Chapter 3.
Optical Waveguides
Most of the photonic devices that we will study in PC421 incorporate optical
waveguides - structures that confine light in one or two dimensions while allowing
for propagation in the remaining dimension(s). This geometry permits light to be
split, re-routed, and recombined. In this chapter, we will explore the relation between
a waveguides geometry (size and refractive index) and the properties of the optical
fields that it guides.
Most of the chapter will be concerned with slab waveguides, in which confinement
exists in only one dimension. These are relatively easy to analyze - although this still
require numerical techniques (Matlab alert!) Towards the end of the chapter, we will
encounter the more common channel waveguides, in which 2D confinement
occurs. Some approximate solutions will be presented.
Much of the material in this chapter is not in Numais text; we draw it from a variety of
sources.
Table of Contents
1
Bending Loss
Multilayer Waveguides
Channel Waveguides
Marcatili Method
Effective Index Method (EIM)
Examples
(3.2)
For larger incident angles, we have a condition of total internal reflection (TIR), in
which 100% of the the rays power reflects from the boundary.
Now, consider the case of a high-index dielectric core of width h sandwiched
between low-index dielectrics (for historical reasons, these are called the cladding
and substrate; in general they can both be thought of as cladding layers) (Fig. 1).
We now have two parallel boundaries. If the ray angle is greater than c at both
boundaries, then the ray will be trapped within the high-n material; it propagates by
zigzagging between the two boundaries. This is the concept of a dielectric
waveguide, on which all photonic devices in PC421 are based. For the remainder of
this chapter, we will explore the properties of this guided wave.
Cladding Layer
Film
Substrate
Fig. 1:
(a)
Fig. 2:
(b)
(c)
Propagation modes: (a) radiation mode; (b) substrate radiation mode; (c) guided mode. TN Fig. 3.5.
TN 3.1 3.2
We can identify three types of propagation modes, depending on whether or not TIR
exists at one or more interfaces (it is assumed that nf > ns > nc , such that if TIR
exists at the film-substrate interface, it also exists at the film-cover interface). These
modes are shown in Fig. 2. While a few exotic devices make use of substrate
radiation modes, we will only be concerned with guided modes in PC421.
Although there are a continuum of possible ray angles that produce TIR, there are
actually only a discrete number of possible guided modes in any waveguide possibly just a single one.
Consider Figs. 3 and 4. A ray propagating at an angle f to the surface normal will
have a wavenumber k0 nf in the core material. The component of this wavenumber in
the direction of the waveguide axis (z ) is called the propagation constant :
= k0 nf sin f .
(3.3)
Now, lets keep track of the accumulated phase shifts as the ray completes one full
round trip between the two interfaces. These are equal to k0 nf h cos f (propagation
from substrate boundary to cover boundary), 2c (the phase shift upon TIR at the
film-cover interface), k0 nf h cos f (propagation from cover boundary to substrate
boundary), and 2s (the phase shift upon TIR at the film-substrate interface).
Fig. 3:
Fig. 4:
TN 3.1 3.2
To obtain a lightwave propagating through the optical waveguide without decay, the
total phase shift in the round trip must be a multiple of 2:
2k0 nf h cos f 2c 2s = 2m (m = 0, 1, 2, ).
(3.4)
m = k0 nf cos m ,
Nm =
m
k0
(for the latter, another common notation is neff ). It is trivial to show that
nf > N > (nc , ns ) in a TIR waveguide.
(3.5)
Fig. 5:
(3.7)
By placing the origin (x = 0) at the center of the waveguide, we will exploit the
symmetry of the structure to find even and odd mode solutions. The refractive
indices are ncl for |x | h /2 (the cladding), and nf for |x | < h /2 (the core).
The guided modes can have two polarizations, TE (transverse electric) and TM
(transverse magnetic). In the former, the E field is parallel to the core-cladding
boundary, while in the latter, the H field is parallel to this boundary.
A waveguide mode - also called a normal mode or eigenmode - is defined as a
wave solution to Maxwells equations with all boundary conditions satisfied, for which
the transverse spatial profile of the fields and the polarization remain unchanged
during propagation. As we have seen, each mode is characterized by a propagation
constant m . It is also characterized by a particular spatial distribution of the EM
field; this latter information is not provided by the ray method of waveguide analysis.
