Semiconductor Optoelectronics (Farhan Rana, Cornell University)

Chapter 8

Integrated Optical Waveguides

7.1 Dielectric Slab Waveguides
7.1.1 Introduction:
A variety of different integrated optical waveguides are used to confine and guide light on a chip. The
most basic optical waveguide is a slab waveguides shown below. The structure is uniform in the y-
direction. Light is guided inside the core region by total internal reflection at the core-cladding
interfaces.

x
Cladding
n1
z
Core d n2

n1
Cladding

Most actual waveguides are not uniform and infinite in the y-direction but can be approximated as
slab waveguides if their width W is much larger than the core thickness d, as shown below.

A better description of the guided light is in terms of the optical modes. The slab waveguide supports
two different kinds of propagating modes:
i. TE (transverse electric) mode: In this mode, the electric field has no component in the
direction of propagation.
ii. TM (transverse magnetic) modes: In this mode, the magnetic field has no component in the
direction of propagation.
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

To study these modes we start from Maxwell’s equations. The complex form of Maxwell’s equations
is,
 
  E  io H
 
  H  i o n 2 ( x )E
 2 2 
 E  n ( x )E
c2
    
Since     E  (  E )   2E . In general   E  0 . Rather   [n 2 ( x )E ]  0 . But if index

is piecewise uniform in different regions then inside each region one may assume   E  0 . So we
have in each region,
 2 2 
  2E  n ( x)E
2
c 
Similarly, with the assumption of piecewise uniform index we can write for the H field,
 2 2 
  2H  n ( x)H
c2
7.1.2 TE Modes:
For TE modes, the electric field can be written as,

E ( x, z )  yˆE o  ( x )e iz
In each region of piecewise uniform index (core and cladding),  (x ) satisfies,
 2 2 2  2
  2  2 n ( x ) ( x )     ( x )
  x c 
Given a value for the frequency  , we can find all solutions of the above equation which is an
eigenvalue equation with eigenfunction  (x ) and eigenvalue   2 . Once we have the electric field,

the magnetic field H can be found as follows,

  xˆ   izˆ   E ( x, z )
   E  x 
H ( x, z )  
io i0
 zˆE  ( x ) E  
 o  xˆ o  ( x )e iz
 io x o 
The boundary conditions needed to solve the eigenvalue equation above are as follows:
i) y-component of the electric field is continuous at the core-cladding interfaces
ii) z-component of the magnetic field is continuous at the core-cladding interfaces
iii) x-component of the magnetic field is continuous at the core-cladding interfaces (this is
automatically satisfied when (i) above is satisfied)

The solutions are labeled with the integer index m (m  0,1,2,3,). So the field for the TE m mode is,

E( x, z)  yˆEom ( x, )e i m ( )z
where the dependence of the eigenfunctions and the propagation vector  on the frequency  is
explicitly indicated. For the TE modes we assume the solution,

C1e  ( x  d / 2) x  d / 2
cos(kx)
 ( x )  C2  | x | d / 2
 sin(kx )
C1e ( x  d / 2) x  d / 2
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

Plugging the above solutions in the equation,

 2 2 2  2
  2  2 n ( x ) ( x )     ( x )
 x c 
we get,
2 2 
 2  k2  n2 
c2  2 2
   (n2  n12 )  k 2
2 2
 2 c
2 2  n
2 1
c 
Using the boundary conditions give,
 kd  
tan  (cosine solutions  even TE modes)
 2  k
 kd  
 cot  (sine solutions  odd TE modes)
 2  k
A solution can be obtained graphically by plotting the left and right sides of the above equations, as
shown below where the left side is plotted using blue lines and right side using red lines. Below a
certain frequency m the m-the mode ceases to exist. This cut-off frequency is,
 c
m  m
d n2  n2
2 1

The values of  m ( ) for each mode are sketched in the Figure below.
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

x
Cladding
TE0 TE1 n1
z
Core n2 d
E Ey
y n1
Cladding


The propagation vectors  m ( ) behave as n1 near then cut-off frequencies, and asymptotically
c

approach the n 2 curve at high frequencies. In most applications, waveguide dimensions are chosen
c
such that it supports only one mode. Unless necessary we will usually drop the mode index “m“.

