08b

HELMHOLTZ EQUATIONS
PIOTR KOWALCZYK
MICHAL MROZOWSKI
Gdansk University of
Technology
Gdansk, Poland
1. INTRODUCTION
The Helmholtz equation
1
is one of the most fundamental
equations of mathematical physics widely used in many
areas [4,5,16,17], including classical mechanics (waves
and vibrations), thermodynamics [28] (diffusion and heat
transport), quantum mechanics [24] (for special forms of
potential and scattering theory), and nally electricity
and magnetismespecially electromagnetic wavesin
problems such as radiation, scattering, and guided and
unguided waves as well as oscillations in cavities and res-
onators. The electromagnetic Helmholtz equation is de-
rived from the Maxwell equations under either static or
time-harmonic conditions. It allows one to determine how
the geometry of a given physical structure interacts with a
time-harmonic (or static) electromagnetic eld. The solu-
tion of the Helmholtz equation provides information such
as the resonant frequencies of an electromagnetic cavity,
wavelength of waves guided along transmission lines or
waveguides, attenuation of an electromagnetic wave due
to nite conductivity of a metal or lossy medium, spatial
distribution of elds inside electromagnetic structures, ra-
diation loss in open circuits, and eld conguration pro-
duced by antennas of different shapes or scattered by
objects made of various media. This information is essen-
tial for design of RF and microwave circuits and systems.
The Helmholtz equation has the form of a partial-differ-
ential equation involving spatial derivatives, and its solu-
tion is inuenced by the shape of the structure under
investigation (boundary conditions), media properties,
and parameters such as frequency. Depending on the
problem, the Helmholtz equation is dened in one, two,
or three spatial dimensions.
2. SCALAR AND VECTOR HELMHOLTZ EQUATIONS
The Helmholtz equation results from the separation of
space and time variables in partial-differential equations
such as wave and diffusion equations. In general, it in-
volves vector elds (or functions) and has the following
form
r
2
Mr k
2
Mr 0 1
where M and r can be vectors
2
in any multidimensional
space and r
2
is the vector Laplacian [2,16]. Examples of
the physical quantity represented by M include electric
eld, magnetic eld, or vector potentials. Whenever pos-
sible, the problem is reduced to the scalar form
r
2
Fr k
2
Fr 0 2
which is more amenable for analytical and numerical
treatment. Here r
2
is the scalar Laplacian operator, and
the examples of a physical quantity involved are the elec-
tric potential, temperature, or acoustic pressure or a
wavefunction (in quantum mechanics). The formula for
evaluationg the scalar Laplacian depends on coordinate
system; for instance, in Cartesian coordinates it reads
r
2
f x; y; z
@
2
f
@x
2

@
2
f
@y
2

@
2
f
@z
2
3
and in spherical coordinates, an analogous formula takes
the following form:
r
2
gr; Y; f
@
2
g
@r
2

2
r
@g
@r

1
r
2
sin
2
f
@
2
g
@Y
2
cos F
r
2
sin f
@g
@f

1
r
2
@
2
g
@f
2
4
Homogeneous equations (1) and (2) with the right-hand
side equal to zero describe the physics of processes that
take place when no external excitation is present, such as
unforced vibrations.
3
If some external force drives the
system, this is represented by a source term that appears
at the right-hand side of the equationthis is character-
istic of nonhomogeneous equations. The nonhomogeneous
vector and scalar Helmholtz equation are hence given by
r
2
Mr k
2
Mr Jr 5
r
2
Fr k
2
Fr rr 6
These forms of the equation are very general and can rep-
resent many physical problems, for instance
*
k0 static problems
*
k
2
40 wave motion and oscillations
*
k
2
o0 diffusion and heat transport
In general, the solution to the Helmholtz equation
depends on the boundary condition and the value of para-
meter k. It has to be noted that the solution to the
homogeneous or nonhomogeneous equation may not exist
for all values of parameter k. In general, the nonhomoge-
neous equation cannot be solved for values of parameter k
1
Hermann Ludwig Ferdinand Helmholtz [born Aug. 31, 1821,
Potsdam, Prussia (Germany); died Sept. 8, 1894, Charlottenburg,
Berlin, Germany]; German scientist and philosopher who rst
expressed the relationship between mechanics, heat, light, elec-
tricity, and magnetism in the form of the law of conservation of
energy [9].
2
In this article vector quantities are denoted by bold face.
3
An example of unforced vibration is a string that is deected and
then let free to oscillate. The vibrations of the string are not due to
a driving force which sustains them but to the initial state of the
system.
1936 HELMHOLTZ EQUATIONS
Previous Page
for which the homogeneous equation is soluble. The Helm-
holtz equation may be expressed in different coordinate
systems, and one usually selects the system that leads to
the separation of variables. Separation is particularly im-
portant for the analytical treatment of partial-differential
equations. The scalar Helmholtz equation is separable in
all basic orthogonal coordinate systems [16]. For the vec-
tor equation there are fewer coordinate systems where the
separation is possible [17].
2.1. Helmholtz Equation from the Wave Motion Perspective
To see the relevance of the Helmholtz equation in inves-
tigating wave motion, it is instructive to consider the re-
sult of the separation of variables in a one-dimensional
wave equation
@
2
u
@x

1
v
2
@
2
u
@t
2
0 7
where v
2
40 is the wave velocity squared, t is time, and x is
the space variable.
Assuming that u(x, t) can be expressed as u(x, t)
S(x)T(t), we get
d
2
S
dx
2
l
2
Sx 0 8
and
d
2
T
dt
2
l
2
v
2
Tt 0 9
where l
2
is the separation constant. In both equations one
readily recognizes the form of Eq. (2), namely, of the scalar
Helmholtz equation in one dimension. General solutions
to these equations for a given l
2
40 are
Sx A cos lx j
S
10
and
Tt B cos lvt j
T
11
where A, B and j
S
, j
T
are arbitrary constants.
4
Function
u(x, t) is a product of S(x) and T(t), so, in general, one has
to consider products for all l. In other words, u(x, t) has the
form
ux; t
_
dlAB cos lx j
S
cos lvt j
T
_
dl
AB
2
cos lx vt j
S
j
T
cos lx vt j
S
j
T
Ex vt Fx vt
12
Each term represents a wave, one traveling in the pos-
itive and the other in the negative x direction (Fig. 1). This
observation is true for higher dimensions and other coor-
dinate systems.
3. HELMHOLTZ EQUATION IN ELECTROMAGNETICS
3.1. Static Problems
Static problems come into play when k0. In this case the
Helmholtz equation transforms into the Laplace or Pois-
son equation and function F(r) has usually a physical in-
terpretation of the electrostatic or magnetostatic potential.
For instance, in electrostatics we have rE(r) 0
so E rf(r). Substituting this relationship into the
Gauss law in a homogeneous medium, we get the Poisson
equation for the electric charge density r(r)
r
2
fr
rr
e
13
which, for r(r) 0, becomes the Laplace equation
r
2
fr 0 14
It is seen that both equations are a particular case
(k 0) of the scalar Helmholtz equation.
For k
2
40 the Helmholtz equation describes wave mo-
tion, and hence it is instrumental in investigating radia-
tion and scattering problems, guided and unguided
electromagnetic waves, as well as oscillations in cavities.
3.2. One-Dimensional Helmholtz Equation:
Transmission Line
One of the most elemental problems whose mathematical
model is given by the one-dimensional scalar Helmholtz
equation is a transmission line. In the frequency domain,
the voltage distribution in an innitely long trans-
mission line is given by the telegraphists equation
E(x) E(xvt)
F(x+vt) F(x)
X
X
Figure 1. Two waves traveling in opposite directions.
4
AA(l), BB(l), j
S
j
S
(l), j
T
j
T
(l).
HELMHOLTZ EQUATIONS 1937
[10,20], which reads
@
2
Ux
@x
2
o
2
LC joRCGL
RGUx 0
15
where o is the angular frequency, j
1
p
is the imagi-
nary unit, and L, R, C, G denote, respectively, the per unit
length series inductance, series resistance, shunt capaci-
tance, and the shunt conductance (Fig. 2).
If one denotes k
2
o
2
LCjo(RCGL) RG, the
telegraphists equation assumes the form of (2), specical-
ly, of the homogeneous Helmholtz equation. For a lossy
line k is complex k a jb. The general solution is
Ux U
0
e
ax
e
jbx
U
0
e
ax
e
jbx
16
In the term U
0
e
ax
e
jbx
we recognize a voltage wave trav-
eling in the positive x direction, whereas U
0
e
ax
e
jbx
corre-
sponds to a voltage wave going in the opposite sense.
3.3. Electromagnetic Waves: Auxiliary Potentials
Using time-harmonic Maxwells equations for the isotropic
homogeneous media with the source in the form of electric
and magnetic currents and charges [2,12]
rEr joBr J
m
r
rHr joDr J
e
r
r
.
Dr r
e
r
r
.
Br r
m
r
_
_
17
and eliminating the magnetic eld H(r) vector we get a
wave equation for the electric eld E(r)
r
2
Er k
2
Er jomJ
e
r rJ
m
r
rr
e
r
e
18
In a similar vein, elimination of the electric eld yields
r
2
Hr k
2
Hr
joeJ
m
r rJ
e
r
rr
m
r
m
19
Equations (18) and (19) both can be regarded as a nonho-
mogeneous vector Helmholtz equation with k
2
meo
2
. In
these equations r is the curl, r- is the divergence, r
2
is
the vector Laplacian, D(r) is the electric ux density, B(r)
is the magnetic ux density, r
e
(r) and r
m
(r) are the electric
and magnetic charge density, J
e
(r) and J
m
(r) are the elec-
tric and magnetic current density, while e and m stand for
the permittivity and permeability of the medium, respec-
tively. It is seen that the right-hand side of these nonho-
mogeneous Helmholtz equations involves all types of
sources. When only electric sources are present, the
result is
r
2
Er meo
2
Er jomJ
e
r
rr
e
r
e
20
r
2
Hr meo
2
Hr rJ
e
r 21
3.3.1. Vector and Scalar Potentials. Because electric and
magnetic elds are, in general, generated by sources con-
sisting of electric and magnetic currents as well as electric
and magnetic charges, it is often convenient to operate
with auxiliary functions that are related to one type of
source. This simplies the right-hand side of the resulting
nonhomogeneous Helmholtz equations. One popular
choice [1,14,29] is the generalized electrostatic potential
j
e
(r) and the vector magnetic potential A(r). Potential
A(r) is dened as
Br rAr 22
With these definitions, one gets from Maxwells equations
rEr joAr 0 23
This implies
Er joAr rj
e
r 24
where j
e
(r) is an arbitrary function.
Imposing the Lorentz gauge
r
.
Ar joj
e
r 25
we get the following relationships, which have to be sat-
ised by potentials j
e
(r) and A(r):
r
2
Ar k
2
Ar mJ
e
r 26
r
2
j
e
r k
2
j
e
r
r
e
r
e
27
L R
C
U
G
Figure 2. Lumped-element model of a small segment of a trans-
mission line.
From these equations it is seen that the electric scalar
potential is due to electric charges, whereas the vector
magnetic potential is generated by electric currents. In a
similar manner one can introduce the magnetic scalar po-
tential j
m
(r) and the electric vector potential F(r), which
are due to magnetic charges and currents. Electric vector
potential F(r) is dened as
Er rFr 28
which implies that
Hr joFr rj
m
r 29
The Lorentz gauge is
r
.
Fr joj
m
r 30
Considering Faradays and Gauss laws with magnetic
currents J
m
(r) and charges r
m
(r), one gets two nonhomo-
geneous Helmholtz equations that are satised by mag-
netic scalar potential j
m
(r) and electric vector potential
F(r) [12]:
r
2
Fr k
2
Fr J
m
r 31
r
2
j
m
r k
2
j
m
r
r
m
r
m
32
To sum up, by introducing the auxiliary vector and
scalar potential, one gets the Helmholtz equations in a
canonical form with each potential associated with
one type of the source. Once the potentials have
been found the electric and magnetic eld can be obtained
from [2,12]
Br rAr jomeFr
1
jo
rr
.
Fr 33
Er joAr
1
jome
rr
.
Ar rFr 34
3.3.2. Hertz Potentials. There are also other ways to
dene potential. For instance, one may dene the electric
potential as [8]
Hr joerP
E
r 35
For time-harmonic elds there is no significant mathe-
matical difference between this potential and the vector
potential introduced previously. However, this choice is
popular for investigating, for instance, guided waves, so
before proceeding we assume there are no sources in the
region of interest. Under this assumption we get from
Faradays law
rEr meo
2
rP
E
r 36
and hence
Er meo
2
P
E
r rcr 37
where c(r) is an arbitrary function.
Assuming
r
.
P
E
r cr 38
and substituting into the Maxwells equation (Ampe`res
law), we have
r
2
P
E
r meo
2
P
E
r 0 39
Note that this expression has the formof the homogeneous
vector Helmholtz equation. Substituting (38) into (37), one
gets the expression for electric eld involving only P
E
(r):
Er meo
2
P
E
r rr
.
P
E
r
rrP
E
r
40
The magnetic vector potential can be introduced in a sim-
ilar manner by assuming that
Er jomrP
H
r 41
The potential thus introduced has to fulll the vector
Helmholtz equation
r
2
P
H
r meo
2
P
H
r 0 42
and the magnetic eld can be retrieved from
Hr meo
2
P
H
r rr
.
P
H
r
rrP
H
r
43
3.4. Homogeneous Helmholtz Equation in
Electromagnetic Problems
Two of the most important categories of electromagnetic
problems involving homogeneous Helmholtz equation are
electromagnetic cavities (resonators) and waveguides. For
simple cases (structures lled with homogeneous or lay-
ered isotropic media), the most elegant approach to the
solution, which yields modal eld patterns, involves vector
potentials. This is because the complete vector eld can be
found from a single component of vector potential. To
illustrate this solution technique, we shall briefly discuss
the mathematics related to guided waves based on Hertz
potentials.
3.4.1. Cylindrical Waveguides. The problem is ex-
pressed by a 2D homogeneous Helmholtz equation. As-
sume that the wave propagates in the z direction inside a
homogeneous and uniform cylinder of arbitrary but
homogeneous cross section [8,10]. This type of a hollow-
pipe problem can often be solved analytically in a 2D cur-
vilinear system. It is convenient to choose the coordinate
system in such a manner that the waveguide boundary L
(Fig. 3) coincides with the constant value surfaces of
the system.
We shall consider the Helmholtz equation for electric
and magnetic Hertz potential. Because of the uniformity
of the structure along the z direction, the potentials have
only one component, which, for a perfectly conducting cyl-
inder, has to fulll either the homogeneous Dirichlet or
Neumann boundary conditions on the boundary L. The
procedure is identical for electric and magnetic potentials,
so we can omit the subscript H or E. When the Hertz
potential has only the z component [i.e., P(r) P(r)a
z
,
where a
z
denotes the unit length vector in the z direction],
then the vector Helmholtz equation reduces to the scalar
one:
r
2
Pr meo
2
Pr 0 44
Decomposing operator r
2
into two terms, r
2
z
along the z
direction and r
2
t
acting on u
1
and u
2
, assuming that P(r)
f(u
1
, u
2
)g(z), and applying the separation of variables
technique, one gets
@
2
gz
@z
2
g
2
gz 0 45
and
r
2
t
f u
1
; u
2
b
2
f u
1
; u
2
0 46
where b
2
g
2
meo
2
, from which it follows that
gz G
e
gz
G
e
gz
47
where subscript / refers to waves propagating in op-
posite senses. Function f(u
1
, u
2
) depends on the boundary
condition imposed on L. For a complete solution, one has to
consider two cases: one with the Dirichlet boundary con-
dition and the other with the Neumann boundary condi-
tion. For perfectly conducting cylinders the Dirichlet
condition has to be imposed on the electric vector Hertz
potential. This leads to the TM modes. The TE modes are
obtained by considering the Helmholtz equation for mag-
netic Hertz potential P
H
(r) P
H
(r)a
z
, with the boundary
condition (qP
H
(r)/qn) 0 on L.
To explain the solution procedure based on the Hertz
potential in more detail, it sufces to consider a simple
geometry with a rectangular cross section. In this case one
assumes the rectangular coordinate system (u
1
, u
2
)-
(x, y). If the width and the height of the waveguide are
denoted by a and b, respectively, then (using the separa-
tion of variables technique for the scalar Helmholtz equa-
tion), we get
f
nm
x; y A cos
pnx
a
cos
pmy
b
48
for the Neumann boundary conditions, whereas for the
Dirichlet boundary conditions we obtain
f
nm
x; y A sin
pnx
a
sin
pmy
b
49
where
b
2
nm
pn
a
_ _
2
pm
b
_ _
2
50
To get the TE modes, we set
P
H;nm
r B
e
g
nm
z
cos
pnx
a
cos
pmy
b
a
z
51
where g
2
nm
pn=a
2
pm=b
2
meo
2
. The electric and
magnetic elds are then obtained directly from (35)
and (40):
E
nm
r E
0
pm
b
e
g
nm
z
cos
pnx
a
sin
pmy
b
a
x
E
0
pn
a
e
g
mn
z
sin
pnx
a
cos
pmy
b
a
y
52
H
nm
r E
0
pn
a
jg
nm
mo
e
g
nm
z
sin
pnx
a
cos
pmy
b
a
x
E
0
jg
nm
mo
pm
b
e
g
nm
z
cos
pnx
a
sin
pmy
b
a
y
E
0
jb
2
nm
mo
e
g
nm
z
cos
pnx
a
cos
pmy
b
a
z
53
Likewise, setting
P
E;nm
r B
e
g
nm
z
sin
pnx
a
sin
pmy
b
a
z
54
u
2
u
1
a
1
a
2
L
Figure 3. Cross section S of a cylindrical waveguide with the
curvilinear system coordinates u
1
, u
2
.
and applying relationships (41) and (43), one immediately
gets the elds for the TM modes:
H
nm
r H
0
pm
b
e
g
nm
z
sin
pnx
a
cos
pmy
b
a
x
H
0
pn
a
e
g
nm
z
cos
pnx
a
sin
pmy
b
a
y
55
E
nm
r H
0
pn
a
jg
nm
eo
e
g
nm
z
cos
pnx
a
sin
pmy
b
a
x
H
0
jg
nm
eo
pm
b
e
g
nm
z
sin
pnx
a
cos
pmy
b
a
y
H
0
jb
2
nm
eo
e
g
nm
z
sin
pnx
a
sin
pmy
b
a
z
56
For other cross-sections the procedure is analogous.
There are other wave guidance problems where the Hertz
potentials are instrumental. They include parallel plate or
rectangular waveguides loaded with transversely inhomo-
geneous media including stratied isotropic (Figs. 4a,b)
[8,18] or gyrotropic (Fig. 4c) media [18]. For these prob-
lems the solution can be obtained by considering the Hertz
magnetic or electric potential with a single component
that is parallel to the inhomogeneity direction (e.g., nor-
mal to the strata). For isotropic media or media with
diagonal anisotropy, this leads to the classical Sturm
Liouville equation [16] (in general, a different equation
is obtained for magnetic and electric potential) with the
Hertz potential as its eigenfunctions. The SturmLiouville
equation can be regarded as a generalized version of one
dimensional Helmholtz equations with variable coef-
cients. From the two sets of eigenfunctions two families
of modes (LSEthe longitudinal section electric modes
and LSMthe longitudinal section magnetic modes [8])
are derived from the relationships (41) and (43). Hertz
potentials can also be dened for the gyrotropic media.
Such generalized Hertz potentials can be used to derive a
vector analog of the SturmLiouville equation for, for ex-
ample, a stratied gyromagnetic parallel plate waveguide
magnetized perpendicularly to the strata (Fig. 4c) [18].
More complex waveguide geometries, such as cylindri-
cal waveguides loaded with inhomogeneous media (other
than the categories described above), in general do not
support the TE, TM or LSE, LSM modes. For these prob-
lems one has to deal with a much more complex two-di-
mensional partial-differential equation with coefcients
depending on the transverse space coordinates [18]. These
equations may be regarded as a generalization of a two-
dimensional vector Helmholtz equation.
3.4.2. Cavities (Resonators). Most microwave resona-
tors used in practice are in the form of either empty cav-
ities or cavities loaded with isotropic media. For the latter
case the relationships for nding natural elds (modes)
and resonance frequencies are obtained by taking the curl
of one of the equations Maxwell equations and using the
other to eliminate either the electric or magnetic eld.
This yields
r
1
mr
rEr o
2
erEr 0 57
and
r
1
er
rHr o
2
mrHr 0 58
In the simplest case of cavities lled with a homoge-
neous medium and no sources inside the cavity volume,
one may use the vector identity rrMrr
.
Mr
2
M
and the Gauss law to transform the relations given above
to the Helmholtz equations.
r
2
Er k
2
Er 0 59
r
2
Hr k
2
Hr 0 60
These equations can be solved by techniques similar to
those used for cylindrical waveguides. The simplest for-
mulation is to use vector potentials instead of elds E
and H. For homogeneously lled structures the complete
eld can be derived for a single component of the poten-
tial. Since only one component is involved, the technique
(a)
(b)
(c)
Ferrite
H
Figure 4. Examples of the stratied waveguiding structures that
can be conveniently analyzed using the Hertz potential.
involving potentials effectively converts a vector problem
to a scalar one. If the volume considered is bounded by a
perfectly conducting simply connected boundary, then two
scalar Helmholtz equations have to be solvedone for the
Neumann condition and the other for homogeneous
Dirichlet conditions. One family of solutions gives the
TE modes, while the other results in the TM modes.
The scalar Helmholtz equation is also used to nd a
complete set of vector wavefunctions that are required to
represent the eld in the cavity with sources. The eld in
such a cavity consists of two parts [6,12,17]. The rst part
is irrotational (lamellar), which means that the curl van-
ishes and the divergence is nonzero; the other part is
rotational (solenoidal), and for this part the curl is nonzero
and the divergence vanishes. To determine the vector
wavefunctions that span both parts, one rst solves the
scalar Helmholtz equation for both Neumann and homo-
geneous Dirichlet conditions:
r
2
cr k
2
cr 0 61
The solution of this equation gives a set of generating
functions cc(r). The rotational part (consisting of di-
vergence-free wavefunctions) is then obtained by substi-
tuting generating functions to the formulas [6]
rcc 62
and
1
k
rrcc 63
where (c) is called the piloting vector. This vector can be
chosen in such a way that the vector wavefunctions thus
obtained have the eld distribution corresponding to the
electric or magnetic eld associated with relevant TE and
TM modes. The lamellar wavefunctions are generated also
from c by the taking the gradient
rcr 64
If one uses the generating functions calculated from the
scalar Helmholtz equation with the Neumann conditions,
then Eq. (64) gives the lamellar vector wavefunctions
corresponding to the electric eld. Likewise, substituting
into (64) generating functions calculated from the scalar
Helmholtz equation with the homogeneous Dirichlet con-
ditions, one gets the wavefunctions corresponding to the
magnetic eld.
4. SOLUTION OF A NONHOMOGENEOUS HELMHOLTZ
EQUATION: GREENS FUNCTION
4.1. Greens Function for Scalar Helmholtz Equation
A nonhomogeneous Helmholtz equation can readily be
solved if one knows the Green function for the problem at
hand. Let us consider a scalar problem [3,2527]
r
2
k
2
Cr f r 65
where f(r) is an excitation. If k
2
belongs to the resolvent
set [12], then the solution to this equation can be written
in as a sum of two terms
Cr fr cr 66
where c(r) is any particular solution of (65) and f(r) is a
solution of the homogeneous problem
r
2
k
2
fr 0 67
Greens function G(r, r
0
) for the scalar Helmholtz equation
satises the following relationship
r
2
k
2
Gr; r
0
dr r
0
68
where d(r r
0
) is the Dirac function. If G(r, r
0
) is known,
then c(r) can be found as
cr
_
V
dV
0
Gr; r
0
f r
0
69
In general, nding the Green function for various prob-
lems of practical significance is a topic of intensive re-
search [12,13,15], but for open-space conguration,
Greens function for the scalar Helmholtz equation has a
simple representation collected [3,21] in Table 1.
For bounded regions, the free-space Green function has
to be supplemented with a term F(r, r
0
) that fullls the
homogeneous equation
r
2
k
2
Fr; r
0
0 70
Greens function for bounded region G
B
(r, r
0
) becomes
G
B
r; r
0
G
0
r; r
0
Fr; r
0
71
The techniques for constructing F(r, r
0
) are described in
several other sources [2,3,5,21].
4.2. Solution of a Nonhomogeneous Vector Helmholtz
Equation: Dyadic Greens Function
The vector Helmholtz equation is by far more difcult to
solve as both the solution and the source have many com-
ponents. It has to be noted that a particular component of
the solution vector may be due to source vector compo-
nents that are oriented in other directions. In this case the
Table 1. Free-Space Green Functions for the Helmholtz
Equation
1D x 2 R 2D (q 2 R
2
) 3D (r 2 R
3
)
j
2k
exp jkjx x
0
j
j
4k
H
2
0
kjr r
0
j
a
j
4k
jr r
0
j
1
exp jkjr r
0
j
a
H
2
0
( ) is the Hankel function of the second kind.
Green function has to describe this complex excitationre-
sponse relationship. A convenient way to do this is by using
the dyadic representation. Each component in the dyadic
Green function corresponds to the response to a single com-
ponent of a delta-type vector source. The construction and
application of the dyadic Greens function are discussed ex-
tensively in the literature [6,29] and often constitute an
open research problem. For current planar technologies
used in the RF and microwave circuits and antennas, of
particular interest are dyadic Green functions involving
multilayer and possibly anisotropic substrates [15].
4.3. Application of the Scalar Green Function Technique to a
Radiation Problem for a Finite-Length Dipole
An example of a nonhomogeneous problem that can be
solved in an elegant way by means of a vector potential
formulation of Helmholtz equation and Greens function
technique is the derivation of the far-eld radiated in open
space by a dipole of nite length l [1]. Let us assume an
innitesimally thin dipole is oriented along the z direction
and the current distribution is given by
Iz
0
I
0
a
z
sin k
l
2
z
0
_ _ _ _
;
l
2
oz
0
o0
sin k
l
2
z
0
_ _ _ _
; 0oz
0
o
l
2
_
_
72
The simplest formulation of the problem is obtained by
considering the nonhomogeneous Helmholtz equation for
vector potential A(r). With the source current having only
one component, one gets A(r) A(r)a
z
, and the vector
Helmholtz equation reduces to the scalar one
r
2
Ar k
2
Ar mIz
0
rx; y 73
where k
2
meo
2
. To nd the solution, one uses the 3D free-
space Green function
Ar
_
l=2
l=2
dz
0
mIz
0
e
jkR
4pR
74
where R
x x
0
2
y y
0
2
z z
0
2
_
. Far from the
origin REr z
0
cos Yand hence
Ar
_
l=2
l=2
dz
0
mIz
0
e
jkr
e
jkz
0
cos Y
4pr
mI
0
4pr
e
jkr
_
0
l=2
dz
0
e
jkz
0
cos Y
sin
kl
2
kz
0
_ _
_
_
l=2
0
dz
0
e
jkz
0
cos Y
sin
kl
2
kz
0
_ _
_
75
Both of these integrals can be integrated using
_
dx e
ax
sinbx g
e
ax
a
2
b
2
a sinbx g
b cos bx g
76
After some mathematical manipulations, potential A(r)
takes the form of
Ar
mI
0
2pr
cos
kl
2
cos Y
_ _
cos
kl
2
_ _
k sin
2
Y
e
jkr
a
z
77
or, in spherical coordinates
Ar
mI
0
2pr
cos
kl
2
cos Y
_ _
cos
kl
2
_ _
k sin
2
Y
e
jkr
cos Ya
r
sin Ya
Y

