Micron-size bending radii in silica-based waveguides

Milos Popovica , Kazumi Wadab , Shoji Akiyamab , Hermann A. Hausa and J

urgen Michelb
a Research

Laboratory of Electronics, Massachusetts Institute of Technology,

50 Vassar St, Cambridge, MA 02139, USA
b Department of Materials Science and Engineering, Massachusetts Institute of Technology,
77 Massachusetts Ave, Cambridge, MA 02139, USA
ABSTRACT
Sharp bends in low index contrast waveguides using tapered air trenches are proposed. To minimize claddingtrench junction loss, cladding tapers are designed to provide adiabatic mode shaping between low and high
index contrast regions. Drastic reduction in eective bend radius is predicted. We present 2D FDTD/EIM
simulations of bends in representative silica index contrasts. Substrate loss in air trenches of nite depth is
investigated, and the required trench depth, given an acceptable substrate loss, is calculated. Fabrication steps
are described.
Keywords: air trench bend, cladding taper, silica waveguide, sharp bend loss, low index contrast, enhanced
lateral mode connement

1. INTRODUCTION
Low index contrast silica bench technology has gained widespread use in practice in the fabrication of passive
integrated optical components, by virtue of its use of well-tested IC industry manufacturing processes and
technology. Large waveguide cross-sections oer low ber-to-chip coupling and propagation losses. A major
drawback is the relatively large component size and thus low density of integration, where a limiting factor
is the minimum waveguide bend radius. Bend radii can be in the millimeters in the low index contrasts
( = 0.25 1.5%) found in silica.1 On the other hand, high index contrast such as in silicon-on-insulator
(SOI), while oering dense integration, poses challenges of ber-to-chip coupling due to mode mismatch and
misalignment, and sensitivity to fabrication defects.
A technology that allows a drastic reduction in the bending radius would overcome one of silicas major
obstacles to attaining large-scale optical integration. We propose a scheme using tapered air trenches around
bends to provide locally enhanced lateral mode connement. We use adiabatic tapering to avoid mode mismatch
and Fresnel reection losses at abrupt junctions in order to miniaturize waveguide bends while maintaining
broadband low-loss performance.2
Air trenches have been proposed for suppressing bend radiation in several ways.3, 4 When they replace
the cladding to enhance lateral mode connement, mode mismatch-induced junction loss is incurred at points
of abrupt change in refractive index and limits the success of the approach in low index contrast. To our
knowledge, no attempt has been made to use air trenches to design small, low-loss bends in low index contrast
by properly addressing the mode mismatch issue introduced by the present air trench. We use judiciously placed
air trenches for sharp bending, with adiabatic transition to mitigate junction loss.
We introduce a pair of inverted or cladding tapers as an integral part of the air trench at the bend
(Fig. 1), in order to provide fast mode transition to and from the high index contrast trench region with low
radiation loss and low reection. The high index contrast region contains the bend, resulting in a reduction in
bending radius by a factor of 10-1000, and in total bend structure edge length by a factor of 4-60 over a range
of silica index contrasts. We present two-dimensional (2D) nite dierence time-domain (FDTD) simulations
of air trench bends (ATBs) with index contrast between = 0.25% and = 7%.
Further author information: (Send correspondence to M.P.)
M.P.: E-mail: milos@mit.edu, Telephone: (617)253-8302
K.W.: E-mail: kwada@mit.edu, Telephone: (617)252-1104

54

Integrated Optics: Devices, Materials, and Technologies VI, Yakov S. Sidorin, Ari Tervonen, Editors,
Proceedings of SPIE Vol. 4640 (2002) 2002 SPIE 0277-786X/02/$15.00

air trench

cladding
core

air trench

modal
field

WB
WH
LHB

vertical

core

LLB

lateral

LA

R2

Si substrate

taper-bend junction
LHA

LS

(c)
air
trench

"cladding"
or
"inverted"
taper

cladding

longitudinal

R1

III
II

cladding

LLA

cladding

core

cladding
ridges

cladding

dimensions

WL

(a)

Si substrate

(d)

(b)

Figure 1. Air trench bend schematic: labeled (a) and dimensioned (b) plan view, and cross-sectional views in the low
index (c) and air trench (d) regions. Contour plots representative of the mode pattern are superimposed on cross-sections.

