JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO.

20, OCTOBER 15, 2009 4407

A Numerically Stable Analysis Method for

Complex Multilayer Waveguides Based on
Modified Transfer-Matrix Equations
Min-Suk Kwon

Abstract—This paper discusses a method of analyzing complex method (TMM) [1] is used to form a CE at the first step. At the
one-dimensional multilayer waveguides in a numerically stable second step, diverse numerical methods have been employed; a
way. When a multilayer waveguide contains a quite thick layer or
a layer whose refractive index has a very large real or imaginary few of them are the downhill method [2], the iteration method
part, conventional analysis methods tend to fail to find out its [3], and the argument-principle-based method (APM) [4], [5].
modes. In order to solve such a problem, the discussed analysis Until recently, the existing methods of analyzing 1-D multilayer
method is based on modified transfer matrix equations. The waveguides based on the conventional TMM were good enough
method consists of three steps. At the first step, two types of
modified transfer matrix equations are used to derive two types to analyze waveguides consisting of low-index materials such
of characteristic equations. At the next step, the iteration method as polymer and silica. However, they are confronted with a
comprised of initial root-finding and root-tracking is used to find stability problem [6]–[8] when they are applied to waveguides
the roots of the characteristic equations. Since the two charac-
containing quite thick low-index materials, relatively thick
teristic equations have different features, the initial root-finding
is carried out with one of them, and the root-tracking with the high-index materials like silicon, or complex-index materials
other. Finally, the field profiles of waveguide modes are calculated like metal. For such waveguides, the stability problem means
by using the found roots and employing a stable routine based on that, if a CE is given by , numerically evaluated values
the modified transfer matrix equations. A few presented examples
show that the discussed method works well. In addition to the of become too large and sometimes reach overflow. Such
method, it is shown that one of the characteristic equations can be large values of may cause serious round-off errors during
expressed as an explicit form rather than a conventional recursive a process of finding the roots of such a CE. Consequently, some
matrix relation because of the simple form of one of the transfer root-finding algorithms may fail. Since the silicon photonics
matrix equations.
and the plasmonics have become more and more important for
Index Terms—Characteristic equations, iteration method, multi- the last decade, it is required to develop a numerically stable
layer waveguides, transfer-matrix equations (TMEs).
analysis method for complex 1-D multilayer waveguides.
A few studies of numerically stable analysis method have
I. INTRODUCTION been reported. Rather than the TMM, to derive a CE that does
not incur the stability problem, Hsueh et al. used the graph
method [6], Visser et al. used the scattering matrix method [7],
C LASSICAL is the problem of analyzing 1-D multilayer
waveguides. However, still, it is very important since
we can study quickly, efficiently, and (or) quite accurately 2-D
and Mehrany et al. used the polynomial expansion method [8].
In this paper, provided is a different, numerically stable anal-
ysis method for complex 1-D multilayer waveguides. For that
waveguides by analyzing the 1-D multilayer waveguides to
purpose, first, in Section II, the author introduces two modi-
which they are approximated with the effective index method.
fied transfer-matrix equations (TMEs) and, using them, derives
In other words, the analysis of 1-D multilayer waveguides may
stable CEs in the same way as doing with the conventional
be the first step of that of 2-D waveguides. It has been studied
TMM. Hence, the derivation of the CEs based on them may be
very actively [1]–[8]. It consists of three steps: the first step
more intuitive and friendly than the previous methods. Because
of constructing a characteristic equation (CE), the second step
of the simple form of one of the two TMEs, the correspond-
of finding its roots, and the last step of determining the field
ingly derived CE is shown to be expressed as an explicit form
distributions of found modes. On the whole, the transfer matrix
rather than a recursive matrix expression. Second, in Section III,
the author explains what we should deliberate when we find the
Manuscript received March 06, 2009; revised May 14, 2009. First published roots of the derived CEs by using either the iteration method or
June 02, 2009; current version published August 28, 2009. This work was sup- the APM. Third, in Section IV, the author provides an algorithm
ported by the Korea Research Foundation Grant funded by the Korean Govern- with which the field distributions of found modes in a wave-
ment (KRF-2008-313-D00725).
The author is with the Department of Optical Engineering, Sejong University, guide are numerically evaluated in a stable way. In Section V,
Seoul 143-747, Korea (e-mail: [email protected]). appropriate examples are presented such that they show that the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. suggested analysis method works well and stably as claimed.
Digital Object Identifier 10.1109/JLT.2009.2024089 Finally, conclusions are given in Section VI.
0733-8724/$26.00 © 2009 IEEE
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
4408 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

