For
each
of
the
six
ellipsoids,
two
of
the
masses
are
the
light,
transverse
effective
mass,
mt*
and
the
mass
along
the
axis
is
the
heavier,
longitudinal
effective
mass,
m* .
Fig.
1
Constant
energy
surfaces
of
silicon.
1
The
density-‐of-‐states
is
an
important
quantity
that
can
be
derived
from
the
bandstructure.
For
a
parabolic
band
described
by
(2),
the
result
in
3D
is
(m ) 2 ( E − EC )
3/2
*
D3D ( E) = π 23
( E > E ) .
C
(4)
For
ellipsoidal
bands
described
by
(3),
the
result
is
more
complicated,
but
we
can
make
it
look
simple
by
defining
a
“density-‐of-‐states
effective
mass”
so
that
eqn.
(4)
becomes
(m ) 2 ( E − EC )
3/2
*
D3D ( E) = DOS
π 2 3 ( E > E )
C
(5)
where
for
ellipsoidal
bands
m*DOS = ( gV ) (m m )
2/3 1/3
* *2
t
.
(6)
For
the
conduction
band
of
Silicon,
gV = 6
,
so
m*DOS = ( 6 ) (m m )
2/3 1/3
2
t
(7a)
For
the
conduction
band
of
Ge,
gV = 4 ,
so
m*DOS = ( 4 ) (m m )
2/3 1/3
2
t
(7b)
For
other
types
of
parabolic
bandstructures,
appropriate
density-‐of-‐states
effective
masses
could
be
defined
to
make
the
correct
density-‐of-‐states
look
as
simple
as
(5).
For
example,
see
R.F.
Pierret
(Advanced
Semiconductor
Fundamentals,
2nd
Ed.,
2003,
p.
96)
for
the
valence
band
density-‐of-‐state
effective
mass.
One
can
also
define
a
conductivity
effective
mass.
For
a
spherical,
parabolic
energy
band,
the
conductivity
is
q τm
σ = nq ,
(8)
m*
where
τ m
is
the
average
momentum
relaxation
time.
For
isotropic
scattering,
τ m ,
and
τ m ,
so
we
use
the
density-‐of-‐state
effective
mass
to
determine
τ m .
The
carrier
2
density
involves
an
integration
of
the
density-‐of-‐states,
for
which
we
also
use
the
density-‐
of-‐state
effective
mass.
But
what
should
we
use
for
the
effective
mass
in
the
denominator
of
(8)?
Assume
there
is
an
electric
field
in
the
x-‐direction;
we
expect
electrons
to
respond
to
the
electric
field
with
the
effective
mass
in
direction
of
the
electric
field.
Near
equilibrium,
one
sixth
of
the
electrons
are
in
each
of
the
six
ellipsoids.
For
ellipsoids
one
and
two
n q τm
σ 1,2 = q
(9a)
6 m*
while
for
ellipsoids
three
through
six,
n q τm
σ 3−6 = q .
(9b)
6 mt*
The
total
conductivity
is
σ = 2σ 1 + 4σ 3
(9c)
or
⎡ 1 2 ⎤
σ = nq ⎢ * + * ⎥ q τ m .
(9d)
⎣ 3m 3mt ⎦
We
can
write
(9d)
in
a
simple
form
like
(8)
by
defining
a
conductivity
effective
mass
q τm
σ = nq
(10a)
mc*
where
1 1 2
*
= * + * .
(10b)
mc 3m 3mt
( )
Now,
what
about
the
distribution
of
channels,
M 3D E ,
which
for
spherical,
parabolic
bands
is
given
by
3
m*
M 3D ( E ) = ( E − EC ) ( E > E ) .
(11)
2π 2 C
What
effective
mass
do
we
use
when
the
bands
are
ellipsoidal?
The
answer
is
m*DOM ,
the
distribution-‐of-‐modes
effective
mass.
( )
To
compute,
m*DOM ,
we
begin
with
the
general
description
of
M E for
a
general
band:
h
M 3D ( E ) ≡ ∑ υ δ ( E − Ek )
.
(12)
2L k x
(See:
Jeong,
Changwook;
Kim,
Raseong;
Luisier,
Mathieu;
Datta,
Supriyo;
and
Lundstrom,
Mark
S.,
"On
Landauer
versus
Boltzmann
and
full
band
versus
effective
mass
evaluation
of
thermoelectric
transport
coefficients,”
J.
Appl.
Phys.,
107,
023707,
2010.)
When
(12)
is
evaluated
for
ellipsoidal
bands,
we
find
fo
each
ellipsoid,
m*DOM = m*y mz*
(13a)
where
we
have
assumed
transport
in
the
x-‐direction.
For
Si,
we
add
the
channels
in
each
ellipsoid
to
find
m*DOM = 2mt* + 4 mt*m* .
(13b)
It
is
instructive
to
put
numbers
in.
For
Si,
we
find
m*DOS = ( 6 ) (m m )
2/3 1/3
2
t
= 1.06m0
(14a)
−1
⎡ 1 2 ⎤
m = ⎢ * + * ⎥ = 0.26m0
*
c
(14b)
⎣ 3m 3mt ⎦
m*DOM = 2mt* + 4 mt*m* = 2.04m0 ,
(14c)
which
shows
that
the
numerical
value
of
these
masses
can
be
quite
different.
4
Now
let's
ask
another
question.
We
know
that
for
a
non-‐degenerate
semiconductor,
we
can
estimate
the
mean-‐free-‐path
from
the
diffusion
coefficient,
which
we
obtain
from
the
mobility
with
the
Einstein
relation.
kT υ λ
Dn = B µ n = T 0
(15)
q 2
where
the
uni-‐directional
thermal
velocity
is
2k BT
υT = .
(16)
π m*
What
effective
mass
do
we
use?
To
answer
this
question,
we
should
begin
at
the
beginning.
*
2q 2 2q 2 m k BT
σ= λ0 M 3D = λ0 F (η )
(17a)
h h 2π 2 0 F
which
is
(A.34)
in
Near-‐Equilibrium
Transport
by
Lundstrom
and
Jeong.
The
effective
mass
in
(17)
must
be
m*DOM
*
2q 2 mDOM k BT
σ= λ0 F 0 (η F ) = nqµ n .
(17b)
h 2π 2
We
also
know
that
3/2
⎡ m* k T ⎤
n = N CF 1/2 (η F ) = 2 ⎢ DOS 2B ⎥ F 1/2 (η F ) ,
(18)
⎣ 2π ⎦
so
from
(18),
we
find
1 2q m*DOM k BT
µn = λ F 0 (η F )
n h 0 2π 2
(19)
⎡ 2π 2 ⎤ 2q m*DOM k BT F 0 (η F )
3/2
=⎢ * ⎥ λ0
⎣ mDOS k BT ⎦ h 2π 2 F 1/2 (η F )
Assuming
MB
statistics
and
simplifying,
we
find
( )
2
λ0 2k BT m*DOM
µn =
(20)
( k BT q ) π m*DOS 3 ( )
5
Now
let’s
re-‐write
this
as
1 υ T λ0
µn =
(21)
( k BT q ) 2
2k BT
υT =
(22)
π m *
2
⎛ m*DOS ⎞
m ≡ m
* *
DOS ⎜ 2m* ⎟
(23)
⎝ DOM ⎠
Putting
in
numbers
for
Si,
we
find
2 2
⎛ m*DOS ⎞ ⎛ 1.06 ⎞
m ≡ m
* *
DOS ⎜ 2m* ⎟ = 1.06m0 ⎜⎝ 2 × 2.04 ⎟⎠ = 0.072m0
⎝ DOM ⎠
6