78
Using relation (22) and neglecting terms inversely pro-
portional to r
2
(far-eld region), the magnetic eld takes
the following form
Hr
jI
0
2pr
cos
kl
2
cos Y
_ _
cos
kl
2
_ _
sin Y
e
jkr
a
f
79
Then, the electric eld can be obtained from Ampe`res law
Er
jZI
0
2pr
cos
kl
2
cos Y
_ _
cos
kl
2
_ _
sin Y
e
jkr
a
Y
80
where Z
m=e
_
.
There is also a different way to determine elds gener-
ated by a nite-length dipole [1]. As a rst step one solves
the radiation problem for innitesimally short dipole of
length dl that is placed at the origin. Assuming that the
current is constant I(z
0
) I
0
a
z
, one gets
r
2
Ar k
2
Ar mI
0
81
Using the 3D free-space Green function
Ar
_
dl=2
dl=2
dz
0
mI
0
e
jkR
4pR
82
where R
x x
0
2
y y
0
2
z z
0
2
_
. Far from the or-
igin REr and hence
Ar mI
0
e
jkr
4pr
dl 83
which, expressed in the spherical coordinates, becomes
Ar mI
0
e
jkr
4pr
dlcos ya
r
sin ya
y
84
The magnetic eld is obtained directly from (22) as
Hr
jkI
0
dl siny
4pr
e
jkr
a
f
85
and then
Er
1
joe
rHr
jkZI
0
dl sin y
4pr
e
jkr
a
y
86
Having obtained the elds for a small dipole one can
easily get the solution for a nite-length structure with
sinusoidal current distribution. To this end one has to di-
vide the structure into short segments and integrate the
contributions from each segment. One nds that for a seg-
ment dz located at point z
0
along z the elds can be
expressed as
dHr
jkIz
0
dz
0
siny
4pr
e
jkr
a
y
87
and
dEr
jkZIz
0
dz
0
sin y
4pr
e
jkr
a
y
88
where relationship R
x
2
y
2
z z
0
2
_
% r z
0
cos y
has been used to derive the formula presented above
(Fig. 5).
Assuming the current in the form given by (72) and in-
tegrating (87) and (88) over z
0
, (z
0
A[ (l/2), (l/2)]), one
nally gets expressions for the magnetic and electric
elds, identical to (79) and (80).
5. NUMERICAL TECHNIQUES FOR SOLVING THE
HELMHOLTZ EQUATION
The class of the problems described by the Helmholtz
equation that can be solved analytically is very limited so
in most cases the only practical approach is to apply nu-
merical methods. Many methods can be used to this end
[13,21]; however, two of them have gained the widest ac-
ceptance due to generality. For problems where the Green
or the dyadic Green function is known, the method of mo-
ments is the most popular choice [11,13,30]. More general
problems can be handled with the nite-element method
[19,22]. The method of moments
5
transforms an integral
equation into a system of linear equations with a dense
coefcient matrix (often called the impedance or admit-
tance matrix). The technique is very powerful as it often
allows one to reduce a three-dimensional problem to two
dimensions. For this reason the method of moments is
very popular for problems such as scattering from per-
fectly conducting bodies, radiation from wire antennas,
slots and patches on multilayered substrates, or planar
microwave circuits. On the other hand, because the coef-
cient matrix is dense, the numerical costs associated
with the method of moments increase rapidly with the
number of unknowns. The nite-element method solves
the Helmholtz equation by discretizing the whole compu-
tational space into small volumes and representing the
eld over each such element by a superposition of func-
tions with local support (typically simple polynomials). By
using the variational principle the Helmholtz equation is
transformed into, again, the system of linear equation, but
this time, due to the local character of the differential op-
erators and local support of functions, the coefcient ma-
trix is highly sparse. This system is then solved by means
of iterative methods of numerical linear algebra. The -
nite-element method is generally used for problems that
cannot be easily handled by the method of moments, such
as three-dimensional structures of complicated shape in-
volving inhomogeneous media. The current trend is to de-
velop techniques that combine the advantages of both
methods or lead to matrix systems having the coefcient
matrices that allows one to use faster computational al-
gorithms [7,23]. These new schemes are aimed at han-
dling larger structures or treating objects that contain
geometric features that differ significantly in size.
6. OPERATOR FORMULATION
The Helmholtz equation can be put in the framework
of the theory of linear operator equations [12,18]. For
instance, the homogeneous Helmholtz equation can be
x
z
dz
z r
r
R
l/2
l/2
+
y
Figure 5. Dipole of nite length l divided into segments dz.
5
The method of moments is one of the approximate techniques of
the functional analysis that can be applied to solve any equation
with a linear operator. Variational techniquessuch as the nite-
element methodcan also be formulated in the framework of the
method of moments. However, in the computational electromag-
netics community, the term method of moments is usually as-
sociated with a technique for solving the integral equation
involving the Green function. This term is used in this article in
this (narrower) sense.
recast in the form
r
2
Fr lFr 89
Written like that, it can be seen as a spectral (eigenval-
ue) problem for a linear operator r
2
. From this perspec-
tive it is seen that the homogeneous Helmholtz equation
has the solution only for l belonging to the spectrum [4,12]
of r
2
, while for the nonhomogeneous case l has to be
chosen from the resolvent set [12] of this operator. When
the problem is dened in the appropriate functional space
setting, dened by the inner product and the boundary
condition, it can easily be shown that operator r
2
has
many useful properties. For instance, for bounded regions
with homogeneous Dirichlet conditions, which in electro-
magnetics corresponds to oscillations in an perfectly con-
duction cavity, the following is true for problem (89):
1. Operator r
2
is Hermitian and positive definite.
2. Eigenvalues of r
2
are positive.
3. Eigenfunctions are orthogonal and form a complete
set.
4. Operator r
2
is invariant under certain symmetries.
The operator approach is a very powerful tool of anal-
ysis. It allows one to predict the properties of the solution
and often provides a uniform way to treat problems for-
mulated in a different manner. For instance, properties 1
and 2 lead to the conclusion that the resonances in a per-
fectly conducting homogeneous electromagnetic cavity oc-
cur only for positive real frequencies. Furthermore,
property 3 implies that all modes oscillate independent
of one another and that an arbitrary eld inside an empty
cavity can be decomposed into modes. Finally, property 4
explains why in certain geometries, such as spheres, sev-
eral modes may oscillate with the same resonant frequen-
cy. The operator approach can also be applied to the
Helmholtz equation with a source term providing means
to construct Greens function and is very instrumental in
investigating complex problems [3,12,18].
For instance, Greens function for any Hermitian oper-
ator L can be found [3] by rst solving the eigenvalue
problem:
LCr YCr 90
Once the eigenvalues Y
n
and eigenfunctions C
n
(r) have
been found, the Green function can be constructed in the
form of a series
Gr; r
0

n
C
n
r
0
C
n
r
Y
n
91
For the scalar Helmholtz equation, Lr
2
k
2
, so if the
boundary conditions are such that r
2
is Hermitian, one
can use the eigenvalues and eigenfunctions of Eq. (89),
namely, the negative scalar Laplacian: C(r)F(r) and
Y
n
k
2
l
n
. Greens function is then
Gr; r
0

n
F
n
r
0
F
n
r
k
2
l
n
92
BIBLIOGRAPHY
1. C. A. Balanis, Antenna Theory: Analysis and Design, Wiley,
New York, 1982.
2. C. A. Balanis, Advanced Engineering Electromagnetics, Wiley,
New York, 1989.
3. G. Barton, Elements of Greens Functions and Propagation
(Potentials, Diffusion, and Waves), Clarendon Press, Oxford,
1989.
4. F. W. Byron and R. W. Fuller, Mathematics of Classical and
Quantum Physics, Vol. 1, Dover, New York, 1969.
5. F. W. Byron and R. W. Fuller, Mathematics of Classical and
Quantum Physics, Vol. 2, Dover, New York, 1969.
6. C.-T. Tai, Dyadic Green Functions in Electromagnetic Theory,
IEEE Press, New York, 1994.
7. W. C. Chew, J. M. Jin, E. Michielssen, and J. M. Song, eds.,
Fast and Efcient Algorithms in Computational Electromag-
netics, Artech House, Boston, 2001.
8. R. E. Collin, Field Theory of Guided Waves, McGraw-Hill,
New York, 1960.
9. Encyclopedia Britannica, from Encyclopedia Britannica Pre-
mium Service; http://www.britannica.com/eb/article?eu
109225, Nov. 28, 2003.
10. F. E. Gardiol, Introduction to Microwaves, Artech House,
Dedham, MA, 1984.
11. R. F. Harrington, Field Computation by Moment Methods,
Macmillan, New York, 1968.
12. G. W. Hanson and A. B. Yakovlev, Operator Theory for Elec-
tromagnetics, Springer, New York, 2002.
13. T. Itoh, ed., Numerical Techniques for Microwave and Milli-
meter-Wave Passive Structures, Wiley, New York, 1989.
14. J. D. Jackson, Classical Electrodynamics, Wiley, NewYork, 1975.
15. K. A. Michalski and J. R. Mosig, Multilayered media Greens
functions in integral equation formulations, IEEE Trans. An-
ten. Propag. AP-45:508519 (March 1997).
16. P. M. Morse and H. Feshbach, Methods of Theoretical Physics,
Vol. 1, McGraw-Hill, New York, 1953.
17. P. M. Morse and H. Feshbach, Methods of Theoretical Physics,
Vol. 2, McGraw-Hill, New York, 1953.
18. M. Mrozowski, Guided Electromagnetic Waves: Properties
and Analysis, Research Studies Press, Taunton, MA, 1997.
19. G. Pelosi, R. Coccioli, and S. Selleri, Quick Finite Elements for
Electromagnetic Waves, Artech House, Boston, 1998.
20. D. M. Pozar, Microwave Engineering, Wiley, New York, 1998.
21. M. N. O. Sadiku, Numerical Techniques in Electromagnetics,
CRC Press LLC, Boca Raton, FL, 2001.
22. M. Salazar-Palma, T. K. Sarkar, L. E. Garcia-Castillo, T. Roy,
and A. Djordjevic, Iterative and Self-Adaptive Finite-Elements
in Electromagnetic Modeling, Artech House, Boston, 1998.
23. T. K. Sarkar, M. Salazar-Palma, and M. C. Wicks, Wavelet
Applications in Engineering Electromagnetics, Artech House,
Boston, 2002.
24. L. Schiff, Quantum Mechanics, Wiley, New York, 1989.
25. I. Stakgold, Boundary Value Problems of Mathematical Phys-
ics, Vol. 1, Macmillan, New York, 1967.
26. I. Stakgold, Boundary Value Problems of Mathematical Phys-
ics, Vol. 2, Macmillan, New York, 1968.
27. D. C. Stinson, Intermediate Mathematics of Electromagnetics,
Prentice-Hall, Englewood Cliffs, NJ, 1976.
28. A. N. Tikhonov and A. A. Samarskii, Equations of Mathemat-
ical Physics, Dover, New York, 1963.
29. J. Van Bladel, Electromagnetic Fields, McGraw-Hill, New
York, 1964.
30. J. J. H. Wang, Generalized Moment Methods in Electromag-
netics. Formulation and Computer Solution of Integral Equa-
tions, Wiley, New York, 1991.
HETEROJUNCTION BIPOLAR TRANSISTOR
JAYASIMHA PRASAD
Maxim
San Jose, California
1. INTRODUCTION
Heterojunction bipolar transistor (HBT) technology has
advanced very rapidly since the mid-1980s and has re-
placed the traditional bipolar transistor in many high-
speed applications. An increasing number of commercial
HBT products are being introduced into the market al-
most daily. Foundry services are even available for anyone
who wants to design HBT circuits to take advantage of its
speed.
The theory behind the HBT is well known and dates
back to the time when Shockley invented the transistor
[1]. The heterojunction bipolar transistor differs from con-
ventional silicon bipolar junction transistor (BJT) in that
at least one of the junctions is formed between dissimilar
semiconductor materials and has been widely studied
[28]. HBTs have been fabricated in IIIV semiconductors
and silicon. The main advantage of HBTs is an improve-
ment in DC and AC performance. HBT device physics is
covered in Section 2.1. The complexities in fabrication of
HBTs and reliability issues are discussed in Section 3. The
performance of both Si and IIIV HBTs are covered in
Section 3.6. Some issues in modeling are covered in Sec-
tion 4. A wide variety of applications are covered in Sec-
tion 5. Finally, the future of HBT technology is discussed
in Section 6.
2. ADVANTAGES OF THE HBT OVER THE BJT
2.1. Wide-Bandgap Emitter
A simple HBT is formed by using a wide-bandgap material
for the emitter and a narrow-bandgap material for the
base and collector. This is called a single heterojunction
bipolar transistor (SHBT). If the collector is also made of a
wide-bandgap material, then it is called a double hetero-
junction bipolar transistor (DHBT). The difference in
bandgap DE
g
between the two materials is accommodat-
ed at the junction by discontinuities in the conduction and
valence bands. With abrupt emitterbase junctions, the
band offsets between the wide-bandgap emitter and the
narrow-bandgap base lead to a spike DE
c
in the conduction
band and a step DE
v
in the valence band, as shown in
Fig. 1 with the constraint DE
g
DE
c
DE
v
. The valence
band offset prevents the back injection of holes from base
to emitter and improves the emitter injection efciency.
On the other hand, the conduction band spike reduces
emitter efciency as the electrons have to surmount that
energy to be injected into the base to eventually reach the
collector. This also produces a lateral shift of the I
C
V
CE
curves on the V
CE
axis by an amount V
offset
from the origin.
In other words, I
C
40 only for V
CE
4V
offset
, which is unde-
sirable.
There are two ways to reduce V
offset
. One is to reduce
the conduction band spike, and this can be achieved by
smoothly by varying the bandgap of the emitter by vary-
ing the composition of the material, in which case DE
c
almost disappears and the valence band offset DE
v
ap-
proaches DE
g
as shown in Fig. 2a (dotted lines). The other
option is to use a widegap material for the collector
(resulting in a symmetric DHBT) and create a conduction
band spike at the collector to compensate for the DE
c
at
the emitter. This is illustrated in Fig. 2c (dotted lines).
Thus, under the application of a forward bias, the back-
ward injection of holes into the emitter is reduced by a
greater extent than the forward injection of electrons into
the base; this leads to two benecial effects: (1) a reduction
in the minority-carrier charge stored in the emitter under
forward bias, and hence a reduction in the emitter-base
storage capacitance and, consequently, an improvement in
high-speed and high-frequency performance; and (2) an
improvement in the electron injection efciency, which
impacts directly and favorably on b, the forward common-
emitter current gain of the device.
A simple numerical example will illustrate the
point. Equations (1) and (2) compare the current gain of
a Si BJTwith the current gain of an AlGaAs/GaAs HBT. In
these equations, we assume that the device has thin
base and thin emitter. We also assume that the base
Ec
Ev
Wide gap
Emitter
Eg=1.8eV
Narrow gap
Base
Eg=1.4eV
Narrow gap
Collector
E
g
=1.4eV
N-AlGaAs
p-GaAs
n-GaAs
E
F
Ec
Ev
Figure 1. Energy band diagram for a single-heterojunction N-p-n
transistor with an emitter having a bandgap wider than that of
the base. Let us assume the emitter is doped to 10
17
cm
3
, and the
base is doped to 10
19
cm
3
. The hole ow from base to emitter is
exponentially reduced by DE
v
, improving the emitter efciency.
1946 HETEROJUNCTION BIPOLAR TRANSISTOR
thickness W
B
is approximately equal to emitter thickness
W
E
. D
n
and D
p
refer to the diffusion constants for electrons
and holes, respectively. It is seen that the gain of the HBT
is enhanced by the difference in bandgaps between emitter
and base:
b BJT N
E
W
E
D
n
=N
B
W
B
D
p
10
20
6:5=10
18
1:3 500
1
b HBT N
E
W
E
D
n
=N
B
W
B
D
p
expDE
g
=kT
10
17
25=10
19
510
5
5000
2
Additionally, the current gain cutoff frequency f
T
is related
to transit time of electrons from emitter to collector t
EC
by
the following relation [6]
1=2pf
T
t
EC
t
E
t
B
t
CSCR
t
C
3
where t
E
is the emitter charging time, t
B
is the base tran-
sit time, t
CSCR
is the transit time in the collector depletion
region, and t
C
is the collector charging time. The emitter
charging time t
E
has two components; the rst is associ-
ated with the charge stored in the emitterbase depletion
region and the second is due to the holes stored in the
emitter as a result of backward-injected holes from the
base. In HBTs, the second component is very small, and t
E
can be approximated by the charging time for the emitter
base depletion capacitance C
BE
t
E
C
BE
kT=qI
C
4
The base transit time in the absence of any electric eld in
the base is given by
t
B
W
2
B
=2D
n
W
B
=v
m
5
where v
m
is the velocity at which the electrons leave the
base and v
m
(kT/2pm*)
1/2
.
Typical values for v
m
are 5 10
6
cm/s for Si and 1 10
7
for GaAs. If the base has an electric eld due to a doping
gradient or a bandgap variation, then the carriers move by
drift and
t
B
W
2
B
=gD
n
6
where g depends on the electric eld present. For a uni-
form electric eld E
0
g 21qE
0
W
B
=2kT
3=2
7
Therefore, by creating an electric eld, t
B
can be substan-
tially reduced.
The transit time in the collector space charge region is
given by
t
CSCR
W
C
=2v
sat
8
where W
C
is the collector depletion layer width for a given
V
CB
and v
sat
is the saturation velocity for electrons. In Si,
v
sat
is 10
7
cm/s. The collector charging time t
C
is the RC
time constant of the collector junction capacitance and all
the series resistances R
E
, R
C
and dynamic junction resis-
tance r
e
(kT/qI
C
).
t
C
C
BC
R
E
R
C
kT=qI
C
9
The frequency f
max
at which the power gain becomes unity
is given by
f
max
f
T
=8pR
B
C
BC
1=2
10
where R
B
is the base resistance [9].
Since the reduction in hole ow is large, the emitter
doping density N
E
can be decreased, and the base doping
density N
B
can be increased (contrary to that of a BJT),
while still maintaining b at an acceptably high value as
shown in (2). A lightly doped emitter leads to a reduction
in C
BE
and emitter charging time t
E
, improving f
T
, as seen
from Eqs. (4) and (3). More importantly, the ability to
employ a highly doped base opens up a large number of
advantageous possibilities:
1. A reduction in the lateral base resistance R
B
, thus
improving the power gain at high frequencies, as
seen by the expression for f
max
in Eq. (10)
2. Reduction in emittercurrent crowding and also im-
provement in the noise gure by reducing the ther-
mal noise due to R
B
in the base
3. A thinner base in meeting a particular R
B
goal, re-
sulting in a shorter base transit time t
B
and, there-
fore, an improved f
T
as seen from Eqs. (6) and (3)
(a) (b) (c)
Base Emitter Collector
Figure 2. Examples of bandgap engineering in HBT: (a) emitter
grading to reduce conduction band spike; (b) base grading to in-
troduce a eld to accelerate the electrons; (c) composite collector
a narrow-bandgap collector is used close to the base and a wide-
bandgap material is used for the remainder of the collector to
improve the breakdown voltage; (d) if all of the collector is made of
the same wide-bandgap material as that of the emitter, it is a
DHBT. In this case, the conduction band spike occurs exactly at
the basecollector junction (not shown) and compensates the DE
c
at the emitter, reducing the offset voltage V
offset
. In these gures,
the solid lines represent the normal conditions and the dotted
lines represent the modications to the band diagram.
HETEROJUNCTION BIPOLAR TRANSISTOR 1947
4. A reduction in the collector depletion region en-
croachments into the base due to heavy doping,
leading to higher output conductance and Early
voltage, which is often benecial in analog circuit-
ry, and a reduced susceptibility to device breakdown
via punchthrough
It is clear from the discussions above that a HBT gives
freedom to independently optimize DC and AC perfor-
mance. In a BJT, DC and AC parameters are closely cou-
pled and some compromises have to be made in DC
performance to meet a given AC performance.
For high-speed digital circuits, bipolar devices are com-
monly used in two logic families. The current-mode logic
(CML) is a simple differential pair with a current source in
the emitter leg. The emitter-coupled logic (ECL) is essen-
tially a CML with emitter followers at the outputs. Gene-
rally, ECL is faster than CML because of the buffering
action of the emitter followers at the expense of additional
power dissipation in the emitter followers. The accurate
prediction of gate delays of CML and ECL gates is fairly
complex and involves several time constants [10]. Howev-
er, for the purposes of understanding the major factors
that contribute to the propagation delay, we can use the
approximate expressions given in [11]
t
pdCML
t
EC
R
L
2R
B
2C
BC
C
L
0:5
11
and
t
pdECL
1:7 t
EC
R
L
1:4R
B
3C
BC
C
L
0:5
12
where R
L
is the load resistance and C
L
is the parasitic ca-
pacitance connected to the output node.
Table 1 shows commonly used material systems, and
Table 2 shows the bandgaps and band discontinuities. It is
interesting to note from Table 2 that for Si/SiGe system,
DE
c
is almost zero and emitter grading is not necessary.
2.2. Regional Bandgap Engineering
In nearly all the HBTs listed in Table 1, alloy materials are
used in some region of the device. By varying the compo-
sition of the alloys, it is possible to introduce features
that enhance performance of the abrupt-junction HBT.
Figure 2 shows some possibilities, which are discussed in
the following subsections.
2.2.1. Emitter Grading. Figure 2a illustrates a situation
where the wide-bandgap material of the bulk emitter is
compositionally graded close to the junction with the base.
An example would be an Al
x
Ga
1 x
As emitter with the
mole fraction x varied from 0.3 to 0 over a distance of
B30 nm from a GaAs base. The diagram depicts a linear
grading, as would follow from a linear variation in Al mole
fraction. Parabolic grading is also often used. In either
case, the objective of the emitter grading is to reduce, or
eliminate, the conduction band spike present at abrupt
junctions. There are at least two favorable consequences of
this. First, because the electron ow across the junction is
increased, the collector current and the current gain are
enhanced. One advantage of this is a lower turnon volt-
agethat is, a lower baseemitter bias for a given collec-
tor current. This is advantageous for the implementation
of HBTs in low-power-consumption circuitry. Second, the
ideality factor of the emitter current reduces to unity as
thermionic emission takes over from tunneling as the
dominant barrier transport mechanism. This fact, when
coupled with the increase in emitter current, leads to a
lower dynamic emitter resistance and, when coupled with
the increase in collector current, leads to an improvement
in transconductance. A possible third benet of a graded
emitter is that electrons are injected into the base at near-
equilibrium energies, rather than at the elevated kinetic
energies possessed by electrons that have tunneled
through a high potential barrier, or have been thermion-
ically emitted over it. While these latter hot electrons
may make a faster transit of the base, they are more likely
in GaAs and InP HBTs to be scattered into the lower-
mobility, upper-conduction-band valleys on entering the
Table 1. Examples of Material Combinations Frequently Used to FormHBTs with Either a Single Heterojunction (SHBT) or
Two Heterojunctions (DHBT)
a
Type Emitter Base Collector Substrate
SHBT AlGaAs GaAs GaAs GaAs
SHBT GaInP GaAs GaAs GaAs
DHBT AlGaAs GaAs AlGaAs GaAs
SHBT InP GaInAs GaInAs InP
SHBT AlInAs GaInAs GaInAs InP
DHBT AlInAs GaInAs InP InP
SHBT a-Si Si Si Si
SHBT SiC Si Si Si
DHBT Si SiGe Si Si
a
Generically, the devices are grouped according to the substrate material.
Table 2. Properties of most Commonly Used Material
Systems for HBTs
Material System E
g
(eV) DE
c
(eV) DE
v
(eV)
Al
0.3
Ga
0.7
As/GaAs 1.8/1.42 0.243 0.131
In
0.49
Ga
0.51
P/GaAs 1.88/1.42 0.160 0.290
InP/In
0.53
Ga
0.47
As 1.35/0.75 0.234 0.366
Al
0.48
In
0.52
As/In
0.53
Ga
0.47
As 1.49/0.75 0.525 0.215
Si/Si
0.88
Ge
0.12
1.12/0.99 0.020 0.110
high-eld, collector space charge region. Thus, the overall
emittercollector delay time t
EC
may in some instances be
lower in a graded-junction device.
2.2.2. Base Grading. In base-graded n-p-n HBTs the
bandgap is progressively reduced from the emitter to the
collector by an appropriate variation in the composition of
the base material. As shown in Fig. 2b, the bandgap
change is taken up by the conduction band alone. This is
because the high hole conductivity precludes any signi-
cant variation in the valence band [6].
Thus, the grading has the effect of producing an electric
eld to aid the passage of electrons across the base. The
obvious benet of this is an improvement in the base tran-
sit time t
B
. If, because of the very narrow base that is al-
lowed by having a high base doping density, t
B
is not a
major contributor to the overall delay time t
EC
, then base
grading can be used to achieve an acceptably low value of
t
B
in a wider base. The resulting smaller lateral base re-
sistance enables an improvement in f
max
.
Base grading can also have a benecial effect on the
current gain b; the reasons for this are twofold:
1. Recombination in the bulk quasineutral base, which
can be one of the most significant contributors to the
base current, is diminished, with respect to that in a
uniform, narrow-bandgap base, because the alloy
eld increases the electron velocity and, for a given
collector current, this means a reduction in the base
charge. Of course, this will reduce the base recom-
bination current only if the minority-carrier lifetime
is not overly shortened by using an alloy material for
the base.
2. Recombination at surface regions of the base is re-
duced by the tendency of the base alloy eld to sweep
electrons to the collectorthat is, in a direction per-
pendicular to the base surface. This effect can be
particularly benecial in HBTs with small-dimen-
sion emitters, because these devices necessarily
have a large emitter perimeter/area ratio.
However, there is a limit to the base grading, beyond
which b starts to decrease. Increasing the bandgap of the
base material at the emitterbase junction serves to re-
duce the band offsets, and to make the junction become
more and more like a homojunction made from wide-band-
gap material. The advantages of a wide-bandgap emitter
is lost and current gain degrades to that of a homojunction
emitter.
In a wide-bandgap collector DHBT, a further limitation
to the permissible amount of base grading is imposed by
the need to trade off some of the base grading against the
height of the barrier to electron ow into the collector as
shown in Fig. 2c. For a given bandgap in the base at the
emitter end of the device, decreasing the base bandgap at
the collector end increases the amount of base grading, but
also increases the barrier to electron ow into the collec-
tor. Both b and t
B
are affected by this tradeoff [12].
2.3. Collector Engineering
There are four main motivations for modifying the collec-
tor region of a SHBT: (1) to reduce the transit time t
CSCR
of
electrons in the collector space charge region, (2) to design
for a given breakdown voltage BV
CBO
, (3) to reduce Kirk
effect or base pushout effect, and (4) to reduce the collector
resistance R
C
.
2.3.1. Reduction of t
CSCR
. Historically, in bipolar tran-
sistors the major transit-time component of t
EC
has been
t
B
. In modern HBTs, the ability to realize very narrow
bases and, to a lesser extent, to implement composition-
ally graded bases, has led to significant reductions in t
B
,
with the result that the dominant transit-related time is
t
CSCR
. Clearly, reducing the width of the basecollector
space charge region W
C
would shorten the collector signal
delay time, but, if this were accomplished by increasing
the collector doping density, then the breakdown voltage
and collector junction capacitance C
BC
would be adversely
affected. Trading off speed for improvement in breakdown
performance, and vice versa, presents opportunities for
novel collector designs. Two structures intended to reduce
t
CSCR
in GaAs devices are shown in Fig. 3. In the inverted
eld structure of Fig. 3b [12], the eld pattern of a con-
ventional collector Fig. 3a is reversed, with the intention
of delaying the onset of electron transfer to the lower-mo-
bility, upper conduction bands. This transfer occurs more
readily in a conventional collector, as electrons immedi-
ately enter a high-eld region, in which they are acceler-
ated and rendered more likely to scatter into the upper
valleys. This phenomenon is more likely to occur in GaAs
than in InP because of the smaller separation between the
G and L valleys [7]. In the intrinsic collector structure
shown in Fig. 3c, the high-eld zone is restricted to a nar-
row p
-n
region, and the eld remains favorably low in

the remaining, weakly doped part of the collector. The rst
GaAs HBT to register an f
T
in excess of 100GHz was fab-
ricated using such a collector structure [13].
2.3.2. Breakdown Voltage. Turning now to breakdown
voltage considerations, the attainment of acceptably high
values is difcult in HBTs employing collectors made from
(a)
E
p
+
n n
+
(b)
E
p
+
p
n
+
(c)
E
p
+
p
+
i n
+
Figure 3. Electric eld proles in different collector structures.
The p
base is on the left, and the collectors are classied as (a)

conventional, (b) inverted, and (c) intrinsic.
narrow-bandgap material, for which the ionization coef-
cients for electrons and holes are generally high. This is
the situation in InP HBTs that have collectors of lattice-
matched In
0.53
Ga
0.47
As, which has a bandgap of only
0.75 eV. Solutions to this problem involve making all, or
part, of the bulk of the collector from a wider-bandgap
material, such as InP, for which the impact ionization co-
efcients are low. When all the collector is InP, the device
is a DHBT. When the collector comprises GaInAs near the
base and then InP for the remainder of the layer, as illus-
trated in Fig. 2c, the device is labeled as a composite-col-
lector HBT. This design is a renement of the DHBT, with
the objective of using just enough n-GaInAs in the collec-
tor to ensure that the conduction band spike at the base
collector junction is reduced below the level of the conduc-
tion band in the base. In this way, the stored base charge,
which strongly inuences t
B
and b, and the collector cur-
rent are not adversely affected by the presence of the con-
duction band spike.
2.3.3. Kirk Effect. Of all the aspects of collector design,
base pushout or the Kirk effect is the most important de-
sign consideration. When the electrons injected into the
collector becomes comparable to the background donor
concentration, the electric eld in the collector rst de-
creases and as the injected carriers continue to increase
with collector current density J
C
, the eld reverses and f
T
falls off rapidly. An approximate expression for the critical
current density J
K
at which the onset of Kirk effect occurs
is given by
J
K
qv
sat
N
C
13
where N
C
is the collector doping concentration. Above this
current density J
K
, the collectorbase junction becomes
forward-biased and holes are injected into the collector. At
this point, the base widens into the collector, t
B
increases
and b decreases. Even if the base and emitter are designed
properly, if the collector is not at its optimum, the device
will not yield the best f
T
. To overcome this effect and to
meet a given f
T
, J
K
has to be increased, which requires
higher collector doping as predicted by Eq. (13). Higher N
C
results in lower BV
CBO
, lower collector depletion layer
width W
C
, lower t
CSCR
, and higher t
C
, due to higher C
BC
.
To prevent large increases in C
BC
, selectively implanting
the collector just below the emitter is often used in silicon
technology. The overall result is that f
T
improves at the
expense of BV
CBO
. To regain the some of the lost BV
CBO
,
the collector epitaxial thickness can be increased.
2.3.4. Reduction of R
C
. If the collector epitaxial-layer
(epi) thickness is increased too much, the undepleted por-
tion of the epi contributes significantly to R
C
and t
C
in-
creases. Further, the voltage drop in R
C
would cause the
collectorbase junction to be forward-biased, leading to a
drop in f
T
with increasing I
C
. Attention should be paid to
the doping and sheet resistance of the subcollector region
as this also contributes to R
C
. Therefore, the parameters
N
C
, J
K
, f
T
, and BV
CBO
have to be optimized for a given
application.
3. FABRICATION OF HBTs
HBTs using GaAs or InP substrates, or SiGe base layers,
have all progressed beyond the experimental device stage
to the point that they are being used in commercial cir-
cuits. Examples of the fabrication procedures of HBTs rep-
resentative of various material systems are discussed in
this section.
3.1. AlGaAs/GaAs HBTs
HBTs, at least those of the IIIV material variety, are in-
variably fabricated from a stack of epitaxial layers grown
on a semiinsulating substrate (see Table 3), and the de-
vices are built by etching away or implanting unwanted
regions. Growing the epitaxial layers on a semiinsulating
substrate has the benecial effect of eliminating the par-
asitic collectorsubstrate capacitance, which is present in
most silicon BJT structures. It also provides an inert
platform, which may allow fabrication with equal ease of
emitter-up and collector-up structures. Collector-up struc-
tures are of interest because the smaller dimension of the
collector leads to a reduction in collectorbase junction
capacitance [14]. A more realistic use of collector-up de-
vices might be as the sole transistor type in exceptionally
high-speed ICs [15]. We will now focus on the most fre-
quently used emitter-up HBT structures.
AlGaAs/GaAs HBTs were the rst to be fabricated by
several companies to take advantage of the high frequency
capabilities of these devices [4,5,16]. The bandgap of
AlGaAs is given by
E
g
x 1:424 1:247x for xo0:45 14
where x is the Al mole fraction. AlGaAs is lattice-matched
to GaAs for a wide range of Al mole fractions. With high
electron mobility in AlGaAs, this material is well suited
for HBTs. Since LEDs and lasers have been fabricated in
this material before, this material system has been well
researched. The current gain achieved in this material is
about 50100 (unlike 5000 predicted earlier), due to large
surface recombination velocity and significant recombina-
Table 3. Typical Epitaxial-Layer Structure for AlGaAs/
GaAs HBTs
Layer Al Composition Dopant
Doping
(cm
3
)
Thickness
(A
)
Cap 0 Si 510
18
1000
Grade 00.25 Si 510
17
300
Emitter 0.25 Si 510
17
1000
Grade 0.250 Si 510
17
300
Spacer 0 none 100
Base 0 Be 210
19
1000
n
collector 0 Si 310
16
5000
n
collector 0 Si 510
18
5000
Substrate 0 none Semiinsulating
tion in the emitterbase space charge region. An example
of a manufacturable process for AlGaAs/GaAs HBTs from
NORTEL is illustrated in Fig. 4 [17]. The epitaxial layers
comprising the device, shown in Table 3, can be grown by
molecular-beam epitaxy (MBE) or organometallic chemi-
cal vapor deposition (OMCVD), and are shown schemati-
cally on the gure by the alternating dark and light bands.
From the top, the layers are the heavily doped cap to fa-
cilitate ohmic contacting to the emitter metal, the emitter,
the base, the collector, and the subcollector.
A dummy dielectric emitter (Si
3
N
4
) is employed for self-
alignment to the base metal, which completely surrounds
the emitter in this case (see Figs. 4a and 4b). A deep He
ion implantation serves to deactivate the collector layer