7.1.3 TM Modes:
Similar analysis can be done for the TM modes. Assume,

H ( x, z)  yˆHo ( x )e iz

 H  Ho  Ho  
E ( x, z )   zˆ  xˆ  ( x )e iz .
 i o n 2 ( x )   i o n 2 ( x ) x  o n 2 ( x ) 
For piecewise uniform index n (x ) ,  (x ), satisfies,
 2 2 2 
  n ( x )  ( x )    2 ( x )
 x 2 c 2 
 
Assume solution of the form,
 2
C e   ( x  d / 2) x  d / 2   2  k 2   n 2
 1  c2
2
 cos(kx )
 ( x)   C2  | x | d / 2 
2 2
  sin(kx )  2   2  n1
C1e  ( x  d / 2) x  d / 2  c2

The boundary conditions needed to solve the eigenvalue equation above are as follows:
i) y-component of the magnetic field is continuous at the core-cladding interfaces
ii) z-component of the electric field is continuous at the core-cladding interfaces
iii) x-component of the electric field weighted by the square of the index is
continuous at the core-cladding interfaces (this is automatically satisfied when (i)
above is satisfied)

Using the boundary conditions give,

2
 kd  n 2 
tan  2 (cosine solutions  even TM modes)
 2  n1 k
2
 kd  n 2 
 cot   2 (sine solutions  odd TM modes )
 2  n1 k
The above relations are similar to those obtained for the TE modes other than the factors containing
the squares of the core and cladding refractive indices. The general behavior of the TM modes is
similar to the TE modes.
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

7.1.4 Effective Index and Group Index:

The effective index neff ( ) of a mode is defined by the relation,

 ( )  neff ( )
c
The effective index of each mode equals n 1 at their cut-off frequencies and approaches n 2 at high
frequencies. The group velocity of a mode is,
1 

v g ( ) 
v g is written as,
c
v g ( ) 
ng ( )
where n g ( ) is called the group index of the mode.

7.2 Two Dimensional Waveguides

7.2.1 Introduction:
In slab waveguides, light is confined in only one dimension (the x-dimension in our notation) as it
travels in the z-direction. In most actual waveguides light is confined in two dimensions (x and y-
dimensions) and travels in the z-direction. For example, the cross-section of a rectangular waveguide
is shown below.

Unfortunately, the exact modes of such 2D waveguides are not as easy to complete as those  of slab

waveguide. The modes are neither TE nor TM. All modes  have a small component of E and H
fields in the direction of propagation. The E and the H fields can be written as,

   

E ( x, y , z)  xˆE x ( x, y )  yˆE y ( x, y )  zˆE z ( x, y ) e iz  Et x, y   zˆE z ( x, y ) e iz

   

H ( x, y , z)  xˆH x ( x, y )  yˆH y ( x, y )  zˆH z ( x, y ) e iz  Ht x, y   zˆH z ( x, y ) e iz
We have defined the transverse components of the fields as follows,
Et ( x, y )  xˆE x ( x, y )  yˆE y ( x, y )

Ht ( x, y )  xˆH x ( x, y )  yˆH y ( x, y )
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

Since the exact solution is difficult and cumbersome, several levels of approximations are commonly
used and are discussed below.