that light propagates along the -axis and that the light propa-
gation has time-dependence, they are expressed by

(2)

(3)

for between and , where

is either the electric field of TE modes or the mag-
netic field of TM modes. In (3), is a propagation constant;
, where is the wave number in a vacuum;
is a dielectric constant equal to is equal to 1 for
TE modes or for TM modes. The (3) is called the field
expression of type I. By applying the boundary conditions that
and are continuous at , where
in the layer , the conventional TME of type I is
obtained, which is
Fig. 1. Schematic diagram of a 1-D multilayer waveguide. The layer p has a
refractive index n , a dielectric constant " , and thickness d . It exists between
x and x . The waveguide is bounded by the PECs (PMCs).

(4)
II. TWO MODIFIED TRANSFER MATRIX EQUATIONS
Applying (4) recursively from to , subject to the
condition , we obtain the CE of type I expressed
by , where
A. Derivation of Two Modified TMEs and CEs

A 1-D multilayer waveguide consisting of layers is ana- (5)

lyzed, which is schematically shown in Fig. 1. Its layer has a
thickness and a complex refractive index , and exists be- Instead of (3), can also be given by the field expression of
tween and . It is assumed that the structure type II
is uniform along the -axis, i.e., , and bounded by
perfect electric conductors (PECs) for analyzing TE modes or
perfect magnetic conductors (PMCs) for analyzing TM modes. (6)
Using such a closed waveguide has an advantage that the contin-
uous spectrum of radiation modes is discretized such that radi- When (6) is used, the sign of should be chosen, and it is done
ation modes can be analyzed in the same way as guided modes. here such that . Obtained from (6), the conventional
Although the closed waveguide is handled here, the provided TME of type II is
analysis method is also applied to open waveguides without dif-
ficulty. Since undesired reflections from the PECs or PMCs af-
fect simulation of light propagation through the waveguide, they (7)
are removed by using perfectly matched layers (PMLs) [5]. The
PMLs are introduced in the structure by the complex coordinate
stretching [9]. The complex coordinate stretching transforms the where . Another CE of
coordinate into a new complex coordinate. The transforma- type II expressed by , where
tion is given by
(8)

(1) is achieved by recursive evaluations of (7) from to

under the condition . In (4) and
(7), , and increase exponentially as the
where is the thickness of the PMLs [10]. By (1), and absolute value of increases and (or) increases.
are transformed into and , Consequently, numerically evaluated values of and may
respectively. For other . become extremely large in such a situation, and the aforemen-
Solutions of the Maxwell’s equations in the transformed coor- tioned stability problem occurs.
dinate system are easily obtained by substituting transformed In order to solve this stability problem, new amplitudes
and for and in those in the original coordinate system and are defined as follows: and
(for a closed waveguide without PMLs). Under the assumptions and
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
KWON: NUMERICALLY STABLE ANALYSIS METHOD FOR COMPLEX MULTILAYER WAVEGUIDES 4409

. By inserting the new amplitudes into (4), it is easily checked

that they satisfy the modified TME of type I

(9)
The modified TME of type I results in the modified CE of type
I expressed by , where

(10)