under most of the base metal area, thereby eliminating
most of the extrinsic basecollector capacitance. Obtaining
a good ohmic contact to the thin, p-type base, without al-
lowing metal penetration through to the collector, is a
crucial step. The procedure here is to rst etch off the
emitter cap and then deposit PdZnPtAuPd, which is then
alloyed through the AlGaAs to the p
base. The penetra-

tion depth during alloying is controlled by the thickness of
the Pd layer below the Pt barrier. The next step (Fig. 4c) is
to etch off the cap layer between the edge of the base metal
and the emitter stack. This denes the emitter dimensions
and, very importantly, leaves an AlGaAs shelf layer on
top of the base, thereby reducing the surface recombina-
tion velocity in this peripheral region. This helps maintain
a high b in small-dimension devices. The peripheral region
is protected from the subsequent etch, which exposes the
subcollector (Fig. 4d), by a SiON sidewall spacer, which
also provides alignment tolerance for the emitter metal-
lization. The fabrication sequence is terminated by pro-
viding two levels of metal interconnects and a planarizing
layer of benzocyclobutene (BCB).
There are many ways to build HBTs, and a simpler
process with fewer masks incorporating 1.4-THz Schottky
diodes has been described [16]. Generally, Be is used as a p-
type base dopant. There were some initial reliability con-
cerns of the device due to dopant migration. These have
been resolved and a median time to failure of 410
8
h for
discrete devices have been shown [18]. Carbon is considered
superior to Be in terms of reliability, and most HBTs today
employ carbon doping for the base [1921]. The shift in HBT
device parameters is a function of collector current density
[22]. The highest reported f
max
for AlGaAs/GaAs HBT was
350GHz and this was achieved using a 50nm base doped to
610
19
cm
3
using carbon [23].
He implant
P-ohmic metal
Resist Si
3
N
4
He isolation implant and
dummy emitter formation.
SiON sidewall
N-ohmic metal
2nd interconnect Post
SiON dielectric
1st interconnect
Back end processing:
SiON deposition and via etch
NiCr resistor deposition
ME1 deposition
ME1 to ME2 post deposition
BCB planarization and ME2
deposition
Selective removal of dummy
emiter and N-ohmic metal
deposition.
Thick sidewall deposition
and RIE etch to sub collector.
Field Si
3
N
4
removal and
second etch to AlGaAs emitter.
Selective dry etch AlGaAs
emitter and P-ohmic
metal deposition.
(f)
(e)
(d)
(c)
(b)
(a)
BCB
Figure 4. Fabrication sequence for an AlGa-
As/GaAs HBT.
3.2. InGaP/GaAs HBTs
A significant event in AlGaAs/GaAs HBT development
was the replacement of the AlGaAs emitter by In
0.49-
Ga
0.51
P. Owing to small conduction band offset (0.16eV)
and large valence band offset (0.29 eV), compositional
grading of the emitterbase junction is not required (since
V
offset
is small), and this simplies the epitaxial layer
growth. Also, there are selective etchants to etch the In-
GaPemitter and stop on GaAs base, unlike the case for the
AlGaAs/GaAs system [24]. Oxygen complexes, which are
often incorporated in aluminum-containing emitters, are
less prevalent when using InGaP. This reduces the emit-
terbase junction recombination current and allows b41
to be maintained down to very low current densities,
showing ideal behavior. In fact, InGaP HBTs provide su-
perior current gains compared to AlGaAs HBTs for a given
doping prole and almost independent of device size
[24,25], due to reduced surface recombination. A rst dem-
onstration of a 60-GHz InGaP/GaAs IC technology with
28 ps ECL gate delay was reported in 1993 by Prasad [26].
A YIG-tuned InGaP/GaAs HBT microwave oscillator with
low phase noise was rst reported in 1994 [27]. The ease of
epitaxial growth, simpler fabrication, better temperature
stability, lower 1/f noise, and improved D.C properties of
InGaP HBTs have led to rapid growth of this technology.
Today, InGaP HBTs are widely used for power amplier
applications in cellphones [28,29]. Hitachi has developed a
high-speed InGaP HBT IC technology with f
T
of 156 GHz
and f
max
of 255 GHz [30]. As described before, carbon dop-
ing of the base results in highly reliable devices with no
change in the device performance for 10
4
h of operation at
a junction temperature of 2641C at J
C
2.5 10
4
A/cm
2
[31,32]. The mean time to failure (MTTF) is an order of
magnitude better than that for AlGaAs HBTs, suggesting
that the InGaP emitter is the best choice for GaAs HBTs
[33,34].
3.3. InP/InGaAs HBTs
It is well known that InGaAs, as used for the base material
in InP HBTs, has many desirable properties [7]. For ex-
ample, with respect to GaAs and Si, GaInAs possesses
higher electron mobility, higher electron peak velocity,
higher electron saturation velocity, and, compared to
GaAs, a higher separation of the G and L conduction
band minima. These properties are very useful for build-
ing high-speed devices. Another variation of this device is
to use AlInAs emitter instead of InP emitter.
Another benecial property of InGaAs is its low band-
gap (0.75eV), which results in a higher minority-carrier
electron concentration for a given base doping density.
When combined with a graded-emitter junction, a signif-
icantly lower emitterbase turnon voltage, V
BE
can be
achieved. For example, V
BE
of InP/InGaAs HBTs is 0.7 V
compared to 1.4V for AlGaAs/GaAs HBTs. A low V
BE
per-
mits lower supply voltages to be used, resulting in less
power consumption and a more favorable gate delaypow-
er product. GaInAs also has a very low surface recombi-
nation velocity, which permits achievement of near-ideal
IV characteristics over a large range of forward bias [7].
A typical process sequence for an InP HBT is illustrat-
ed schematically in Fig. 5 [35]. Three mesa etches are re-
quired: one to dene the emitter, one to dene the base
and collector regions, and one to etch down to the semiin-
sulating substrate in order to provide device isolation. The
mesa etches are necessary because the InP material sys-
tem lacks an ion implantation damage process capable of
rendering GaInAs sufciently resistive to provide isola-
tion. To reduce the parasitic basecollector capacitance in
a mesa structure, the basecollector mesa must be made
as small as possible, necessitating tight self-alignment of
the base and emitter metallizations. This is achieved by a
slight undercutting of the emitter metal contact, which
allows this metal to serve as a mask for the medial edge of
the subsequently deposited base metal. The etch used in
the undercutting must also be selective to avoid destruc-
tion of the thin base material; for GaInAs bases, this can
present a problem. If the collector is InP, a selective etch
can also be used to dene the basecollector mesa. If the
collector is GaInAs, the same material as the subcollector,
then this convenient etch-stop method cannot be em-
ployed. However, the etching is less critical than in the
Emitter contact
Base contact
Base and collector
mesa
collector
contact
Emitter mesa
Emitter
Base
Collector
Sub
Collector
Semi-insulating substrate
Second-level interconnect
Polyimide
C
B B
C
E
Figure 5. Fabrication sequence for an InP HBT.
case of the emitter mesa formation because the subcollec-
tor is relatively thick. The device structure resulting from
a triple-mesa process is highly nonplanar, but can be pla-
narized, to facilitate device interconnection and terminal
access, by using a suitable polyimide. This coating also
provides some passivation of the exposed GaInAs surfaces.
The highest reported f
T
for InP HBT is 452GHz [36], and
the highest reported f
max
is 478GHz [37]. Recently, Vitesse
Semiconductor reported a InP DHBD process with f
T
and
f
max
of 325 GHz with the lowest CML gate delay of 1.95 ps
[126]. Using a transferred-substrate technology, Lee and
others have shown a f
max
of 820GHz for AlInAs/InGaAs
HBT [38]. Reliability studies on InP HBTs have shown
that there are no shifts in device parameters based on ac-
celerated tests at 2051C for 6500 h [39,40].
InP technology allows the monolithic integration of
long-wavelength photodetectors with HBTs. Inevitably,
there are difculties associated with the simultaneous
realization of high-quality optical and electronic devices
from a given stack of epitaxial layers. However, in opto-
electronic integrated-receiver front ends, for example,
good results have been obtained when using the
base-collector of an InP HBT as the photodiode [41]. An
example of this embodiment is shown in Fig. 6. The
top-illuminated photodiode is constructed from the HBT
layers by removing the emitter layers and adding an
antireection coating.
3.4. GaN HBTs
There has been a lot of research in GaN materials and
devices since the late 1990s. GaN has a bandgap of 3.4eV,
electron saturation velocity of 210
7
cm/s, and a high
breakdown eld of 2MV/cm. This material is well suited
for high breakdown voltage and high temperature applica-
tions. A typical GaN HBT would use Si-doped Al
0.15
Ga
0.85
N
as the emitter (E
g
3.85eV) and Mg-doped GaN as base
and GaN collector grown on sapphire substrate. At this
time, GaN HBT technology is not as mature as other HBTs
that we have considered so far. The technology suffers from
inability to dope the base heavily, very high V
offset
of 15V
due to high base contact resistance, low current gains
(o10), and high collectoremitter leakage [42].
3.5. SiGe HBTs
Unlike the IIIV HBTs discussed in the preceding sub-
sections, the starting material for SiGe HBTs are gene-
rally p-type silicon substrates. Since the SiGe is the base
layer, SiGe HBTs are normally DHBTs. Further, most
SiGe HBTs are fabricated by growing a SiGe base layer
and depositing a polysilicon layer on top of it to form
an emitter, in sharp contrast to IIIV HBTs, where all
the device layers are grown at once, forming the starting
material.
3.5.1. Material Properties. Germanium has a lattice
constant of 5.65 A
, whereas silicon has a lattice constant

of 5.43 A
. Therefore, the SiGe layer is not lattice-matched

to silicon. However, if the epitaxial lm is not too thick and
the growth temperature is not too high, the SiGe base
layer will conform to the Si collector material on which it
is grown. Such a layer is referred to as pseudomorphic or
coherently strained. The strain decreases the SiGe band-
gap, fortuitously improving the bandgap differential be-
tween the base and the subsequently deposited Si emitter,
allowing one to build a build an HBT with tailored band-
gap differences. The bandgap of Si
1 x
Ge
x
varies according
to the relation [43]
E
g
x 1:124 1:22x 0:88x
2
for xo0:6 15
Furthermore, the strain in the SiGe lm breaks the sixfold
degeneracy of the conduction band minima and the two-
fold degeneracy of the valence band maxima; these phe-
nomena can be exploited to improve the electron mobility
(reducing t
B
) and the hole mobility (reducing R
B
) [6]. The
critical thicknessbeyond which the SiGe lm becomes
unstable, and sufciently defective for the associated rise
in recombination centers to render it useless for bipolar
applicationsis inversely proportional to the Ge content
or, for graded SiGe lms, to the integrated Ge content.
Maintenance of the coherently strained nature of the lm,
SiO
2
A.R.coating
SiO
2
side wall
p cont.
h
n
contact
n
+
InGaAs
n

InAlAs
n

InGaAs n

InGaAs
n
+
InGaAs n
+
InGaAs
p-i-nPD HBT
InP S.I. substrate
TF-R S-Ind
p
+
InGaAs
C
B
E
Spiral
inductor
Ti Res.
Figure 6. Schematic cross section of the ele-
ments of a positiveintrinsicnegative (p-i-n)/
HBT OEIC photoreceiver.
by not allowing it to relax and assume its bulk lattice con-
stant, is dependent on minimizing the exposure of the lm
to subsequent high-temperature environments.
3.5.2. Base Proles. The constraints imposed by the
considerations of SiGe lm stability lead to a Si/SiGe
HBT fabrication process, which is more complicated than
those described previously for IIIV HBTs. A review of the
procedures used to date to achieve high-performance
Si/SiGe HBTs can be found in Refs. 4446, along with a
detailed description of the technology that is employed at
IBM to produce HBTs in a manner compatible with stan-
dard CMOS processing resulting in a BiCMOS technology.
With reference to the HBT base, there are two approaches:
(1) using a box SiGe prole with a xed Ge concentration
along with a heavily doped base, similar to the IIIV
HBTs, where a lightly doped Si emitter is also grown,
making it a true HBT (the box prole is used by a few
companies, e.g., TEMIC); and (2) using a graded Ge prole
(lowest Ge at emitterbase junction and highest Ge at
basecollector junction) with a lightly doped base and con-
tact it with a heavily doped polysilicon emitter, which in-
tegrates well in standard bipolar process. Here, the speed
advantage is due to the accelerating eld in the base and
the device is a drift base transistor, often abbreviated by
the misnomer HBT. The graded base approach is widely
used in the industry as it has several advantages over the
box prole [44].
3.5.3. SiGe Epitaxial Growth. Turing our attention now
to the epitaxial (epi) growth of the base, we have two op-
tions: (1) nonselective epi and (2) selective epi. In the sim-
plistic nonselective epi, the SiGe base grows as a single
crystal on the exposed areas of silicon (where we want the
base to be formed) and as polysilicon on oxide. The poly-
silicon naturally contacts the base and facilitates connec-
tion to the base. Also, this polysilicon layer can be used as
a resistor if needed. In the selective epi, SiGe base is
grown only on the exposed areas of silicon and no growth
occurs on the oxide. An additional layer of polysilicon is
necessary to contact the base. Each of these techniques
has its own advantages and disadvantages.
The SiGe base layers in the HBTs from IBM are grown
using a UHV/CVD system, but in this case it is of the hot-
wall variety [47]. This equipment has been developed spe-
cifically for blanket Si and SiGe epitaxy and is consistent
with a low-temperature, commercially feasible process. In
this process, hydrogen-passivated Si wafers are admitted
to the system, in which a vacuum of around 10
9
torr is
maintained. The residual gas is predominantly hydrogen;
other species, which may be chemically active with silicon,
are not present at sufcient partial pressures to violate
the hydrogen passivation of the wafer. Films are subse-
quently deposited under vacuum by CVD at temperatures
in the range 4005001C. Precise dimensional control, of
the order of one or two atomic layers, is possible with the
UHV/CVD process. Reduced-pressure CVD reactors with
pressures of the order of 10 torr are also being used for
SiGe epi [48]. These are typically single-wafer machines,
unlike the high-vacuum batch reactor described earlier.
3.5.4. SiGe HBT Flow. Here we describe a selective epi
technology from NEC, Japan which is geared toward high-
performance bipolar circuitry and therefore has the same
goal as that of the processes described earlier for IIIV
HBTs. A partial process sequence is illustrated in Fig. 7
[49,50]. To prepare for the SiGe growth and to begin the
emitter definition, the top dielectric lm (Si
3
N
4
) and the
large-grain p
polysilicon layer shown in Fig. 7a is etched