7.2.2 Mode Solutions for 2D Waveguides:

We need to solve the equation,

 2 2      E 
    E ( x, y , z )  n ( x, y )E ( x, y ) H  
c2  i o 
 
subject to all the proper boundary conditions for the E and H fields at all the interfaces.
The operator  is,
  
  xˆ  yˆ  zˆ
x y z
Define,
  
t  xˆ  yˆ    t  zˆ
x y z
 
From   E  io H , it follows that,
 
(  E ).zˆ  io H.zˆ  i o H z e iz

 (t  Et ).zˆ  io H z

(t  Et ).zˆ
 Hz  (1)
io
Similarly it can be shown that,

( t  Ht ).zˆ
Ez  (2)
 i o n 2
Equations (1) and (2) show that knowing the transverse components of the field is enough since the z-
components can be determined from the transverse components. We need to solve,
 2 2 
E  n ( x, y )E
c2

 
   . E   2E 
 2 2
n ( x, y )E

c2
Taking the transverse component gives,
   2 2  iz
t (.E )   2Et e iz  n Et e 0 (3)
c2
We need to find convenient and useful expressions for the first two terms on the right hand side. Now,
 
.E  t .Et e iz  i E z e iz
but,

.(n 2 E )  0

  t .(n 2 Et )  i n 2 E z  0

t .(n 2 Et )
 i E z  
n2

  i z i z
 iz  .(n 2 E ) iz
 .E  t .Et e  i E z e   t .Et e  t t e
2
n

  .(n Et )  iz
 
  2
  t (.E )   t  t . Et e iz   t  t e
 n2 
The other term in Equation (3) is,
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

  
 2 Et e iz  t2 Et e iz   2 Et e iz
Using the results above, Equation (3) becomes,
   1   2 2  
 t2Et  t t .Et  t .(n 2Et )  n Et    2Et
 n2  c2
The above eigenvalue equation is what one needs to solve to get the exact solution. This equation can
be put in the form,
Pˆxx Pˆxy  E x  E x 
      2   ( 4)
ˆ ˆ E
Pyx Pyy   y  E y 
where the differential operators are.
  1 ( n 2 E x )   2 E x  2 2
Pˆxx E x      n Ex
x  n 2 x  y 2 c2

  1 (n E y )   E y
2 2
Pˆxy E y   
x  n 2 y  xy
 
  1 (n E y )   2 2
2 2
 Ey
Pˆyy E y     n Ey
x 2 y  n 2 y  c2
 
  1 ( n 2 E x )   2 E x
Pˆyx E x    
y  n 2 x  yx
The above equation
 is an eigenvalue equation and its solution gives the transverse components of the
electric field E for the mode and the corresponding propagation constant  ( ) . E z ( x, y ) can be

obtained from E x ( x , y ) and E y ( x , y ) as already explained, and H field can be obtained from the
relation, 
 (  E )
H
i 0
For piecewise uniform indices, all derivatives of the index n ( x , y ) can be dropped provided
appropriate boundary conditions are used at all the interfaces.

7.2.3 The Semi-Vectorial Approximation:

In many cases of practical interest one transverse component of the electric field (either E x or E y )
dominates over the other component. In such cases, we may assume that the other transverse field
component is zero. For example, if we know a priori that E x dominates then we may assume that
E y is zero and solve the much simpler eigenvalue equation,

Pˆxx E x    2E x
 1  (n 2 E x )   2 E x  2 2

  2   n ( x, y )E x )    2E x .
x
 n x  y 2 c2
The above equation is called the semi-vectorial approximation. For piecewise uniform dielectrics we
can also write it as,
2 2 2
 E x Ex  n 2 ( x, y )E x    2E x
x 2 y 2 c2
2
 t2 E x  n 2 ( x, y )E x    2 E x
2
c
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

provided we take care to impose the boundary conditions on E x ( x, y ) at all the dielectric interfaces
as appropriate for the x-component of the electric field. Once the dominant E x ( x, y ) component has
been found, the remaining field components can be found as follows,

  
 . n 2E  0

 Ez 
i 
1  n 2E x 
 n2 x

 (  E )
H
i o

 Hy 

Ex 
1   1  n 2E x 
 
 
 o  o x  n 2 x 

 Hx 
1 
 2

 1  n 2E x 
 o y  n x 

.H  0
1 E x
 Hz  
i o y
When the horizontal electric field component dominates the modes are sometimes called HEpq modes
or TEpq modes (with a slight abuse of terminology). When the vertical electric field component
dominates the modes are called EHpq modes or TMpq modes (again with a slight abuse of
terminology). The two subscripts p and q indicate the number of nodes the dominant electric field
component has in the horizontal and vertical directions, respectively.