Polarization insensitivity was not a design consideration. It is inevitably poor in the high aspect ratio
trench in Fig. 1d. Polarization insensitivity of optical processing is to be achieved by splitting the polarizations
and rotating one of them, the signal being processed by identical circuits. Following, the polarizations are
recombined after another rotation. We use the vertical (out of the wafer plane, Fig. 1) electric eld polarization
for our designs.

2. AIR TRENCH BENDS WITH CLADDING TAPERS

The bend we propose is shown in Fig. 1. Use of air trenches for sharp bending radii in silica waveguides has been
discussed previously in the context of optical ber and slab4 and ridge waveguide bend loss.3 For tight bends,
trenches are placed immediately against the waveguide core, and provide higher lateral mode connement at the
expense of some junction loss.3 In typical silica index contrasts, the abrupt junction (Figs. 2a) loss becomes
intolerable. To mitigate this junction loss, we use an adiabatic taper of the trench.

2.1. Suppression of Junction Loss

In a junction of two straight waveguides (with a junction similar to the one in Fig. 2a) the trench waveguide
width can be optimized for surprisingly low junction loss (<0.1dB even for ber-like 0.25%). At optimum,
it is wider than the low index input guide. Unfortunately, with a bent trench waveguide as in Fig. 2a this
requirement for low junction loss is in direct conict with the objective of minimizing the bend radius. At
small radii, the modal width is determined by the bend radius, not by the waveguide width (whispering gallery
regime, e.g. see Ref. 5). Bending loss may still be very small, but a good mode match to the straight waveguide
is no longer possible. A straight waveguide has a minimum mode width determined by the index contrast, and
thus a priori cannot match well to much narrower modes in the air trench bend. Thus, junction loss sets a
lower bound on the bend radius of a trench bend with abrupt junctions (e.g. 15mm for = 0.1%). To make
use of small radii, a low loss mechanism to compress the input mode is needed. We propose using an adiabatic
cladding taper (Figs. 1, 2c). The taper allows reduction of the radius in the trench region to the point where
it is limited by bending loss rather than the modal distribution width required for acceptable junction loss.
A progression from an abrupt junction to a cladding taper is illustrated in Fig. 2. An abrupt junction (Fig.
2a) requires a large bending radius. A taper after the abrupt junction can be used to compress the mode so that
a tighter bend radius can be used (Fig. 2b), but a more natural way to build the transition is by introducing
the taper as a slow perturbation of the waveguides index prole, as shown in Fig. 2c. This eliminates the
Fresnel reection at the junction of Fig. 2b, and allows an adiabatic evolution of the mode to a shape that is
better matched to the leaky bend mode. A better design would be to taper both the air trench and the core
toward the bend region, but fabrication would be more dicult, and we dont consider that design here.

Proc. SPIE Vol. 4640

55

(a)

(b)

(c)

Figure 2. Air trenches in waveguide bends: a) an abrupt junction, b) an abrupt junction followed by a taper, and c) the
proposed adiabatic cladding taper. The bend radius is junction loss limited in (a), but curvature loss limited in (b),(c).
Junction loss is present in (a) and (b), but is virtually eliminated in (c).
Bend size vs. index contrast for 98% transmission

10

Conventional bend radius

and total size
4

bend

10

Air Trench Bend

total size
2

Air Trench Bend

radius only

Example: A

10

10
-3
10

taper

10

bend

Edge length (m)

10

-2

10
Index contrast = nco/ncl - 1

C
-1

10

Figure 3. Bend (radius only and total) size for regular waveguide bends and air trench bends, showing the reduction in
size achieved by an ATB relative to a conventional bend, as a function of the index contrast of the waveguide.