and the edge of the resulting feature is covered with a
Si
3
N
4
sidewall. The bottom dielectric (SiO
2
) is then etched
and allowed to laterally undercut the sidewall. From this
point on, attention must be paid to the thermal budget, so
as not to destroy the pseudomorphic nature of the SiGe
base. This layer is grown on the exposed Si substrate, si-
multaneously with a polysilicon lm which descends from
the underside of the polysilicon layer exposed by the ear-
lier lateral etch. Growth is stopped when the two growing
lms touch (see Fig. 7b). Growth is carried out by CVD in a
(100) Sub
Selective epi (base)
Dielectric film
P
+
polysilicon
(base electrode)
Selective poly
(graft base)
BPSG-filled trench SiGe intrinsic base
Base
Collector
Emitter
BSG film
In situ phosphorus
doped poly-Si
(b)
(a)
N
+
N
+
P
P
+
N
N
Figure 7. Partial fabrication sequence for SiGe HBT (a) and (b)
with the nished device.
cold-wall, ultra-high-vacuum (UHV) system at B6501C.
Because lms grown by this selective epitaxial growth
(SEG) process are very sensitive to the condition of the Si
surface, the collector is formed subsequently, using ion
implantation. The remaining fabrication steps comprise
(1) coating of the Si
3
N
4
sidewall with borosilicate glass
(BSG), (2) doping of the external SiGe base by driving in
boron from the glass at 8001C for 10 min, (3) deposition of
phosphorus-doped polysilicon for the emitter, and (4) driv-
ing in the emitter at 9501C for 10 s. The nished device,
complete with BSG trench isolation, is shown at the
bottom of Fig. 7.
3.5.5. SiGe BiCMOS Flow. Most SiGe bipolar processes
today incorporate CMOS to take advantage of the high
functionality, density, and popularity of several working
circuits in standard CMOS. This also helps designing a
system on a chip (SoC). In Fig. 8 we show a method of
adding bipolar steps to an existing CMOS ow to generate
a BiCMOS process ow. The reason for processing the
PMOS after the NPN is not to add additional thermal cy-
cles to change the characteristics of the PMOS. It should
be noted that if a nonselective epi is used for SiGe base,
the SiGe poly layer must be removed from the CMOS ar-
eas during base definition. A typical BiCMOS ow with
process cross sections is presented in Ref. 45.
One problem associated with thin boron-doped base
layers grown by epi is that the dopant diffuses during
subsequent device processing, making the base wide and a
consequent reduction in f
T
, f
max
, and b. To overcome this
problem, the diffusion of boron must be reduced. It has
been found that if carbon is added to the SiGe base in
concentration of o10
20
cm
3
(B0.2%), the diffusion coef-
cient of boron is reduced by more than one order of mag-
nitude [51]. Adding carbon to the base thus results in
stable DC and AC performance and improved yield. Ac-
celerated testing on SiGe HBTs has shown that the device
is very robust and no parametric shifts occur for 410
6
h
[52,53].
3.6. Device Performance
The purpose of this subsection is to briefly mention some
of the high-performance HBTs that have been reported in
the literature and to relate them to the materials consti-
tuting the device. Using f
T
and f
max
as gures of merit,
Table 4 lists the high-frequency performance of various
HBTs reported in the literature. The table also includes
data reported in the early 1990s to enable the reader to
assess the progress made since that time. It is seen that
the InP HBT leads with highest f
T
and f
max
of 370 and
459 GHz, respectively. The highest f
max
of 820 GHz is ob-
tained on InAlAs/InGaAs HBT transferred to GaAs host
substrate and mounted in the collector-up conguration.
InGaP HBTs have made a steady progress in f
T
from 60 to
156 GHz. The rst reported ECL gate delay for InGaP
HBT is 28ps. In the early 1990s, AlGaAs HBTs have
shown a f
max
of 350GHz. SiGe HBTs have shown perfor-
mance comparable to IIIV HBTs with the lowest CML
gate delay of 3.6ps. More recently, with the regrown emit-
ter process, InP HBTs have shown a f
T
of 183GHz.
The large bandgap of GaAs is also helpful in the at-
tainment of acceptable breakdown voltages. For example,
values of BV
CEO
in the range of 915 V are typical of Al-
GaAs/GaAs HBTs in which the collectors have a doping
density of 3 10
16
cm
3
and thicknesses in the range
300700nm. Comparable breakdown voltages in InP
HBTs can be obtained only with composite-collector or
DHBT structures since the bandgap and breakdown eld
in InGaAs are smaller than in GaAs. For really high break-
down voltages, InGaP DHBTs are preferred with BV
CEO
of
47V [63]. For a given f
T
, SiGe HBTs generally have a lower
breakdown voltage compared to GaAs HBTs since the col-
lector has to be doped heavily to avoid the Kirk effect.
4. MODELING
The importance of modeling cannot be overemphasized as
it forms crucial link between the device physics, device
CMOS Core Flow
Added Bipolar Steps
Buried Layer, n-epi and deep trench isolation
Shallow Trench Isolation
Nwell, Pwell implants
Gate ox & Poly Gate
NLDD implants
Spacer formation
NMOS S/D/G implant
Anneal
Collector sinker implant
Base window
SiGe epi base growth
Extrinsic base implant
Emitter poly dep & anneal
PMOS S/D/G implant
Anneal
Silicide and contacts
Metal backend
Figure 8. Addition of bipolar steps to a CMOS
ow to generate a BiCMOS ow.
structure, process technology, and circuit design. The bi-
polar compact models have evolved since the Gummel
Poon model to MEXTRAM, high-current model (HICUM),
vertical bipolar intercompany (VBIC) model [6466], and
others. Newer models are more complex and take addi-
tional computational time, but the advantage is that they
are more accurate. This helps in designing complex cir-
cuits and better prediction of the performance of a circuit
or a system. Some of the newer models have taken into
account impact ionization in the collector, quasisatura-
tion, self-heating effects, parasitic substrate transistor ef-
fects, and other factors that were not present in the
original GummelPoon model. A detailed consideration
of all these models is beyond the scope of this article.
However, in this section, we will present some of the major
differences between silicon BJT and HBT models.
4.1. DC and Thermal Effects
Let us now focus on DC properties. IIIV HBTs have high-
er offset voltage (V
offset
) compared to SiGe or silicon BJT.
As mentioned before, V
BE
for AlGaAs/GaAs and InGaP/
GaAs HBT is B1.4 V, whereas V
BE
for InP/InGaAs and
SiGe HBT is close to 0.7V, comparable to silicon BJT. Due
to the presence of the heterojunction at the emitterbase
region, the relation between I
C
and V
BE
in a HBT is not as
simple as in a BJT, and the ideality factor is greater than
unity and varies with V
BE
. As seen fromEq. (2), the current
gain of a HBT decreases with temperature while in a sil-
icon BJT, the current gain increases with temperature.
This has to be modeled properly. Typical I
C
V
CE
character-
istics of a silicon BJT slants upward because of (1) lower
doping in the base and (2) higher b at higher currents. In
HBTs, the I
C
V
CE
characteristics are generally at because
of higher doping in the base and/or compositional grading
across the base. At high currents, the IV curves droop
downward, due to self-heating, which results in a lower
current gain and negative Early voltage [67]. Thermal ef-
fects are more severe in GaAs HBTs since the thermal con-
ductivity of GaAs is 0.46W/cm 1C compared to silicon
1.2W/cm 1C. It should noted that the thermal effects mod-
eled for DC effects should also model AC thermal effects as
it is very important for power amplier applications.
4.2. AC Effects
Turning our attention now to AC properties, we nd that
there are new effects in DHBTs. It well known that at high
currents, f
T
falls off as a result of the familiar Kirk effect
[68], and this also results in a reduction of b with I
C
. It is
observed that f
T
falls off much more rapidly in DHBTs
than what is predicted by theory. In addition to the Kirk
effect, there is the heterojunction barrier effect, which
causes this rapid fall of f
T
at high currents. In DHBTs,
when there is a transition from narrow-bandgap base to
wide-bandgap collector, there is a valence band offset at
the heterointerface as explained in Fig. 1. Since this bar-
rier is masked by the band bending and the depletion re-
gion caused by collectorbase (CB) reverse bias, it has
negligible effect during low-injection or low-current oper-
ation. Under high injection levels, the CB electric eld de-
creases and nally collapses exposing the valence band
barrier. This barrier prevents the ow of holes into the
collector to maintain charge neutrality when the Kirk ef-
fect has set in. These holes pile up at the heterointerface,
causing charge imbalance, which results in a conduction
band barrier. The conduction band barrier opposes the
electron ow into the collector, causing a sudden decrease
of f
T
[69,70]. This effect is observed in IIIV DHBTs and
SiGe HBTs. At this point, it is also important to mention
another important factor, the heterojunction barrier effect,
which occurs at low currents as a result of improper base
prole design or diffusion of base dopants during subse-
quent processing, altering the intended base prole. If the
transition from narrow-bandgap base to wide-bandgap
collector occurs inside a neutral base, parasitic barriers
are produced at the conduction band edge, preventing the
electron ow into the collector and severely degrading f
T
[71,72]. This can be avoided by placing undoped spacers
between base and collector. In the case of a SiGe-graded
base HBT, if the boron outdiffuses past the Ge prole, it
could result in a parasitic conduction band barrier. One
method is to set the boron back a certain distance from Ge
peak to account for the boron diffusion or use carbon in the
base to suppress boron diffusion as explained before.
So far, we have discussed the variation f
T
with I
C
. We
will now consider the variation of f
T
with V
CE
. For a given
I
C
, as we increase V
CE
, the collector depletion region width
W
C
increases and C
BC
decreases, causing a reduction in t
C
as seen from Eq. (9). This would imply that we have a
lower transit time t
EC
from Eq. (3) and f
T
would increase
with increasing V
CE
. This is what is typically observed in
silicon BJTs. However, there is an opposing effect. The
transit time in the collector depletion region t
CSCR
in-
creases, as seen from Eq. (8). Therefore, the overall vari-
ation of f
T
with V
CE
is mixed. Initially, f
T
increases with
increasing V
CE
since there is a considerable reduction C
BC
.
Further increase in V
CE
increases the collector transit
time, and f
T
decreases with V
CE
. This has been observed in
IIIV HBTs. SiGe HBTs do not show this behavior since
the collector is very heavily doped.
Table 4. A Sample of High-Frequency Performance of
Various HBTs
Material Afliation
f
T
(GHz)
f
max
(GHz)
t
pd
(ps) Ref.
InP/InGaAs UCSB
a
370 459 54
InP/InGaAs Univ. Illinois,
UrbanaChampaign
452 155 55
InP/InGaAs POSTECH, Korea 154 478 56
InP/InGaAs UCSB 183
b
165
b
57
InAlAs/InGaAs UCSB 300 235 58
InAlAs/InGaAs UCSB 162
c
820
c
38
InGaP/GaAs Hitachi 156 255 30
InGaP/GaAs Tektronix 60 45 28 26
AlGaAs/GaAs Rockwell 60 350 62
Si/SiGe IHP, Germany 190 243 3.6 60
Si/SiGe IBM 180 338 3.9 61
Si/SiGe IBM 350 170 4.2 59
a
University of California at Santa Barbara.
b
Regrown emitter.
c
Transferred-substrate technology.
Since HBTs are increasingly used as power ampliers
in cellphones, the nonlinear behavior such as output
power, gain compression, power-added efciency (PAE),
inter-modulation distortion (IMD), and adjacent-channel
power ratio (ACPR) have to be modeled for optimizing
power amplier design. Some of the abovementioned
effects have been successfully modeled for both IIIV and
SiGe HBTs [7378].
4.3. Nonequilibrium Electron Transport
As device dimensions continue to shrink in the never-end-
ing quest for higher speed, transport in the shortened bulk
regions of the HBTcan no longer be faithfully described by
the classical processes of drift and diffusion. The base of
modern HBTs is one such region whose width is approach-
ing the length of a mean free path and, therefore, is a
region in which quasiballistic transport can be expected
to prevail. Quasiballistics transport in the base and velo-
city overshoot effect in the collector is observed in IIIV
HBTs. At high values of J
C
, the electric eld is perturbed
by the traveling space charge, leading to very high velocity
(even as high as B7 10
7
cm/s) near the basecollector
junction, leading to very short collector transit time t
CSCR
.
This results in very high f
T
just before onset of the Kirk
effect [7678].
4.4. Parameter Extraction
Once the model is chosen, the next task is to extract device
parameters for the model. This task is by no means is
trivial and creates problems due to the nature of the HBTs
and the complexity of the chosen model. For example, the
base resistance of a HBT is very low, and conventional
DC methods to extract R
B
and R
E
do not yield correct
results [79]. New AC methods have to be used for direct
extraction of the model parameters, and there are a
variety of papers in the literature [8082].
5. APPLICATIONS
When AlGaAs/GaAs and SiGe HBTs rst came into exis-
tence, the main applications were A/D and D/A converters,
low-noise ampliers (LNAs), and power ampliers. The
development and performance of these circuits are de-
scribed very well in the literature [47,45,46]. At the time
of this writing, approximately two decades have passed
since the rst demonstration of these circuits and in the
meantime, several new circuits have been developed.
Since the applications are widely varied, it will be impos-
sible to cover all of them. We will take a systematic ap-
proach to discuss these circuits on the basis of device
performance.
5.1. Frequency Dividers
One of the rst major building blocks in a digital circuit is
the ip-op, and the speed at which it can be toggled gives
a feel for the performance of the circuit. Table 5 shows a
summary of various HBT frequency dividers. Static fre-
quency dividers are simple ip-ops. Dynamic frequency
dividers make use of a mixer and a lowpass lter in a
regenerative loop. The output of the lter is fed back to
the mixer so as to generate an output whose frequency is
one-half of the input frequency. These generally run at
higher frequency, and the output swing is small. In the
dynamic frequency dividers, InP HBT from NTT stands
tall at a record frequency of 150GHz. The SiGe counter-
part from Inneon is equally impressive, running at
110 GHz.
Coming now to the static dividers, InP HBT from
Hughes Research Labs once again sets the record at
100 GHz. On the other hand, a SiGe divider clocking at
96 GHz was reported by IBM at the 2003 GaAs IC Sym-
posium. InGaP HBT dividers have moved from the hum-
ble beginning of 12.5 GHz to 39.5 GHz. The emphasis in
InGaP HBT technology today is mostly for power ampli-
ers for cellphones and base stations. Table 5 also shows a
0.18-mm in CMOS dynamic frequency divider operating at
40 GHz.
5.2. Oscillators and VCOs
The next circuit that can be benchmarked against the
others is a simple oscillator or voltage-controlled oscillator
(VCO). Table 6 compares oscillators from various technol-
ogies. Shown in the table is also the phase noise at a given
offset frequency. The highest-frequency oscillator reported
so far operates at 134GHz, and the transistor used in this
oscillator is a InGaP/InGaAs HBT with a f
max
of 170GHz.
This record is yet to be broken! The SiGe process from In-
neon with a f
max
of 200 GHz has shown a VCO running at
98 GHz. This is the highest reported frequency for an
oscillator in silicon-based technology. The next-highest
reported frequency for an InP HBT VCO is from TRW
running at 62.4GHz. The very rst YIG-tuned InGaP
HBT oscillator was reported by Prasad and Haynes [27].
InGaP HBT VCOs are currently running at 34.2 GHz
as reported by Hilsenbeck in the 2003 GaAs IC Sympo-
sium. A 0.12 m CMOS VCO from Inneon is running
at 51 GHz, which is very good benchmark for a standard
CMOS process.
Table 5. Summary of Various HBT Frequency Dividers
Material Afliation f (GHz) Type Ref.
InP/InGaAs NTT 150 Dynamic 91
InP/InGaAs Hughes Research
Labs
4100 Static 87
AlInAs/InGaAs Hughes Research
Labs
72.8 Static 88
InGaP/GaAs Hitachi 39.5 Static 89
InGaP/GaAs Tektronix 12.5 Static 24
AlGaAs/GaAs NTT 34.8 Static 90
SiGe Inneon 110.0 Dynamic 83
SiGe Inneon 99.0 Dynamic 84
SiGe IBM 96.0 Static 92
SiGe Inneon 86.0 Static 83
SiGe Hitachi 81.0 Static 85
SiGe Inneon 71.8 Static 86
CMOS 0.18mm UCLA 40 Dynamic 93
5.3. Power Ampliers
The superior high-frequency and high-voltage capabilities
of HBTs make them very good candidates for RF power
ampliers. The performance levels of power ampliers
and power transistors in different technologies are listed
in Table 7. It is clear that InGaP HBTs tend to dominate
RF power amplier market. A 10-W (40-dBm) InGaP
DHBT with a gain of 13 dB at 2 GHz and power-added ef-
ciency (PAE) of 50% was reported by Kurpas et al. [63].
An equally good contender for this was a 230-W (53.6-
dBm) SiGe power HBT from NorthropGrumman. This
device had a gain of 6.9dB at 2.8 GHz and a PAE of 46%.
For comparison, we have shown a 1-W AlGaAs HBT with
6 dB gain and 72% PAE from Mitsubishi. These high-pow-
er transistors are used mainly in power ampliers for cell-
phone base stations.
Turning our attention to power ampliers, we see a 60-
GHz power amplier with 15 dB gain and 13 dBm power
output from Sharp built on InGaP HBT technology. We
can compare this with a SiGe HBT power amplier from
Inneon operating at 718GHz. This power amplier has
14 dB gain, 11% PAE, and an output power of 17.5dBm. A
more recent IS-95B power amplier for cellphones from
Skyworks boasts a 27 dB gain and 40% PAE. The data
shown in Table 7 represent only a small sample of power
ampliers and power HBTs reported in technical papers
and available on the market.
5.4. LNA and Wideband Ampliers
The most important functional block in a receiver chain is
the LNA as it determines the minimum detectable input
signal level. Since the HBTs have high current gain, high
f
T
, and low base resistance, the noise gure of the device is
low. Minimum noise gure as low as 0.2 dB and associated
gain of 14 dB at 10 GHz has been reported for SiGe HBTs
[108]. Table 8 summarizes LNA and wideband amplier
performance. The SiGe LNA from STMicroelectronics
with 22dB gain and a noise gure (NF) of 1.6dB operat-
ing at 8.2 GHz is a good example of the performance that
can be expected from SiGe HBT technology. An example of
CMOS LNA built on 0.18m technology is also shown in
Table 8. At a frequency of 24 GHz, the LNA has a gain of
12.9 dB and noise gure of 5.6dB. Coming to wideband
ampliers, the highest record is again set by InP HBT. The
amplier boasts a bandwidth of 140200 GHz and has a
gain of 8.5dB.
5.5. Digital Communication Circuits
In beroptic communication systems, several digital cir-
cuit blocks can be built with HBTs because of these tran-
sistors superior high-frequency properties. InP HBT is
considered to be the best candidate for this purpose since
it is a suitable material for the fabrication of OEICs [124].
Beginning with the receiver chain, these include the
transimpedance amplier (TIA), the limiting amplier,
clock and data recovery (CDR), and the demultiplexer
(DMUX). On the transmitter chain, we have the multi-
plexer, the clock multiplication unit, and the voltage driv-
er and the laser modulator. Over the years, the data rates
have steadily increased, and we show some recent exam-
ples of circuits operating at 40Gbps. Table 9 shows digital
communication circuits fabricated in InP HBT, SiGe HBT,
silicon bipolar, and CMOS. We have also included one
example of a TIA in InP and SiGe HBT technologies since
it is an integral part of the system. It is interesting to note
that 50 Gbps 21 multiplexer was demonstrated as early
as 1996 in silicon bipolar technology. Some of these high-
speed circuits were once considered to be only in the realm
Table 6. Performance Summary of HBT Oscillators and VCOs
Material Type Afliation f (GHz) Phase Noise (dBc/Hz) f
offset
(MHz) Ref.
InGaP/InGaAs Oscillator Teratec 134 72 1 94
InGaP/GaAs VCO FBIH 34.2 108 1 95
InGaP/GaAs VCO KAIST 22.3 108 1 96
InGaP/GaAs YIG oscillator Tektronix 7.8 135 1 27
InP/InGaAs VCO TRW 62.4 104 1 97
SiGe VCO Inneon 98.0 97 1 98
SiGe VCO IHP 76.0 99
SiGe VCO RUB 46.9 108.5 1 100
CMOS 0.12mm VCO Inneon 51.0 85 1 101
Table 7. Performance Summary of HBT Power Ampliers
Material Type Afliation f (GHz) P
out
(dBm) Gain (dB) Power-Added Efciency (%) Ref.
InGaP/GaAs Power HBT FBIH 2 40 13 50 63
InGaP/GaAs Power HBT TI 3 31.8 7 52 28
InGaP/GaAs PA
a
Sharp 60 13.0 15 102
InGaP/GaAs WCDMA PA KAIST 1.9 28 28 40 103
InGaP/GaAs IS-95B PA Skyworks 0.837 28 27 40 104
AlGaAs/GaAs Power HBT Mitsubishi 12 30 6 72 107
SiGe Power HBT Northrop 2.8 53.6 6.9 46 105
SiGe PA Inneon 718 17.5 14 11 106
a
Power amplier.
of compound semiconductors. Note that the standard
CMOS technology is also advancing rapidly and demon-
strating successful 40-Gbps circuits.
6. FUTURE TRENDS AND INDUSTRY OUTLOOK
The heterojunction bipolar transistor has traditionally
been the workhorse for ultra-high-speed circuits and will
continue to maintain its place at least in the foreseeable
future. The HBT is here to stay in spite of the competition
from silicon bipolar and deep sub micrometer CMOS.
HBTs will dominate in wireless, microwave, beroptic
communications and test instruments.
The emitter widths of IIIV HBTs have scaled down
from several micrometers in the mid-1990s to current val-
ues of 0.4mm. On the other hand, the emitter widths of
SiGe HBTs have shrunk a lot faster and have reached
0.12 mm. This aggressive scaling of SiGe HBTs is due to
the advances in CMOS scaling, lithography, and advanced
processes. Base widths will continue to become thin to
meet the never-ending need and greed for high f
T
. As the
base gets thinner, the extrinsic portion of the base also
contributes to high R
B
. To reduce extrinsic base resis-
tance, a raised extrinsic base (REB) has been employed in
200-GHz SiGe processes [61]. Compared to SiGe HBTs,
IIIV HBTs will continue to have higher breakdown volt-
age because of its higher bandgap and breakdown eld.
The AlGaAs/GaAs HBT, which was dominant at one time,
has now been taken over by InGaP HBT. These days,
GaAs HBT means InGaP HBT, which will play a major
role in wireless applications. InP HBT will continue to
dominate in high-speed circuits and has a great potential
for developing 160-Gbps beroptic communication circuits
[124,125].
IIIV HBTs have traditionally employed mesa etching
of the emitter and a metal liftoff process to form base con-
tacts self-aligned to the emitter. This requires a base (and
emitter) that is large enough to be contacted by the ohmic
metal and results in high C
BC
. Higher C
BC
results in poor
device performance, despite the superior material proper-
ties of IIIV semiconductors. To overcome this problem,
regrown emitters are now being used to emulate the dou-
ble-polytransistor structures in silicon [57].
SiGe HBTs, on the other hand, have used heavily doped
polysilicon to form the emitter and contact the SiGe base.
This has resulted in lower BV
EBO
, higher R
E
, and a slight-
ly higher 1/f noise due to the presence of the interfacial
oxide between emitter and base. By using a regrown emit-
ter with the correct amount of doping, BV
EBO
can be in-
creased and R
E
can be decreased [83] with improved 1/f
noise. Standalone SiGe HBT processes will disappear, and
SiGe BiCMOS will be the mainstream process with very
high integration levels.
In this fast-moving space age, cost is a deciding factor
in making technology decisions. SiGe BiCMOS, which is
currently running on 200-mm wafers, with expectations to
move to 300-mm wafers, will continue to drive the cost
down with improved performance over newer generations.
Along with CMOS, BiCMOS offers a very high level of in-
tegration and functionality. This fact cannot be denied.
Realizing this, many silicon houses (IBM, Inneon, Jazz,
National, Maxim, Micrel, Motorola, etc.) have developed
their own SiGe processes. For those companies, which do
not have their own fabrication facilities, SiGe BiCMOS
foundry services are available from IBM, Jazz, TSMC,
UMC, and other manufacturers. Comparing this with
Table 9. Performance Summary of Digital Communication Circuits
Material Circuit Afliation Speed (Gbps) Ref.
InP/InGaAs 41 MUX/DMUX
a
NTT 40.0 114
InP/InGaAs CDR
b
/DMUX Inphi 43.2 115
InP/InGaAs 161 MUX Vitesse 40.0 116
InP/InGaAs TIAc Vitesse 40.0 117
SiGe 41 MUX/DMUX IBM 50.0 118
SiGe CDRRx/Tx IBM 43.0 119
SiGe TIA Lucent 40.0 120
Si bipolar 21 MUX Siemens 50.0 121
CMOS 0.12mm 21 MUX/DMUX Inneon 40.0 122
CMOS 0.18mm CDR/14 DMUX UCLA 40.0 123
a
Multiplexer/demultiplexer.
b
Clock and data recovery.
c
Transimpedance amplier.
Table 8. Performance Summary of LNA and Wideband Ampliers
Material Type Afliation f (GHz) Gain (dB) NF (dB) Ref.
InP/InGaAs Wideband amplier UCSB 140220 8.5 109
SiGe VG LNA ETH 16 14.5 3.8 110
SiGe LNA STMicro 8.2 22 1.6 111
SiGe DA LNA UIUC
a
0.123 14.5 6 112
CMOS 0.18mm LNA UCLA 24 12.9 5.6 113
a
University of Illinois, UrbanaChampaign.
IIIV HBTs running on 6- or 4-in. wafers with medium to
large-scale integration and higher cost, we nd that IIIV
HBTs have a place in some special applications where
SiGe BiCMOS cannot compete. For those competing in the
IIIV HBT applications who do not have a fabrication fa-
cility, foundry services are available from TRW, Knowl-
edge*On, Win Semiconductors, and other companies.
CMOS will continue to catch up with HBTs. However,
in applications where 1/f noise matters, such as in oscil-
lators, HBTs will dominate. The high-transconductance,
high-Early-voltage, and high-current-handling capabili-
ties of bipolar technology cannot be ignored. Also, aggres-
sively scaled CMOS can stand only lower supply voltages,
and this may be a problem in applications where voltage
swing is important. We believe that all the three technol-
ogies will coexist and have their own roles to play. It is not
too unrealistic for us to witness terahertz transistors in
the years to come.
Acknowledgment
The author is deeply grateful to his undergraduate profes-
sor, Dr. Sainath, who was instrumental in steering him
into the eld of semiconductor devices, which was consid-
ered a black art at that time (and still is!). The author
would also like to express his sincere thanks to D. L.
Pulfrey for providing an earlier version of this article,
which formed a framework on which to build. The author
also expresses his heartfelt thanks to the following people
for many fruitful discussions on issues pertaining to HBTs:
Peter Asbeck, Madhu Gupta, Frank Chang, S. J. Pearton,
David Roulston, C. R. Selvakumar, John Cressler, David
Harame, Tak Ning, Michael Schroter, Hans-Martin Rein,
and the late Greg Stillman. The author was given an
opportunity to work and contribute to IIIV HBTs at
Tektronix with the encouragement and support of Jack
Sachitano and Bruce Murdock. Thanks are also due to
Reda Razouk (National Semiconductor), Barry Small
(Micrel Semiconductor), Viktor Zekeriya, and Vijay Ullal
(Maxim) for giving the author an opportunity to contribute
to SiGe BiCMOS. Finally, the author would like to thank
his wife Chaya and daughter Rama for providing plenty
of time during evenings and weekends to complete the
deceptively simple task of writing this article.
BIBLIOGRAPHY
1. W. Shockley, Circuit Element Using Semiconductor Materi-
al, U.S. Patent 2,569,347 (1951).
2. H. Kroemer, Theory of a wide-gap emitter for transistors,
Proc. IRE 45:15351538 (1957).
3. H. Kroemer, Heterostructure bipolar transistors and inte-
grated circuits, Proc. IEEE 70(1):1325 (1982).
4. P. M. Asbeck, M. F. Chang et al., GaAlAs/GaAs heterojunc-
tion bipolar transistors: Issues and prospects for applica-
tions, IEEE Trans. Electron Devices 36(10):20322042
(1989).
5. M. E. Kim, A. K. Oki et al., GaAs heterojunction bipolar
transistor device and IC technology for high-performance
analog and microwave applications, IEEE Trans. Microwave
Theory Tech. MTT-37(9): 12861303 (1989).
6. P. M. Asbeck, Bipolar transistors, in S. M. Sze, ed., High-Speed
Semiconductor Devices, Wiley, New York, 1990, pp. 335397.
7. B. Jalali and S. J. Pearton, eds., InP HBTs: Growth, Processing
and Applications, Artech House, Boston, 1995, pp. 89133.
8. C. K. Maiti and G. A. Armstrong, Applications of Silicon-
Germanium Heterostructure Devices, Institute of Physics
Publishing, 2001.
9. A. P. Laser and D. L. Pulfrey, Reconciliation of methods for
estimating f
max
for microwave heterojunction bipolar tran-
sistors, IEEE Trans. Electron. Devices 38:16851692 (1991).
10. W. Fang, Accurate analytical delay expressions for ECL and
CML circuits and their applications to optimizing high-speed
bipolar circuits, IEEEJ. Solid State Circ. 25(2):572583 (1990).
11. J. M. C. Stork, Bipolar transistor scaling for minimum
switching delay and energy dissipation, IEDM Tech. Digest,
1988, pp. 550553.
12. O. S. Ang and D. L. Pulfrey, The cut-off frequency of base-
graded and junction graded AlGaAs DHBTs, Solid State
Electron. 34:13251328 (1991).
13. T. Ishibashi and Y. Yamauchi, A possible near-ballistic col-
lection in an AlGaAs/GaAs HBT with a modied collector
structure, IEEE Trans. Electron. Devices 35:401404 (1988).
14. C. G. Fonstad, Consideration of the relative frequency per-
formance potential of inverted heterojunction n-p-n transis-
tors, IEEE Electron. Device Lett. 5:99100 (1984).
15. B. Agarwal et al., A 277GHz f
max
transferred-substrate he-
terojunction bipolar transistor, IEEE Electron. Device Lett.
18:228231 (1997).
16. S. J. Prasad et al., A 45GHz AlGaAs/GaAs HBT IC technol-
ogy, GaAs IC Symp. Tech. Digest, 1991, pp. 121124.
17. T. P. Lester et al., A manufacturable process for HBT cir-
cuits, Proc. Int. Symp. GaAs and Related Compounds, 1993,
pp. 449454.
18. D. C. Streit et al., High-reliability GaAs-AlGaAs HBTs by
MBE with Be base doping and InGaAs emitter contacts,
IEEE Electron. Device Lett. 12:471473 (1991).
19. S. J. Prasad and E. Hultine, Reliability of Carbon doped base
AlGaAs/GaAs HBTs as a function of collector current, MRS
Symp. Proc. 300: 609614 (1993).
20. S. J. Prasad and E. Hultine, Safe operating current density
and failure modes of carbon doped base AlGaAs/GaAs HBTs,
Proc. European Solid-State Device Research Conf. (ESS-
DERC93), 1993, pp. 643646.
21. D. C. Streit et al., Comparison of MOCVD and MBE for
GaAs-AlGaAs HBT manufacturing, Int. Conf. GaAs Manu-
facturing Technology Tech. Digest, 1997, pp. 162165.
22. N. Bovolon, R. Schultheis et al., Analysis of the short-term DC
current gain variation during high current density-low tem-
perature stress of AlGaAs/GaAs heterojunction bipolar tran-
sistors, IEEE Trans. Electron. Devices 47(2):274281 (2000).
23. W. J. Ho, N. L. Wang et al., Self-aligned, emitter-edge pass-
ivated AlGAs/GaAs heterojunction bipolar transistors with
extrapolated maximum oscillation frequency of 350GHz,
IEEE Trans. Electron. Devices 39(11): 2655 (1992).
24. S. J. Prasad et al., 35GHz Ft. and 26GHz-Fmax GaInP/
GaAs heterojunction bipolar transistors, IEE Electron. Lett.
28(25):23412343 (1992).
25. W. Liu and S.-K. Fan, Near-ideal I-V characteristics of
GaInP/GaAs heterojunction bipolar transistors, IEEE Elec-
tron. Device Lett. 13:510512 (1992).
26. S. J. Prasad et al., First demonstration of 60GHz Ft GaInP/
GaAs HBT IC technology with 28ps ECL gate delay, IEE
Electron. Lett. 29(3):320321 (1993).
27. S. J. Prasad and C. Haynes, First demonstration of a InGaP/
GaAs HBT microwave oscillator, IEEE MTT-S Digest, 1994,
pp. 8790.
28. W. Liu, E. Beam III, et al., 1.5W CW S-band GaInP/GaAs/
GaInP double heterojunction bipolar transistor, IEEE Elec-
tron. Device Lett. 15(6):215217 (1994).
29. T. Iwai, S. Ohara et al., High efciency and high linearity
InGaP/GaAs HBT power ampliers: Matching techniques of
source and load impedance to improve phase distortion and
linearity, IEEE Trans. Electron. Devices 45(6):11961200
(1998).
30. T. Oka, K. Hirata et al., High-speed small scale InGaP/
GaAs HBT technology and its application to integrated
circuits, IEEE Trans. Electron. Devices 48(11):26252630
(2001).
31. S. R. Bahl, L. H. Camnitz et al., Reliability investigation
of InGaP/GaAs heterojunction bipolar transistors, IEEE
Electron. Device Lett. 17(9):446448 (1996).
32. N. Pan, J. Elliott et al., High reliability InGaP/GaAs HBT,
IEEE Electron. Device Lett. 19(4):115117 (1998).
33. T. S. Low, C. P. Hutchinson et al., Migration from an AlGaAs
to an InGaPemitter HBT IC process for improved reliability,
GaAs IC Symp. Digest, 1998, pp. 153156.
34. M. G. Alderstein and J. M. Gering, Current induced degra-
dation in GaAs HBTs, IEEE Trans. Electron. Devices
47(2):434439 (2000).
35. M. Hazi and W. E. Stanchina, Device and circuit fabrica-
tion, device characteristics and reliability, in B. Jalali and S.
J. Pearton, eds., InP HBTs: Growth, Processing and Appli-
cations, Artech House, Boston, 1995, pp. 135194.
36. W. Hafez, J. W. Lai et al., Vertical Scaling of 0.25mm Emitter
InP/InGaAs single heterojunction bipolars with f
T
of
452GHz, IEEE Electron. Device Lett. 24(7):436438 (2003).
37. D. Yu, K. Lee et al., Ultra high-speed InP-InGaAs SHBTs
with f
max
of 478GHz, IEEE Electron. Device Lett. 24(6):384
386 (2003).
38. Q. Lee et al., Submicron transferred-substrate heterojunc-
tion bipolar transistors, IEEE Electron. Device Lett.
20(8):396398 (Aug 1999).
39. R. Lee, Accelerated life test results on InGaAs/InP hetero-
junction bipolar transistors, GaAs IC Symp. Digest, 2003,
pp. 7477.
40. S. Bahl, N. Moll et al., Be diffusion in InGaAs/InP hetero-
junction bipolar transistors, IEEE Electron. Device Lett.
21(7):332334 (2000).
41. K. Yang et al., Design, modeling and characterization of
monolithically integrated InP-based (1.55mm) high-speed
(24Gb/s) p-i-n/HBT front-end photoreceivers, IEEE J. Light-
wave Technol. 14:18311838 (1996).
42. L. S. McCarthy et al., GaN HBT: Toward an RF device, IEEE
Trans. Electron. Devices 48(3):543551 (2001).
43. R. People, Physics and applications of Ge
x
Si
1x
/Si strained
layer heterostructures, IEEE J. Quantum Electron.
22:16961710 (1986).
44. D. L. Harame et al., Si/SiGe epitaxial-base transistorsPart
I: Materials, physics and circuits, IEEE Trans. Electron. De-
vices 42:455468 (1995).
45. D. L. Harame et al., Si/SiGe epitaxial-base transistorsPart
II: Process integration and analog applications, IEEE Trans.
Electron. Devices 42(3):469482 (1995).
46. J. D. Cressler, SiGe HBT Technology; A new contender for
Si-based RF and microwave circuit applications, IEEE
Trans. Microwave Theory Tech. MTT-46(5):572589 (1998).
47. B. S. Meyerson, UHV/CVD growth of Si and SiGe alloys:
Chemistry, physics and device applications, Proc. IEEE
80:15921608 (1992).
48. C. A. King, J. L. Hoyt et al., Si/Si
1x
Ge
x
heterojunction bi-
polar transistors produced by limited reaction processing,
49. F. Sato et al., A super self-aligned selectively grown SiGe
Base (SSSB) bipolar transistor fabricated by cold-wall type
UHV/CVD technology, IEEE Trans. Electron. Devices
41:13731378 (1994).
50. F. Sato et al., Sub-20ps ECL circuits with high-performance
super self-aligned selectively grown SiGe Base (SSSB) bipolar
transistors, IEEE Trans. Electron. Devices 42:483488 (1995).
51. H. J. Osten, D. Knoll et al., Carbon doped SiGe heterojunc-
tion bipolar transistors for high frequency applications,
Proc. BCTM Conf., 1999, pp 109116.
52. J. S. Rieh, K. Watson, F. Guarin, Z. Yang et al., Wafer level
forward current reliability analysis of 120GHz production
SiGe HBTs under accelerated current stress, Proc. Reliabil-
ity Physics Symp., 2002, pp. 184188.
53. G. Freeman, Z. Yang, F. Guarin et al., Reliability character-
istics of 200GHz f
T
285GHz f
max
SiGe HBTs, GaAs IC Symp.
Digest, 2003, pp. 187190.
54. Z. Grifth et al., InGaAs-InP Mesa DBHTs with simulta-
neously high f
T
and f
max
with low Ccb/Ic ratio, IEEE Elec-
tron. Device Lett. 25(5):250252 (2004).
55. W. Hafez et al., Vertical scaling of 0.25mm emitter InP/In-
GaAs single heterojunction bipolar transistors with f
T
of
452GHz, IEEE Electron. Device Lett. 24(7):436438 (2003).
56. D. Yu et al., Ultra high-speed InP-InGaAs SHBTs with f
max
of 478GHz, IEEE Electron. Device Lett. 24(6):384386
(2003).
57. D. Scott et al., A183GHz f
T
and 165GHz f
max
with abrupt InP
emitter, IEEE Electron. Device Lett. 25(6):360362 (2004).
58. Y. Betser, D. Scott et al., InAlAs/InGaAs HBTs with simu-
ltaneously high values of f
T
and f
max
for mixed analog/
digital applications, IEEE Electron. Device Lett. 22(2):56
58 (2001).
59. J. S. Rieh et al., SiGe HBTs with cut-off frequency of 350GHz
IEDM Tech. Digest, 2002, pp. 771774.
60. H. Rucker et al., SiGe:C BiCMOS technology with 3.6 ps gate
delay, IEDM Tech. Digest, 2003, pp. 121124.
61. B. Jagannathan, M. Meghelli, K. Chan, J.-S. Rieh, K.
Schonenberg, D. Ahlgren, S. Subbanna, and G. Freeman,
3.9ps SiGe HBT ECL ring oscillator and transistor design
for minimum gate delay, IEEE Electron. Device Lett.
24(5):324326 (2003).
62. W. J. Ho et al., Self-aligned, emitter-edge passivated AlGa-
As/GaAs heterojunction bipolar transistors with extrapolat-
ed maximum oscillation frequency of 350GHz, IEEE Trans.
Electron. Devices 39(11):2655 (1992).
63. P. Kurpas, A. Maassdorf et al., 10W GaInP/GaAs power
HBTs for base station applications, IEDM Tech. Digest,
2002, pp. 681684.
64. H. C. de Graaf and F. M. Klassen, Compact Modeling for
Circuit Design, Springer-Verlag, 1990.
65. J. J. X. Feng et al., A physics-based HBT SPICE model for
large-signal applications, IEEE Trans. Electron. Devices
42:814 (1995).
66. C. McAndrew, VBIC95, the vertical bipolar inter-company
model, IEEE J. Solid State Circ. 31:14761483 (1996).
67. W. Liu, Handbook of III-V Heterojunction Bipolar Transis-
tors, Wiley, New York, 1998.
68. C. T. Kirk, Theory of transistor cutoff frequency fall of at
high current densities, IEEE Trans. Electron. Devices 3:164
170 (1962).
69. S. Tiwari, A new effect at high currents in heterostructure bi-
polar transistors, IEEE Electron. Device Lett. 10:2105 (1989).
70. Z. Yu, P. E. Cottrell, and R. W. Dutton, Modeling and simu-
lation of high-level injection behavior in double heterojunc-
tion bipolar transistors, Proc. BCTMConf., 1990, pp. 192195.
71. J. W. Slotboom, G. Streutker et al., Parasitic energy barriers
in SiGe HBTs, IEEE Electron. Device Lett. 12(9):486488
(1991).
72. E. J. Prinz, P. M. Garone et al., The effects of base dopant
outdiffusion and undoped SiGe junction spacer layers in
Si/SiGe/Si heterojunction bipolar transistors, IEEE Elec-
tron. Device Lett. 12(2):4244 (1991).
73. G.-M. Park and S. Hong, A novel temperature dependent
large-signal model of heterojunction bipolar transistor with
a unied approach for self-heating and ambient tempera-
ture effects, IEEETrans. Electron. Devices 49(12):20992106
(2002).
74. Q. Liang, J. D. Cressler et al., A physics-based high injection
transit time model applied to barrier effects in SiGe HBTs,
IEEE Trans. Electron. Devices 49(10):18071813 (2002).
75. P. Asbeck, III-V HBTs for microwave applications: Technol-
ogy status and modeling challenges, Proc. BCTM Conf.,
2000, pp. 5257.
76. L. H. Camnitz and N. Moll, Compound Semiconductor Tran-
sistors, Physics and Technology, IEEE Press, 1993, pp. 2146.
77. T. Ishibashi, Nonequilibrium electron transport in HBTs,
78. M. Iwamoto, D. E. Root, J. B. Scott, A. Cognata, P. M. Asbeck,
B. Hughes, and D. C. DAvanzo, Large-signal HBT model
with improved collector transit time formulation for GaAs
and InP technologies, IEEE MTT Symp. Digest, June 2003,
pp. 635638.
79. S. J. Prasad, A method of measuring base and emitter resis-
tances of AlGaAs/GaAs HBTs, Proc. BCTM Conf., 1991, pp.
204207.
80. M. Sotoodeh, L. Sozzi, A. Vinay, A. H. Khalid, Z. Hu, A. A.
Rezazadeh, and R. Menozzi, Stepping toward standard
methods of small-signal parameter extraction for HBTs,
81. B. Sheinman, E. Wasige, M. Rudolph, R. Doerner, V. Sidorov,
and S. Cohen, A peeling algorithm for extraction of the HBT
small-signal equivalent circuit, IEEE Trans. Microwave
Theory Tech. MTT-50(12):28042810 (2002).
82. M. Dvorak and C. Bolognesi, On the accuracy of direct ex-
traction the heterojunction bipolar transistor equivalent-cir-
cuit model parameters Cp, CBC and RE, IEEE Trans.
Microwave Theory Tech. MTT-51(6):16401649 (2003).
83. T. F. Meister, H. Schafer et al., SiGe bipolar technology with
3.9 ps gate delay, Proc. BCTM Conf., 2003, pp. 103106.
84. J. Bock, H. Shafer et al., Sub 5 ps SiGe bipolar technology,
85. K. Washio, E. Ohue et al., Ultra-high speed scaled down
self-aligned SEG SiGe HBTs, IEDM Tech. Digest, 2002, pp.
767770.
86. M. Wurzer, J. Bock et al., 71.8GHz static frequency divider
in a SiGe bipolar technology, Proc. BCTM Conf., 2002, pp.
216219.
87. M. Mokhtari, C. Fields et al., 100 GHz static divide-by-2
circuit in InP-DHBT technology, IEEE J. Solid State Circ.
38(9):15401544 (2003).
88. M. Sokolich, C. H. Fields et al., A low-power 72.8GHz static
frequency divider in AlInAs/InGaAs HBT technology, IEEE
J. Solid State Circ. 36(9):13281334 (2001).
89. T. Oka, K. Hirata et al., High-speed small-scale InGaP/GaAs
HBT technology and its application to integrated circuits,
90. Y. Yamauchi, O. Nakajima, K. Nagata, H. Ito, and T. Ishi-
bashi, A 34.8GHz 1/4 static frequency divider using AlGaAs/
GaAs HBTs, GaAs IC Symp. Digest, 1989, pp. 121124.
91. S. Tsunashima, K. Murata, M. Ida et al., A 150GHz dynamic
frequency divider using InP/InGaAs HBTs, GaAs IC Symp.
Digest, 2003, pp. 284287.
92. A. Rylyakov and T. Zwick, 96GHz static frequency divider in
SiGe bipolar technology, GaAs IC Symp. Digest, 2003, pp.
288290.
93. J. Lee and B.Razavi, A 40 GHz frequency divider in 0.18m
CMOS technology, IEEE J. Solid State Circ. 39(4):594601
(2004).
94. K. Uchida, I.Aoki et al., 104 and 134GHz InGaP/InGaAs
HBT oscillators, Proc. GaAs IC Symp., 1999, pp. 237240.
95. J. Hilsenbeck, F. Lenk, W. Heinrich, and J. Wur, Low phase
noise MMIC VCOs for Ka band applications with improved
GaInP/GaAs HBT technology, GaAs IC Symp. Digest, 2003,
pp. 223226.
96. J.-G. Kim, D.-H. Baek, S. Jeon et al., A K-Band InGaP/GaAs
HBT Balanced MMIC VCO, IEEE Microwave Wireless
Compon. Lett. 13(11):478480 (2003).
97. H. Wang, K.-W. Chang et al., A 62GHz monolithic InP-based
HBTVCO, IEEEMicrowave Guided Wave Lett. 388390 (1995).
98. W. Perndl, H. Knapp et al., A 98GHz voltage controlled os-
cillator in SiGe bipolar technology, Proc. BCTM Conf., 2003,
pp. 6769.
99. W. Winkler, J. Borngraber et al., 60 GHz and 76GHz oscil-
lators in 0.25mm SiGe:C BiCMOS, IEEE ISSCC Digest,
2003, pp. 454455.
100. H. Li and H.-M. Rein, Millimeter wave VCOs with wide
tuning range and low phase noise, fully integrated in a SiGe
production technology, IEEE J. Solid State Circ. 38(2):184
191 (2003).
101. M. Tiebout, A 51GHz VCO in 0.13u CMOS, ISSCC Digest,
2002, pp. 372373.
102. S. Handa, E. Suematsu et al., 60GHz-band low noise am-
plier and power amplier using InGaP/GaAs HBT tech-
nology, GaAsIC Symp. Digest, 2003, pp. 227230.
103. J. Kim, J. Kim et al., An InGaP-GaAs HBT MMIC smart
power amplier for W-CDMA mobile handsets, IEEE J.
Solid State Circ. 38(6):905910 (2003).
104. Y. Yang, K. Choi et al., DC boosting effect of active bias cir-
cuits and its optimization for class-AB InGaP-GaAs HBT
power ampliers, IEEE Trans. Microwave Theory Tech.
MTT-52(5):14551463 (2004).
105. P. A. Potyraj, K. J. Petrosky et al., A 230W S-band SiGe
heterojunction bipolar transistor, IEEE Trans. Microwave
Theory Tech. MTT-42(11):23922397 (1996).
106. W. Bakalski, A. Vasylyev et al., A fully integrated 718GHz
power amplier with on-chip output balun in 75 GHz f
T
SiGe
bipolar, Proc. BCTM Conf., 2003, pp. 6164.
107. T. Shimura, M. Sakai et al., 1 W Ku band AlGaAs/GaAs
power HBTs with 72% peak power added efciency, IEEE
Trans. Electron. Devices 42(11):18901896 (1995).
108. D. R. Greenberg et al., Noise performance of low base resis-
tance 200GHz SiGe technology, IEDM Tech. Digest, 2002,
pp. 787790.
109. M. Urtega, D. Scott et al., G-band (140220GHz) InP-based
HBTamplier, IEEEJ. Solid State Circ. 38(9):14511456 (2003).
110. F. Ellinger and H. Jackel, Low-cost BiCMOS variable gain
LNA at Ku-band with ultra-low power consumption, IEEE
Trans. Microwave Theory Tech. MTT-52(2):702708 (2004).
111. G. Gramegna, A. Magliarisi, and M. Paparo, An 8.2 GHz,
14.4mW, 1.dB NF SiGe bipolar LNAwith DC current reuse,
Proc. BCTM Conf., 2003, pp. 4952.
112. Q. He and M. Feng, Low-power, high-gain and high-linearity
SiGe BiCMOS wide-band low-noise amplier, IEEE J. Solid
State Circ. 39(6):956959 (2004).
113. K. Yu, Y. Lu et al., K-band low-noise ampliers using
0.18mm CMOS technology, IEEE Microwave Wireless Co-
mpon. Lett. 14(3):106108 (2004).
114. K. Ishii, H. Nosaka et al., 4-bit multiplexer/demultiplexer chip
set for 40-Gbit/s optical communication systems, IEEE Trans.
Microwave Theory Tech. MTT-51(11):21812186 (2003).
115. S. Nielsen, J. C. Yen et al., A fully integrated 43.2Gb/s clock
and data recovery and 1:4 demux IC in InP technology,
IEEE J. Solid State Circ. 38(12):23412346 (2003).
116. A. Hendarman, E. Savaro et al., STS-768 multiplexer with
full-rate output data retimer in InP HBT, IEEE J. Solid
State Circ. 38(9):14971503 (2003).
117. C. Wu and E. A. Sovero, 40GHz transimpedance amplier
with differential outputs using InP-InGaAs heterojunction
bipolar transistors, IEEE J. Solid State Circ. 38(9):1518
1523 (2003).
118. M. Meghelli, A. Rylyakov, and L. Shan, 50Gb/s SiGe BiC-
MOS 4:1 multiplexer and 1:4 demulitplexer for serial
communication systems, IEEE J. Solid State Circ. 37(12):
17901794 (2002).
119. M. Meghelli, A. Rylyakov et al., A 0.18mm SiGe BiCMOS re-
ceiver and transmitter chipset for SONETOC-768 transmission
systems, IEEE J. Solid State Circ. 38(12):21472154 (2003).
120. H. S. Weiner, A. Leven et al., SiGe differential transimped-
ance amplier with 50GHz bandwidth, IEEE J. Solid State
Circ. 38(9):15121517 (2003).
121. A. Felder, M. Moller et al., 46Gb/s DEMUX, 50Gb/s MUX
and 30GHz static frequency divider in silicon bipolar tech-
nology, IEEE J. Solid State Circ. 31(4):481486 (1996).
122. D. Kehrer, H.-D. Wohlmuth et al., 40Gb/s 2:1 multiplexer
and 1:2 demultiplexer in 120nm standard CMOS, IEEE J.
Solid State Circ. 38(11):18301837 (2003).
123. J. Lee and B. Razavi, A 40Gb/s clock and data recovery cir-
cuit in 0.18mm CMOS technology, IEEE J. Solid State Circ.
38(12):21812190 (2003).
124. H. Jackel, U. Hammer et al., High speed InP-based HBTs
and OEICs, IEDM Tech. Digest, 2002, pp. 8386.
125. J. C. Zopler, Super-scaled InP HBTs for 150GHz circuits,
126. G. He, J. Howard et al., Self-aligned InP DHBT with f
T
and
f
max
over 300GHz in a new manufacturable technology,
HETEROSTRUCTURES DEVICES
VICTOR RYZHII
University of Aizu
Japan
All semiconductor devices are made of building blocks,
consisting of semiconductor structures with different
properties and supplied by some nonsemiconductor parts.
This relates to discrete devices and integrated circuits
alike. Junctions of semiconductors doped by different im-
purities are the most important structures of semiconduc-
tor devices. If a junction is formed by regions of the same
semiconductor material with one of them doped by accep-
tor impurities, and the other one by donor impurities, such
junction is called the p-n junction, or the p-n homojunc-
tion. Semiconductor structures consisting of homojunc-
tions are known as homostructures. More generally,
homostructures are dened as semiconductor structures
in which doping changes with position. Semiconductor
structures with a nonuniform chemical composition are
called heterostructures. The simplest heterostructure in-
cludes a single heterojunction, which is an interface be-
tween two different semiconductor materials. The size of
the region where the chemical composition changes can be
both small (about a few periods of the crystal lattice) and
relatively large (comparable with the size of the device).
The rst case corresponds to so-called abrupt heterostruc-
ture, in contrast to graded (or graded-gap) heterostruc-
ture, in which the chemical composition and, in turn, the
semiconductor energy bandgap change smoothly.
The abrupt heterostructures can be formed by almost
all combinations of semiconductor materials, although not
all combinations yield heterostructures with desirable
properties and quality. Examples of heterostructures in-
clude SiSi
1x
Ge, GaAsAl
x
Ga
1 x
As, GaSbInAs, and
other heterostructures. Graded heterostructures are
formed by many semiconductor compounds, in particular,
A
3
B
5
compound system.
The main property of heterostructures used in most
heterostructure devices is the nonuniformity of spatial
distributions of the bandgap and the edges of the valence
and conduction bands. This leads to the formation of the
so-called quasi-electric eld in heterostructure bulk affect-
ing charge carriers, as has been pointed out in a pioneer-
ing paper by H. Kroemer (see, e.g., Ref. 1 and references
cited therein). The quasi-electric eld in graded hetero-
structures forces electrons and holes to move in the same
direction, despite their opposite charges (Fig. 1). In het-
erostructures with abrupt heterojunctions, the band off-
sets can form potential barriers or ramps for electrons and
holes, shown in Fig. 2. The band offsets are the result of
abrupt variations of the chemical composition. Thus, the
Electrons
Valence band
Conduction band
E
n
e
r
g
y
Holes
Figure 1. Forces on electrons and holes in a graded-gap hetero-
structure. The forces in electrons and holes are in the same
direction.
HETEROSTRUCTURES DEVICES 1963
energy of the carriers at the band edges must change as
those carriers pass through the heterojunction.
An important practical constraint is the necessity to
select materials with small differences in lattice constants
at the required fabrication and operating temperatures. A
good lattice match is needed to minimize the density of
interface states and strain elds at the heterojunction. A
heterostructure with a double heterojunction can form a
potential well. If such a well is narrow enough, the energy
spectrum of electrons (or holes) becomes quantized
(Fig. 3), due to the connement of the latters in the direc-
tion perpendicular to the heterojunction plane. Such po-
tential wells are known as quantum wells (QWs).
Heterostructures are the elements of many of the most
advanced semiconductor devices currently being devel-
oped and fabricated. They are essential parts of modern
optoelectronic devices, such as semiconductor lasers, light-
emitting diodes, and photodetectors with highest perfor-
mance. Heterostructures are being employed increasingly
in high-speed digital and high-frequency analog devices
[2,3]. The advantages of heterostructures are that they
provide effective control over the states and motions of
charge carrierselectrons and holes in active regions of
heterostructure-based devices.
1. HETEROSTRUCTURE BIPOLAR TRANSISTORS
The principle of a bipolar transistor with heterojunctions
is old as the transistor itself. A heterostructure (hetero-
junction) bipolar transistor (HBT) is the rst semiconduc-
tor device incorporating heterostructures as it has been
patented by W. Shockley as early as 1948. HBTs differ
from ordinary bipolar transistors by the utilization of a
wide-bandgap semiconductor material for the transistor
emitter and, in some cases, for the collector, instead of the
same material as for the base. In particular, HBTs are
formed by a N-p-n heterostructure with a N-p heterojunc-
tion serving as the emitter junction, and a p-n junction for
the collector. Such HBTs are single-heterostructure bipo-
lar transistors (SHBTs). SHBTs can be made of P-n-p het-
erostructures as well. Double heterostructure bipolar
transistors (DHBT) consists of both the emitter and col-
lector heterojunctions. They have N-p-N or P-n-P struc-
tures. Symbols N and P denote the wide-bandgap
semiconductor portions doped by donors and acceptors,
respectively, while the symbols n and p correspond to the
narrow-bandgap regions with related types of doping.
The basic idea of an HBT is as follows. Consider the
energy band structure of an N-p-n HBT with an abrupt or
graded-emitter heterojunction, as in Fig. 4. The incorpo-
ration of the wide-gap emitter leads to the formation of an
additional barrier for the carriers in the base (holes in the
example under consideration in Fig. 4) inhibiting their
escape to the emitter region. This decreases the current of
holes injected from the base into the emitter. A gure of
merit for an HBT is the ratio of the collector current I
c
to
the base current I
b
:
b
I
c
I
b
o
I
n
I
p
b
max
1
Here I
n
and I
p
are the currents of electrons injected from
the emitter into the base and holes injected from the base
into the emitter. The ratio of the electron and hole cur-
rents, that is, the parameter b
max
, is given by the following
expression
b
HBT
max