7.2.4 The Scalar Field Approximation:

If one component of the transverse electric field dominates over the other transverse component and
the index differences among different regions in the structure are also relatively small, then one may
use the semi-vectorial approximation and also do away with the boundary conditions on the normal
component of the electric field at all dielectric interfaces. For example, if we know a priori that
E x ( x, y ) dominates then we may solve the equation,
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

2 2 2
 E x ( x, y )  E x ( x, y )  n 2 ( x, y )E x ( x, y )    2 E x ( x, y )
2 2 2
x y c
2

  t2 E x ( x, y )  n 2 ( x, y )E x ( x, y )    2 E x ( x, y )
2
c
assuming that the field and its derivative are continuous across all dielectric interfaces. This is called
the scalar field approximation. Once the dominant electric field component has been found, the
remaining field components can be found as in the case of the semi-vectorial approximation. Scalar
field approximation seems crude but it gives very accurate answers for the propagation vector  ( )
(or the effective index n eff ( ) ) as long as one is for away from the mode cut-off frequency. It is also
very accurate in calculating mode confinement factors – as we will see in later Chapters. For most
part of this course we will use the scalar field approximation to keep the computational overhead low.
One disadvantage of the scalar field approximation is that it does not tell accurately whether the
single-mode condition holds since the scalar field approximation is not accurate near mode cut-off.

7.2.5 Slab Waveguide Approximation:

If the aspect ratio of the waveguide is such that one dimension is much larger than the other
dimension, then the modes and the corresponding wavevectors and effective indices can be
approximated by those of the corresponding slab waveguide as discussed earlier.

7.2.6 Energy and Power in Waveguides:

The energy flow for electromagnetic fields is given by the complex Poynting vector,
  1    
S(r )  Re[E (r )  H * (r )]
2
  1        
S(r )  [E (r )  H * (r )  E * (r )  H (r )]
4
For waveguides electric and magnetic fields are of the form,
 

E ( x, y , z )  xˆE x ( x, y )  yˆE y ( x, y )  zˆE z ( x, y ) e iz
 

H ( x, y , z )  xˆ Hx ( x, y )  yˆH y ( x, y )  zˆH z ( x, y ) e iz
The total energy flow (or power) in a waveguide is obtained by integrating the Poynting vector over
the cross-section of the waveguide,
 1     1    
P (z)   S . zˆ dxdy   [E  H * E * H ] . zˆ dxdy   [Et  Ht * Et * Ht ] . zˆ dxdy
4 4
Assuming the medium to be dispersive, the energy per unit length of the waveguide is,
 1  o     1   
W z     E.E *   o H.H * dxdy
 4  4 
1
4
   1
  
    o n 2n gM  n E.E *   o H.H * dxdy
4 
Here, ngM   is the material group index, defined earlier as,
dn  
ngM    n    
d
The superscript “M“ is intended so as not to cause confusion with the group index of the optical mode.
The following relation can also be proven,
1 2
  1  
  o n E.E * dxdy    o H.H * dxdy
4 4
Therefore, the energy per unit length can be written as,
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

1  
W z   M
  o n ng E.E * dxdy
2
The effective index neff ( ) of a mode is,

 ( )  neff ( )
c
The group velocity of a mode is,
1 

v g ( ) 
v g is frequently expressed in terms of the group index of the mode,
c
v g ( ) 
ng ( )
where n g ( ) is called the group index. One can prove the following relation between the power
P (z ) and the energy per unit length W (z ) ,
P(z)  v g W (z)
which can also be written as,
M
 
ng W (z)   o n ng E.E * dxdy
c
  

P (z)  Re Et  Ht * . zˆ dxdy 
In the slab-waveguide approximation, assuming TE modes with the transverse component of the
electric field given by  x, y  , one obtains the following expressions for various quantities of interest,
1   1
W z     o n ngM E.E * dxdy    o n ngM  dxdy
2
2 2
1     1  2
P (z)   [Et  Ht * Et * Ht ] . zˆ dxdy    dxdy
4 2 o
M 2
 n n g  dxdy
n g neff 
2
  dxdy