E = E y ( x ) e i z y
(3.8)
i t
is suppressed here and for most of PC421). We
(the time dependence e
automatically know the H field as well, using Maxwells equations for a
time-harmonic field - specifically, E = i H, which gives us
!
Ey
1
(3.9)
z e i z ,
i Ey x +
H=
i 0
x
where we assume that = 0 everywhere. Even and odd modes are sketched in the
figure below.
,
(3.14)
C0 =
C1 sin
i 0
i 0
2
which simplifies a bit to
C0 =
C1 sin
.
0
0
2
(3.15)
We now have two equations in the two unknowns C0 and C1 . These can be written
"
#
#"
# "
1
cos 2h
C0
0
.
(3.16)
=
sin 2h
0
C1
0
0
h
.
2
(3.17)
Because the tangent function is periodic, there may exist multiple solutions (that is,
pairs of [, ]) to this equation. Each represents a guided TE mode with even
symmetry. The first two of these even modes - m = 0 and 2 - are illustrated in Fig. 6.
By the way, we now know the ratio of C0 to C1 , but we dont know either of
their absolute values. Knowing the ratio allows us to eliminate one of these
constants. The remaining constant is proportional to the amplitude of the field
that exists in the waveguide, and is irrelevant to this discussion.
C0 e (x h /2)
C1 sin x
Ey (x ) =
C0 e (x +h /2)
x h /2,
|x | h /2,
x h /2.
(3.18)
Using the same procedure of matching the tangential field components (Ey and Hz )
at x = h /2, we obtain the characteristic equation
= cot
h
.
2
(3.19)
This is very similar, but not identical, to the characteristic equation for the TE even
modes.
Before the advent of modern computing, the characteristic equations were solved
graphically. We will now show how this is done. Then, we will introduce an algorithm
for numerical solution.
2h tan 2h ,
TE even modes,
h
(3.20)
=
Now, subtract Eqn. (3.12) from Eqn. (3.11) to eliminate , and multiply by (h /2)2 .
This results in
!2
2
2
h
h
k0 h 2 2
h
(3.21)
= 2 0 0 (nf2 ncl 2 )
+
=
(nf ncl 2 ),
2
2
2
2
where k0 = /c . Now we will plot the curves of Eqns. (3.20) and (3.21) on the same
graph, with normalized coordinates X = h /2 and Y = h /2; that is, we plot
(
X tan X , TE even modes,
Y=
and X 2 + Y 2 = R 2 .
(3.22)
X cot X , TE odd modes
q
k h
nf2 ncl 2 .
Note that the latter equation appears as a circular arc of radius R = 02
Every point at which the lines intersect represents a guided TE mode, alternating
even and odd and labeled, in turn, TE0 , TE1 , TE2 , and so on, the index indicating the
number of zeros in the field profile.
h/2
0
0
4
h/2
Fig. 7: A graphical solution to determine and for TE modes. This particular waveguide supports five TE modes, shown
by the intersection of the black line with the blue (even) and red (odd) lines.
(3.23)
Therefore, finding the discrete Xm or Ym from the graphical method leads easily to
the effective indices Nm and thus the propagation constants m .
Also, as we shall soon see, the effective index is bounded below by ncl and above by
nf . Therefore, the easiest numerical solution proceeds as follows:
1
Create a vector of trial N that extends from nf to ncl with the desired resolution
(usually 104 ). Call this N.
Use the above equations to create corresponding vectors and . These will
both be functions of N.
Using Eqns. (3.17) or (3.19), create
vectors equal to
h
h
and + cot
(3.24)
tan
2
2
Use any appropriate root finder to locate the values of N for which either of the
above equations are equal to zero. These values of N are the effective indices
of the guided modes, Nm . Note that the functions in step 3 have poles. It may
take a while for you to figure out how to deal with these poles when solving for
the roots.