N
e
p
b
v
nb
v
pe
exp
e
v
kT
_ _
b
0
max
exp
e
v
kT
_ _
2
for HBTs with an abrupt emitter heterojunction and
b
HBT
max
b
0
max
exp
e
v
e
c
kT
_ _
3
in the case of HBTs with a graded heterojunction. Here k
is the Boltzmann constant, T is the temperature, N
e
and
p
b
are the electron and hole concentrations in the emitter
and the base, respectively, v
nb
and v
pe
are the mean ve-
locities of electrons and holes in the related regions, e
v
and
e
c
are the band edge discontinuities related to the valence
and conduction bands in the case of an abrupt heterojunc-
tion. For a graded heterojunction, e
v
and e
c
are the frac-
tions of the change of the bandgap in the emitter and base
regions related to the valence and conduction bands. In
Conduction band
Heterojunctions
Valence band
E
n
e
r
g
y
(a) (b)
Figure 2. Energy barriers (a) and ramps (b) for electrons and
holes in an abrupt heterostructure. Electrons and holes are re-
jected from the heterojunction if their energy is smaller than the
barrier height. They acquire the energy passing the ramp at the
heterojunction.
Conduction band
Quantum well for electrons
E
n
e
r
g
y
Quantum well for holes
Valence band
(a) (b)
Figure 3. Quantum wells for electron and holes (a) and quantum
well for electrons only (b) in double heterostructures.
1964 HETEROSTRUCTURES DEVICES
the general case e
v
e
c
De
g
, where De
g
is the difference
between the bandgaps in the different parts of the hetero-
structure. In Eqs. (2) and (3) b
0
max
corresponds to a homo-
structure bipolar transistor (e
v
e
c
0), with the same
doping of all its parts as the HBT under consideration.
For a good transistor, a value b
max
should be large. In
conventional bipolar transistors the large value of b
max
is
achieved by significantly higher doping level of the emit-
ter, in comparison to the base (N
e
bp
b
). However, very
high values of b
max
can be realized in HBTs, almost re-
gardless of the doping ratio, due to large value of the ex-
ponent in Eqs. (2) and (3). Indeed, if De
g
0.2 eV at T
300K, one has b
HBT
max
=b
0
max
expDe
g
=kT 3000b1.
Thus, high b values can be obtained without a high emit-
ter-to-base doping ratio. The increase of the base doping
level results in lower base resistance. Simultaneously, a
lightly doped emitter region provides smaller capacitance
of the emitter. Both high base and low emitter doping pro-
mote better high-frequency performance of HBTs, in com-
parison to homostructure bipolar transistors. This is due
to lower base resistance and smaller emitterbase capac-
itance. High base doping leads to lower noise as well. In
addition, high base doping results in higher punch-
through voltage.
The conduction band discontinuity in the emitter het-
erojunction provides the injection of hot electrons from the
emitter to the base. If the base is thin enough, the injected
electrons can pass it without scattering or enduring a few
collisions with impurities and phonons. In the case of such
ballistic or near-ballistic transport of electrons in the HBT
base, their delay time can be very short. This also con-
tributes to the advantages of HBTs over standard bipolar
transistors. Incorporation of the graded-gap base with
quasi-electric eld yields the acceleration of the injected
electrons (or holes) in the base. Such a design provides
higher performance as well.
Apart from HBTs with a single heterojunction (usually
in the emitter), that is, SHBTs, DHBTs are also considered
as prospective components, especially for digital circuits
and novel functional devices. The utilization of an addi-
tional heterojunction (in the collector) opens up opportu-
nities for separate optimization of the base and the
collectorthe interchangeability of their functions in
some circuits.
The incorporation of more sophisticated heterostruc-
tures into the emitter of an HBT or its other parts can
substantially extend the diversity of the HBT character-
istics. For example, HBTs with a double-barrier hetero-
structure in the HBT emitter providing resonant-
tunneling injection can exhibit N-shape current-voltage
characteristics. The utilization of resonant tunneling
HBTs (RTHBTs) promises a significant reduction in the
logic circuits complexity [4]. HBTs without base contact
can be used as phototransistors with very high photoelec-
tric gain, making them rather effective photodetectors for
optical ber communication systems and other applica-
tions.
2. HETEROSTRUCTURE HOT-ELECTRON TRANSISTORS
Heterostructure hot-electron transistors (HETs) are made
on the base of a double heterojunction structure. Wide-gap
regions form the HET emitter and collector. A narrow-
bandgap region, sandwiched between the wide-bandgap
emitter and collector regions, serves as the HET base. In
contrast to HBTs, the HET base is doped by the same type
of dopants as the emitter and the collector. Hence HETs
are unipolar devices. The energy band diagram of a HET
with a N-n-N structure is shown in Fig. 5.
The electron injection from the emitter to the base, and,
further, to the collector in HETs with the structure of
Fig. 5, is associated with thermionic emission of electrons
overcoming the barrier at the N-n interface. Electrons in-
jected from the wide-bandgap emitter have excess kinetic
energy in the base. Their motion is directed primarily
perpendicular to the heterojunction plane. The directed
velocity of electrons significantly exceeds the thermal
velocities of both the injected and thermalized electrons
in the base. That is why such transistors are called the
HETs. The thermalized (or cold) electrons provide electri-
cal neutrality of the base. As the scattering of the injected
electrons in the HET base leads to their capture in the
latter, and prevents them from the injection into the col-
lector, the electron transport in the base should be either
ballistic or near-ballistic.
Different types of HETs have been proposed. The injec-
tion of electrons (or holes in P-p-P HETs) from the emitter
Figure 4. Energy band diagrams of N-p-n
HBTs with an abrupt (a) and graded (b) emit-
ter and schematic view of their structure.
can be associated with thermionic or tunneling processes.
The energy band diagram of a HET, with tunneling injec-
tion of electrons, is shown in Fig. 6.
The main problem connected with the development of
high-performance HETs is the tradeoff between the cap-
ture rate of hot electrons into the base, which decreases
the HETcurrent gain, and the base resistance, which lim-
its high-speed potentials of HETs. The point is that the
enhancement of the base doping level necessary for
the lowering of the base resistance leads to the increase
of the capture probability of hot electrons, due to the rein-
forcement of their scattering on donors and cold electrons.
In HETs with a thin base, electrons in the latter can be
quantized. The quantization of the electron spectrum in
the HET base adds an additional complexity to the HET
operation. The existence of a bound state in the electron
spectrum can result in significant contribution of resonant
tunneling processes to the injection of electrons. Further
development of the idea of hot-electron transport resulted
in the proposal of a HET with a resonant tunneling emit-
ter. The operation principle of resonant tunneling HETs
(RTHETs) is demonstrated in Fig. 7. The most important
feature of the RTHET operation is that at some collector
base voltage, the injected electron current has a maxi-
mum. The further increase of this voltage leads to a sharp
drop of the injected current. Thus, RTHETs are transis-
tors exhibiting negative differential resistance. This prop-
erty is considered as very promising for future
applications in different circuits.
Although HETs of different types are still under inves-
tigation, they have no commercial significance, despite
very promising features of their characteristics.
3. HETEROSTRUCTURE FIELD-EFFECT TRANSISTORS
Heterostructure eld-effect transistors (HFET) are eld-
effect, three-terminal devices, akin to the metal semicon-
ductor eld-effect transistors (MESFETs) with a Schottky
gate and metal insulator eld-effect transistors (MISFET),
taking advantage of electron transport in heterostructure
channel. A general name HFET is used for a family of
eld-effect transistors on the base of different heterostruc-
tures. This family includes the modulation-doped eld-ef-
fect transistor (MODFET), which is also known as the
high-electron-mobility transistor (HEMT), the hetero-
structure insulated-gate eld-effect transistor (HIGFET),
and some others.
In MODFETs, the wide-bandgap layer beneath the me-
tallic gate is doped, and carriers transfer to the layer of an
undoped narrow-gap material. The narrow-bandgap ma-
terial layer forms the MODFET channel, which is usually
undoped. The result of the modulation doping is that elec-
trons (or holes) in the channel are spatially separated
from the doped layer. Because of this, they can have ex-
tremely high mobility along the heterojunction due to the
elimination of impurity scattering. The most common
Figure 5. Energy band diagram of a N-n-N HET and schematic
view of its structure.
Figure 6. Energy band diagram and structure of a HET with
tunneling injection of electrons.
Figure 7. Operational principle and schematic view of the struc-
ture of a resonant tunneling HET.
MODFETs utilize Al
x
Ga
1 x
AsGaAs heterostructures. A
typical view of the MODFET structure cross section is
shown in Fig. 8.
Electrons (holes) in the MODFET channel are conned
by the heterojunction from one side and by the electric
potential creating the electric eld, forcing them against
the heterointerface. Such connement of electrons may
lead to the quantization of their energy spectrum. If the
electron connement is strong, so that the width of the
channel is small enough, electrons form a two-dimension-
al (2D) electron gas, located near the heterojunction.
Sometimes, MODFETs with a 2D electron gas in the chan-
nel are called the two-dimensional electron gas eld-effect
transistors (TEGFETs).
The MODFET performance is strictly dependent on the
thickness and quality of a very thin undoped layer of a
wide-gap material, separating the doped region and the
narrow-bandgap channel. This so-called spacer is usually
made of iAL
x
Ga
1x
As (see Fig. 8).
HIGFETs differ from MODFETs, in that both the wide-
bandgap and the narrow-bandgap layers are undoped. In
some HFETs, the narrow-bandgap channel is doped. Such
HFETs lose the advantage of high electron mobility. Their
operation is similar to that of MISFETs. Along with the
HFETs, in which the channel is formed by a heterojunc-
tion and an electrostatic barrier, in some HFETs a double
heterostructure is used to form the channel. In the latter
case, the HFET channel can be a QW, because of strong
electron connement.
4. HETEROSTRUCTURE LASERS AND LIGHT-EMITTING
DIODES
First semiconductor lasers began as homostructure devic-
es comprising a p-n homojunction. Today, semiconductor
lasers are usually made of a heterostructure, forming a
single or multiple QW. The incorporation of a heterostruc-
ture and, especially, a QW in the laser structure provides
significant advantages of heterostructure laser diodes
(HLDs) over lasers with homojunctions [2,57]. The
same is true for heterostructure light-emitting diodes
(HLEDs) as well. The main of such advantages are much
lower threshold current of lasing and higher operational
temperatures. The implementation of heterostructures in
lasers resulted in the development and wide applications
of HLDs operating at room temperature.
The energy band diagrams of HLDs are shown in Fig. 9.
The wide-bandgap and N and P regions provide the elec-
tron and hole injection into a narrow-bandgap active
region (Fig. 9).
The lasing power is given by the following equation:
P
h% o
e
I I
th
4
where _o is the energy of the lasing photons, _ is the re-
duced Planck constant, o is the photon angular frequency,
e is the electron charge, I is the pumping current, and I
th
is
the threshold current. The latter is dened by the prop-
erties of the HLD structure materials, the HLD geometry,
quality of the mirrors reecting generated radiation or
reecting property of the HLD surfaces serving as the
mirrors, and so on.
In HLDs with a QW, electrons and holes are captured
in the latter and occupy 2D states. Due to the existence of
two barriers, there is the electronhole connement with-
in a narrow-gap region. So the barriers prevent the leak-
age of carriers from the HLD active region. It results in
higher electron and hole concentrations in the active nar-
row-bandgap region, in comparison with homostructure
lasers at the same values of the injected current. This
means that the threshold concentration of electrons and
holes sufcient for the change of the absorption coefcient
from negative (absorption of photons with the energies
close to the energy gap of the active-region material) to
positive (amplication) is achieved at lower injected cur-
rent. As a result, the injected current corresponding to the
threshold of lasing (threshold current I
th
) reduces if a
double heterostructure is used. The incorporation of a
more narrow-bandgap layer into the HLD active region,
which forms a QW, yields an additional improvement in
Figure 8. Cross-sectional view of a MODFET with n-channel.
Figure 9. Energy band diagrams of HLDs with a narrow-gap
active region (a) and narrow-gap active region with a quantum
well (b).
the laser performance. Smaller density of states in a QW,
due to the 2D nature of the latter, leads to their more full
occupation by electrons and holes at given value of the
injected current. This effect also contributes to the
achievement of the lowest possible threshold current I
th
.
Apart from the electronhole connement, HLDs ben-
et of the connement of lasing modes, due to the wave-
guide effect. This effect (termed optical connement) is
connected with the larger crystal lattice refraction index
of a narrow-gap part of the HLD structure playing a role of
the active region. In homostructure lasers, the effect of
optical connement plays some useful role in reducing the
diffraction losses of emitted radiation. In such lasers, the
optical connement is associated with the nonuniformity
of the electron and hole concentrations near the p-n junc-
tion. In HLDs, the optical connement is much more ef-
fective.
The spectrum of radiation emitted by HLDs and
HLEDs is determined primarily by the value of the ener-
gy gap of the active region material. The energy of the
emitting photons _o is close to the energy gap of the
active-region semiconductor E
g
:
h% o E
g
5
This equation can be rewritten as
l
2p h% c
E
g
6
where l is the lasing wavelength and c is the velocity of
light in vacuum. If E
g
is expressed in electronvolts, the
lasing wavelength l in micrometers according to Eq. (6) is
given by lC1.24/E
g
. For HLDs with a GaAs active region
(E
g
1.42 eV) and Al
1x
Ga
x
As contact regions, one has
lC0.87 mm. Utilizing different semiconductor materials,
especially binary, ternary, and quaternary compounds,
one may fabricate HLDs and HLEDs operating from mid-
infrared to blue range of the spectrum. In HLDs with QWs
in the active region, the emitted photons are generated
due to the transitions from the electron quantum levels to
the hole quantum levels in the QWs (see Fig. 9). In such a
case, the effective bandgap dening the energy of the las-
ing photons E
QW
g
and their wavelength l 2ph% c=E
QW
g
de-
pend also on the positions of the quantum levels, with
respect to the bottom of the conduction and the top of the
valence bands in the QWs. One may obtain
h% o E
QW
g
E
g
p
2
h%
2
2w
2
1
m
n
1
m
p
_ _
7
where w is the width of the QW and m
n
and m
p
are the
effective masses of electrons and holes, respectively. Grow-
ing the HLD structure with appropriate width of the QW
in the action region, one may control the lasing wave-
length.
HLDs and HLEDs nd numerous applications. They
are used in consumer electronics, for example, in CD play-
ers, optical communications, navigation devices, and other
systems. Thanks to the development of HLDs and HLEDs,
operating in relatively long-wavelength range of spectrum
from one side, and heterostructure devices, emitting very
short-wavelength light (blue light), their applications can
be extended to atmosphere monitoring systems, displays,
trafc lights, and so on.
5. QUANTUM-WELL PHOTODETECTORS
Conventional photodetectors utilize the transitions of elec-
trons from the valence band ground state to the conduc-
tion band excited state. The photocarriers (electrons and
holes) created as a result of such interband transitions
produce a photocurrent in photodetectors. To create an
electronhole pair, the photon energy _o should be greater
than the energy gap E
g
of a semiconductor material used
for a photodetector. By controlling E
g
, one may fabricate
photodetectors for different ranges of spectrum. It is pos-
sible by using a chemical composition chosen in the proper
way. For visible or near-infrared ranges of spectrum, A
3
B
5
and some other semiconductor materials are used.
Carrier multiplication, which results from impact ion-
ization initiated by electrons and holes generated due to
optical interband transitions at high electric elds across
the photodetector active region, is used to achieve internal
photoelectric gain and, as a result, higher performance.
Avalanche photodiodes (APDs), which utilize impact ion-
ization at bias near the breakdown voltage, can be built
using both homo- and heterostructures. Two of the crucial
performance characteristics of APDs, the gainbandwidth
product and the excess noise arising from the random na-
ture of the avalanche multiplication of electrons and holes,
are determined by the electron and hole ionization coef-
cients and, what is more important, by the ratio of the
latter. One approach to achieving low multiplication noise
in APDs is the use of heterojunctions to articially en-
hance the ionization rate of either electrons or holes. The
most successful APD of this type is the APD with a mul-
tiple-QW structure. The point is that, for low-noise and
high gainbandwidth product, the ratio of the electron and
hole ionization coefcients k should be either large or
small (k51 or kb1). It means that a large difference in
the ionization rates is necessary. As electrons emerge from
the wide-bandgap region between the QWs into the nar-
row-bandgap portion (into the QW), the discontinuity in
the conduction band provides sufcient additional energy
to initiate ionization. This enhances the ionization rate of
electrons. The ionization rate for holes, on the other hand,
is not enhanced to the same degree, since the valence band
offset is smaller than that of the conduction band [2,8] in
many practically important heterostructures.
Heterostructure APDs with QWs are successfully fab-
ricated in GaAsAl
1x
Ga
x
As and InPIn
0.53
Ga
0.47
As com-
pound material systems, and some others.
To satisfy the condition _o4E
g
for very important far-
and midinfrared ranges of spectrum, corresponding to
wavelengths l 220 mm, semiconductor alloys such as
Hg
1x
Cd
x
Te are used. There are substantial technologi-
cal difculties to grow, process, and fabricate photodetec-
tors made of such materials. The transitions from the
impurity states to the conduction or valence band can also
be utilized in photodetectors operating in far and near in-
frared ranges. However, these photodetectors also have
some disadvantages.
Quantum-well intersubband photodetectors (QWIPs),
based on semiconductor heterostructures, are considered
as a very prospective alternative to both Hg
1x
Cd
x
Te in-
terband photodetectors, as well as impurity photodetec-
tors. QWIPs utilize the intraband electron transitions in
the conduction band (in n-type QWIPs) or the intraband
hole transitions in the valence band (in p-type QWIPs). By
absorbing photons, electrons transfer from the bound
states in QWs into states above the barriers between the
QWs (continuum states); that is, they transfer between
the subbands within a band. Such intersubband transi-
tions result in the occurrence of electrons (holes) in con-
tinuum states, where they can freely move, producing a
photocurrent. Conventional QWIP consists of a hetero-
structure with a single or multiple QW, doped either by
donors or acceptors. The QW structure is supplied by con-
tact regions of the same type of doping as the QWs. These
contacts serve as the QWIP emitter and collector [9]. The
conduction band edge prole of the QWIP structure is
shown in Fig. 10. Usually QWIPs are made of A
3
B
5
or
Si
1 x
Ge
x
compounds.
QWIPs operate in the range of spectrum, in which the
energy of incident photons is sufcient to provide electrons
absorbing such photons energy to escape from a QW:
h% o > E
i
8
Here E
i
is the ionization energy of the QW, which is the
difference between the energy of the barrier top and the
bottom of the 2D subband in the QW (see Fig. 10). The
ionization energy E
i
depends on the depth of the QW,
which, in turn, is dened by the difference in the chemical
compositions of the barrier and QW materials, and the
QW width. Both the depth and width can be easily varied
during the QWIP structure growth process, to adjust the
range of the sensitivity of the QWIP.
The photoexcited electrons are collected, thereby pro-
ducing a photocurrent. The escape of electrons from QWs
due to their photoexcitation leads to some redistribution of
the potential across the QWIP structure and, in turn, to
the increase of the electric eld at the QWIP emitter con-
tact. This results in the injection of extra electrons from
the emitter. The current created by the injected electrons
can significantly exceed the current produced by the pho-
toexcited electrons, so that QWIPs can exhibit a photo-
electric gain. The latter can be markedly greater than
unity. The photocurrent in a QWIP is given by the follow-
ing formula:
I
ph
es
F
p
c
9
Here s is the cross section of the electron photoionization
from the QW, S is the electron sheet concentration in each
QW dened by the sheet concentration of donors, F is the
photon ux, and p
c
is the probability of the electron cap-
ture into QWs. It is instructive that the photocurrent ac-
cording to Eq. (9) does not depend on the number of QWs
in the QWIP N. Usually the value p
c
is small, so that the
photoelectric gain g (Np
c
)
1
can be large, even in QWIPs
with multiple-QW structure. From Eq. (9), one may obtain
the following expression for the QWIP responsivity:
R
es
p
c
h% o
10
For a QWIP with typical parameters s 2 10
15
cm
2
, S
10
12
cm
2
, p
c
0.010.05, and _o0.1 eV (l12 mm),
from Eq. (10), one has R0.4 to 2A/W.
Relatively simpler and cheaper QWIP technology is not
the only advantage making QWIPs potential competitors
to standard infrared photodetectors. The QWIP advanta-
ges are connected also with their intrinsic high-speed op-
eration and the feasibility of their integration with other
A
3
b
5
and Si
1 x
Ge
x
devices.
Despite the novelty of QWIPs, they already nd appli-
cations as components for infrared imaging devices with
large staring arrays of QWIPs.
6. OTHER HETEROSTRUCTURE DEVICES
Among heterostructure devices not discussed above, there
are some others that are considered as very prospective in
the future. One may point out the real-space transfer
transistors (RSTTs) utilizing the real-space transfer of
electrons or holes between two semiconductor materials.
The RSTT operation requires a heterostructure in which
the semiconductor layer with wider energy gap has much
reduced mobility. The GaAsAl
1x
Ga
x
As heterostructures
have been used so far largely for the RSTTs. For detailed
discussions on RSTTs, readers are referred to Ref. 4.
Solar cells are also an example of devices in which the
utilization of heterostructures provides marked advanta-
ges. The advantages of heterostructure solar cells (HSCs)
over conventional p-n homojunction SCs are as follows [5]:
First, HSCs exhibit enhanced short-wavelength response,
if the energy gap of the HSC wide-bandgap layer exceeds
the energy of photons to be absorbed inside the depletion
region in the narrow-bandgap portion. Apart from this,
they have lower series resistance, if the wide-bandgap
Figure 10. Conduction band edge proler of a n-type QWIP un-
der biasing voltage. Arrows show injected, captured, and photo-
excited electrons.
region is heavily doped without affecting its transparency.
In addition, HSCs can exhibit higher radiation tolerance.
Many of novel devices utilize quantum properties of
electrons in heterostructures with QWs. As an example,
one may mention resonant tunneling diodes (RTDs) for
logic and ultra-high-frequency applications [2,10]. Com-
bining RTDs with HBTs or HFETs allows the fabrication
of compact high-speed circuits that operate at room tem-
perature. In the early 1990s, the electron intersubband
transitions were utilized for laser generation of midinfra-
red radiation in QW structures, called quantum cascade
lasers [11]. The intersubband lasers have many potential
applications in the midinfrared range of spectrum.
The integration of heterostructures utilizing both in-
tersubband (intraband) and interband transitions open up
additional prospects for the development of new functional
QW devices. For example, integrated QWIP-LED devices
[12] can be used for effective conversion of far- or midin-
frared signals or images into near-infrared or, possibly,
visible signals and images. In QWIP-LED devices, their
QWIP part, utilizing intersubband electron transitions,
serves as an element sensitive to infrared radiation. It
produces a photocurrent that, being injected into the LED
active region, results in the generation of relatively short-
wavelength output radiation, due to radiative recombina-
tions of the injected electrons. High-performance discrete
devices and pixelless imagers can be fabricated using in-
tegrated QWIP-LED heterostructures.
In QWs electrons or holes are spatially conned in one
direction. The energy corresponding to their motion in this
direction is quantized, while in other directions it can be
considered as classical. Advances in microfabrication tech-
nology now permit the building of heterostructures, in
which electrons and holes are conned in two or even
three directions. In the rst case, electrons and holes are
conned in a narrow-bandgap semiconductor material re-
gion extensive in one direction and buried in a wide-band-
gap material. In such heterostructures, usually called
quantum wires (QWRs), the electron (hole) energy spec-
trum is characterized by two discrete quantum numbers
and one continuous quantum number. The latter is the
electron (hole) momentum. Thus, electrons and holes in
QWRs are propagating as one-dimensional particles. If the
region of a narrow-bandgap semiconductor, material has a
form of a small box electrons, and holes in the narrow-
bandgap box exhibit fully discrete energy spectrum. Small
boxes of a narrow-bandgap semiconductor surrounded by
a wide-bandgap material with discrete energy spectrum
are called the quantum boxes or quantum dots (QDs).
QDs are similar to real atoms, because electrons (holes) in
them have discrete energy spectrum as it takes place in
atoms. A QD is said to be a zero-dimensional structure.
However, sizes of QDs are substantially larger than those
of real atoms. This is due to large number of atoms of
semiconductor material involved in the formation of a QD.
The energy spectrum of a QWR or a QD is dened by the
difference in the energy gaps of the QWR or QD material
from one side, and surrounding material from another.
The most crucial are the QWR or QD sizes. If the QD size
is small enough, the QD can have the only one quantum
level. Usual size of QWRs (in the direction perpendicular
to the direction of classical electron or hole motion) and
QDs is about a few nanometers. Due to energy spectra
significantly different from those for conventional hetero-
structures and even for QWs, the one- and zero-dimen-
sional structures constituting QWRs and QDs are very
promising for electronic and optoelectronic devices, espe-
cially lasers. More detailed description of the QWR and
QD properties, and their possible applications, can be
found in Refs. 2 and 13.
BIBLIOGRAPHY
1. H. Kroemer, Proc. IEEE 70:1325 (1992).
2. N. G. Einspruch and W. R. Frensley, eds., Heterostructures
and Quantum Devices, Academic Press, San Diego, 1994.
3. S. Tiwari, Compound Semiconductor Device Physics, Academ-
ic Press, San Diego, 1992.
4. S. M. Sze, ed., High-Speed Semiconductor Devices, Wiley, New
York, 1990.
5. S. M. Sze, Physics of Semiconductor Devices, Wiley, New York,
1981.
6. G. P. Agrawal and N. K. Dutta, Semiconductor Lasers, Van
Nostrand Reinhold, New York, 1993.
7. P. Bhattacharya, Semiconductor Optoelectronic Devices,
Prentice-Hall, Englewood Cliffs, NJ, 1994.
8. P. Bhattacharya, ed., Properties of III-V Quantum Wells and
Superlattices, INSPEC, London, 1996.
9. B. F. Levine, Quantum-well infrared photodetectors, J. Appl.
Phys., 74:R1R81 (1993).
10. H. Mizuta and T. Tanoue, The Physics and Applications of
Resonant Tunneling Diodes, Cambridge Univ. Press, Cam-
bridge, UK, 1995.
11. J. Faist, F. Capasso, D. L. Sivco, A. L. Hutchinson, and A. Y.
Cho, Quantum cascade laser, Science 264:553557 (1993).
12. V. Ryzhii, H. C. Liu, I. Khmyrova, and M. Ryzhii, Analysis of
integrated quantum-well infrared photodetector and light-
emitting diode for implementing pixelless imaging devices,
IEEE J. Quantum. Electron. 33:15271531 (1997).
13. C. Weisbuch and B. Vinter, Quantum Semiconductor Struc-
tures, Academic Press, San Diego, 1991.
HIGH-FIELD EFFECTS
K. HESS
University of Illinois
Electrons and holes contribute to the charge transport in
semiconductors, while ordinary charge transport in met-
als is restricted to electrons only. Another distinctive
feature of semiconductor transport is its inherent non-
linearity in high electric elds. It is difcult to generate
high electric elds in metals because of their high electri-
cal conductivity and the necessity of the presence of high
currents according to Ohms law. In semiconductors, high
elds can exist with current densities of the order of 10
5
A/cm
2
, or even much lower, depending on their conductance,
which at low temperatures may approach zero. Ryder [1]
1970 HIGH-FIELD EFFECTS
and Shockley [2] noticed that semiconductor electron
transport was extremely nonlinear in high electric elds.
Strong deviations from the proportionality of current den-
sity to elds were measured in germanium at room tem-
perature for elds as small as 1000 V/cm.
High-eld transport in semiconductors then became an
area of considerable research. It was found that the non-
linear behavior (the deviation from Ohms law) was due to
an elevation of the energy of the charge carriers caused by
the accelerating force of the electric eld. Interactions
with the lattice vibrations lower the energy of the charge
carriers as Joule heat is transferred to the crystal lattice.
However, electric elds always cause a nite rise of charge
carrier energy above the equilibrium energy (correspond-
ing to the temperature of the crystal lattice). This excess
energy can sometimes also be described by a temperature,
the temperature T
c
of the charge carriers, which is larger
than the temperature T
L
of the crystal lattice. One speaks
therefore of hot-carrier transport in semiconductors.
The rise in charge carrier energy changes the conduc-
tance for two reasons. For one, a higher energy gives rise
to significant changes in the interactions of the charge
carriers with crystal imperfections that form scattering
centers. The scattering rate inuences the conductance
directly. Increases or decreases of this rate lead to de-
creases or increases of the conductance, respectively. The
second reason for variations in conductance with charge
carrier energy arise from changes in the E(k) relation.
This function gives the connection between the energy E
and the wavevector k of the charge carriers, which corre-
sponds in classical mechanics to the energymomentum
relation. Since conductance is a sensitive function of
@Ek=@k, it changes with the energy of the electrons (or
with T
c
whenever a temperature of the carriers is well
dened).
The most complete existing theory of high-eld trans-
port in semiconductors involves the solution of a Boltzm-
ann-type equation [3]. The use of this equation can be
justied by invoking the dephasing of quantum coherence
over distances that correspond to the feature sizes of semi-
conductor structures and devices. Consequently, the
charge carriers do behave classically and can, in a way,
be understood from the principles of classical mechanics.
Quantum mechanics has then only a background role and
determines, for example, the effective mass or the velocity
of the electrons or gives justication for the existence of
holes and hole transport. Fermis golden rule of quantum
mechanics is used to calculate the scattering rates and
thus represents another quantum contribution. It is also
easy to include the Pauli principle in such an equation and
thus to approach some properties of a Fermi liquid. How-
ever, the assumption of weak perturbational interaction of
the particles, which is basic to Boltzmanns derivation,
must remain true, and indeed is a good approximation for
most of the important semiconductor materials.
The basic phenomena of high eld transport are non-
linear conductance [3]; changed responses to magnetic
elds [4]; changed high-frequency response, including
the dielectric function [5]; and changed connement in
potential wells or at heterojunctions between different
semiconductors [6].
In semiconductor devices, the high-eld effects are
based on the same phenomena. For example, in transis-
tors of the eld effect family, the electric elds can be as
high as 10
5
V/cm
2
, and the corresponding carrier temper-
atures may reach T
c
10
4
K [7]. Such temperatures cause
changes in the conductivity by orders of magnitude (most-
ly reductions) [3,5]. Remember that T
c
is the temperature
of the charge carriers and not of the crystal lattice, which
would melt at such temperatures. T
L
, the temperature
of the crystal lattice, can also be raised, but is usually
raised orders of magnitude less. This difference is possible
because the systems of electrons and crystal lattice are
only weakly coupled. Under certain circumstances the
conductance can also be raised by hot-electron phenome-
na, leading to speed advantages in devices (e.g. velocity
overshoot phenomena) [8]. Devices made of semicon-
ductor heterojunctions also exhibit various forms of hot-
electron transfer (e.g., real-space transfer) between the
different materials. These effects often give rise to a
device performance degradation, but also can be used
advantageously [6].
In the following, a general theory of high-eld trans-
port is developed. The next section describes high-eld
transport in bulk semiconductors (theory and experimen-
tal results), and nally an overview of major effects in de-
vices is given.
1. THE BOLTZMANN EQUATION FOR HIGH-FIELD
SEMICONDUCTOR TRANSPORT
The following is a topdown approach to the theory of
high-eld phenomena. The next section derives a modern
Boltzmann equation including the most important quan-
tum effects as derived from the energy band structure. The
relevance of the various terms of this equation to hot-elec-
tron phenomena are discussed and general ways to obtain
solutions are briefly reviewed and referenced. Subse-
quently a more phenomenological discussion and analyt-
ical approximation of several high-eld effects is given.
1.1. Derivation of a Modern Boltzmann Equation
The Boltzmann equation derived here is more general
than the original equation derived by Boltzmann. The
only cases of weakly interacting charge carriers in solids
that cannot be understood from this modied equation are
those that involve macroscopic feature sizes (e.g., device
boundaries) that are smaller than the quantum dephasing
length under the given conditions. This dephasing length
in silicon at room temperature is of the order of several
hundred angstroms, but can be much smaller in high elec-
tric elds, approaching 20 A
for electron energies of about