7.2.7 Properties of Waveguide Modes and Orthogonality of Modes:

Frequently, solutions in various cases involve expansions in terms of all the waveguide modes. In
such cases, knowledge of the orthogonality of the modes is useful. The electric and the magnetic
fields for the m-th mode can be written as,

 
E m  Etm x, y   zˆEzm ( x, y ) e i m  z 

 
H m  Htm x, y   zˆHzm ( x, y ) e i m  z 
Some useful properties of the waveguides modes are listed below:

i) When the indices are real, the propagation vectors are also real, and the transverse field components
can be chosen to be real as well. Equations (1) and (2) show that in this case the z-components of the
fields are purely imaginary.

ii) When the indices are real, the complex conjugate of the electric field mode gives the field for the
mode propagating in the opposite (time-reversed) direction. For example, if,


E m  Etm x, y   zˆEzm ( x, y ) e i m  z 
represents the field for the forward propagating mode then,
  
 
E m *  Etm x, y   zˆEzm ( x, y ) e i m  z 
represents the field for the backward propagating mode.
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

iii) When the indices are real, the negative of the complex conjugate of the magnetic field mode gives
the field for the mode propagating in the opposite (time-reversed) direction. For example, if,

 

H m  Htm x, y   zˆHzm ( x, y ) e i m  z
represents the field for the forward propagating mode then,
  
 

 H m *   Htm x, y   zˆHzm ( x, y ) e i m  z
represents the field for the backward propagating mode.

iv) Consider two different modes, “m“ and “p“ with the same propagation vector but different
frequencies then the orthogonality between the mode fields is expressed as,
m p
 
 dxdy n x, y  E . E *  0 for m  p
2
m p
 
 dxdy H . H *  0 for m  p

v) Consider two different modes, “m“ and “p“ with the same frequency but different propagation
vectors then the orthogonality between the mode fields is expressed as,

m  p
 
 dxdy Et  Ht * . zˆ  0 for m  p

vi) The most general way of expanding a time harmonic field of a particular frequency inside a
waveguide is in terms of the waveguide modes,
 
  
 
E x, y, z    am Etm x, y   zˆEzm ( x, y ) e i m  z   bm Etm x, y   zˆEzm ( x, y ) e i m  z
m m

H x, y, z    am
m
 H
m
t x, y   zˆHzm ( x, y ) e 
i m  z
 b  H
m
m
m
t x, y   zˆHzm ( x, y ) e i m  z

7.2.8 Longitudinal vs Transverse Modes of a Waveguide:

Consider the model dispersion relations shown below for the first three modes of a waveguide. The
HE and EH modes are called the transverse modes of the waveguide since these modes describe the
field profile in dimensions x and y that are transverse to the direction of propagation.

EH00 mode

HE01 mode

HE00 mode

The field profile in the direction of propagation is described by the propagation vector  . The
dispersion for the lowest HE00 mode is plotted below in a slightly different way that also shows the
dispersion for negative values of the propagation vector (propagation in the –z-direction). Different
values of  correspond to different longitudinal modes of the waveguide.
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

HE00 mode
HE00 mode



Periodic boundary conditions can be used to find the number of different longitudinal modes
corresponding to a transverse mode in an interval  of frequency. The problem is identical to
finding the density of states for photons in one dimension. If the length of the waveguide is L then
there are L  2 different longitudinal modes in an interval  . Since v g      , there are
 
2  L v g  2 different longitudinal modes in an interval  . The factor of 2 accounts for the
forward and backward propagating modes.

