EE 223

Applied Quantum Mechanics II

David Miller
k·p theory
k·p theory

k·p method
k·p method

For devices
we want to know semiconductor behavior
near band maxima or minima
We can construct a simple, semi-empirical
model
with only a few parameters
that can calculate properties and
relate different phenomena, such as
effective masses, optical absorption
strengths and band gap energies
k·p method

We start by formally writing down a single electron

wavefunction in a crystal in Bloch form
for a state in band n
with wavevector k
 nk  r   unk  r  exp  ik  r 
Note we explicitly indicate that the unit cell function
is in general different in different bands and for
different k
k·p method

We will be substituting this Bloch-form wavefunction

 nk  r   unk  r  exp  ik  r 
into the Schrödinger equation
2 2
   nk  r   VP  r  nk  r   En  k  nk  r 
2mo
where we have chosen the notation En  k  for the
corresponding eigenenergy of the state of
wavevector k in band n
We will need to take the appropriate derivatives of  nk  r 
 k·p method

 
With  unk  r  exp  ik  r    unk  r    ikunk  r  exp  ik  r  and

 2 unk  r  exp  ik  r     2unk  r   2ik unk  r   k 2unk  r  exp  ik  r 

then 2 2
   nk  r   VP  r  nk  r   En  k  nk  r 
2mo
becomes
ˆ    2k 2 
Ho  k  pˆ  unk  r    En  k    unk  r 
 mo   2mo 
  2
where pˆ  i and Hˆ o   2  VL  r 
2mo
k·p method

Note that, mathematically, if we choose k = 0, then

ˆ    2k 2 
Ho  k  pˆ  unk  r    En  k    unk  r 
 mo   2mo 
becomes Hˆ u  r   E  0  u  r 
o n0 n n0

These un 0  r  are therefore the solutions for the unit

cell functions at k = 0
Now comes a key step
Because the un 0  r  are the solutions to an eigenequation
they form a complete set for describing unit cell
functions
k·p method

So we can expand any unk  r  as

unk  r    annk un0  r 
n

We are expanding E
the unit cell function in band
n and wavevector k
in the set of unit cell
functions of all the bands
indexed by n -p/a 0 p/a
for k = 0 k
k·p method

The expansion
unk  r    annk un0  r 
n
is sometimes known as the E
Luttinger-Kohn representation
When we add in some amount of
the unit cell function un0  r 
from a particular band n
we say we are mixing in some
-p/a 0 p/a
of that "band" k
k·p method

Thus far, we have merely rewritten the Schrödinger

equation for the electron given its known Bloch form
Now presume (or pretend) that we know the
wavefunctions un0(r) and energies En(0) at k = 0
Though we do not really know the wavefunctions
we will only need some specific matrix elements
using these wavefunctions
and these matrix elements can be deduced from
other experimental measurements
k·p method

With this presumption

we will pretend also that
we only need to consider a small number of basis
functions, , to analyze the problem,
This is an analytic “finite basis set approximation.”
giving exact results for a somewhat artificial problem
If we have chosen our limited basis set well
we will have a good first approximation result
and we could add other terms as perturbations
 k·p theory

Setting up the k·p method

Setting up the k·p method

Now we can formally set up the k·p mathematics

Substituting the expansion unk  r    annk un0  r 
n

ˆ    2k 2 
into  H o  k  pˆ  unk  r    En  k    unk  r 
 mo   2mo 
gives
ˆ    2k 2 
Ho  k  pˆ   anun0  r    En  k     anun0  r 
 mo  n  2mo  n
 Setting up the k·p method

Now Hˆ oun0  r   En  0  un0  r 

ˆ    2k 2 
so  H o  k  pˆ   anun0  r    En  k     anun0  r 
 mo  n  2mo  n
becomes
  2k 2 
n an En  0  un0  r   m k  pˆ n anun0  r    En  k   2m  n anun0  r 
o  o 

Now we are going to multiply on the left by un0  r 

and integrate over a unit cell
Setting up the k·p method

Remember the un 0  r  for different n

are all orthogonal
because they are the solutions of the
eigenequation Hˆ oun 0  r   En  0  un 0  r 
and we can choose them to be normalized
Therefore  un0  r  un0  r  d 3r   nn
unit cell
We can also define the matrix element
p nn   un0  r  pˆ un0  r  d 3r
unit cell
 Setting up the k·p method

With  un0  r  un0  r  d 3r   nn and p nn   un0  r  pˆ un0  r  d 3r

unit cell unit cell

then multiplying
  2k 2 
n an En  0  un0  r   m k  pˆ n anun0  r    En  k   2m  n anun0  r 
o  o 

on the left by un0  r  and integrating over a unit cell

gives
    2k 2 
n an En  0   nn  m k  n anp nn    En  k   2m  n an nn
o  o 
Setting up the k·p method

Rearranging
    2k 2 
n an En  0   nn  m k  n anp nn    En  k   2m  n an nn
o  o 
gives
 2k 2   
n  En  0   2m   nn  m k  p nn  an  En  k  an
 o  o 
Note this is just a restatement the Schrödinger wave
equation
for a periodic potential
with no other approximations
 k·p theory

k·p model for a two-band semiconductor

k·p model for a two-band semiconductor

Now we presume that only two bands are important

We also presume (as is usually the case) that
the un 0  r  have definite parity
Now, the spatial derivative in p̂
will flip definite parities
e.g., if un0  r  is even in some direction
then pˆ un0  r  is odd in that direction
So, from p nn   un0  r  pˆ un0  r  d 3r
unit cell
we have p nn  0
k·p model for a two-band semiconductor