1.5 eV. This means that for conventional devices operating
with high elds, hot-electron transport can be understood
extremely well with the theory developed below even if the
feature sizes are below 100A
. Quantum effects such as

tunneling can often be added to this theory by the Bardeen
transfer Hamiltonian formalism [3,9].
The following derivation lacks complete mathematical
rigor but is valid under widely varying circumstances. For
example, we need not assume conservation of the number
HIGH-FIELD EFFECTS 1971
of particles, which is important in that electrons and holes
can annihilate each other or can be created by light. We
follow Boltzmann, but replace the velocity in seven-di-
mensional phase space by the wavevector k in the deni-
tion of a distribution function f(k, r, t), meaning as usual
the probability of nding charge carriers at k, r and at
time t in the volume elements dk, dr, dt. Consider then a
cube in r space and in k space as shown in Fig. 1.
We rst calculate how many electrons arrive from the
left and enter the cube through the left dydz plane, and
how many leave at the corresponding plane on the right,
all during a time period dt. Since the x-direction travel
distance of electrons with velocity v is v
x
dt, we have
Incoming : f k; r; t dkdy dz v
x
dt 1
Outgoing : f k; x dx; y; z; t dkdy dz v
x
dt 2
and the net particle gain is
v
x
f k; x dx; y; z; t f k; r; t dy dz dkdt
v
x
@f
@x
dx dy dz dkdt
v
.
rf dkdr dt
3
in three dimensions. Note that the velocity v here is
equal to the group velocity and is related to k by
v r
k
Ek=h% (3).
This balance depends on the velocity and k vector of the
electrons and therefore on how hot the electrons are. It
also describes diffusion, since concentration gradients en-
ter this term. The relation between diffusion and electron
heating is actually a complicated one and will be discussed
below phenomenologically.
In an analogous manner, we obtain the change of the
number of electrons at k in k space because of accelera-
tions. Replacing dx by dk
x
and so on as illustrated in
Fig. 1, and replacing dx/dt v
x
as used in Eq. (1) by
dk
x
/dt, one gets
dk
dt
.
r
k
f dkdr dt 4
where dh% =dt eF (3) and F is the electric eld. This
term represents the major energy supply to the electrons
and therefore the root cause of the heating and hot-elec-
tron effects.
There is still another possibility to change the number
of electrons with wavevector k at r. The electrons can be
scattered and change their wavevector from k to k
0
at a
given point r in real space. Figure 2 shows the two inn-
itesimal volumes in k space to illustrate the scattering
events. The outgoing (out of state k) electrons are
out
k
0
Sk; k
0
f k; r; t dkdr dt 5
The factor f (k, r, t) is necessary because an electron
has rst to be in the k state to be scattered out. In degen-
erate systems (Fermi statistics), an additional factor
1 f k
0
; r; t arises from the Pauli principle. The incoming
(into the k state) electrons are
in
k
0
Sk
0
;k f k
0
; r; t dkdr dt 6
Again, the Pauli principle will call for a factor 1 f k; r; t.
The in and out scatterings lead to a (more or less) random
distribution of the k vector. The energy gained from the
Figure 1. Cubes in (a) r space and (b) k space to illustrate the
balance of incoming (e.g., into the yz face) and outgoing (e.g., at
x dx) electrons (or change carriers in general). The changes in r
space are due to carrier velocity, those in k space due to acceler-
ation by a force.
dk
y
dk
y
dk
z
dk
z
dk
x
dk
x
k
x
k
y
k
z
k
k
Figure 2. Schematic of a scattering process. The charge carrier
scatters froma volume dk
x
dk
y
dk
z
at k to another such volume at k
0
.
eld is therefore distributed in k space and thus receives a
random temperaturelike component. The interaction with
lattice vibrations is inelastic and also changes the energy
(magnitude of k). This energy loss represents Joule heat
and determines, together with the force, the actual aver-
age electron energy (temperature).
The Boltzmann equation describes all of these heating
cooling dynamics and is obtained by balancing the particle
numbers and the change in f given by the net change of
incoming and outgoing particles. Therefore, we have
@f k; r; t
@t
v
.
rf k; r; t
1
h%
F
0
.
r
k
f k; r; t
k
0
f k
0
; r; tSk
0
; k f k; r; tSk; k
0
7
where F
0
is the force (eF for an electric eld F).
If we include the factors arising from the Pauli princi-
ple as discussed above, we arrive at
@f k; r; t
@t