Similarly, the amplitudes and in (7) are replaced

by new amplitudes and
for
and for . The resultant modified TME of type
II is

(11)

and it results in the modified CE of type II expressed by Fig. 2. (a) Evaluated absolute values of f and f with respect to an effective
index (= =k ) for three values of d : 2 m (black), 4 m (red), and 6 m
(green). (b) and (c) show the field profiles of the modes corresponding to the
first and second groups of the dips in (a), respectively.
(12)

Although the absolute value of increases,

and are bounded such that and are bounded ex- centered at such simple poles. In contrast, is neither even
cept for some values of explained below. Therefore, the use with respect to each nor real-valued for a real . Moreover,
of the modified TMEs resolves the stability problem. it has branch points at . Actually, it can be deduced
As an example, a silicon channel waveguide made on a from (7) and (8) that is even with respect to each and
silicon-on-insulator wafer [11] was considered. The silicon meromorphic although it is not real-valued for a real . The
channel with a width of 450 nm was approximated to a layer simple poles and the branch points arise from the modification
with an effective index. The five-layer waveguide with this of the TMEs for stabilizing the CEs. Because of the differences
layer has the following parameters: , and are between and , they are used differently in finding the roots
3.5, 1.45, 3.10659, 1.45, and 1, respectively; , and of the CEs, which is shown in the following section.
are 4, 2, 0.23, and 4 m, respectively; is 2, 4, or 6 m.
Fig. 2(a) shows the calculated values of and as functions C. Closed Form of
of an effective index . In the calculation, the wavelength
Actually, in (10) is the (1, 2) element of a matrix
was 1.55 m; and were evaluated for TE polarization;
, where is the 2 2 matrix in (9). Since the diag-
the PML parameter was set to 0. As anticipated, the absolute
onal elements of are equal to 1, it is relatively easy to carry
values of are small except for two values of the effective
out the matrix multiplication for . After executing a few steps
index in the figure, and all the curves of for the three values
of such multiplication, we can find the explicit expression of
of coincide and cannot be discriminated. In contrast, the
every element of , which can be proved by induction. Then, it
absolute value of increases by a few orders of magnitude
gives an explicit form of , which is
as the absolute value of increases. (The absolute
value of increases with the effective index.) The first
group of dips in the curves is centered at the effective index (13a)
of a silicon channel waveguide mode, and the second group
of dips is centered at that of a dicretized radiation mode. The
field profiles of the two modes are shown in Fig. 2(b) and (c),
respectively.

B. Characteristics of and
As easily checked from (9) and (10), , is even
with respect to each , real-valued for a real , and meromor-
phic since it has only removable poles at and simple
poles at , where
(13b)
is an integer. In Fig. 2(a), the two peaks in the curves of are
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
4410 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

where , i.e., is the greatest in- bracketing it by applying the Brent method [12] to the modified
teger less than or equal to . In (13b), the sequence CE of type I. In the bisectioning process, the simple poles
is a descending array of 2 numbers as well as the roots are bracketed. Therefore, such poles should
between 0 and . For example, if and be excluded before the process or the intervals bracketing them
, where and should be discarded after the process.
. At the second step of the iteration process, the Newton
The explicit form of shows straightforwardly the in- method can be applied to both the modified CE of type I and
fluence of each layer on itself. That is an advantage of using it. II. However, when it is applied to the modified CE of type I,
However, there is a disadvantage with respect to the efficiency of the tracking process that starts from often ends in
evaluation. The number of terms in the summation for or rather than if is close to one of the simple
is equal to for poles of . Therefore, rather than should be used in the
. The evaluation of requires times multi- tracking process.
plication and times addition if or After finding the roots of the CEs for the waveguide with ini-
is considered as one term. Consequently, the tial and at an initial wavelength, it is necessary to calcu-
total number of multiplication necessary for the evaluation of late how they change as one of those parameters changes. Those
(13) is and that of addition is changes are also calculated by applying the Newton method to
. In contrast, (10) can be evaluated by executing 4 times the modified CE of type II. In this study, the author employed
multiplication and 2 times addition which are required for the globally convergent Newton method complemented with the
carrying out (9) consecutively from to . It can line search algorithm [12].
be easily checked that the total number of operations for (13)
is much larger than that for (9) and (10) if . Therefore, B. Comment on Using for the APM
using (9) and (10) is more efficient in analyzing a multilayer If we employ the APM to find the roots of the modified CE
waveguide consisting of five or more layers. of type I, we should calculate a contour integral