Mode Cutoff
A mode is said to be cut off when the wavelength is sufficiently long (i.e. is
sufficiently low) that the mode is no longer supported by the waveguide. From the
graph, it is clear that the cutoff condition for the TEm mode occurs at R = m/2.
q
That is,
k0 h
nf2 ncl 2 = m , m = 0, 1, 2, .
(3.25)
2
2
The implication is that the fundamental (TE0 ) mode has no cutoff frequency; it
exists for any values of h , , nf , and ncl , as long as the WG is symmetric.
In most applications, we desire single-mode operation. That is, we require that our
waveguide can support exactly one TE mode. In this case, we must have
q
k0 h
(3.26)
nf2 ncl 2 < .
2
2
This is accomplished by an appropriate choice of the operating frequency , the
core thickness h , and the material parameter nf2 ncl 2 ; for a fixed , single-mode
operation is forced by having a small index difference, a small h , or both.
For example, if we must operate at a wavelength 0 , with a core thickness h , then
single-mode operation requires that
2
0
.
(3.27)
nf2 ncl 2 <
2h
(3.28)
20
8nf h 2
(3.29)
Note that all of these equations (indeed, this entire section) makes the simplification
that nf and ncl are independent of frequency. Of course, this is not true, as n = n()
or n(). The dispersion of the material indices will be important for some concepts
later in the course.
For all TIR waveguides, the effective indices get smaller as the mode number
increases. That is,
nf > N0 > N1 > N2 > > ncl .
(3.30)
= 0 0 ncl = ncl /c .
(3.31)
Therefore, the propagation constant of the mode approaches that which would
be encountered in the bulk cladding material. This is to be expected; at cutoff,
the mode power is no longer localized at the core - nearly all of it propagates in the
cladding.
In the high-frequency
limit, R . The graphical solution shows that, for the TEm
h
(m+1)
mode, 2h 2 and 2 . Therefore, from eq. (3.11),
= 0 0 nf = nf /c .
(3.32)
Here, virtually all of the power is localized to the core, so the mode propagation
constant is identical to that of the bulk core material.
These two extreme cases, and the transition between them, are illustrated on the
following slide.
Fig. 8: Dispersion curves of the guided TEm modes. They are confined to the region between the light lines of the cladding
and core material. In the low-frequency limit, they approach the former and in the the high-frequency limit, they approach the
latter. The light lines should actually be slightly curved because the core and cladding material is dispersive. The region above
the cores light line is forbidden, as such modes would not have an oscillatory profile in the core. The region below the
claddings light line represents the continuum of radiation modes. The cutoff frequencies of modes 1 and 2 are also shown.
Normalized Parameters
Earlier, we saw that the propagation constant is related to the effective index as
N=
.
k0
(3.33)
Furthermore, we have just seen that nf > N > ncl ; that is, the effective index is
bounded by the cladding and core material indices. This is a property of all TIR
waveguides.
Another widely-used parameter is the normalized propagation constant b . It takes
a value between zero and one:
b=
N 2 ncl 2
.
nf2 ncl 2
(3.34)
(3.35)
Fig. 9: (a) Effective index of the TEm modes vs. frequency. The cut-off frequencies of the TE1 and TE2 modes are
indicated. Normalized propagation parameter b vs. normalized frequency V for the TEm modes.
Normalization Constant
Many slides ago, we learned that the ratio of C0 to C1 must take certain values in
order to satisfy the boundary conditions on the fields at the core-cladding interface.
The absolute value of these constants does not affect the mode condition. However,
it is sometimes helpful to set the constants such that the modes are normalized.
That is, so that their total guided power is unity.
Because the mode propagates in the z -direction, we can write this power using the
Z
Poynting vector:
1
(E H )
zdx = 1.
P = Re
(3.36)
2
1
2
P=
Ey Hx dx =
Ey dx .
2
20
(3.37)
Carrying out the integral (which must be separated into its core and cladding
portions), we find that for both even and odd modes,
v
t
4
.