1
h%
r
k
Ek
.
rf k; r; t
1
h%
F
0
.
r
k
f k; r; t
k
0
ff k
0
; r; t1 f k; r; tSk
0
; k
f k; r; t1 f k
0
; r; tSk; k
0
g
8
This equation is very general and includes automatically,
for example, effects of the energy band structure as de-
scribed by E(k). Of course, it describes transport in one
band only [3]. The functional form of this band, however,
can be arbitrary. Effects of strain in the solid, for example,
need only be included in the band structure and then are
automatically taken into account in the distribution func-
tion f once the Boltzmann equation is solved. From the
distribution function, one can in turn obtain macroscopic
quantities such as the electronic current in the usual pre-
scribed way.
To summarize, all terms of this equation have special
significance for hot-electron effects. The second term on
the right-hand side (RHS) of Eq. (8) represents the driving
force, the electric eld F, and signies the heating of the
electrons. The third term on the RHS describes the scat-
tering and how momentum and energy are distributed.
This term signies the inuence of hot electrons on the
classical conductivity. Electrons are scattered out of a giv-
en range of the wavevector k or scattered into that range
from all other possible values of k, denoted by k
0
. This
scattering term makes the Boltzmann equation an int-
egrodifferential equation. Therefore, to solve it, one needs
a tenfold integral over time, space r, and k, as well as k
0
.
The tenfold numerical integration can probably be done
best by Monte Carlo methods, although approximations in
lower dimensions may permit the use of computationally
more efcient numerics. The rst term on the RHS repre-
sents space-dependent effects such as diffusion and indi-
cates that the hot-electron diffusion not only is dependent
on the heating of the electrons but in turn inuences the
heating. This term is relevant to the question of how the
force heats the electrons. For example, a conning force
that does not give rise to a current (as, e.g., at a potential
minimum) may not heat the electrons, since for this case
the accelerating drift (second term on RHS) and diffusion
(rst term on RHS) can cancel exactly [3,10,11].
The energy band structure enters the Boltzmann equa-
tion through the E(k) relation in the rst term on the
RHS. It also enters indirectly through the sum over k,
since this sum includes the density of states. Finally, if one
wants to calculate a current, one needs to integrate the
product of velocity and distribution function over all k
space, where the velocity is given by v r
k
Ek=h%.
A full solution of the Boltzmann equation as derived
above does describe all hot-electron phenomena currently
known in semiconductors. It can be and has been achieved
numerically in bulk semiconductors [12] and in devices
[13]. We refer the reader to these references and to soft-
ware and explanations available on Websites [14]. Here
we continue with discussions of important experimental
results in terms of approximate concepts and solutions.
1.2. Approximate Solutions of the Boltzmann Equation and
Hot Electrons
Since the Boltzmann equation is an integrodifferential
equation, precise and explicit solutions can be found only
under very special circumstances. The best-known exam-
ple is the time-independent solution for homogeneous sys-
tems (no space-dependent terms) in the relaxation-time
approximation. Under the assumption of weak forces
(electric elds), one can write the distribution function
as a sum of a function f
0
that is even in the wavevector k
and an odd function f
1
. Assuming that the Pauli terms are
negligible, as they are for not too high carrier concentra-
tions, the whole collision integral of Eq. (8) reduces to
Collisionintegral f
1
=t
tot
9
with
1
t
tot
k
0
Sk; k
0
10
and f
0
is equal to the equilibrium Boltzman distribution
given by
f
0
expE
F
=kT
L
expE=kT
L
11
The odd part of the distribution function that determines
the electric current is then
f
1
t
tot
F
0
h%
.
r
k
e
E
F
E=kT
12
where E
F
is the Fermi level. The electric current density j
is then obtained from
j
e
4p
3
_
vf
1
dk 13
The high eld or hot-electron term is neglected in this
approach, which describes only low elds and ohmic
behavior. The reason is the approximation of f
0
by the
equilibrium distribution. To allow for high electric elds,
one needs to solve the full Boltzmann equation or at least
the coupled equations resulting for both f
0
and f
1
in the
relaxation-time approximation. As described above, the
force accelerates the electrons (holes), and the scattering
randomizes, thus causing f
0
to contain more energetic
electrons and therefore changing its form away from the
equilibrium.
As mentioned, the complete solution of Eq. (8) has been
achieved by so-called Monte Carlo methods, which are re-
lated to the Monte Carlo integration known from numer-
ical mathematics [3,5]. These solutions have been
described at length in the literature and form a vast eld
[5,1214].
Approximate solutions of the Boltzmann equation for
high electric elds are also well known. Of particular im-
portance is the electron temperature approximation,
which is described in the next section together with typ-
ical experimental results for high eld transport in bulk
(homogeneous) semiconductors.
2. HOT-ELECTRON EFFECTS IN SEMICONDUCTORS AND
APPROXIMATE THEORY
2.1. Electron Temperature and Scattering Rate
A rst glimpse of how important hot electrons would be
in semiconductor transport was given by the work of
Shockley (1951) [2] and Ryder (1953) [1], who found a sat-
uration of the electron current in germanium at electric
elds around 1000 V/cm as shown in Fig. 3. This current
saturation is basic to hot-electron phenomena and has
been shown to arise from the increase of average
electron energy hEi. Under certain assumptions that are
approximately satised in n-type silicon and germa-
nium for intermediate electric elds [3], one can repre-
sent the average energy by a temperature T
c
, which,
for not too high electron densities, can be approximated
by a Boltzmann-type formula
hEi
3
2
k
B
T
c
14
where k
B
is Boltzmanns constant. The actual solution of
the Boltzmann equation to arrive at the electron temper-
ature concept is involved, and the reader is referred to
Ref. 3 for detailed information.
An approximate formula for T
c
is [3]
T
c
% T
L
1
F
F
c
_ _
2
_ _
15
F
c
is a critical electric eld that is around 10
4
V/cm for
silicon at room temperature. The carrier temperature T
c
can therefore become extremely high. For example, at the
eld F2 10
4
V/cm, which can easily be reached and in-
deed is routinely reached in modern transistors, we have
T
c
1500K for T
L
300K. Such temperatures have in-
deed been measured by various methods that can sense
the electron energy inside the semiconductor [15]. An out-
side touch does not reveal a temperature increase of the
electrons because of the large workfunction that the elec-
trons would need to overcome in order to propagate out of
the semiconductor [16]. Equation (11) is not valid for elec-
tric elds much higher than 3 10
4
V/cm for silicon be-
cause of band structure effects such as non-parabolicity
[5]. For very high elds, a full-band Monte Carlo approach
is again necessary to calculate the average energy and
distribution functions of the electrons reliably.
The rise of the electron temperature (or energy) causes
a change in the scattering rate that enters the mobility m
of the charge carriers and therefore the conductivity s
enm (n being the carrier concentration). The mobility de-
rives from the relaxation time as dened in Eq. (10) if the
scattering in randomizing (independent of the wavevec-
tor). Otherwise t
tot
needs to be replaced by the so-called
momentum relaxation time [3]. This relaxation time usu-
ally exhibits a significant energy dependence. For phonon
scattering it decreases typically with increasing energy,
while for scattering by weakly screened Coulomb charges
it increases. The mobility m is proportional to a weighted
energy average of the relaxation time, and the conductiv-
ity therefore increases or decreases proportionally.
A typical dependence of the mobility on the electron
temperature is then (for the case of scattering by
phonons) [3]
m % m
0
T
L
T
c
_ _
1=2
16
where m
0
is the mobility at zero electric eld, that is, for T
L
T
c
. Using, then, the equation for the current density of a
homogeneous semiconductor, j enm, one gets from Eqs.
(16) and (15)
j enm
0
F
1F=F
c
2
_ _
1=2
17
Figure 3. Current density (or drift velocity) versus electric eld
in bulk n-type germanium for three temperatures (as indicated).
Notice the saturation at electric elds above 10
3
V/cm. (After
Ref. 1.) Reprinted with permission, r 1953 by the American
Physical Society.
This equation shows immediately the essential feature
of current saturation (nonlinear transport) at high electric
elds and is quite general. For example, the square root in
Eq. (16) derives from the density of states. Equation (17) is
valid even for a more general density of states; it can be
proved easily for any density of states that is proportional
to E
p
for any p40. The dependence of the electron phonon
scattering rate on the density of states is illustrated in Fig.
9. Note, however, that in devices other scattering mecha-
nisms such as surface roughness scattering are of great
importance. Some of these mechanisms are reviewed in
Ref. 16.
2.2. Effects of Band Structure and the Gunn Effect
The derivation of Eq. (17) contains only changes in the
scattering rate with carrier temperature. As mentioned
above, however, the band structure inuences not only the
scattering rate (via the density of states) but also the ve-
locity v of the electrons, since in the one-band approxima-
tion we have [3]
v
1
h%
r
k
Ek 18
For the typical band structure of some IIIV compounds,
this leads to pronounced effects in the currentvoltage
characteristics. For GaAs, the effective mass is small at
low energies (m
0.067m
0
) and Ek h%
2
k
2
=2m
. Corre-
spondingly, the velocity of electrons becomes very high at
moderate energies. However, the band structure changes
drastically only 0.3eV above the conduction band edge,
exhibiting there additional minima with much higher ef-
fective mass (close to the germanium type conduction),
and above 0.5 eV, even higher effective masses (close to the
silicon conduction mass). In simple terms, GaAs contain
three highways, each one becoming available as the en-
ergy increases and each one exhibiting higher mass. If the
electrons are heated by electric elds, they initially are on
the fastest highway, exhibiting a high mobility and con-
ductance. At higher elds they are heated enough to
transfer to the germanium like conduction band minimum
(highway 2) and at still higher elds to the siliconlike
minimum (highway 3). This decreases their speed so dras-
tically that GaAs actually shows a range of negative dif-
ferential resistance, that is, the current drops as the
electric eld is increased. This phenomenon leads to the
Gunn effect, which manifests itself by high-frequency cur-
rent oscillations, since the situation of negative differen-
tial resistance is not stable [7]. These phenomena have
received considerable attention, and a large framework of
research exists [7,18].
The currenteld characteristic of GaAs is shown in
Fig. 4 and compared with the characteristics of silicon
(both for electron transport). It is evident that the low-
eld current behavior of GaAs transforms into silicon like
behavior at high electric elds, with a region of negative
differential resistance in between. The reader is referred
to the literature [3,5,7] for more information.
Aword of caution should be added here. There is a lack
of direct experimental verication of band structure and
related semiconductor parameters for high-eld trans-
port. Optical measurements and femtosecond spectrosco-
py [19] do give detailed information on the density of
states.
2.3. Real-Space Transfer
The negative differential resistance of GaAs and the Gunn
effect are determined mostly by the structure of the E(k)
relation, that is, by effects in k space. The terms of the
Boltzmann equation signifying real-space operations have
not been discussed yet. From the discussions following
Eq. (8) it is clear, however, that real-space effects compli-
cate hot-electron transport significantly. The prime reason
for this complication is the real-space transfer effect [6],
which is described here.
The transfer of electrons between two different solids is
known from Bethes thermionic emission theory [3]. This
type of transport includes only electric elds perpendicu-
lar to the different layers of semiconductors. However,
electric elds parallel to semiconductors can energize the
charge carriers (hot-electron effect) and lead to a redistri-
bution of them in the different layers that depends sensi-
tively on the difference of the conduction band edges of the
various materials. (Readers not used to solid-state con-
cepts should think of the conduction band edge as the
minimum kinetic energy of conduction electrons, which is
different in different materials, the electrons thus having
different potential energy in each material.) The effect of
redistribution of electrons or holes due to electric elds
parallel to different layers of semiconductors is called real-
space transfer (RST) and is shown schematically in Fig. 5.
This transfer of electrons (heated by parallel elds) over
barriers is more complicated and more difcult to under-
stand than other effects basic to nonlinear semiconductor
transport and device operation. The reason is that RST
can be visualized only the combination of two concepts
related to the energy distribution of electrons. The rst
concept is that of quasi-Fermi levels [3], and the second is
the concept of a charge carrier temperature T
c
as already
Figure 4. Schematic of current density versus electric eld for
homogeneous (bulk) GaAs and silicon. Notice the negative differ-
ential resistance of GaAs (which leads to instabilities [7] and in-
homogeneities of the carrier density). Also note that at very high
elds the GaAs and silicon curves approach each other. Depend-
ing on temperature and electric eld, they can even cross. How-
ever, they are always close at very high elds.
discussed above. For RST problems, both concepts matter,
and both the carrier temperature and the quasi-Fermi
levels are a function of space coordinate and time.
Imagine, for example, electrons residing in a layer of
high-mobility GaAs neighboring, on either side, two layers
of low-mobility AlAs (Fig. 5). The GaAs equilibrium dis-
tribution function f
0
is
f
0
expE=kT
L
19
while in the AlGaAs we have
f
0
/ exp
DE
c
E
kT
L
_ _
20
Here the energy is measured from the GaAs conduction
band edge, and rE
c
is the band edge discontinuity be-
tween AlAs and GaAs.
If now the electrons are heated by an external eld
parallel to the layers, we have to replace T
L
in Eqs. (19)
and (20) by a space-dependent carrier temperature T
c
. It is
clear that for T
c
-Nthe difference between the AlAs and
the GaAs population densities vanishes. In other words,
the electrons will spread out into the AlAs layers. This also
means that even perpendicular to the layers (z direction) a
constant Fermi level cannot exist, and E
F
has to be re-
placed by the quasi-Fermi level E
QF
(z) as the density of
electrons becomes a function of T
c
(z). This is unusual,
since commonly the quasi-Fermi levels differ only in the
direction of the applied external voltage V
ext
(by the
amount eV
ext
). In the present case, a voltage is applied
parallel to the layers, the electrons redistribute them-
selves perpendicularly to the layers, and a eld (and volt-
age perpendicular to the layers) develops owing to the
carrier redistribution. Basic to the calculation of this pro-
cess are the thermionic emission currents [3] of hot elec-
trons from one layer to the other. Since the external
voltage is applied parallel to the layers, we have in steady
state a precise balance of currents owing from left to
right and right to left, which determines the z-dependent
carrier population.
A complication of the theory is presented by the neces-
sity (in most cases) of having to solve Poissons equation as
charge is transferred. For typical parameters of the GaAs
AlAs material system and electric elds of the order of 10
3
to 10
4
V/cm parallel to the layers, one obtains time
constants of the order of picoseconds for the transfer,
which gives the RST effect importance for device applica-
tions (RST transistors as developed by [6]).
The real-space transfer effect is also of general impor-
tance in all situations when electrons are conned in po-
tential wells and parallel elds are applied (and accelerate
the charge carriers), even if the electrons do not propagate
out of the wells but merely redistribute themselves within
each well. This is of relevance for the understanding of the
inuence of transverse elds (such as the gate eld) in a
transistor [10]. The RST effect and the spreading of the
electrons are then determined by the transverse eld. The
quantum analog of this classical picture is the redistribu-
tion of hot electrons in the different quantized subbands of
a quantum well.
2.4. Time Dependences, Velocity Overshoot, and
Ballistic Transport
As complex as the considerations of nonlinear transport in
r space and k space are already, time dependences add to
the richness of hot-electron effects. All the discussions
above were essentially valid for the steady state only. On
short timescales, however, the mode of transport changes
its type because of transitions from the ballistic to the
overshoot regime and to diffusive transport.
Diffusive transport is the well-known mode for which
the mobility is proportional to the average of t
tot
. This
proportionality implies the validity of the Boltzmann
equation, which, as discussed at the beginning, involves
the dephasing of the wave function. For times much short-
er than t
tot
, the transport is essentially ballistic, which
means that in crystalline semiconductors the accelera-
tions are described by the equation
h%
dk
dt
eF 21
where F is the electric eld and h% k is the crystal momen-
tum. This equation becomes invalid at extremely short
times, as described in Ref. 20, but is applicable under most
circumstances. Neglecting the basic thermal motion of the
electrons (due to T
L
), this means that all electrons are ac-
celerated in the direction of the force and gain speed ac-
cording to Eqs. (18) and (21). Since all electrons move in
the same direction, the average velocity is very large com-
pared to the thermal case where electrons move in all di-
rections and have just a small average drift velocity on top
of the thermal motion. In high electric elds, and with
high values of T
c
, the thermal velocity (pointing randomly
in all directions) can be as high as 10
8
cm/s, while the sat-
urated drift velocity is (in silicon) a factor of 10 below this
value. A visualization of these facts is given in Fig. 6.
The range between the high-velocity ballistic transport
and the low-velocity diffusive transport is called the range
of velocity overshoot. Imagine transport in a semiconduc-
tor switched on at time t
0
0 by application of a high eld
F. The electrons are then accelerated ballistically for about
100 fs and may reach a velocity well above 10
7
cm/s (the
value of the saturated velocity in silicon and other impor-
tant semiconductors). Then, as time goes on, scattering
Figure 5. Electrons in a material with lower conduction band
edge (e.g., GaAs) neighboring two layers with higher conduction
band edge (e.g., AlAs) accelerated by a force F
0
. The electrons gain
energy and are scattered and then transfer out of the central ma-
terial layer, thus exhibiting real-space transfer.
events randomize the velocity to all directions, which
leads, typically after a picosecond or so, to the saturated
average velocity. At the times in between, the velocity is
higher and exhibits the overshoot. In GaAs these effects
are accentuated by the changes in effective mass at high
energy, and the overshoot can be considerable [8]. This is
shown in Fig. 7.
A similar effect is also achieved for transport in short
semiconductor sections sandwiched in between contacts.
As soon as the electrons enter the semiconductor from the
contact, they are accelerated by high electric elds into the
overshoot or even the ballistic regime and leave the semi-
conductor, entering the second contact, before scattering
can randomize their motion. Thus the velocity in short
semiconductor diodes can overshoot the saturated value
(now for all times), which in some devices gives a speed
advantage [22].
2.5. Changes of Carrier Concentration and Hot Electrons:
Impact Ionization
Up to now, all the conductivity changes we have discussed
have arisen from changes in the carrier velocity and a re-
distribution of charge carriers in energy. The total number
of electrons or holes was not affected by hot-electron ef-
fects. There exists, however, an important phenomenon
that causes considerable changes of conductance because
of changes in the total number of charge carriers. This is
the occurrence of impact ionization, the exact inverse of
the Auger effect [3]. A heated electron in the conduction
band (the same can be argued for holes) gains energy from
the applied electric eld and the collides with an electron
in the valence band, lifting this electron up to the conduc-
tion band, with the net result of two conduction electrons
and a hole. The primary electron and the secondary elec-
tron plus hole all now contribute to the conduction, and
the electronic current density j therefore increases with
time according to the equation
dj
dt
a
t
j 22
This increase of current with time also gives rise to a spa-
tial increase of the current as
rj a
r
j 23
The coefcient a
r
is in general a matrix. However, for a
constant electric eld impact ionization is mostly isotropic
and a
r
is a scalar [12,23].
We will concentrate, in the following on the theory of a
t
.
The theory of a
r
proceeds very similarly, and one can
almost always use a
t
v
d
a
r
, where v
d
is the average (drift)
velocity of the electrons. The multiplication of electrons
(and holes) with time depends physically on two quanti-
ties. One is the actual ionization probability, that is, the
probability for an electron with given energy to create an
Figure 6. Schematic of the development of average drift velocity
versus time, assuming that a high electric eld F is suddenly ap-
plied at t 0. The electrons are accelerated to high velocities and
high average drift velocities. Their random velocity (indicated by
arrows pointing in all directions) is at rst small but is steadily
increased because of randomizing scattering. At a certain point
the average drift velocity approaches a maximum. From here on
the random component of the velocity increases because of the
very strong scattering at the high energies that the charge car-
riers now have, and the average drift velocity decreases. Finally,
steady state is reached, where the scatterings and accelerations
balance each other. The large random arrows indicate then a high
electron temperature, and the drift velocity is saturated
(at 10
7
cm/s in silicon).
Figure 7. Overshoot of electrons injected with certain wavevec-
tors k
x
in the x direction and with certain energies E
0
into GaAs. A
eld F subsequently accelerates them and leads to velocity over-
shoot (19).
electronhole pair. This probability per unit time is called
the impact ionization rate R
I
. The second quantity is the
actual probability of having such an electron at a given
energy, which is, of course, given by the distribution func-
tion. a
t
is then the average product of these two probabil-
ities.
a
t
_
1
1
dkR
I
f
_
1
1
dkf
24
The calculation of R
I
involves the quantum mechanics of
three particles (two electrons, one hole) in two different
bands (conduction and valence) with different (Bloch)
wavefunctions and E(k) relations. Even using the approx-
imate golden rule of time-dependent perturbation theory
involves then a difcult numerical simulation with multi-
ple integrations. Most of the theories of the past are there-
fore oversimplied and of very limited use, including the
formula derived by Keldysh [24]. The only theory in rea-
sonably close agreement with experiments is the Monte
Carlo integration of the golden rule formulae by Kane
[25], the result of which is shown in Fig. 8.
An even greater numerical problem is the calculation of
the distribution function entering Eq. (24). The reason is
that R
I
starts to become appreciably large only at rela-
tively high energies, as shown in Fig. 8. Typically, in sil-
icon, ionization becomes important only for electron
energies of 3 eV above the conduction band edge. If this
were the average electron energy, it would correspond to a
temperature of 24,000K, which is seldom reached before
catastrophic damage occurs. Therefore it is the high-energy
tail of the distribution function that matters for impact
ionization. This high-energy tail depends sensitively on
the band structure and also on spatial and temporal
changes of the electric eld and can be reliably obtained
only from a full-band solution of the Boltzmann equation
such as the full-band Monte Carlo [12]. An example of the
distribution at high energies is given in Fig. 9 for a con-
stant electric eld. Figure 9 also shows the phonon scat-
tering rate in silicon at room temperature as a function of
conduction band energies. Not that this rate becomes of
the order of 10
14
s
1
at energies above 1.5eV.
A discussion of all of these complexities and the corre-
sponding results for a
r
are given in Refs. 15 and 27. The
best results are obtained for theories of R
I
a la Kane and
Figure 8. Impact ionization rate a
t
(multiplied by h% ) as a function
of conduction electron energy according to the theory of Kane [25].
(Reprinted with permission, r 1967 by the American Physical
Society.)
Figure 9. Electron energy distribution as a function of electron
energy (in the conduction band), according to Ref. 24, for various
constant electric elds. Also shown is the electronphonon scat-
tering.
Figure 10. Theoretical electron impact ionization coefcients a
t
versus 1/F for constant electric elds F and various materials af-
ter Bude and Hess [28]. Solid lineGaAs; dashed lineInAs;
dotteddashed lineInP; dotted lineGa
0.43
In
0.57
As. The elec-
tron ionization coefcient for silicon is roughly a factor of 2 above
the coefcient of GaAs. The hole ionization coefcients of GaAs
and InP are close to the electron ionization coefcients, while the
hole ionization coefcient for silicon is much below the curve for
electrons (factors of 410). (Reprinted with permission, r1992 by
the American Physical Society.)
for distributions from full-band Monte Carlo integrations
of the Boltzmann equation (25). Typical theoretical results
of a
r
are shown in Fig. 10.
The presence of layers of different semiconductors, as
occurs in quantum-well structures and superlattices, adds
another dimension to the science and engineering of im-
pact ionization and has been discussed extensively in the
literature. Examples are given in Refs. 29 and 30.
A simplied approach of great usefulness and para-
metrized expressions for a
r
have been derived by Baraff
and others using his theory [7,31]. These are excellent for
estimates and for an appreciation of contributory physical
parameters. Note, however, that all the simplied analyt-
ical expressions given in most textbooks for the threshold,
phonon scattering mean free path, and other para-
meters are quantitatively incorrect and cannot replace
the integration of Boltzmanns equation, including the
band structure.
3. HOT ELECTRONS IN DEVICES
The special status of hot electrons in semiconductor devices
arises from the time and space dependences of electric eld
and carrier concentrations. These dependences require
special care in the theory of average velocity (overshoot,
real-space transfer, and transport including abrupt inter-
faces) and impact ionization. The nonlocality of these ef-
fects deserves special attention [32]. Consistency with
Gauss law also needs to be satised. In fact, numerical
solutions of Poissons equation are a xed part of any pre-
dictive device simulation. The eld of hot electrons in de-
vices is therefore a vast one and cannot be reviewed within
this limited space. For a general appreciation the reader is
referred to discussions of Gunn devices, IMPATT diodes,
hot-electron diodes and transistors, real-space transfer
devices, and avalanche photodiodes in Ref. 18. All of these
devices are based on hot-electron effects. Many of the most
important devices, such as metaloxidesemiconductor
eld-effect transistors (MOSFETs), involve hot electrons
in their operation. Often, hot electrons are felt to cause
great disadvantage because they reduce speed (current
and velocity saturation), lead to degradation and aging of
devices [33], and can cause various instabilities (e.g.,
through negative differential resistance). However, there
are two counts in favor of hot electrons in devices that
have led to the continual involvement of hot electrons in
chip technology. For one, hot electrons lead to a large scat-
tering rate by phonons as discussed above. This large rate
gives rise to dephasing of the wavefunction on the length
scale of 0.003 mm. Therefore it is possible to reach feature
sizes down to 0.1 mm without major transitions from clas-
sical to quantum transport. The second point in favor of
hot electrons arises from the need for aggressive designs,
particularly with respect to switching speed. One needs to
use the highest possible current densities, and this means in
semiconductors also high electric elds. Frequent predic-
tions that hot-electron effects will be scaled away soon
have therefore often not come true. Since hot-electron eff-
ects are important in so many devices and of such variable
consequences, we will discuss here only major effects that
appear in many variations.
3.1. Hot Electrons in Field-Effect Devices
Field-effect transistors exhibit a large number of hot-elec-
tron effects. It has been known since the 1960s [34] that
velocity saturation is important in these devices. The sat-
uration was at rst only encountered close to the drain, in
the so-called pinchoff region [7]. As the device sizes de-
creased, the velocity saturation spread all over the chan-
nel, at least for the highest drain voltages used [7]. This
effect reduces some of the gures of merit of the transis-
tors (as, e.g., the transconductance) and was therefore
seen as an unwelcome side effect; it was simulated by use
of Eqs. (16) and (15) or similar equations but with space-
dependent electric elds. This local dependence on a vary-
ing electric eld is, of course, valid only if the eld varies
so slowly that the transport is always diffusive without the
nonlocal velocity overshoot or ballistic components. These
latter effects become important for transistor channel
length of the order of 0.1mm [20]. Then the overshoot ef-
fects counteract and undo part of the velocity saturation
and generally contribute to higher device speed. This has
been proved in MOSFETs, as discussed in Ref. 20.
The documentation of overshoot effects in silicon-based
devices is a nontrivial task, since they always appear in
connection with velocity degradation in other regions of
the devices, and they are small (around a factor of 2 for T
L
300K). Overshoot effects are larger in IIIV compound
eld-effect transistors such as metalsemiconductor tran-
sistors (MESFETs). Ordinary MESFETs exhibit much
scattering in the conducting channel due to the charged
donors or acceptors, in addition to the always present
scattering by the polar optical phonons [3,7]. The over-
shoot effects are therefore more pronounced in modula-
tion-doped eld effect transistors (MODFET), which
contain the dopants in a layer of a different semiconduc-
tor (e.g., AlGaAs) neighboring the channel (e.g., GaAs) as
described in Ref. 35. These transistors exhibit, therefore, a
significant speed advantage over MESFETs, as shown by
numerous works [36,37]. Typical values of the overshoot in
such devices, as deduced from Monte Carlo simulations,
correspond to those shown in Fig. 7. We note that real-
space transfer can, of course, be also of importance in
these devices and may reduce some of their advantages.
Impact ionization is an important limiting factor in
eld-effect transistors, since its presence usually will dis-
turb device operation. The theory of impact ionization in
devices proceeds along the lines discussed above. Howev-
er, nonlocal effects are of great significance because of the
presence of rapidly varying electric elds in the conduct-
ing channels of eld-effect devices. One then needs to allow
for dead spaces of the ionization [32,38].
As discussed above, the probability of impact ionization
is very small for electron energies below a certain energy
in the conduction band. Of course, the minimum energy
the electron needs is the energy of the gap E
G
. Even above
this energy, the threshold for significant ionization is often
not reached for energies of 2 or 3 times E
G
, as can be seen
from Kanes results in Fig. 8. This energy needs to be
reached to start significant ionization.
With a space-dependent electric eld F(z) in the z di-
rection, the electron needs to traverse a certain distance d
to reach effective threshold. Typically d can be obtained
from an equation such as
_
d
0
Fz dz cE
G
25
where c is a constant determining the effective threshold
(normally of the order of 2). Over such a distance, ioniza-
tion cannot occur, even if the electric eld is very high.
Ionization in very short eld-effect transistors with highly
peaked electric elds is therefore smaller than one would
expect from an integration of the ionization probability
over all elds not counting the dead space of length d. This
explains the fact that ionization is not a function of the
local electric eld alone (nonlocality).
Anisotropies of impact ionization, which have not been
found for constant electric elds [23], can occur when bal-
listic acceleration over short distances is important, main-
ly because of anisotropies of the effective threshold [39].
These further complicate the simulation of impact ioniza-
tion in devices. It is the conviction of this author that a
quantitative understanding of impact ionization in devices
is possible only with a full-band solution of the Boltzmann
equation consistent with the solution of Poissons equa-
tion, as can be done with various simulation tools [13,14].
Some of the controversies in the literature can be tracked
to oversimplied simulation.
The hot-electron effects discussed above are reversible
in the sense that after turning off the electric elds, the
hot electrons cool down to the ambient temperature within
picoseconds without any structural changes of the crystal
lattice. However, hot electrons can also cause structural
changes. A particularly well-known hot-electron degrada-
tion occurs at the MOSFET interface between the silicon
and the silicon dioxide [33]. This damage is linked to the
breaking of siliconhydrogen bonds that are always pre-
sent at this interface. A clear proof was given by damage
measurements involving the isotope deuterium [40].
When the siliconhydrogen complex was replaced by sili-
condeuterium, a much-reduced hot-electron degradation
of MOSFETs was found, which has the benecial effect of
increasing the device lifetime [41]. A large framework of
experimental and theoretical contributions to this area
exists in the literature and is reviewed in Ref. 42.
3.2. Hot Electrons in Quantum-Well Laser Diodes
Hot electrons in quantum-well laser diodes have an origin
and consequences different from what they have in eld-
effect devices. The electrons here are not heated by the
electric eld but by other electrons propagating into the
quantum well and exhibiting suddenly high kinetic ener-
gy as shown in Fig. 11. These electrons, originating from
outside the well, can now transfer their energy in essen-
tially two ways to the electrons in the quantum well. For
one, they can transfer the energy through direct electron
electron interaction (electronhole interactions can also be
important in laser diodes because of the presence of both
electrons and holes with approximately equal density).
The incoming electron beam therefore heats the quantum-
well electrons. Another pathway of energy transfer is by
polar optical phonons. Electrons at high energies in the
quantum wells emit polar optical phonons (within about
10
13
s). These phonons cannot propagate out of the quan-
tum well and decay relatively slowly (typically within 3
10 ps). The phonons therefore accumulate, giving rise to a
nonequilibrium (heated) phonon distribution. The quan-
tum-well electrons at low energy can then in turn absorb
phonons and heat up themselves. This gives rise to a non-
equilibrium electron temperature T
c
.
A completely consistent calculation of these effects has
been made [43] and shows that the modulation of electron
Figure 12. Modulation response of quantum-well laser diodes for
various driving currents corresponding to the power indicated: (a)
theory without hot-electron effectsthe agreement with experi-
ments is weak, particularly at higher power levels; (b) theory in-
cluding hot-electron effects (temperatures indicated above curves)
shows excellent agreement with experiments [38].
Figure 11. Electron propagating over a semiconductor hetero-
junction and acquiring significant kinetic energy (in the GaAs).
density in quantum-well laser diodes is always accompa-
nied by a modulation of the electron temperature and
therefore inuences the laser performance and modula-
tion response sensitively through very small temperature
rises. While led-effect devices operate at electron tem-
peratures of several thousand kelvins caused by the heat-
ing of the electric eld, laser diodes cease to operate for
dynamic electron heating of a few kelvins, as shown in
Fig. 12, which depicts the modulation response of a typical
semiconductor laser diode. The static heating is also im-
portant in lasers and degrades their performance. For the
static case electron temperatures around 100K above
room temperature may be toleratedstill much less
than the T
c
values reached in eld-effect devices.
BIBLIOGRAPHY
1. E. J. Ryder, Phys. Rev. 90:766769 (1953).
2. W. Shockley, Bell Syst. Tech. J. 30:9901040 (1951); Solid
State Electron. 2:3567 (1961).
3. K. Hess, Advanced Theory Semiconductory Devices, Prentice-
Hall, Englewood Cliffs, NJ, 1988.
4. G. Bauer, in D. K. Ferry, J. R. Barker, and C. Jacoboni, eds.,
Physics of Nonlinear Transport in Semiconductors, Plenum,
New York, 1979, pp. 175224.
5. D. K. Ferry, Semiconductors, Macmillan, New York, 1991, pp.
502557.
6. Z. S. Gribnikov, K. Hess, and G. A. Kosinovsky, J. Appl. Phys.
77:13371373 (1995).
7. S. M. Sze, Physics of Semiconductor Devices, Wiley, New York,
1981.
8. J. G. Ruch, IEEE Trans. Electron. Devices ED-19:652659
(1972).
9. P. D. Yoder and K. Hess, in N. Balkan, ed., Negative Differen-
tial Resistance and Instabilities in 2-D Semiconductors, Ple-
num, New York, 1993, pp. 99107.
10. K. Brennan and K. Hess, IEEE Electron. Device Lett. 7:8688
(1986).
11. K. Hess, in N. Balkan, ed., Hot Electron Physics and Devices,
Oxford Univ. Press, 1997, pp. 1333.
12. H. Shichijo and K. Hess, Phys. Rev. B 23:41974207 (1981).
13. S. E. Laux and M. V. Fischetti, in Monte Carlo Device Sim-
ulation: Full Band and Beyond, Kluwer Academic, Boston,
1991, pp. 126.
14. DEGAS, http://www.ise.ch.
15. Z. Selmi et al., IEDM Tech. Digest, 1995, pp. 293296.
16. D. K. Ferry, K. Hess, and P. Vogl, in N. G. Einspruch, ed.,
VLSI Electronics, Vol. 2, Academic Press, New York, 1981, pp.
67103.
17. H. M. J. Boots et al., Appl. Phys. Lett. 57:24462448 (1990).
18. K. K. Ng, Complete Guide to Semiconductor Devices, Mc-