2.2. Bend Loss

To demonstrate a reduction in overall size, we compare conventional waveguide bends and air trench bends
designed for the same total loss (98% or 0.1dB/90). For the chosen transmission, optimization of the present
structure - consisting of a bend and two tapers - for minimum total size is a complex problem in many parameters.
For simplicity, we choose a bend rst, and optimize the taper parameters with respect to the chosen bend design.
In lower index examples, the bend contributes little to overall structure size, so we make the radius larger than
needed for low bend loss in order to also lower the taper-bend junction loss (Fig. 1a).
In a conventional curved waveguide, bending loss is minimized by coupling to the lowest order leaky mode.510
The total loss in a 90waveguide bend comes from the bending loss of the chosen leaky mode in the bend and
from mode mismatch (junction loss) at the interfaces between the straight and bent waveguides. In the air

56

Proc. SPIE Vol. 4640

Total size box (253 m)

(a)

(b)

Figure 4. Electric eld plot from FDTD simulation: (a) Example C (Table 1), (b) Example A (Table 1). The total
size, as referred to in the text, is outlined by a box (dash-dot). The inset in (b) shows the 180bend simulation used for
the throughput eciency of the low index contrast Example A. The key in the lower right corner refers to transmission
data in Fig. 5.

trench bend, additional loss is incurred by propagation through each cladding taper, and at the junction where
each taper meets the low index contrast waveguide (Fig. 1). Radiation loss due to curvature in regular and
air trench bends is evaluated numerically according to the approaches in Refs. 79, by the WKB method or by
Airy functions. Conventional bend radii for 98% transmission and ATB radii in the bending region are shown
in Fig. 3 and Table 1. In low index contrast conventional waveguide bends, bending loss is dominant and the
junction loss can be ignored.
Table 1: Regular and Air Trench Bend radii and total sizes for 98% transmission.
Property
Contrast ()
Radius
a
Total Size
Radius
Air Trench Bend
Total Sizea
by Length
Size Reduction
by Area

Index

Regular Bend

A
0.25
15700
15708
15.35
252.60
62
3867

Example
B
0.68
2565
2570
9.25
95.85
27
719

C
6.85
76
78
7.25
19.75
4
16

units
%
m
m
m
m

Dened as the edge length of a box enclosing the entire bend structure and accommodating >99.9% of the input and
output mode power. The minimum possible bend size is a square with an edge equal to this 99.9% mode width.

2.3. Cladding Tapers

The cladding tapers adiabatically shape the fundamental mode of the low index contrast input waveguide (at
the input of section I, Fig. 1a) to the shape of the fundamental mode of the high index contrast output (at the
output of section II, Fig. 1a). In low index contrast, this output mode can be much smaller than the minimum
possible width of the input waveguide mode. The output waveguide width is xed by the optimal bend design.
The cladding tapers we consider are piecewise linear.
The tapers are designed using some simple ray optics as a starting point and characterized by 2D FDTD

Proc. SPIE Vol. 4640

57

Table 2: Air Trench Bend loss budget.

Property

A
Index Contrast ()
0.25
Bending
0.0020
Taper-Bend Junction 0.0090
ATB
Taper
0.0232
Losses
Total Lossa
0.074 b
Transmission
98.3 b

Example
B
0.68
0.0024
0.0355
0.0187
0.061
98.6

units
C
6.85
%
0.0023
dB/90
0.0394 dB/jctn
0.0089 dB/taper
0.052
dB
98.8
%

Total loss is due to 2 tapers, 2 junctions and 90of bending. Estimated individual losses do not exactly add to equal the
total loss, because the latter is the result of a simulation of the entire structure and as such is more accurate. Junction
loss at the interface between the low index waveguide and the cladding taper is deemed negligible and ignored.
b
Loss from 180ATB (note that bending loss is only 0.002 dB/90).

simulations. Individual taper eciencies are listed in Table 2, with equal throughput in either direction dictated
by reciprocity (assuming lossless dielectrics). We split the total loss between the cladding tapers and the bending
region in such a way that the tapers (which dominate the size of lower index contrast structures, Fig. 4b) carry
a larger part of the losses. Since bends constitute a small part of the ATB structure in silica index contrasts,
the exponentially radius-sensitive bend loss5 can be practically removed and taper-bend junction loss reduced
with a small radius increase, then redistributed among the tapers to reduce overall structure size.