With our two bands (so n  1, 2; n  1, 2 )

and with p nn  0
 2k 2   
n  En  0   2m   nn  m k  p nn  an  En  k  an
 o  o 
can be rewritten in matrix form as
 2k 2  
 E1  0   2m k  p12 
 o mo   a1   E  k   a1 
    a 
 2 k 2   a2   2
 k  p 21 E2  0   
 mo 2mo 
k·p model for a two-band semiconductor

To solve this equation for the eigen energies

we set the appropriate determinant to zero, i.e.,
2k 2 
E1  0    E k  k  p12
2mo mo
0
 k
2 2
k  p 21 E2  0    E k 
mo 2mo
The operator p̂ is Hermitian
and so p12  p21
which will simplify the determinant somewhat
 k·p model for a two-band semiconductor

Let us also presume for simplicity here

(i) that we know for some reason that p12 is isotropic
so k  p12  kp12 and k  p 21  k  p12

 kp12
(ii) that  2 k 2 / 2mo is negligible compared to E  k 
as will be the case if the resulting bands turn out to
have very light effective masses
2k 2  
E1  0    E k  k  p12 E1  0   E  k  kp12
2mo mo mo

 2
k 2
 *
k  p 21 E2  0    E k  kp12 E2  0   E  k 
mo 2mo mo
k·p model for a two-band semiconductor

So our simplified determinant equation becomes


E1  0   E  k  kp12
mo
0
 *
kp12 E2  0   E  k 
mo
Choosing the energy origin as E1  0   0
2
defining Eg  E2  0   E1  0  and E p 
2
p12
m0
we have
2k 2
 E  k   Eg  E  k    E p 0
2mo
k·p model for a two-band semiconductor
2k 2
Solving the quadratic  E  k   Eg  E  k    E p 0
2mo
for any given value of k
gives two solutions for E(k)
Rewriting the quadratic in standard form, we have
2k 2
 E  k    E  k  Eg  E p 2m  0
2

which has roots o

 2 2
k
Eg  Eg2  4 E p
2mo Eg  4E p 2k 2 
E k    1  1  2 
2 2  Eg 2mo 
k·p model for a two-band semiconductor

Using 1  x  1  x / 2 for small x

we approximating the roots for small values of k
Eg  4 E p  2 k 2  Eg   2 E p  2 k 2  
1  1  2  1  1  2  
2  Eg 2mo  2   Eg 2mo  

so the roots become, approximately

E p 2k 2 E p 2k 2
E  k   Eg  and E  k   
Eg 2mo Eg 2mo
k·p model for a two-band semiconductor

We can rewrite these results

E p 2k 2 E p 2k 2
E  k   Eg  E k   
Eg 2mo Eg 2mo
as
2k 2 2k 2
E  k   Eg  E k  
2meff 2 2meff 1
where
Eg Eg
meff 2  mo meff 1   mo
Ep Ep
Two-band k·p model
Hence this model gives “conduction”
a conduction-like band starting at band
energy Eg
Eg
with effective mass meff 2  mo
Ep
and a valence-like band starting at
energy 0 Eg
with a negative (i.e., hole-like) E
effective mass meff 1   mo g
Note for constant Ep
Ep “valence”
band
these effective masses are  Eg
Two-band k·p model
“conduction”
Note the effective mass is "caused"
band
by the "interaction between the
bands"
If the p12 and p21 matrix elements
were zero
the original matrix would be Eg
diagonal
and both bands would simply
have the free electron mass “valence”
band
Two-band k·p model

Knowing the eigenenergies “conduction”

we can work out the eigenvectors band
or eigenfunctions
With the simplifying approximations
and definitions we made
the matrix equation is now Eg
  
 0 mo
kp12 
a  a 
   1   E k   1  “valence”
    a
 m kp12 Eg   2   a2  band
 o 
Two-band k·p model

Substituting the now-known E(k) into “conduction”

   band
 0 kp12 
mo  a1   a1 
     E k   
     a2   a2 
 m kp12 Eg 
 o  Eg
lets us to solve for a1 and a2 for a
given k in a given band
giving us the unit cell function
“valence”
in terms of the two basis unit cell band
functions
Two-band k·p model

E.g., for the upper band, we have “conduction”

 E p 2k 2   band
  E g  E 2m kp12 
 g o mo   a21k   0
   E p  2 k 2   a22 k 
 kp12  
 mo Eg 2mo 
Eg
Presuming we know p12 (and hence Ep)
for a specific k, we solve the two
simultaneous equations here to get the
“valence”
eigenvector elements a21k and a22k in
band
u2k  r   a21k u10  r   a22 k u20  r 
Two-band k·p model

E.g., for the simple case of k = 0 “conduction”

 E p 2k 2   band
  E g  E 2m kp12 
 g o mo   a21k   0
   E p  2 k 2   a22 k 
 kp12  
 mo Eg 2mo 
Eg
the only non-zero term here is the
upper left term, equal to –Eg
so  Eg a21k  0 so only a22k is non-zero
“valence”
and the unit cell function is u20(r) band
as expected
Two-band k·p model

Away from k = 0, however, “conduction”

the coefficients a21k and a22k in band
u2k  r   a21k u10  r   a22 k u20  r 
are both, in general, non-zero
We would find a growing admixture of
u10(r) as we move away from k = 0 Eg
This is a simple example of "band mixing"
As we move away from k = 0
unit cell wavefunctions no longer have “valence”
the exact symmetries found at k = 0 band