Graw-Hill, New York, 1995.
19. C. J. Stanton, D. W. Bailey, and K. Hess, IEEE J. Quantum
Electron. 24:16141627 (1988).
20. G. J. Iafrate and K. Hess, Proc. IEEE 519532 (1988).
21. J. Y. Tang and K. Hess, IEEE Trans. Electron. Devices ED-
29:19061910 (1982).
22. M. S. Shur and L. F. Eastman, IEEE Trans. Electron. Devices
ED-26:16771683 (1979).
23. G. E. Stillman, V. M. Robbins, and K. Hess, Proc. 4th Int. Conf.
Hot Electrons Semiconductors, Inst. Phys. 134B/C:241246
(1985).
24. L. V. Keldysh, Soviet Phys. JETP 21:1135 (1965).
25. E. O. Kane, Phys. Rev. B 159:624 (1967).
26. P. D. Yoder and K. Hess, Semiconduct. Sci. Technol. 9:852
854 (1994).
27. M. V. Fischetti, S. E. Laux, and E. Crabbe, in K. Hess, J. P.
Leburton, and U. Ravaioliu, eds., Hot Carriers in Semicon-
ductors, Plenum, 1996, New York, pp. 475480.
28. J. D. Bude and K. Hess, Impact ionization, J. Appl. Phys.
72:35543561 (1992); J. D. Bude, K. Hess, and G. J. Iafrate,
Semiconduct. Sci. Technol. 7:506508 (1992); N. Sano, M.
Tomizawa, and A. Yoshi, in K. Hess, J. P. Leburton, and U.
Ravaioli, eds., Hot Carriers in Semiconductors, Plenum, New
York, 1996, pp. 337342.
29. K. Kim, K. Hess, and F. Capasso, Appl. Phys. Lett. 51:508510
(1987).
30. F. Capasso, Science 235:172 (1987).
31. G. A. Baraff, Phys. Rev. 128:25072517 (1962).
32. K. Kim, K. Hess, and F. Capasso, Appl. Phys. Lett. 51:508510
(1987).
33. C. Hu et al., IEEE Trans. Electron. Devices ED-32:375385
(1985).
34. J. A. Cooper and D. F. Nelson, J. Appl. Phys. 54:1445 (1983).
35. T. Mimura, K. Taniguchi, and C. Hamaguchi, Semiconduct.
Sci. Technol. 7:379381 (1992).
36. H. Morkoc and P. M. Solomon, Modulation-doped eld-effect
transistors, IEEE Spectrum 21(2):2835 (1984).
37. I. C. Kizilyalli et al., in H. Daemblees, ed., Modulation-Doped
Field-Effect Transistors, IEEE Press, Piscataway, NJ, 1990,
pp. 350355.
38. J. M. Higman, I. C. Kizilyalli, and Karl Hess, IEEE Electron.
Device Lett. 9:399401 (1988).
39. T. Kunikiyo, J. Appl. Phys. 75:297 (1994).
40. J. W. Lyding, K. Hess, and I. C. Kizilyalli, Appl. Phys. Lett.
68:25262528 (1996).
41. I. C. Kizilyalli. J. W. Lyding, and K. Hess, IEEE Electron. De-
vice Lett. 18:8183 (1997).
42. E. Takeda, C. Y. Yang, and A. Miura-Hamada, Hot-Carrier
Effects in MOS Devices, Academic Press, New York, 1995.
43. M. Grupen and K. Hess, IEEE J. Quantum Electron. 34:120
140 (1998).
HIGH-FREQUENCY BROADCASTING
KIM ANDREW ELLIOT
High-frequency (HF) broadcasting, more commonly
known as shortwave broadcasting, employs the long-dis-
tance capabilities of skywave propagation for transmis-
sions over long distances. HF is used for international
broadcasting. Some international broadcasters addition-
ally use the mediumwave broadcast band (standard AM)
to reach audiences in nearer target countries. HF is also
used for domestic broadcasting in countries with large re-
motely populated regions (e.g., Russia, China, Canada, Aus-
tralia) and in tropical countries, where the effectiveness of
HIGH-FREQUENCY BROADCASTING 1981
mediumwave broadcasting is reduced by poor soil conduc-
tivity and static from lightning.
International HF broadcast stations generally transmit
in a number of languages, including the language of the
broadcasting country, plus a selection of major world lan-
guages or languages of countries of particular interest to
the broadcasting country. For example, the Broadcasting
Service of the Kingdom of Saudi Arabia has international
HF transmissions in Arabic, Bambara, Bangali, Indone-
sian, Persian, Somali, Swahili, Turkmen, Turkish, and
Urdulanguages of major Islamic populations. Unlike do-
mestic radio, which remains on the air throughout the day,
international HF broadcasts in a particular language usu-
ally have a duration of a half-hour to 90 min because most
international radio stations have a limited number of
broadcasters procient in each language. The uid nature
of the ionosphere also requires that frequencies be chan-
ged several times during the day.
1. HISTORY
1.1. Early Years
By the 1920s, the establishment of voice modulation and
the discovery of the long-distance properties of the HF
frequencies enabled the development of HF broadcasting.
In 1927 the Philips company in the Netherlands estab-
lished a permanent high-frequency broadcasting station
to send programs to the Dutch colonies [1]. Soon after,
Britain and France began HF broadcasts to their colonies.
During the 1930s, Italy and Germany conducted the
rst sustained use of radio for international propaganda.
Italy transmitted in Arabic to the Middle East in an at-
tempt to turn opinion in that region against the British.
Germany broadcast to German communities abroad, add-
ing Spanish, Portuguese, and English to the Americas,
and other languages to other targets. The British Broad-
casting Corporation (BBC) reacted by supplementing its
Empire Service with broadcasts in many of the same lan-
guages already transmitted by Germany and Italy [2].
1.2. World War II
Axis broadcasting during World War II had two main strat-
egies. One involved attempts to convince neutral countries
to remain neutral or to join with the Axis. The second was
to demoralize or confuse the civilian populations or armed
forces of countries that had already joined the war against
the Axis powers [3]. The BBC countered the Axis broad-
casts with a more factual presentation of news. This was
also largely the policy of the Voice of America (VoA), created
by the United States government in 1942 by consolidating
the several private HF broadcast stations that existed in
the United States before the war [4].
1.3. The Cold War Years
At the end of World War II, there was a sudden reduction
in the amount of international HF broadcasting. However,
the advent of the Cold War led to a growth of international
broadcasting that was sustained through the 1980s. The
Soviet Unions Radio Moscow set the pace with an increas-
ing number of transmitters, transmission hours, and
broadcast languages through the Cold War years. All oth-
er communist countries, and even some Republics of the
Soviet Union, maintained their own international radio
services. Chinas Radio Peking (later, Radio Beijing, and
now Voice of China) and its ally Radio Tirana (Albania)
were among the largest HF broadcast operations.
From the West, Voice of America and BBC overseas
broadcasts expanded. In 1953, West Germany created its
Deutsche Welle, which grew to be one of the major inter-
national radio services. Also in the early 1950s, the United
States created Radio Liberation, later Radio Liberty, di-
rected to the Soviet Union, and Radio Free Europe direct-
ed to the other communist countries of Europe. These
stations were later merged to form Radio Free Europe/
Radio Liberty, Inc.
1.4. The PostCold War Period
Radio Moscow, which led the expansion of international
broadcasting after World War II, also led the postCold
War reduction of international broadcasting. In 1990, Ra-
dio Moscow broadcast in 61 languages. In 1998, its suc-
cessor, Voice of Russia, had 33 language services. Western
broadcasters also reduced some of their operations and
shifted operations from direct HF broadcasts to rebroad-
casting within the target country (rebroadcasting is de-
scribed later). Radio Free Europe/Radio Liberty moved
from Munich to Prague. The station closed its Hungarian
service and began privatization of its Czech and Polish
services. Voice of America HF broadcasts in Russian were
reduced from 16 h per day in 1989 to 6 h per week in 1998.
Many of the HF transmitters active during the Cold
War are now leased to religious and other radio program-
mers. In 1997, BBC sold its UK-based HF broadcast trans-
mitters to Merlin Communications International Ltd.,
which leases the transmitters back to the BBC. Merlin
leases spare transmitter time, as do Deutsche Telekom in
Germany, Sentech in South Africa, and the transmission
companies in many of the former Soviet republics.
Other countries have been considering the future of
their international radio services in the postCold War
era. In late 1996, the Canadian government announced
that Radio Canada International would be closed. Public
and political support kept the station on the air. In 1997,
the Australian government planned to close Radio Aus-
tralia, the international service of the Australian Broad-
casting Corporation. After much debate, Radio Australia
remained on the air, but its operations were reduced by
about half. Its broadcasts in Cantonese, Thai, and French
were dropped, and its main HF transmission site near
Darwin was put into mothballs.
The United States has created new international
broadcasting services that reect the shift from the old
U.S.Soviet Cold War theater. Radio Mart to Cuba began
broadcasting in 1985. Radio Free Asia was inaugurated in
1997, transmitting to China, Tibet, North Korea, Vietnam,
Laos, Cambodia, and Burma (now Myanmar). Congress
has allocated funds for new Radio Free Europe/Radio Lib-
erty broadcasts to Iran and Iraq and for a new Radio De-
mocracy for Africa to be operated by the Voice of America.
1982 HIGH-FREQUENCY BROADCASTING
The use of HF for domestic broadcasting has receded in
recent decades. For example, the 1973 World Radio TV
Hand book listed about 250 stations in the 90-m tropical
broadcast band (32003400kHz). The 1998 issue of the
Handbook lists about 150 stations in that same band [5].
However, investment in domestic HF broadcasting contin-
ues in some countries. In 1994, the Zimbabwe Broadcasting
Corporation revived its domestic HF broadcasting by putting
the rst of four new 100kW HF transmitters into service.
2. PRESENT BROADCASTING ACTIVITY
International radio broadcasting can be divided into three
broad types: government-funded, religious, and commer-
cial. Government-nanced stations comprise, by far, the
largest share. According to the 1998 World Radio TVHand-
book [5], some 85 national governments operate interna-
tional radio stations. Most religious international stations
are Protestant evangelical, but there are also Catholic and
Islamic stations. Commercial HF international broadcast-
ing has not been successful in selling spot advertisements
because of a lack of audience ratings data, but a number of
stations make some profit by selling blocks of time to re-
ligious and special-interest organizations.
Table 1 shows the top forty international radio stations
by frequency hours.
2.1. International Broadcasting from the United States
The United States originates every type of international
HF broadcasting. The Voice of America is a government
Table 1. International Radio Broadcasters: Top 40 Stations Ranked by Program Hours per Week
a
Country Station Type
b
Hours per Week
1 USA Trans World Radio R 939
2 United Kingdom BBC World Service G 929
3 China China Radio International G 893
4 USA Voice of America G 870
5 USA Far East Broadcasting Co R 809
6 Germany Deutsche Welle G 708
7 USA WWCR C 672
8 USA World Harvest Radio R 665
9 Iran Voice of the Islamic Republic of Iran G 637
10 USA Adventist World Radio R 618
11 USA WYFRFamily Radio R 616
12 Egypt Radio Cairo/Voice of the Arabs G 587
13 USA Radio Free Europe/Radio Liberty G 553
14 Russia Voice of Russia G 539
15 Ecuador HCJBVoice of the Andes R 534
16 India All India Radio G 497
17 Japan NHK World G 457
18 USA High Adventure Ministries R 443
19 France Radio France Internationale G 415
20 North Korea Radio Pyongyang G 378
21 Turkey Voice of Turkey G 375
22 Netherlands Radio Netherlands G 369
23 Israel Kol Israel G 356
24 USA WJCR Worldwide R 336
25 Vatican City Vatican Radio R 333
26 Romania Radio Romania International G 332
27 South Korea Radio Korea International G 322
28 Bulgaria Radio Bulgaria G 315
29 USA WEWNWorldwide Catholic Ratio R 301
30 Finland YLE Radio Finland G 295
31 Australia Radio Australia G 280
32 Switzerland Swiss Radio International G 270
33 Spain Spanish Foreign Radio G 255
34 Taiwan Radio Taipei International G 252
35 Greece Voice of Greece G 234
36 Vietnam Voice of Vietnam G 231
37 Italy RAI International G 218
38 Ukraine Radio Ukraine International G 203
39 Poland Polish Radio Warsaw G 200
40 Cuba Radio Havana Cuba G 195
a
This table shows program hours rather than frequency hours. Many religious broadcasters transmit in a language on only one frequency. Government-
funded international broadcasters use more than one HF frequency per language broadcast, and as such they may have a better change of delivering an
audible signal in the target country or countries. Hours per week include mediumwave and HF transmissions. Program hours reect simultaneous trans-
mission in more than one language. (Source: BBC Monitoring, World Media Unit.)
b
Station types: Ggovernment-funded; Rreligious; Ccommercial.
agency (at present part of the U.S. Information Agency)
that broadcasts worldwide in 51 languages. Radio Free
Europe/Radio Liberty and Radio Free Asia (to China and
the other communist countries of Asia, plus Burma) are
U.S. governmentnanced corporations that focus their
programming on the domestic affairs of their target coun-
tries. The United States also has more than 20 private
HF broadcast stations. Some, such as WEWN, World-
wide Catholic Radio in Birmingham, Alabama, are
purely religious operations. Others, such as WWCR
in Nashville, sell time to religious and special-interest
program producers.
2.2. Clandestine Broadcasting
A special category of government-funded HF broadcasting
is clandestine broadcasting [6]. These are stations that do
not reveal the location of their transmitters or studios.
Clandestine stations accompany most wars and revolu-
tions and are used by opposition groups (genuine or con-
trived by unfriendly countries) to provide an alternative to
state-controlled domestic media. Often, a clandestine
station claims to be operating from within the target coun-
try, as the voice of the people of that country, but actually
is transmitting across boundaries. During World War II,
the Germans transmitted a station debunk on HF to
the United States. The station attempted to represent
itself as a voice of disenchanted Americans. More
recently, opposition groups have been relying less on secret
transmitters. Instead, they increasingly lease time from
commercial shortwave transmission companies or dissem-
inate their messages via the Internet. In addition to po-
litical clandestine stations, low-powered hobby pirate HF
stations are also active, especially in Europe and North
America.
2.3. Transmitters and Relays
HF broadcasting requires a significant investment in
transmitters and antennas. In the 1990s, HF transmit-
ters typically have power from 100 to 500kW. Large di-
rectional antenna systems are necessary to focus power to
the desired target (Fig. 1). Major transmitter manufactur-
ers include Thomcast (France) and Continental (USA, re-
cently merged with Telefunken of Germany).
Despite the long-distance capabilities of HF, attempts
to broadcast halfway around the world are generally dis-
appointing. As such, major international radio stations
have established relay stations outside their home coun-
tries. Until the 1970s, the signals reached the relays by
way of HF single-sideband feed transmissions. The signal
received by the audience was never better than the weak-
est HF link. Now, with satellite feeds, transmissions from
HF relays begin with a studio-quality signal. A more re-
cent variation of HF relays is transmitter exchanges in
which two international stations use each others trans-
mitters. China Radio International (Beijing) exchanges
transmitter time with Radio Canada International, Radio
France International, Spanish Foreign Radio, Swiss Radio
International, and Voice of Russia.
Figure 1. High-frequency broadcast antenna
arrays at the Voice of Americas Edward R.
Murrow transmitting station near Greenville,
NC. Guyed masts up to 100m support arrays of
folded dipoles in front of wire reecting
screens. (Courtesy of USIA International
Broadcast Bureau.)
3. FREQUENCY ASSIGNMENTS
Table 2 shows the HF frequencies allocated for broad-
casting. Channels within these bands are spaced 5 kHz
apart, that is, 9505, 9510, 9515, and so on. Administra-
tions register frequencies with the Radio Communica-
tions Bureau of the International Telecommunications
Union that HF broadcast stations within their jurisdic-
tions intend to use. The ITU has a number of procedures
for resolving interference. Also useful in preventing -
interference are frequency coordination meetings held
periodically among many international HF broadcasting
stations.
Many stations escape interference by broadcasting on
frequencies outside (but usually near) the allocated HF
broadcast bands. This out of band broadcasting is al-
lowed by a provision of the ITU regulations, if users of
communications services for which the frequencies are
primarily allocated do not complain of interference.
4. PROGRAM CONTENT
HF broadcasters transmit a variety of programs, including
news, current affairs analyses and commentaries, cultural
and special-interest programs, and music. The most im-
portant distinction in programming is propaganda versus
a balanced and objective presentation of news and infor-
mation. One of the pivotal decisions in international
broadcasting was that of the BBC not to follow the Axis
broadcasters propagandistic approach but to maintain
the BBC domestic broadcasting policy of (mostly) balanced
and factual reportage.
A propagandistic approach to international broadcast-
ing would include one-sided news and current affairs re-
porting in conjunction with commentaries that are direct
attempts to persuade the audience to a certain political
viewpoint. This approach, created by the Axis broadcast-
ers before and during World War II, was followed by the
international broadcasting of the Communist stations
through the 1980s. Propaganda continues to existRadio
Havana and Radio Pyongyang are two notable examples
but it is less prevalent now that many former communist
countries and other dictatorships are experimenting with
democracy.
The decision to broadcast balanced, objective news cor-
responds to the main motivation to listen to foreign broad-
casts. Generally, persons seek out foreign broadcasts to
get the news and information they want, especially about
their own countries and regions, which they cannot get
in reliable form from their domestic media. The most
important evidence that a balanced presentation of news
and information has been more successful than propa-
ganda is that Radio Moscow, during its decades of domi-
nance in terms of broadcast hours, number of languages,
and kilowattage, generally had audiences no larger than
about 10% the size of those for the BBC World Service or
the Voice of America [8]. The international broadcasters of
the Western democracies have determined that even if
news is embarrassing or negative to their own govern-
ments, it attracts audiences and counters the misinforma-
tion, disinformation, and withheld information of the
state-controlled media in authoritarian countries.
The use of HF broadcasting to transmit music has al-
ways been a matter of debate. Some maintain that HF is
too unreliable for the enjoyment of music. Many listeners,
however, have depended on shortwave for the types of
music they wish to hear, but cannot receive from radio
stations in their own countries. Willis Conovers jazz pro-
grams on the Voice of America are an important example.
One of the most important programming functions of
HF broadcasting is to provide news, information, and en-
tertainment to compatriots living abroad. Virtually all in-
ternational radio stations transmit in their own languages
for emigrants or citizens living or studying in other coun-
tries. Often this programming consists of relays of domes-
tic radio programming. This programming fullls a
needinformation about the homeland in the language
of the homelandunmet by the domestic broadcasting in
the countries where the expatriates live. In 1998, Radio
Portugal dropped the last of its foreign-language broad-
casts (except for Tetum, a language spoken in former Por-
tuguese East Timor), in favor of a schedule devoted almost
exclusively to broadcasts in Portuguese for Lusophone
communities worldwide.
5. THE AUDIENCE
Most government-funded international radio stations
conduct little or no audience research. They exist largely
as a matter of national obligation. So the documentation
of an audience is not as vital as for a commercial station,
which must provide statistics about its audience to sell
advertising.
The largest audience research ofce is operated by the
BBC World Service. The Voice of America, Radio Free
Europe/Radio Liberty Inc., Deutsche Welle, and Radio
France International also have significant audience
Table 2. HF Frequencies Allocated by the International
Telecommunications Union for Broadcasting
Frequencies (kHz) Meter Band Notes
23002495 120 For domestic broadcasting in
tropical regions; actually in
the medium frequency
(3003000-kHz) range
tropical regions
39004000 75 Europe and Asia only
tropical regions
59506200 49
71007300 41 Eastern Hemisphere only
95009990 31
11,65012,050 25
13,60013,800 22
15,10015,600 19
17,55017,900 16
21,45021,850 13
25,67026,100 11
research ofces. No global estimate of the number of all
international HF broadcast listeners has been attempted
because of the impossibility of conducting research in all
countries. The best benchmark of the size of the interna-
tional radio audience is the audience size of the most pop-
ular international broadcaster. The BBC estimates that
138 million people listen to BBC World Service at least
once a week. This is a conservative estimate because it
does not include countries where BBC is unable to com-
mission surveys. Although the estimate includes all lis-
tening by way of the BBCs own HF, mediumwave, and FM
transmitters, most of the World Service audience still lis-
tens via HF [9].
In general, the main predictor of audience size is the na-
ture of the domestic media. Countries that have media that
are decient because of government control, poor econo-
mies, or both have the largest international radio audiences.
Some international radio stations cite the volume of
audience mail as evidence of an audience. However, there
is no reliable way to determine the size of an audience
from the number of letters received. Audience mail is more
useful to determine audience reaction to programming
and reception quality.
A statistic commonly cited in the international broad-
casting profession is the existence of 600 million short-
wave radios worldwide. This originates from a 1986 study
by the Academy for Educational Development [10]. How-
ever, an accurate and up-to-date estimate is difcult be-
cause of the many radios and boomboxes that contain
perfunctory shortwave bands. There is no doubt a high
mortality rate among these appliances. Radios with short-
wave coverage have become less prevalent in general mer-
chandise stores, even in developing countries [11]. In these
countries, there is a shift from the number of models with
mediumwave and one shortwave band toward those with
mediumwave and an FM band.
The quality of shortwave radios varies widely. The typ-
ical shortwave available and affordable in a developing
country would be a single-conversion model containing a
mediumwave band, one or two shortwave bands, and, in-
creasingly, an FM band. Tuning is across large segments
of the HF spectrum, from, say 412MHz or 322 MHz.
Broadcast stations are crowded into the small dial spaces
that correspond to the HF broadcast bands. A step up in
quality and increasingly available in developing countries
are multiband radios that tune only the segments of the
HF spectrum devoted to broadcasting, plus mediumwave
and FM. This affords better band spread and station sep-
aration. Portable shortwave radios with digital frequency
readout are less available and more expensive in the de-
veloping countries, but they are favored among consumers
in the industrialized countries. The highest level receiver
is the communications receiver, a tabletop model usually
costing from $700 to more than $2000. These have digital
frequency readout and many technical features that facil-
itate the reception of marginal signals.
Many of the users of high-end communications receiv-
ers are DXers. DXers are a significant segment of the HF
broadcast audience, especially in industrialized countries.
The term DX comes from the old radiotelegraph abbre-
viation for distance. DXing listeners try to receive as
many stations as possible in as many countries as possi-
ble. They verify their reception of these stations by send-
ing reception reports and receiving a QSL (another
radiotelegraph abbreviation for conrmation) card.
6. PROBLEMS OF HF BROADCAST RECEPTION
Good, reliable reception of HF broadcasts is limited by the
nature of the ionosphere. Conditions of the ionosphere
vary day-to-day in weatherlike fashion. They also vary in
accordance with the eleven year sunspot cycle. During
ebbs in the sunspot cycle, frequencies above, roughly,
12 MHz are less useful for long distance broadcasting.
Nearer the peaks of the cycle, the higher HF frequencies
become useful, and the lower frequencies remain so. This
gives the many international radio broadcasters more
room to distribute themselves and avoid interference.
One impediment to good reception is the tendency of
many international radio stations to exceed the capabili-
ties of HF. Stations that do not have relay stations or ex-
change agreements may attempt to send their signals to
target countries halfway around the world. These at-
tempts generally produce poor results and, in the process,
cause interference in areas short of the target.
Most of the problems of HF reception are of human
rather than ionospheric origin. There are too many broad-
cast stations for the number of frequencies available. De-
spite some ITU regulation, many stations commence use
of frequencies with impunity. As already mentioned, short-
wave signals are usually heard outside their target areas.
They are not causing interference on paper, but in
reality they are. Interference can be cochannel or adja-
cent-channel. Theoretically, short-wave stations do not
broadcast to the same target on channels less than
10 kHz apart. However, stations 5 kHz apart are heard
on any shortwave radio. Often typical inexpensive short-
wave radios cannot separate strong signals on adjacent
channels.
6.1. Jamming
Most interference is not intentional. Intentional inter-
ference is known as jamming. Jamming dates back
before World War II. During the Cold War, the Soviet
Union maintained an intensive program of jamming the
broadcasts of Radio Free Europe, Radio Liberty, the Voice
of America, the BBC World Service, and other Western
stations. The jamming usually consisted of a raucous
buzzing noise. This would be accompanied by a Morse
code identier used by Soviet engineers to track the
performance of their jamming transmitters. Because of
the tendency of HF signals to be heard better over long
distances than shorter distances, skywave jamming is not
completely effective in blocking transmissions from
abroad. The Soviets remedied this by ringing major cities
with HF transmitters close enough to propagate ground-
wave signals within these cities. Soviet jamming ended
in 1989, a manifestation of Mikhail Gorbachevs glasnost
policy.
Today, China is the country most actively engaged in
jamming. China interferes with the broadcasts of the Voice
of America, the U.S. Radio Free Asia, BBC World Service,
and transmissions from Taiwan. In addition to noise jam-
ming China transmits overmodulated audio from its do-
mestic radio programs on the frequencies of these external
broadcasts. Cuba, North Korea, Vietnam, and Burma
(Myanmar) are also presently of recently jamming exter-
nal broadcasts. Jamming is also often heard from a num-
ber of Middle Eastern countries.
7. PROSPECTS FOR HF BROADCASTING
A number of new media technologies are now vying
with HF to transmit broadcasts over long distances and
across national boundaries. These media are satellite
broadcasting, the Internet, and rebroadcasting. Radio
was the original medium of international broadcasting
because only radio could travel long distances and across
national boundaries. New media allow text, graphics,
video, and the traditional audio. Most international broad-
casters are maintaining most of their efforts in their tra-
ditional radio medium. Nevertheless, they will have to
decide which medium is most appropriate for each of
their broadcasting missions. Since the 1970s or so, inter-
national radio broadcasts have consisted largely of news,
commentaries, and current affairs talks, with a minimum
of production. Such content might more efciently be
transmitted as text.
7.1. Satellite Broadcasting
With terrestrial broadcasting, radio developed rst, then
television. In satellite broadcasting, television has already
started, but direct satellite radio systems for domestic and
international use are still in development. WorldSpace
Corporation of Washington, DC is planning to launch the
rst of its three direct radio broadcasting satellites in late
1998. The WorldSpace system will transmit radio broad-
casts to small receivers in Africa, the Middle East, Asia,
the Caribbean, and Latin America. Each WorldSpace sat-
ellite will have three beams, each with a capacity of 96
AM-quality channels. The company plans to provide
transmissions services for both international and domes-
tic radio broadcasting.
Many international radio broadcasters are already
using audio subcarriers of direct-to-home television
satellite systems. This is most prevalent on the Astra sat-
ellite television system in Europe. World Radio Network
of London combines the programs of several interna-
tional radio stations into 24 hour services, using subcar-
riers of Astra in Europe and other satellites in other parts
of the world.
7.2. The Internet
Now the advent of RealAudio and other audiostreaming
software it possible for persons to receive foreign broad-
casts via the Internet. Audio delity is usually not even
AM-quality, but reception is more reliable than via HF. An
important advantage of Internet audio is that a listener
can hear a specific program, say, a weekly half-hour pro-
gram about science, at any convenient time, rather than
having to be at the radio at the scheduled time of the pro-
gram. Some international broadcasters also present live
streams of their 24-h schedule. The BBC World Service in
English is available via Broadcast.com, a commercial
World Wide Web service.
At present, however, computers with Internet access
are much less common in homes than are shortwave radios,
especially in the developing countries, where audiences for
international radio are largest. Also, access to certain
World Wide Web sets can be interdicted. The Chinese gov-
ernment controls Internet gateways in that country and
currently blocks access to some foreign news-oriented sites,
including those of the Voice of America and the Cable News
Network. Inbound electronic mail is more difcult to stop,
and now the Voice of America has a Mandarin-language
e-mail news service delivered to computer users in China.
This service uses the most popular software in China to
convert from ASCII to Chinese characters.
7.3. Rebroadcasting
Now many international broadcasters use satellites to
feed complete programs of brief reports to radio sta-
tions in the target country, which rebroadcast this content
to local audiences. When given the choice, listeners certain-
ly prefer to hear foreign broadcasts via a nearby FM or
mediumwave transmitter than by way of a distant and
unreliable HF signal. Rebroadcasting also facilitates
international television because terrestrial television
transmission is limited to relatively short distances and
direct-to-home satellite opportunities are still limited
and expensive.
To some extent, however, rebroadcasting is a self-ne-
gating enterprise. The fact that a country could allow for-
eign broadcasts to be transmitted from its own territory
indicates that the country tolerates a free and diverse do-
mestic media environment. This reduces much of the in-
centive to listen to foreign broadcasts, whether via local or
external transmitters. At present, there is no rebroadcast-
ing (at least of news and current-affairs programs) in some
of the most important target countries for international
radio such as China, Nigeria, Indonesia, India, and Cuba.
During crises, private radio stations are also sometimes
banned from rebroadcasting international programs or
are themselves taken off the air. The Voice of America,
BBC, and Radio France International have experienced a
temporary loss of local rebroadcasting during turbulent
periods in the Democratic Republic of the Congo, Liberia,
Niger, and other countries.
7.4. Digital HF Broadcasting
In the 1980s, it was widely assumed that satellites would
replace HF as the medium for international broadcasting.
Two events in the 1990s have caused international broad-
casters to temper their optimism about the new technol-
ogies. BBC World Service Television was broadcast to East
Asia, including China, as part of the Star TV service of the
AsiaSat satellite. The English-language television service
was accompanied by a Mandarin-language translation on
an audio subcarrier. Media entrepreneur Rupert Murdoch
purchased Star TV in 1994. Murdoch, pursuing media op-
portunities in China, yielded to Chinese government pres-
sure and removed BBC World Service Television from
the Star TV beam, which reached most of China. BBCs
Arabic Television Service was shut down in 1996 when the
Saudi-owned Orbit Radio and Television Network, which
beamed the service into the Middle East, objected to BBCs
coverage of Saudi domestic affairs.
Therefore, international broadcasters are reconsider-
ing the future of HF for international broadcasting. The
main incentive for listening to foreign broadcasts is to get
news that the audience wants but cannot get from their
domestic state-controlled media. Therefore international
broadcasting must rely on noninterdictable, direct-to-
home media. HF is not encumbered by the politicocom-
mercial links which affect the owners of satellite trans-
ponders. And HF transmissions are not as easy to block as
World Wide Webpages.
A coalition of international radio broadcasters and
manufacturers of HF transmitters and shortwave receiv-
ers have formed Digital Radio Mondial (DRM) to develop a
system for digital broadcast transmission on HF and in
the mediumwave and longwave broadcast bands [12].
The DRM inaugural meeting was held, remarkably, in
China at Guangzhou City in March 1998. A DRM state-
ment proclaims that HF digital broadcasting will allow
listeners to hear shortwave programs free of fading and
interference. This will require new and, at least initially,
more expensive receivers. It remains to be seen if
digital transmission will overcome the degradations of
HF reception. An important question is whether digital
HF transmission will help overcome or facilitate attempts
by authoritarian governments to block broadcasts
from abroad.
BIBLIOGRAPHY
1. Radio Nederland Wereldomroep, History in Brief; avai-
lable online.
2. D. R. Browne, International Broadcasting: The Limits of the
Limitless Medium, Praeger, New York, 1982, pp. 4861.
3. P. E. Jacob, The theory and strategy of Nazi short-wave pro-
paganda, in H. L. Childs and J. B. Whitton, eds., Propaganda
by Short Wave, Princeton Univ. Press, Princeton, NJ, 1942,
pp. 49108.
4. J. Houseman, Front and Center, Simon & Schuster, New York,
1979, pp. 19104.
5. J. M. Frost, ed., World Radio TV Handbook, 27th ed., Bill-
board, New York, pp. 366368; Andrew G. Sennitt, ed., World
Radio TV Handbook, 1998 ed., Billboard, New York, pp.
511512.
6. K. A. Elliott et al., Unofcial broadcasting for politics, profit,
and pleasure, Gazette 29 (1982).
7. G. Mansell, Let Truth Be Told: 50 Years of BBC External
Broadcasting, Weidenfeld & Nicolson, London, 1982, pp. 40
94; A. Walker, A Skyful of Freedom: 60 Years of the BBC World
Service, Broadside Book, London, 1992, pp. 2635.
8. G. Mytton and C. Forrester, Audience for international radio
broadcasts, Eur. J. Commun. 3:457481 (1988).
9. Telephone interview with Colin Wilding, research services
manager, BBC World Service Marketing & Communications,
May 14, 1998.
10. R. S. Fortner, A Worldwide Radio Receiver Population Anal-
ysis, Academy Educational Development, Washington, DC,
1986.
11. A. Dasgupta, Indian receiver survey 1998, Radio Netherlands
Real Radio; available online.
12. Information about Digital Radio Mondial; available online at
http://www.rnw.nl/DRM/.
FURTHER READING
A description of international and domestic radio broadcasting in
all countries, including schedules for HF broadcasts in all broad-
cast languages, is in the annual World Radio TV Handbook, pub-
lished until 1998 by Billboard Publications. Beginning with the
1999 issue, the book will be published by WRTH Publications,
Milton Keynes, UK (editor@wrth.demon.co.uk).
Schedules of English-language HF broadcasts, plus reviews
of receivers and other information for the listener is in Lawrence
Magne, ed., Passport to World Bank Radio, International
Broadcasting Services Ltd., Penns Park, PA (http:www.passport.
com).
Schedules and general articles about HF broadcasting and oth-
er radio topics are contained in Monitoring Times magazine, pub-
lished monthly by Grove Enterprises, Brasstown, NC (http://
www.grove-ent.com). Schedules and information are also pub-
lished in the monthly Journal of the North American Shortwave
Association (http://www.anarc.org/naswa). English-language
HF broadcast schedules are updated weekly by FineWare.
A comprehensive compilation of news about and schedules of
world broadcasting stations, including HF broadcasting, is con-
tained in the World Media and Schedules publications of BBC
Monitoring, Reading, England (marketing@mon.bbc.co.uk).
A listing of most of the worlds HF broadcast transmitters by
country, location, manufacturer, power, date of inauguration, and
(if applicable) decommission, is in L. Maes, Transmitter Docu-
mentation Project, 5th ed., Rijkevorsel, Ludo Maes, Belgium, 1998
(http://www.ping.be/tdp). A comprehensive discussion of HF re-
ceivers is F. Osterman, Shortwave Receivers Past and Present:
Communications Receivers 1942 to 1977, 5th ed., Universal Radio
Research, Reynoldsburg, OH, 1998 (http:www.universal-radio.
com).
General descriptions and the history of international broad-
casting include D. R. Browne, International Broadcasting: The
Limits of the Limitless Medium, Praeger, New York, 1982; H. L.
Childs and J. B. Whitton, eds., Propaganda by Short Wave,
Princeton Univ. Press, 1942; M. Nelson: War of the Black Heav-
ens: Western Broadcasting in the Cold War, Syracuse Univ. Press,
1997; C. J. Rolo, Radio Goes to War: The Fourth Front, Putnam,
New York, Peter 1942; J. Wood, History of International Broad-
casting, Peregrinus, London, 1993.
Several World Wide Websites contain information for the short-
wave listener about international broadcasting. Among the most
useful (with links to other sites) are TRS Consultants (http://
www.trsc.com), Shortwave/Radio Catalog (http://itre.ncsu.edu/
radio), Association of North American Radio Clubs (http://www.
anarc.org), Radio Netherlands Real Radio (http://www.rnw.nl/
realradio/index.html), and the IBB Monitoring Homepage (http://
voa.his.com/).
Next Page

08b

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

08b

Uploaded by

Copyright:

Available Formats

HELMHOLTZ EQUATIONS

region, and the eld remains favorably low in

base is on the left, and the collectors are classied as (a)

ion implantation serves to deactivate the collector layer

base. The penetra-

, whereas silicon has a lattice constant

. Therefore, the SiGe layer is not lattice-matched

polysilicon layer shown in Fig. 7a is etched

for electron energies of about

. Quantum effects such as

You might also like