3. ATB EXAMPLES: 2D SIMULATION RESULTS

We present 2D simulation results for several example air trench bends, designed in the manner described above
and chosen to demonstrate the proposed idea. Examples A-C range in index contrast from 0.25% to 7% (Table
1), where the last example (Fig. 4a) is above the range of typical index contrasts found in silica waveguides
( < 1.5%),1 but was our rst structure, used because of its small size in terms of wavelengths and thus short
simulation time. The refractive indices that are chosen for all examples are ones that could be produced in
SiOx Ny core/SiO2 cladding waveguides. Waveguide cross-sections are chosen to be square (in the low index
contrast region), and at the cuto of the second guided mode.

3.1. FDTD Simulation Setup

The ATB structure is reduced to two dimensions for simulation by using the eective index method (EIM)
with perturbative correction,11 which works well for our high aspect ratio trench because it resembles a slab,
particularly in low index contrast. In applying the EIM, it is assumed that the mode is always well conned
if the waveguide is straight, and the vertical substrate (or bulk cladding) leakage loss beneath the air trench is
considered separately as a correction (see Section 4).
Finite Dierence-Time Domain (FDTD) results for transmission loss are obtained by launching a short pulse
(on the order of 50fs) into the input waveguide with enough spectral width to span the wavelength spectrum
of interest. A discretization of 20 points per wavelength is used, with the perfectly matched layer (PML)
absorbing boundary condition imposed on the edges of the 2D computational domain.

3.2. Air Trench Bend Results

Figures 4a-b are plan view plots of the electric eld amplitude superimposed on the outline of the structure.
The absence of clear radiation patterns gives qualitative evidence that these bends exhibit low loss. The new
bend radii and total edge lengths including tapers (i.e. eective radii) are listed in Table 1 alongside regular
bend data, and show drastic size reduction for the same performance. As a metric for the size of bends, the
total size, we use a box placed around the bend which encompasses the bend structure as well as >99.9% of
the input and output waveguide modes (Fig. 4a-b).

58

Proc. SPIE Vol. 4640

Back-scattering

Transmission

A ( = 7%)
B ( = 0.7%)
C ( = 0.25%)

-25

98

Back-scattering Loss (dB)

Power Transmission Coefficient (%)

100

96

94

92

90
1530

A ( = 7%)
B ( = 0.7%)
C ( = 0.25%)

1535

1540

1545 1550 1555

Wavelength (nm)

1560

1565

-30

-35

Reflection to source
mode

-40

1570

1530

1535

1540

1545 1550 1555

Wavelength (nm)

1560

1565

1570

Figure 5. (a) Transmitted and (b) backscattered power ratio spectra for examples A (dash-dot), B (dash) and C (solid).
True insertion loss and return loss (into the fundamental mode) are shown at the central wavelength (A - triangle, B circle, C - square).

Example A-C results follow. For the = 7% case (Example C), an equivalent throughput eciency of 98%
is achieved with an air trench bend size of 20m (trench region bend radius of 7m), as compared to a regular
bend of 80m (Fig. 4a). For the = 0.7% case (Example B), reduction from a 2.5mm radius to an air trench
bend of 96m is achieved, with a trench radius of 9m. Finally for the lowest (ber contrast) case with a
of 0.25% (Example A), a 180bend (Fig. 4b, inset) was simulated for loss in order to reduce computational
requirements. It is representative of the 90ATB because bending loss is a small part of the total loss (Table
2). The 90ATB eld plot in Fig. 4b is given from a simulation with coarse discretization.
Figure 5 shows the transmission and back-scattering over the C-band communications window of 15301570nm, exhibiting little wavelength dependence as expected, because the structure makes no use of either
resonance or multi-mode/path interference. The exception is backscattering in the 180turn (Example A),
where weak Fabry-Perot resonances show up due to the proximity of the input and output waveguides. Markers
placed in Fig. 5 at the central wavelength (1550nm) show the values of transmission and reection into the
fundamental waveguide mode (obtained using overlap integrals). These represent the bend transmission loss
values of interest, where 98% was targeted in order to compare the bend sizes. Reection is well below -30dB
in all cases, and thus does not present a problem, at least in theoretical design.
An example is given in Fig. 6 to illustrate the impact of the cladding taper by removing the portion of the
taper in the cladding region. We use the structure of Example B with = 0.7%. A throughput of 69% results
in this case, or a loss of 1.5dB.
In all of the example cases, bend size is signicantly reduced, most drastically in lower index contrasts
which also have better ber-chip coupling (matching to the ber mode). However, in the vertical (out-of-plane)
direction, the aspect ratio of the air trenches for low s may pose fabrication problems.

