse away until a uniform concentration is reached.

n(x)
n(x) n1

n(x)
t1 n2

t2 > t 1
t3 > t 2
x x
0 x x0–L x0 x0+L

To see how we can describe this process mathematically, let’s consider a simpler one-dimensional situation
illustrated in the figures above. The first figure shows how an ensemble of particles originally (t = 0) at x = 0
spreads as time increases. Let’s now consider one of those ’bell-shaped’ curves (they are Gaussians, as we shall
see later) and divide them into the indicated rectangles as follows. Let’s assume that electrons collide with a
mean free path L and a scattering time (or ‘mean free time’) τ (as we have seen discussing the mobility).
Therefore, in any given time step τ , electrons can move with equal probability to the left or to the right by an
amount L. Let’s now divide the x-axis into segments of length L. In each segment there will be a number
nj of electrons, where j is an index labeling the segment. Let’s now consider the final figure showing a pair
of adjacent segments, labeled 1 and 2, on the two sides of the point x0 , containing n1 and n2 electrons
respectively. In a time τ the number of electrons leaving the ‘bin’ number 1 will be n1 AL, where A is the
area of the bin in the y and z directions. Half of them will move to the left, and only a half will move to bin
2. With a similar argument applied to bin number 2, the total net number of electrons going across the point
x0 will be n1 AL/2 − n2 AL/2 = (AL/2)(n1 − n2 ). In the limit in which scattering is very frequent, so

ECE344 Fall 2009 84


that L becomes very small, we can approximate n1 with the electron density at the center of segnment 1 (so,
n1 ≈ n(x0 − L/2)), we can similarly approximate n2 with n(x0 + L/2), so that:

n(x0 − L/2) − n(x0 + L/2) dn(x)


n1 − n 2 ≈ L → −L ,
L dx
the last step being simply a Taylor expansion truncated at the first-derivative term. Therefore, the net flux of
electrons at x0 (that is, electrons per unit area going from bin 1 to bin 2 per unit time) will be

L2 dn(x0 )
Φ(x0 ) = − . (88)
2τ dx

The quantity L2 /(2τ ) is denoted by Dn and is called the electron diffusion constant. Therefore, since
electrons carry negative charge, the current density due to the diffusion process will be:

dn(x)
jn,dif f (x) = e Dn . (89)
dx
Similarly for holes:
dp(x)
jp,dif f (x) = − e Dp . (90)
dx
Note the different sign: Carriers always diffuse in the direction of negative spatial derivative of the concentration
(i.e., decreasing concentration). This ‘minus’ sign is compensated by the negative electron charge, but it is still
present when dealing with holes.
• The drift-diffusion equations
Combining Eqns. (89) and (90) with Eqns. (62) and (78), we finally have the complete expressions for the
electron and hole current densities due to both the drift and the diffusion processes:

dn(x)
jn (x) = eµn n(x)Fx (x) + eDn , (91)
dx

ECE344 Fall 2009 85


dp(x)
jp (x) = eµp p(x)Fx (x) − eDp , (92)
dx
and, for the total current,
j(x) = jn (x) + jp (x) . (93)
Equations (91) and (92) are called the drift-diffusion equations (DDEs) and constitute the basic ingredient
of what we are going to study next. The figure below hopefully clarifies the directions of the current densities
(black, solid lines, jn and jp ) and of the particle fluxes (red, dashed lines, φn and φp ) due to the drift and
diffusion processes.

F(x)

jp(diff) and jp(drift)

n(x) jn(diff)

jn(drift)

p(x) φp(diff) and φp(drift)

φn(diff)

φn(drift)
• Einstein’s relation.
Let’s consider an inhomogeneous situation in which the semiconductor is doped non-uniformly. Of course, this
implies that at equilibrium there will be a gradient of the carrier concentration. From what we have seen above,
a gradient of the carrier concentration will induce a ‘diffusion current’. But since at equilibium we cannot have
current flow, the diffusion component of the current must be compensated exactly by a drift current caused