C1 =
(3.38)
h + 2
Mode Orthogonality
The mode solutions of a waveguide possess the property of orthogonality. We
wont prove it here, but it was probably covered in an earlier math class - for a
related example, review your PC321 notes where it was shown that the solutions to
the Schrdinger equation are also orthogonal. Mathematically, orthogonality of the
modes is expressed as
"
1
(Em Hn ) z dx dy = mn ,
(3.39)
2
where m and n are the indices of any two modes and mn is the Kronecker delta
function (equal to 1 if m = n and 0 otherwise).
Physically, the result of mode orthogonality is that we can consider all guided modes
to evolve along the propagation direction independently from each other. If we
couple light into only the nth mode of a waveguide, only the nth mode will emerge
from the other end. On the other hand, an arbitrary input can be decomposed into a
linear combination of the orthogonal modes, propagated independently, and then
re-composed at the output. Since the modes have different m , they will de-phase
during propagation, and the re-composed output field profile will no longer resemble
the input profile. In the next chapter, we will see that many devices operate on the
principle of modal interference, or modal coupling (wherein a z -dependent geometry
produces a controlled coupling among modes).
Confinement Factor
The optical confinement factor is the fraction of the total power residing in the core.
This is a very important concept in active photonic devices as it is generally only the
core material that provides gain, while the cladding is transparent. is calculated as:
2
R
R
1
E (x ) dx
Re(E H )
z dx
2 core
core
= R
.
(3.40)
= 1R
2
Re(E H )
z dx
E (x ) dx
2 total
total
Substituting the expressions for E and H from the beginning of this chapter, we find
1
!
that
2 cos2 (h /2)
h
i
= 1 +
even TE modes
h 1 + sin(h )
h
(3.41)
1
!
2
2 sin (h /2)
h
i
odd TE modes.
= 1 +
h 1 sin(h )
h
We see that the confinement factor is a function of the core width h , the material
indices, and the wavelength. It is implicitly a function of the mode number m, since
and are m-dependent. For a mode that is very far from cutoff (such as the
fundamental mode in a highly multimode waveguide), is very large and
approaches 1. On the other hand, close to cutoff, rapidly becomes small and is
highly reduced; here, much of the guided mode extends into the cladding.
1
2
1
or MFD = h +
+ ,
1
2
(3.42)
for symmetric and asymmetric guides, respectively (the latter case will be discussed
in a moment). These equations are valid for any mode (as long as you use the
appropriate ), but are commonly applied only to the fundamental mode.
When the core width h is several , the MFD scales with h . However, this is not the
case for core widths that are on the order of or smaller. Here, there is a rapid
increase in MFD, as shown on the next 2 slides. The reasons for this increase are
twofold. First, the portion of the field in the core is subject to the diffraction limit:
MFD > /(2nf ). Second, as the core width decreases, the WG becomes weaker,
increasing .
10
9
8
MFD/
7
6
5
4
3
2
1
0
0
h/
Fig. 10:
Mode field diameter (blue) vs. core thickness (both normalized by wavelength). Here, nf = 1.55 and ncl = 1.54.
The green line is the limit of MFD=h. The exact position of the minimum MFD is a function of the core-cladding index
difference, but it generally lies around h / = 1.
Fig. 11:
As the core width (yellow shaded area) decreases, the spatial extent of the mode decreases as well, up to a limit.
Further decrease results in a rapid expansion of the mode, with a loss of confinement within the core.
TM Modes
Recall from the beginning of this chapter that for TE polarization, we had
!
E y
1
i z
+
z .
E = Ey (x )e y and H =
i Ey x
i i
x
(3.43)
. By
For TM polarization, the magnetic field has only a y component: H = Hy (x )e i z y
Ampres equation,
E
(3.44)
H= ,
t
we can write
!
Hy
1
E=
i Hy x
z .
(3.45)
i
x
The guided modes are found as before, by matching the tangential fields (in this
case, Hy and Ez ) at the core-cladding interface. The resulting characteristic
equations are
2
h
ncl
h
h
=
tan
even TM modes
(3.46)
2
nf
2
2
h
ncl 2 h
h
=
cot
odd TM modes
2
nf
2
2
(3.48)
This indicates that if we input a signal with random polarization into the waveguide,
the TE and TM components will drift out of phase as the signal propagates.