III. ROOT-FINDING
(14)
A. Iteration Method
Because of the PMLs, the roots of the CEs are complex num- where the prime means a differentiation with respect to . Since
bers with a small imaginary part although the dielectric constant is meromorphic, the argument principle [13] gives the fol-
of each layer is purely real. The imaginary parts of the roots lowing relation:
become manifest if the dielectric constants of some layers are
complex. It is well known that the APM is good for finding such
complex roots. However, the APM is neither intuitive nor simple
to implement. Thus, in this paper, the iteration method is used to (15)
find the roots of the modified CEs. The iteration method consists
of two steps [3]. At the first step, the roots of where and are the number of zeros, i.e., the roots of
the modified CE of type I, , are found under the con- the CE of type I, and that of the simple poles inside the contour
dition that the PML parameter and the imaginary parts of all used for the integral. From the integral and (15),
the dielectric constants are set to zero. At the second step, the , are calculated and they are used to construct an
Newton method is used to track the roots as the PML param- equivalent polynomial that has the same zeros as the roots. By
eter and the imaginary parts of the dielectric constants increase determining the zeros of the polynomial, the roots can be found.
from zero to their original values. In this tracking process, at
every step, the temporary PML parameter and imaginary part of
each dielectric constant are increased by and , IV. FIELD PROFILE
respectively, where is the number of total tracking steps. The field profile of a mode can be calculated by using the field
At an th step of the tracking process, the root found expression of type I in (3) after determining the amplitudes
at the previous step becomes an initial guess required to find and . If they are calculated successively
the root by the Newton method. The finally tracked-down from to by using (4) under the condition that
roots are denoted by , and , the magnitude of may become erroneously
gives the propagation constant of the th mode. large in some layers due to a round-off error. In order to find
It is well known that the modes of a 1-D multilayer waveguide the correct field profile, a different method of determining the
whose individual layers have real dielectric constants have a amplitudes is required.
real . Since is real-valued for a real , it is used at the Here, the amplitudes and are calculated as fol-
first step of the iteration method. Because of this property of lows. First, we determine the layer that has the largest
, the bisection method is applied to the modified CE of type absolute value of the real part of the dielectric constant
I such that the possible position of each root is bracketed in an among all the layers, and its number is denoted by . Then,
appropriate interval. Then, each root is found in the interval , are determined by using (9)
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
KWON: NUMERICALLY STABLE ANALYSIS METHOD FOR COMPLEX MULTILAYER WAVEGUIDES 4411

subject to the condition . For the next step, the

field expression of type I is modified as follows:

(16)

for between and . Actually, .

Similar to and and are defined as
and , respec-
tively, for . Then, the relation between and
is obtained by replacing the left and right column
vectors in (9) with the former and the latter, respectively. Now,
, are determined by using
the relation subject to the condition .
For the continuity of at , appro-
priate numbers and are determined such that
.
If , for the continuity of
at , they are determined such that
.
Finally, or for from 0 to is determined by