4. SUBSTRATE LOSS AND MINIMUM AIR TRENCH DEPTH

For our 2D simulations, eective indices were obtained assuming a guided mode, hence a waveguide cross-section
in which the air trench and cladding regions extend innitely above and below the core (Fig. 7b). The air trench
and cladding are in reality of nite vertical extent (Fig. 7a) and there will be substrate loss if the modal index
is lower than the index of the bulk cladding below the trench. This eect is enhanced in lower index contrast
structures, where the trench is of high aspect ratio, and the mode has a proportionately larger vertical extent
(Fig. 7c). In all of these cases, an ideal air trench of innite extent would be guiding, and it is its truncation
at a nite depth that results in leakage loss.

Proc. SPIE Vol. 4640

59

T = 69.3% (-1.5dB)
R = -42 dB

Figure 6. Air trench bend from example B ( = 0.7%) without cladding tapers gives poor performance. The loss is
caused by the abrupt junction. The core taper section is virtually lossless and no gain results if it is lengthened.

In order to suppress substrate loss in the actual ATB of Fig. 7a, the air trench must be etched to a sucient
depth beneath the core for the evanescent tail of the eld to have a small amplitude where it reaches the bulk
cladding. For a modal index lower than that of the lower cladding, the modal eld will be oscillatory in the bulk
cladding region below the trench, and the mode will leak by a tunneling process from the core, evanescently
through the cladding ridge beneath it (Fig. 7a), into the bulk cladding. The mode is weakly conned in the
vertical direction, so the air trenches must be several core heights (equal to WL in Fig. 1b) deep for acceptable
substrate loss. The aspect ratio of the trench increases with lower index contrast, so fabrication issues will place
a lower limit on the index contrast range for which this technique is practical. Making use of the whispering
gallery regime in bending and removing the inner wall of the trench bend completely with appropriate design
modications would ease this diculty.
For low loss values, we calculate, using a perturbative method, the required depth of the air trench in Fig.
7a given an acceptable substrate loss (chosen to be much less than total ATB loss). We dene an equivalent
current sheet at the trench-bulk cladding interface to replace the source mode (of the innite trench in Fig. 7a),
that exactly reconstructs the eld in the region of interest below the interface. We then integrate a cylindrical
vector Greens function (e.g., Ref. 12) over the source for the perturbed substrate region (changed from Fig.
7a to Fig. 7b) to evaluate the far eld radiation, and thus loss per unit length. In this calculation, we assume
that silica cladding continues below the trench - a Si substrate (n = 3.2) below the trench would provide lower
loss due to enhanced reection. We also assume the mode shape remains unperturbed above the trench-bulk
cladding interface with the appearance of the leaky substrate region. This is valid for low loss.
We show substrate loss results obtained for ATB examples A, B and C in Fig. 8, where 2D cross-section
modal solutions from a vectorial mode solver (e.g. see Fig. 7c) were used in the calculation. From these plots
we choose a trench depth for which the substrate loss is much lower than the ATB loss of 0.1dB.
For given examples A, B and C, the mode resembles an air trench mode with large vertical extent only
within the bend region (region III in Fig. 1a), so we take a propagation distance of 2LS (see Fig. 1b) to nd
total substrate loss. For a chosen 0.01dB total substrate loss, this gives loss per unit length requirements of
0.00005, 0.00014 and 0.00039 dB/m for examples A, B, C, respectively. Total trench depth (as in Fig. 7a)
is twice the distance of the trench from the waveguide axis in Fig. 8. Here, weve assumed a trench that is
symmetric about the core axis, but leaks only to the bottom. Using loss values above, Fig. 8 and the core
height (equal to WL , Fig. 1b), we obtain required total trench depths of 28m, 16m and 4m for examples
A, B and C. Clearly, a lower index contrast poses greater challenges for fabricating the air trenches, but also
oers greater reduction in bend size.