Although a few devices exploit this fact, most photonic devices - particularly passive
ones - are desired to operate in a polarization-insensitive fashion. For these
devices, we need to design our waveguides with B as low as possible.
The figure on the next slide shows the confinement factors of the first few modes for
both TE and TM polarizations; note that there is a slight difference between them,
and that the difference reduces to zero at the low- and high-frequency limits. The TE
mode always has slightly better confinement within the core.
Fig. 12:
Optical confinement factors for the TEm modes (solid) and TMm modes (dashed).
C1 e 1 x
x 0,
cos(
sin(
C
x
)
+
C
x
)
0
x h,
Ey (x ) =
(3.49)
2
3
C4 e 2 (x h )
x h.
Because Ey satisfies the wave equation in each of the three regions, we have
21 + 2 = 2 0 0 ns2 = (k0 ns )
2 + 2 = 2 0 0 nf2 = (k0 nf )
2
2
22 + 2 = 2 0 0 nc2 = (k0 nc ) .
(3.50)
Fig. 13:
C1 eh 1 x
C1 cos(x ) +
Ey (x ) =
C e 2 (x h )
4
sin(x )
x 0,
0 x h,
x h,
(3.51)
i
h
cos(x ) + 1 sin(x )
0
x h,
Ey (x ) = C1
(3.52)
i
h
cos(h ) + 1 sin(h ) e 2 (x h ) x h .
Dividing both sides by cos(h ) and rearranging, we are left with the characteristic
equation
1 + 2
.
tan(h ) =
(3.54)
1 1 2 2
(3.56)
(3.57)
s c
h = tan1
nf2 1
ns2
+ tan1
nf2 2
nc2
+ m, m = 0, 1, 2, .
(3.58)
The tan1 term represents a shift (to the right) of the curves that we found in the
graphical solution (Fig. 7). As a result, for an asymmetric waveguide, there exists
a cutoff frequency even for the TE0 mode.
Dispersion curves for the guided modes of an asymmetric waveguide are shown on
the next slide. Note that the curves are confined between the light lines of the core
and the higher-index cladding.
Fig. 14: Dispersion curves for modes of an asymmetric dielectric waveguide. The light lines should actually be curved
because the core and cladding material is dispersive.
Bending Loss
A straight waveguide made from lossless materials will in theory exhibit no
propagation loss - this is why the numerical mode solving technique produced
real-valued or N . Whenever a waveguide is curved, we have what is known as
bending loss. There are two ways to describe this phenomenon qualitatively. Fig.
15 uses the ray picture, in which a ray that bounces along the core-cladding interface
in a curved waveguide will eventually hit the interface at an angle that does not result
in TIR; therefore, it radiates. A weakly-guided WG will suffer more from bending loss
because the range of allowable TIR angles is much smaller.
Fig. 15: Explanation of bending loss by ray analysis (blue/green) and wave analysis (red). R is the radius of curvature of the
bent waveguide. SOK Fig. 2-32.
Rc depends strongly on the index difference between the core and cladding, as well
as the core thickness. For weak WGs (such as doped glass), it can be on the order
of several millimeters, while for strong WGs (such as semiconductor ribs), it can be a
few tens of microns.
0
loss [dB]
10
15
20
0
Fig. 16:
0.5
1.5
R/Rc
2.5
Bending loss for a curve of constant angle, for various bending radii around the critical radius. Note the log scale.
(3.61)
(ir
(0) i + i (1 ) .
(3.63)
2
2
In other words, the real part of the propagation constant is unchanged by the
presence of gain or loss, while the imaginary part depends on how much of the
mode overlaps the lossy and gainy regions. It is most important to note that a mode
propagating along a waveguide with a gain coefficient g in the core only experiences
a gain coefficient of g .