ECE344 Fall 2009 86


by ‘built-in’ fields. These are due to what we saw before (page 78): Even in an inhomogeneous situation
that the Fermi level must remain constant throughout the sample. Therefore, in order to accommodate a
spatially-dependent carrier concentration, we must have spatially-dependent band edges, EC (x) and EV (x)
and, of course, also intrinsic levels Ei (x) ≈ Emg (x). Since by definition the electric field F (x) is defined in
terms of an electrostatic potential F (x) = −dφ(x)/dx and the potential energy V (x) is, also by definition,
V (x) = −eφ(x), we have:

dφ(x) 1 dV (x) 1 dEi (x)


F (x) = − = = . (94)
dx e dx e dx
This is called the ‘built-in field’ (as it is ”built internally” by the doping gradient). Therefore at equilibrium we
have from Eq. (91):
 
dn(x) dEi(x) eDn dEF (x) dEi(x)
0 = jn (x) = eµn n(x)Fx(x) + eDn = µn n(x) + n(x) − .
dx dx kB T dx dx
(95)
Since dEF /dx = 0, we have from the equation above:
e
µn = Dn , (96)
kB T

which is called ‘Einstein’s relation’. It a very important equation which allows us to calculate µ knowing D and
viceversa. That such a relation should exist is obvious from the fact that the ‘mean free time’ (or relaxation
time) τ enters both the definition of the mobility, µn = eτ /m∗ , as well as that of the diffusion constant,
Dn = L2n /(2τ ). Since the meanfree path Ln ∝ τ , then Dn ∝ τ , like the mobility, so that the two quantities
must be related.
• The continuity and diffusion equations.
In the DDEs above, Eqns. (91) and (92), we have ignored recombination effects. These matter, since they
induce a spatial variation of the carrier concentration and, thus, can affect the diffusion current.
In order to see how we can account mathematically for recombination effects, let’s consider a semiconductor

ECE344 Fall 2009 87


sample shaped like a wire of cross sectional area A and doped p-type. Let x be the coordinate running along
the axis of the wire and jp (x) the hole current density in the wire at position x. Consider a chunk of the wire
starting at x and of length ∆x. The number of holes per unit volume A∆x entering the ‘chunk’ per unit time
will be jp (x)/(e∆x), the number of those leaving it will be jp (x + ∆x)/(e∆x). Let’s assume that the hole
lifetime in the sample is τp . Then, the density of holes present in the chunk at a given time will be given by
the net in-flux into the chunk, [jp (x) − jp (x + ∆x)]/(e∆x), minus the number of holes disappearing via
recombination per unit time, δp/τp . Thus,

∂p 1 jp (x) − jp (x + ∆x) δp
= − . (97)
∂t e ∆x τp
(Note the use of partial derivatives because the unknown function, δp, depends both on time t and on position
x). In the limit of ∆x → 0, we can replace the first term on the right-hand side with the derivative
−∂jp (x)/∂x, so that the equation above becomes:
∂p ∂δp 1 ∂jp δp
= = − − . (98)
∂t ∂t e ∂x τp
This is called the continuity equation for holes. Similarly we have for electrons:
∂δn 1 ∂jn δn
= − . (99)
∂t e ∂x τn

If the current is dominated by diffusion, then jp ≈ −eDp ∂δp


δx . Insering this into Eq. (98), we get the diffusion
equation:
∂δp ∂ 2 δp δp
= Dp − , (100)
∂t ∂x2 τp
and similarly for electrons:
∂δn ∂ 2 δn δn
= Dn − . (101)
∂t ∂x2 τn

ECE344 Fall 2009 88


• Two important examples.
1. Steady-state injection.
Suppose that we shine light of very short wavelength on the surface of our semiconductor. This type of
light is absorbed in a very short distance (that is, its absorption length 1/α is very short), so that we may
safely assume that electron-hole pairs are generated only at the surface of the semiconductor. Elsewhere the
generation rate is zero. If we assume that the density of carriers generated by the light is smaller than the
density of free carriers in the sample (so that we may assume that the sample is everywhere charge neutral),
there will be no electric field and only diffusion and recombination will matter. Thus, from Eqns. (100) and
(101) at steady state (i.e., ∂δp/∂t = 0 and similar for electrons) the equation controlling the density of
excess carriers δp(x) and δn(x) as a function of depth x away from the surface of the sample will be:

d2 δp δp δp
= = , (102)
dx2 Dp τp L2p

d2 δn δn δn
= = , (103)
dx2 Dn τp L2n

where Ln = Dn τn and Lp = Dp τp are the diffusion lengths of electrons and holes respectively.
These represent the distance over which electrons and holes diffuse before recombining, as we shall see below.
The general solution of Eq. (102) is:

x/Ln −x/Lp
δp(x) = C1 e + C2 e , (104)

where the integration constants C1 and C2 must be fixed using the required boundary conditions. Since
infinitely away, deep in the sample, we expect vanishing density of excess carriers, C1 must be set to zero.
(Note that if this were not the case, we would have a density of holes growing without limit as we move deep
inside the sample. This is clearly a physically unacceptable solution.) If we know that ∆p is the steady-state

ECE344 Fall 2009 89


surface concentration of holes (determined by the intensity of the light), then

−x/Lp
δp(x) = ∆p e . (105)

As we said above, Lp is the average distance over which the hole will diffuse before recombining. To see this,
let’s go back once more to the concept of ‘weighted average’. We want to compute the average distance at
which holes will be found, weighted by their concentration which is controlled by recombination. Since holes
are generated only at the surface of the sample, this will tell us exactly what we want: How far, on average,
holes will diffuse from the surface inside the sample before disappearing because of recombination. So, the
average position of the holes weighted by their concentration will be:
∞



−y 
dx x δp(x) L2 ∞
dy y e−y −ye  + 0 dy e−y

0 ∞
0 ∞
0∞
p
= = Lp = Lp , (106)
dx δp(x) L dy e−y dy e −y
0 p 0 0

where we have changed dummy integration variable y = x/Lp . This is the result we were after. Finally, if
we want to calculate the steady-state diffusion current flowing ‘vertically’ through the device, we can obtained
it from Eq. (92) in the absence of electric field:

dp(x) p(x)
jp (x) = − eDp = eDp . (107)
dx Lp

2. Time of flight experiments.


An experimental verification of Einstein relation can be obtained by measuring how a ‘bunch’ of photo-
generated electron-hole pairs moves under the action of a (small) electric field. This was done by Haynes and
Shockley in 1951. A brief analysis of this experiment will also help us derive the basic solution of the diffusion
equation.

ECE344 Fall 2009 90


Take a laser (did they exists in 1951?) and shine its narrow-focussed beam onto a spot in a semiconductor
crystal. Let’s assume that the laser beam is turned on only for a very short time, so that we actually generate
electron-hole pairs in a brief ‘pulse’. Let’s assume the beam so narrow that we can ignore its spread compared
to the other length scales of the problem (mainly, the diffusion length and dimension of the sample). Let’s
also apply a dc bias to the sample, so that after the laser pulse electrons and holes begin to drift in opposite
directions. For simplicity we now deal with holes, but an equivalent discussion applies to electrons as well.
From our knowledge of the position of the beam (say, x = x0 ) and of the time of the laser pulse (say,
t = t0 ), we also know the position of the ‘packet’ of holes at t = t0 . Under the action of the dc field these
holes will drift. If we measure with an ammeter the ‘drift time’ td at which we detect the current due to the
drifting holes at some location x = x0 + L (that is, at a distance L away from the original spot), we can
obviously calculate the hole drift velocity,
L
vd = .
td
If V is the voltage drop across the length L covered by the hole packet, then the average electric field will be
F = V /L and we can obtain the hole mobility from:

vd L2
µp = = .
F V td

This reasoning applies exactly only if the hole ‘packet’ drifts while retaining its original narrow shape. However,
diffusion prevents this. Instead, the packet spreads while shifting. Ignoring for now the drift component and
recombination processes, the equation controlling the spreading of the packet is just the diffusion equation:

∂δp ∂ 2p
= Dp , (108)
∂t ∂x2
which has the solution:
δP 2
δp(x) = e−x /(4Dp t) , (109)
2 πDp t