Fig. 3. Analysis results of a five-layer waveguide with a metal layer. The first
four figures show the changes of the effective indexes of the W (the black solid
lines) and S (the red dashed lines) modes of the waveguide in the tracking
(17) process. In them, N denotes the effective index. The real parts of the effective
indexes are plotted in (a) and (b), and so are the imaginary parts in (c) and (d).
When (a) and (c) [(b) and (d)] were obtained, the tracking process was based on
where A denotes either or . or for from to f (f ). (e) and (f) show the field profiles of the W and S modes, respectively.
is determined by
polarization was considered; the PML parameter was set to
0.1, and the PML thickness was set to 0.5 m. These values
of and were used for the other examples.
The first four figures of Fig. 3 show how the effective in-
dexes of two modes of the waveguide, denoted by W and S,
(18) change as the tracking process proceeds (i.e., the changes of
with respect to a step number ). Since the number
If and in (17) and (18) are re- of total tracking steps, , was set to 150, the step number
placed by and , respectively. By using (4) in the layer in the figures increases from 0 to 150. (This value of was
, where , and using (16) in the layer , where , the also used for the other examples). The lines in Fig. 3(a) and (c)
field profile can be correctly calculated. This method was used and those in Fig. 3(b) and (d) were obtained by applying the
to calculate the field profiles of the waveguides in the following Newton method to the modified CE of type I and II, respec-
section. tively, in the tracking process. In Fig. 3(a) and (c), as mentioned
in Section III, both the real and imaginary parts of the effective
V. ANALYSIS EXAMPLES index of the S mode ends erroneously at those of the W mode
since with the simple poles was used in the tracking process.
The explained analysis method based on the modified TMEs If is used and , the tracking process results in
was applied to three different 1-D multilayer waveguides. the same effective indexes of the S and W modes, which are
1.48861–j and 1.48500–j0.01218, respectively.
A. Waveguide With a Metal Layer Actually, the W mode is similar to the fundamental mode of a
The first example is a waveguide consisting of five layers one 3-layer waveguide consisting of the layers 0, 1, and 2. The S
of which is a thin gold film. The waveguide has the following mode is similar to the surface plasmon polariton guided along
parameters: , and are 1.45, 1.50, 1.45, and 1.33, the interface between the layers 2 and 3. The field profiles of the
respectively; , and are 6, 3, 2, 0.03, and 4 m, W and S modes are shown in Fig. 3(e) and (f).
respectively. The layer 2 is the thin gold film whose dielectric The dispersion curves of the W and S modes are shown in
constant is 86.6–j15.1 at the wavelength m. Fig. 4. In order to obtain the lines in Fig. 4(a), the Newton
This value was given by the Drude model. In the analysis, TM method was applied to the modified CE of type I. Around the
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
4412 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

Fig. 4. Dispersion curves of the W (the black solid line) and S (the red dashed
line) modes of the waveguide related to Fig. 3. In order to get (a) and (b), the
Newton method was applied to the modified CEs of type I and II, respectively.

wavelength of 1.5875 m at which the tracking process based on

fails, they cross falsely. However, when the Newton method
was applied to the modified CE of type II, the correct dispersion
curves were obtained and are shown in Fig. 4(b).