60

Proc. SPIE Vol. 4640

Actual Finite Trench Cross-section

air
trench
cladding core

clad.
ridges

Idealized Infinite Trench Cross-section

Trench Modes

air
trench

trench
depth

cladding
ridges

core

cladding

z
bulk cladding (substrate) loss
(a)

q-TE

(b)

Ey

Ex

q-TM

(c)

Figure 7. Air trench cross-section: (a) actual air trench of nite depth; (b) idealized trench, free of substrate loss; and
(c) the dominant E-eld of the fundamental quasi-TE and -TM modes of the ideal trench for example C.

-1

Bulk cladding leakage loss vs. trench depth

10

-2

10

A ( = 0.25%)
B ( = 0.7%)
C ( = 7%)

Loss (dB/m)

-3

10

-4

10

-5

10

-6

10

-7

10

0.5

1
1.5
Interface distance from core axis (core heights, WL)

Figure 8. Bulk cladding (substrate) loss: loss per unit length for examples A, B and C (from Fig. 4), as a function of
displacement of the air trench-bulk cladding interface from the core axis (measured in core heights).

Proc. SPIE Vol. 4640

61

top cladding
core

core

core

Si substrate

Si substrate

Si substrate

Si substrate

(a)

(b)

(c)

(d)

core

core

core

Si substrate

Si substrate

Si substrate

Si substrate

(e)

(f)

(g)

(h)

under-cladding

air
trench

air
trench

Figure 9. Fabrication steps: (a) forming the under-cladding, (b) forming the core layer, (c) patterning the waveguide,
(d) forming the top cladding, (e) surface smoothing, (f) deposition of nitride mask, (g) nitride etching, (h) making the
air trench.

Another concern that emerges with the introduction of air trenches is scattering loss which is common to all
high index contrast structures. We argue that our bends are small enough to have an acceptably low scattering
loss. For example, by crude estimate, a 20m ATB structure (as in case C), assuming a scattering loss of
5dB/cm, would incur a 0.01dB total loss due to scattering per 90bend, as compared to the bending loss of
0.1dB/bend.

5. FABRICATION
The rst step in the fabrication of the present ATB structures is the formation of a cladding layer (Fig. 9a).
There are two choices for the fabrication of this layer: (a) Thermal oxidation and (b) Plasma Enhanced Chemical
Vapor Deposition (PECVD). We started with the former process because it gives the best quality oxide lm.
PECVD will be used later, because it produces a thick oxide layer in a short time, but post-annealing will be
necessary to remove grain boundaries.
For the core layer (in Fig. 9b) we have chosen silicon oxynitride (SiON). The refractive index of this material
can be varied from 1.46 to 2.0 at the 1.55m wavelength,13 by changing the composition of the source-gases
such as SiH4 , NH3 and N2 O. The N-H bonds, which have been reported to have strong absorption bands around
1.5m,14 are removed by annealing.
The waveguide is patterned (Fig. 9c) by an i-line stepper followed by a high density plasma dielectric etch
(HDPDE). The top cladding (Fig. 9d) is formed using PECVD. The quality of the lm is less critical than that
of the waveguide and therefore the faster process of PECVD is used. Post-annealing is done to remove grain
boundaries. Chemical mechanical polishing (CMP) is used to produce a smooth surface (Fig. 9e).
Next comes the fabrication of the trench. A silicon nitride layer (1500
A thick) is deposited rst on the
top cladding using PECVD (Fig. 9f). This layer serves for the etching mask. The i-line stepper denes the
trench pattern (Fig. 9g), which is then etched by a plasma etcher. The trench pattern must be aligned carefully
with the underlying waveguide; otherwise the cross-section of the waveguide is deformed. Once the mask is
fabricated, the air trench is formed using HDPDE (Fig. 9h).