ECE344 Fall 2009 91


where ∆P is the number of holes per unit area created by the laser pulse. This is a Gaussian function
(which we saw before on page 84) centered at x = 0 and with ‘standard deviation’ (or ‘spread’, or ‘width’)
σ = Dp t. As time goes by, the Gaussian gets wider. When drift is included, the Gaussian will spread at
the same rate, but its ‘central location’ will drift with velocity vd, so that the hole density is described by the
drifting packet:
δP 2
δp(x, t) = e−(x−vd t) /(4Dp t) . (110)
2 πDp t
If we hook an oscilloscope to the cathode, we can detect the arrival of the hole packet. By measuring the
arrival time we can measure the mobility, as we saw before. By measuring the width of the packet, we can
measure the diffusion constant, as explained in the text (Chapter 4.4.5). Comparing the two, we can establish
the validity of Einstein relation, as Haynes and Shockley did in 1951.
• Gradients of quasi-Fermi levels.
A convenient trick to reformulte the DDEs is to make use of the quasi-Fermi levels. The total electron current,
given by the DDE Eq. (91), is:

dn(x)
jn (x) = eµn n(x)Fx (x) + eDn . (111)
dx
Now, let’s rewrite the gradient of the electron concentration as:
 
dn(x) d  (En −Ei )/(kB T )
 n(x) dEn dEi
= ni e = − . (112)
dx dx kB T dx dx

Using Einstein’s relation, we can rewrite Eq. (111) as:


   
n(x) dEn dEi dj n dEi
jn (x) = eµn n(x)Fx (x) + eDn − = eµn n(x)Fx (x) + µn n(x) − .
kB T dx dx dx dx
(113)

ECE344 Fall 2009 92


But Eq. (94) tells us that dEi/dx = eFx , the electric field, so that
 
n(x) En dEi
jn (x) = eµn n(x)Fx (x) + eDn − =
kB T dx dx
dEn dEn
eµn n(x)Fx (x) + µnn(x) − eµn n(x)Fx (x) = µn n(x) . (114)
dx dx
In this form we see that both drift and diffusion are lumped into the derivative of the quasi-Fermi level. Finally,
we can reformulate the DDEs as modified Ohm’s law:
dEn dEn
jn (x) = µnn(x) = σn(x) , (115)
dx dx

dEp dEp
jp (x) = µp p(x) = σp (x) , (116)
dx dx
where σn and σp are the position-dependent electron and hole conductivities.
• Summary of the semiconductor equations.
Let’s summarize the basic semiconductor equations which we need in order to calculate the current flowing in a
semiconductor device:
1. The Drift-Diffusion Equations for electrons and holes. They tell us how the charge carriers move under
the action of the external field and by diffusion. In their general three-dimensional version (recall that bold
characters reprent vectors and that ∇ = (∂/∂x, ∂/∂y, ∂/∂z) is the ‘gradient’ differential operator), they
are:
jn (r, t) = eµn n(r, t)F(r, t) + eDn ∇n(r, t) , (117)
and:
jp (r, t) = eµp p(r, t)F(r, t) − eDp ∇p(r, t) . (118)

ECE344 Fall 2009 93


2. The Continuity Equations. They allow us to calculate the concentration of electrons and holes everywhere
in the device. Including recombination and generation processes, in their 3D version they can be written as:

∂δn(r, t) 1 δn(r, t)
= ∇ · jn (r, t) − + gn (r, t) , (119)
∂t e τn

∂δp(r, t) 1 δp(r, t)
= − ∇ · jp (r, t) − + gp (r, t) . (120)
∂t e τp
3. Poisson Equation, finally, allows us to calculate the electric field F(r) appearing in Eqns. (117) and (118)
once we know the boundary conditions (that is, the voltages applied at the edges of the sample) and the
+
spatial distribution of the electrons and holes, n(r) and p(r), and of the ionized impurities, ND (r) and

NA ( r ) :
+ −
− ∇ · [∇φ(r, t)] = e [ ND (r) + p(r, t) − NA (r) − n(r, t) ] , (121)
with
F(r, t) = − ∇φ(r, t) . (122)

Note that we use only Poisson equation (so, electrostatics) and not the full set of Maxwell’s equations (which
gives the electrodynamics of the system), because we are tacitly assuming that all temporal variations are very
slow.
Note that these equations are coupled: The electric field depends on the location of the charges. But the charge
carried by electrons and holes moves under the action of the field itself. Therefore, in most cases, we must solve
them self-consistently. This is often a rather complicated mathematical task. In the following, when studying
electronic devices, we shall make some approximations to simplify this problem.

ECE344 Fall 2009 94

You might also like