7.2 Perturbation Theory for Waveguides

7.2.7 Perturbation Theory for Waveguides:
Often one is interested in the change in the propagation vector of a mode that comes from a small
change in some parameter, such as the frequency or the refractive index of the medium. The change in
the propagation vector can be obtained from the complex electromagnetic variational theorem and the
result is,

 

     
  o n 2 E.E *   0 H.H * dxdy 
 

2  Re Et  Ht * . zˆ dxdy
For example, if the frequency is changed then using the above Equation one obtains the familiar
result,
M
 
 ng W (z )   o n ng E.E * dxdy

 vg 
c
  

P(z)  Re Et  Ht * . zˆ dxdy 
If the refractive index is changed then one obtains,

  

 
 o n n E.E * dxdy 
  

 Re Et  Ht * . zˆ dxdy
In the slab waveguide approximation, assuming a TE mode and the transverse component of the
electric field given by  x, y  , one obtains the following expression for the change in the propagation
vector with the refractive index,
2
 1  n n  dxdy
 
c neff 2
  dxdy
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

7.2.8 Mode Confinement Factors:

Consider the waveguide shown below. The refractive index of the core is n2 and that of the cladding
is n1 .

n2 → n2 + n2

Suppose the refractive index of the core changes by n2 . The change in the propagation vector can
be found as follows,

  


 
 o n n E.E * dxdy 
 

 Re Et  Ht * . zˆ dxdy


  
core  o n n E.E * dxdy 
  n2 n2 
 

core  o E.E * dxdy 
 

 Re Et  Ht * . zˆ dxdy
 

 Re Et  Ht * . zˆ dxdy

   M
 
 n2 n2  core  o n2 n2g E.E * dxdy 
 
 

 n2 n2Mg   Re Et  Ht * . zˆ dxdy 
   M
 
 M
 
 n2 n2  core  o n2 n2g E.E * dxdy   o n ng E.E * dxdy 

  
  

 n2 n2Mg    o n ngM E.E * dxdy  Re Et  Ht * . zˆ dxdy
 

   M
 
 n2  core  o n2 n2g E.E * dxdy  ng  

 
  
 n2Mg    o n ngM E.E * dxdy  c  
 n 
  g 
 2  n2
c  n2Mg 
 
Note that n2Mg is the material group index of the waveguide core, and n g is the group index of the
waveguide mode. The overlap integral 2 , defined as,
 M
 
core  o n2 n2g E.E * dxdy 
2 
 M
 
  o n ng E.E * dxdy 
represents the fraction of the mode energy density confined in the waveguide core. 2 is called the
transverse mode confinement factor for the waveguide core. The change in the propagation vector is,
as expected, proportional to n2 and also to the fraction of the modal energy inside the core.

Suppose now the refractive index of the cladding also changes by n1 then the total change in the
propagation vector can be written as a simple sum,
 Semiconductor Optoelectronics (Farhan Rana, Cornell University)

n   n 
  g    g 
  1   n1  2  M  n2
c M
 n1g  c  n 2g 
   
In the slab waveguide approximation, assuming a TE mode and the transverse component of the
electric field given by  x, y  , one obtains the following expression for the transverse mode
confinement factor for the core,
M 2
core n2 n2g  dxdy
2 
M 2
 n ng  dxdy
The waveguide perturbation theory can be used to calculate the change in the propagation vector in
the presence of material loss (or gain). Suppose the core of the waveguide becomes lossy and the
imaginary part of the core refractive index acquires a non-zero value given by,
c 2
n2  n2  i
 2
In this case, we can take the index perturbation n2 to be,
c 2
n2  i
 2
The change in the waveguide propagation constant becomes,
 n   n   n 
  g    g   c 2   g  2 ~ ~
  2   n 2  2    i   i 2    i 2 2  i
c  n2Mg  c  n2Mg    2   n2Mg  2 2 2
     
where,
 n 
 g 
~2   2 ~  2 ~2
 n2Mg 
 
The propagation vector  acquires a small imaginary part because of optical loss in the waveguide
core. The imaginary part of the propagation vector will cause the wave energy to decay with distance
as it propagates in the waveguide.