6. CONCLUSIONS
By introducing air trenches gradually, away from the core rst, in a conguration which allows for adiabatic
mode transition from low to high (trench) index contrast regions, a dramatic reduction in the bending radius

62

Proc. SPIE Vol. 4640

of otherwise low index contrast waveguides is possible without incurring large junction losses through mode
mismatch and Fresnel reection. The total size (eective radius) of a bend was reduced by a factor of 4-60 in
our silica examples. Because bending radius is one of the primary factors limiting the density of integration
in silica, use of air trench bends such as the ones presented may allow dense integration leading to reduced
cost and better yield, while preserving the good ber coupling and propagation loss properties of silica PLCs.
We have provided 2D simulations of some example structures. Arbitrarily low substrate loss is possible to
within the limitations on fabricating high aspect ratio trenches, ranging in depth between 4m and 28m in
our examples. Complete removal of the inner wall of the ATB (with necessary modications of the design) may
not only further improve bend loss but also the manufacturability, by reducing the vertical structures aspect
ratio. A Si substrate rather than bulk silica cladding below the trench also results in lower substrate loss, and
thus allows a shallower trench to meet a given substrate loss requirement.

ACKNOWLEDGMENTS
Use of the NPACI Cray T-90 supercomputer at SDSC for all FDTD simulations is gratefully acknowledged.
M. Popovic wishes to thank C. Manolatou for use of her FDTD code, and M. J. Khan and M. Watts for many
discussions and suggestions.

REFERENCES
1. T. Miya, Silica-based planar lightwave circuits: passive and thermally active devices, IEEE J. of Sel.
Topics in Quan. Electron. 6, pp. 3845, Jan 2000.
2. K. Wada, M. Popovic, S. Akiyama, H. A. Haus, and J. Michel, Micron-size bending radii in silica-based
waveguides, in Proc. IEEE/LEOS Summer Topical Meeting on WDM Components, pp. 1314, (Copper
Mountain, Colorado), Aug 2001.
3. L. H. Spiekman, Y. S. Oei, E. G. Metaal, F. H. Groen, P. Demeester, and M. K. Smit, Ultrasmall
waveguide bends: the corner mirrors of the future?, IEE Proc.-Optoelectron. 142, pp. 6165, Feb 1995.
4. J. Yamauchi, M. Ikegaya, and H. Nakano, Bend loss of step-index slab waveguides with a trench section,
Microwave and Optical Technol. Lett. 5, pp. 251254, Jun 1992.
5. E. C. M. Pennings, Bends in Optical Ridge Waveguides: Modeling and Experiments. PhD thesis, T. U.
Delft, The Hague, Netherlands, 1990.
6. E. A. J. Marcatili, Bends in optical dielectric waveguides, Bell Sys. Tech. J. 48, pp. 21032132, Sep 1969.
7. M. Heiblum and J. H. Harris, Analysis of curved optical waveguides by conformal transformation, IEEE
J. of Quan. Electron. QE-11, pp. 7583, Feb 1975.
8. D. Rowland, Nonperturbative calculation of bend loss for a pulse in a bent planar waveguide, IEE
Proc.-Optoelectron. 144, pp. 9196, Apr 1997.
9. I. C. Goyal, R. L. Gallawa, and A. K. Ghatak, Bent planar waveguides and whispering gallery modes: a
new method of analysis, J. Lightwave Technol. 8, pp. 768774, May 1990.
10. M. K. Smit, E. C. M. Pennings, and H. Blok, A normalized approach to the design of low-loss optical
waveguide bends, J. Lightwave Technol. 11, pp. 17371742, Nov 1993.
11. C. K. Madsen and J. H. Zhao, Optical filter design and analysis: a signal processing approach, Wiley, New
York, 1999.
12. J. A. Kong, Electromagnetic Wave Theory, EMW Publishing, Cambridge, MA, USA, 1999.
13. G. L. Bona, R. Germann, F. Horst, B. J. Orein, and H. W. M. Salemink, Versatile silicon-oxynitride
planar lightwave circuits for interconnect applications, in Proc. 6th Intl. Conf. on Parallel Interconnects,
pp. 145148, 1999.
14. R. Germann, H. W. M. Salemink, R. Beyeler, G. L. Bona, F. Horst, I. Massarek, and B. J. Orein, Silicon
oxynitride layers for optical waveguide applications, J. Electrochem. Soc. 147, pp. 223741, June 2000.

Proc. SPIE Vol. 4640

63