Fig. 17:
TN 3.3
To solve for the HEpq modes, we write equations for the Hx field component in the 5
regions of the waveguide, and assume that Hx = 0 in the corner regions. Along both
directions, the field is oscillatory in the middle portion of the guide and
evanescently decaying in the outer portions. In other words, we assume that:
C2 e 2 (x w ) cos(y y + y )
region 2,
3 (y h )
cos(
C
x
+
)
e
region 3,
Hx = e i z
(3.64)
3
x
x
C4 e 4 x cos(y y + y )
region 4,
C cos( x + )e 5 y
region 5,
x
where
x2
2y
+ = (k0 n1 ) ,
x2
22
Fig. 18:
TN 3.3
To find the guided modes, as a first step we separate the 9-section waveguide into
columns, each of which can be solved individually as a slab guide to find its own
effective index. Then, in a second step, we treat these three effective indices as the
core and claddings of a slab waveguide, which is solved to give the overall effective
index. This is best explained using the figure on the next slide.
It is important to realize that you must flip your polarization between the two
steps. This is because a polarization that is parallel to the boundaries in step 1 is
perpendicular to the boundaries in step 2, and vice versa.
The EIM isnt used just for rectangular guides. It is also quite useful for finding the
effective indices of rib or ridge guides, which we will encounter later on. In fact, it can
even be used to analyze waveguides that have a smoothly varying index distribution.
Instead of treating them as a 3 - by - 3 problem as in the figure on the previous slide,
we turn it into an N - by - N problem. The slab methods we learned earlier in the
chapter can be extended to the case where there are multiple layers of core.
The EIM can be used to find effective indices, but it is poor at determining mode
profiles. In class, we will see an example to illustrate why this must be the case.
Fig. 19:
Effective index method. (top) slab waveguides corresponding to the left, middle, and right columns of the channel
guide. (bottom) the corresponding horizontal slab waveguide.
Examples
Realistically, all waveguide designers use computational techniques such as the
finite difference method, the finite element method, or the imaginary-distance
beam propagation method. Here, the waveguide cross-section is discretized into a
grid containing thousands of small elements, and variations of Maxwells equations
are solved in each element, pieced together by boundary conditions. The following
figures show some mode intensity profiles obtained using these methods. For all
figures, nf = 1.55, ncl = 1.50, = 1.55 , with core dimensions of 5 10 m.
Fig. 20:
Fig. 21:
Fig. 22:
Fig. 24:
Fig. 23:
Fig. 25:
Examples cont
In actual fact, relatively few channel waveguides have rectangular cross-sections.
Several other geometries are shown in the following figures (in the figures, n1 = n2 ).
The mode is confined to the shaded regions).
Fig. 26 is a strip-loaded waveguide. Transverse confinement is produced in a slab
geometry, while lateral confinement is achieved by a narrow, low-index strip above
the core. Even though the core layer has no transverse structure, the evanescent tail
of the mode sees enough of the strip that lateral confinement is achieved. This
should be obvious if you analyze this WG using the EIM.
Fig. 27 is a rib waveguide. Mmmmnn, ribs... ... Sorry about that...its like a ridge
waveguide except that the strip is made from the core material (we etch away the
core material outside of the core region). The confinement isnt as strong as with the
ridge or channel WGs, but there is less propagation loss, as the mode doesnt
interact very strongly with material boundaries (which produce scattering losses).
Fig. 28 is a ridge waveguide. It is similar to a channel waveguide but with a lower
cladding index (usually air) above and beside the core. This configuration is used
when we want very strong confinement (small core, tight bending radius, etc.)
Fig. 27:
Fig. 26:
Fig. 28:
Fig. 29:
Examples cont
Fig. 29 is a diffused waveguide. A dopant material is diffused into the substrate
through a patterned mask, resulting in a graded-index profile n(x , y ) rather than a
discrete core/cladding boundary. Examples include Ti:LiNbO3 (titanium-diffused
lithium niobate) and ion-exchanged glass (in which network modifiers such as
sodium are replaced by other ions such as silver).
Obviously, the methods described in this chapter can not be used to analyze
diffused waveguides since there is an index gradient. Slab diffused WGs (where
n = n(x )) can be solved using analytical methods for some index profiles, but in
general they must be solved numerically.
Later in this course, we will look at semiconductor lasers and light-emitting diodes
that incorporate optical waveguides. We will see that their geometries are much
more complicated than those shown on the previous slide. This is because we must
confine electric current and charge carriers in addition to the optical mode.