B. Waveguide Containing Two Layers With Gain and Loss,

Respectively
Fig. 5. Analysis results of a four-layer waveguide containing two layers with
The next example is a four-layer waveguide. The waveguide gain and loss, respectively, for TE polarization. (a) The real parts and (b) the
has the following parameters: , and are 3.169355, imaginary parts of the G (the black solid line) and L (the red dashed line) modes
of the waveguide are plotted with respect to a tracking step number. (a) and (b)
3.252398–j0.07, 3.252398 j0.07, and 3.169355, respectively; were obtained by using f in the tracking process. (c) The changes of the real
, and are 4, 0.5, 0.5, 4 m, respectively [7]. and part of the effective index of the L mode in the tracking process depending on
have the same real parts but opposite imaginary parts. In the whether it is based on f (the black solid line), f (the red dashed line), or f
(the dotted blue line). (d) shows the field profiles of the G (the black line) and L
analysis, the wavelength was set to 1.55 m. (the red line) modes, respectively.
Fig. 5(a) and (b) shows for TE polarization how the effective
indexes of two modes of the waveguide, denoted by G and L,
change as the tracking process proceeds. In the tracking process, C. Anti-Resonant Reflecting Optical Waveguide (ARROW)
the Newton method was applied to the modified CE of type II.
The last example is an ARROW consisting of five layers. The
In this case, as long as , the finally tracked-down ef-
waveguide has the following parameters: , and
fective indexes of the G and L modes were always the same,
are 3.85, 1.46, 2.3, 1.46, and 1.0, respectively; , and
and they are 3.18287 j0.01273 and 3.18287–j0.01273, respec-
are 4 m, 3.15 , 0.142 , 6.3 , 4 m, respectively, where
tively. For TM polarization, they are 3.18439 j0.00435 and the wavelength is 0.6328 m [6].
3.18439–j0.000435, respectively. As easily checked from them, The found effective indexes of some modes of the ARROW
the G and L modes have gain and loss, respectively. Also, for are shown in Table I. The first column of the table shows the
either polarization, they have the same real parts but opposite order in which the real parts of the effective indexes decrease;
imaginary parts as explained in [7]. The field profiles of the G the second column shows the names of a few modes, which were
and L modes for TE polarization are shown in Fig. 5(d). A large used for the leaky modes of the open ARROW (i.e.,
proportion of the energy of the G (L) mode is guided through ) in [6]. The effective indexes without mode names belong
the layer 2 (1) with gain (loss). Similar to Fig. 3(a), 5(c) shows to discretized radiation modes. The effective indexes of
that the real part of the effective index of the L mode ends er- and modes almost coincide with those in ([6,
roneously at that of another mode if is used in the tracking Table II]). For example, the differences between the values in
process. Even worse than this, the tracking process based on Table I and those in ([6, Table II]) for , and
terminates on the way since the Newton method fails to con- modes are 0, , and (0.75–j0.535) , re-
verge if it is applied to the conventional CE of type II. spectively. Such differences for the TM modes increase slightly.
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
KWON: NUMERICALLY STABLE ANALYSIS METHOD FOR COMPLEX MULTILAYER WAVEGUIDES 4413

TABLE I VI. CONCLUSION

EFFECTIVE INDEXES OF AN ARROW CONSISTING OF FIVE LAYERS
The author has presented the numerically stable analysis
method for complex 1-D multilayer waveguides that is based
on the modified TMEs of types I and II. It has been explained
that this method consists of three steps. First, the modified
TMEs are used to derive the modified CEs of types I and II,
which have their own pros and cons. The modified CE of type I,
especially, has been shown to be expressed as the explicit form.
Next, the roots of the modified CEs are found by the iteration
method which consists of the initial-root-finding process and
the root-tracking process. For the former, the bisection method
and the Brent method are applied to the modified CE of type
I. For the latter, the Newton method is applied to the modified
CE of type II. At the final step of the analysis, the field pro-
files of found modes are calculated by using the found roots.
Through the three examples, it has been demonstrated that this
analysis method can be applied to diverse types of waveguides.
Moreover, the analysis results related to them have shown that
the presented method is really numerically stable since the
effective indexes and field profiles of the modes in all the tested
waveguides have been successfully found regardless of the
existence of optically thick layers or complex-index layers.

REFERENCES
[1] M.-S. Kwon and S.-Y. Shin, “Simple and fast numerical analysis of
multilayer waveguide modes,” Opt. Commun., vol. 233, pp. 119–126,
2004.
[2] K. H. Schlereth and M. Tacke, “The complex propagation constant of
multilayer waveguides: An algorithm for a personal computer,” IEEE
J. Quantum Electron., vol. 26, no. 4, pp. 626–630, Apr. 1990.
[3] C. A. Hulse and A. Knoesen, “Iterative calculation of complex prop-
agation constants of modes in multilayer planar waveguides,” IEEE J.
Quantum Electron., vol. 28, no. 12, pp. 2682–2684, Dec. 1992.
[4] E. Anemogiannis and E. N. Glytsis, “Multilayer waveguides: Efficient
numerical analysis of general structures,” IEEE J. Lightw. Technol.,
vol. 10, no. 10, pp. 1344–1351, Oct. 1992.
[5] S. B. Gaal, H. J. W. M. Hoekstra, and P. V. Lambeck, “Determining
PML modes in 2-D stratified media,” IEEE J. Lightw. Technol., vol. 21,
Fig. 6. (a) Field profiles of the modes 1 (the black line) and 2 (the red line) of no. 1, pp. 293–298, Jan. 2003.
the analyzed ARROW. (b) The field profiles of TE (the black line) and TE [6] W.-J. Hsueh and J.-C. Lin, “Stable and accurate method for modal anal-
(the red line) of the analyzed ARROW. ysis of multilayer waveguides using a graph approach,” J. Opt. Soc.
Amer. A, vol. 24, no. 3, pp. 825–830, 2007.
[7] T. D. Visser, H. Blok, and D. Lenstra, “Modal analysis of a planar
waveguide with gain and losses,” IEEE J. Quantum Electron., vol. 31,
These small differences exist since the analyzed ARROW is a no. 10, pp. 1803–1810, Oct. 1995.
closed waveguide with the PMLs. The field profiles of the modes [8] K. Mehrany and B. Rashidian, “Polynomial expansion for extraction of
1 and 2 are shown in Fig. 6(a). Their most energy is confined to electromagnetic eigenmodes in layered structures,” J. Opt. Soc. Amer.
B, vol. 20, no. 12, pp. 2434–2441, 2003.
the layer 0. Fig. 6(b) shows those of and modes. The [9] W. C. Chew, J. M. Jin, and E. Michlelssen, “Complex coordinate
most portion of the energy of mode is guided through the stretching as a generalized absorbing boundary condition,” Microw.
layer 3, and mode is similar to the fundamental mode of Opt. Technol. Lett., vol. 15, no. 6, pp. 363–369, 1997.
[10] P. Bienstman, H. Derudder, R. Baets, F. Olyslager, and D. Zutter,
an ordinary slab waveguide. “Analysis of cylindrical waveguide discontinuities using vectorial
Since the real part of the effective index of the mode 1 or eigenmodes and perfectly matched layers,” IEEE Trans. Microw.
2 is very large, the values of which are evaluated near the Theory Tech., vol. 49, no. 2, pp. 349–354, Feb. 2001.
value of determined by the effective index are excessively [11] Y.-B. Cho, B.-K. Yang, J.-H. Lee, J.-B. Yoon, and S.-Y. Shin, “Silicon
photonic wire filter using asymmetric sidewall long-period waveguide
large. Consequently, when the Newton method was applied to grating in a two-mode waveguide,” IEEE Photon. Technol. Lett., vol.
the conventional CE of type II in the tracking process, it failed 20, no. 7, pp. 520–522, Jul. 2008.
due to a round-off error, so that the tracking process terminated [12] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flanery,
Numerical Recipes in C. Cambridge, U.K.: Cambridge Univ. Press,
on the way. The examples in this section make it clear that the
1994, ch. 9.
appropriate use of the modified CEs solves the stability problem [13] J. B. Conway, Complex Variable, 2nd ed. New York: Springer-
due to the conventional CEs. Verlag, 1978, ch. 5.
Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.
4414 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 20, OCTOBER 15, 2009

Min-Suk Kwon was born in Korea on May 4, 1975.

He received the B.S. (summa cum laude), M.S.,
and Ph.D. degrees in electrical engineering from the
Korea Advanced Institute of Science and Technology
(KAIST), Daejeon, Korea, in 1998, 2000, and 2005,
respectively.
He was a Postdoctoral Research Associate at
KAIST and the University of Southern California.
In September 2007, he joined the Department of
Optical Engineering, Sejong University, Seoul,
Korea, as a full-time Lecturer, where he is currently
an Assistant Professor. He has been doing active research on thermooptically
induced long-period gratings. His current research interests are integrated-op-
tical sensors based on micro-ring resonators, long-period waveguide gratings,
and surface plasmon polaritons.

Authorized licensed use limited to: Anhui University. Downloaded on November 29,2024 at 09:17:48 UTC from IEEE Xplore. Restrictions apply.