Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37400
Visible-light silicon nitride waveguide devices
and implantable neurophotonic probes on
thinned 200 mm silicon wafers
W
ESLEY
D. S
ACHER
,
1,2,3,*
X
IANSHU
L
UO
,
4
Y
ISU
Y
ANG
,
2
F
U
-D
ER
C
HEN
,
2
T
HOMAS
L
ORDELLO
,
2
J
ASON
C. C. M
AK
,
2,3
X
INYU
L
IU
,
1
T
ING
H
U
,
5
T
IANYUAN
X
UE
,
2
P
ATRICK
G
UO
-Q
IANG
L
O
,
4
M
ICHAEL
L.
R
OUKES
,
1
AND
J
OYCE
K. S. P
OON
2,3
1
Division of Physics, Mathematics, and Astronomy, California Institute of Technology, Pasadena,
California 91125, USA
2
Department of Electrical and Computer Engineering, University of Toronto, 10 King’s College Rd.,
Toronto, Ontario M5S 3G4, Canada
3
Max Planck Institute of Microstructure Physics, Weinberg 2, 06120, Halle, Germany
4
Advanced Micro Foundry Pte Ltd, 11 Science Park Road, Singapore Science Park II, 117685, Singapore
5
Institute of Microelectronics, A*STAR (Agency for Science, Technology and Research), 11 Science Park
Road, Singapore Science Park 11, 117685, Singapore
*
wesley.sacher@mpi-halle.mpg.de
Abstract:
We present passive, visible light silicon nitride waveguides fabricated on
≈
100
μ
m
thick 200 mm silicon wafers using deep ultraviolet lithography. The best-case propagation losses
of single-mode waveguides were
≤
2.8 dB/cm and
≤
1.9 dB/cm over continuous wavelength
ranges of 466-550 nm and 552-648 nm, respectively. In-plane waveguide crossings and multimode
interference power splitters are also demonstrated. Using this platform, we realize a proof-of-
concept implantable neurophotonic probe for optogenetic stimulation of rodent brains. The probe
has grating coupler emitters defined on a 4 mm long, 92
μ
m thick shank and operates over a
wide wavelength range of 430-645 nm covering the excitation spectra of multiple opsins and
fluorophores used for brain stimulation and imaging.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
As foundry-fabricated silicon nitride-on-silicon (SiN-on-Si) photonic platforms on 200 mm and
300 mm substrates for telecommunication wavelengths have rapidly matured in the past several
years [1–4], the opportunity opens to consider extending the manufacturing technology of the
SiN waveguides to the visible spectrum. SiN is CMOS-compatible and exhibits broadband
transparency that, in principle, extends into the visible spectrum. Visible light integrated
photonics can address new applications in atomic physics and quantum information, fluorescence
excitation and sensing, optogenetics, and imaging and display. However, challenges in fabrication
and characterization arise in realizing integrated photonics devices in the visible spectrum when
compared to telecommunication wavelengths near 1310 nm or 1550 nm. First, the waveguide
and device dimensions are smaller, especially in the blue-end of the spectrum, to maintain the
single-mode or few-mode condition. Second, the mode confinement in the waveguide is also
higher at short wavelengths, which leads to higher sensitivity to surface roughness scattering and
tighter fabrication tolerances. Therefore, low-loss waveguides require excellent control over the
dimensions, sidewall and surface roughness, and material absorption. Beyond the fabrication,
the lack of swept-wavelength laser sources across the entire visible spectrum may also limit the
comprehensive characterization of fabricated devices.
#375224
https://doi.org/10.1364/OE.27.037400
Journal © 2019
Received 26 Aug 2019; revised 6 Nov 2019; accepted 11 Nov 2019; published 11 Dec 2019
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37401
To date, single-mode visible light waveguides have been demonstrated using SiN [5–7] and
alumina (Al
2
O
3
) [8,9] on 200 mm or 300 mm Si wafers, and using aluminum nitride (AlN) with
chip-scale fabrication [10]. However, in all of these demonstrations, waveguide losses were only
reported at discrete wavelengths, and complete, continuous waveguide loss spectra were not
obtained. In [5] on 200 mm wafers, single-mode SiN waveguides formed by plasma enhanced
chemical vapour deposition (PECVD) exhibited a loss of about 1 dB/cm at a wavelength of 532
nm. In [8,9], SiN waveguide losses
≤
20.7 dB/cm and Al
2
O
3
waveguide losses
≤
1.6 dB/cm
were observed at 4 discrete wavelengths between 405 nm and 458 nm; losses as low as 4.8 and
0.6 dB/cm were observed at 458 nm for the SiN and Al
2
O
3
waveguides, respectively. Although
SiN generally exhibited higher losses than Al
2
O
3
in that demonstration, the CMOS compatibility
and fabrication maturity of SiN makes it a promising material for further development for visible
light integrated photonics platforms.
In this article, we present low-temperature PECVD and high-temperature low pressure chemical
vapour deposition (LPCVD) SiN waveguides with SiO
2
cladding formed on 200 mm Si wafers
using the Advanced Micro Foundry (AMF) foundry process and their application in implantable
neuroprobes. The loss spectra of the waveguides were fully characterized in the visible spectrum
using a supercontinuum light source for both orthogonal polarizations. Propagation losses
≤
2.8 dB/cm and
≤
1.9 dB/cm were observed for the best performing single-mode PECVD
SiN waveguides over 466-550 nm and 552-648 nm wavelength ranges, respectively. We also
demonstrate waveguide crossings and power splitters based on multimode interference (MMI)
couplers. The waveguide losses are sufficiently low to be suitable for further demonstrations
and developments of visible light photonic integrated circuits on Si substrates. Finally, as an
example application, we demonstrate implantable neuroprobes for optogenetic stimulation using
the PECVD SiN waveguides.
2. Waveguide geometry and fabrication
The cross-section of the SiN waveguides is shown in Fig. 1(a). A thin SiN waveguide layer
with SiO
2
cladding is defined above a bulk Si substrate. The SiN, bottom cladding, and top
cladding thicknesses are
t
SiN
,
t
clad
,
bot
, and
t
clad
,
top
, respectively.
t
SiN
is chosen to be thick enough
for moderate to high optical confinement across the visible spectrum while thin enough for
single-mode operation at blue wavelengths with waveguide widths
>
200 nm, which are attainable
with deep ultraviolet (DUV) lithography.
t
clad
,
bot
and
t
clad
,
top
are thick enough for negligible
absorption by the Si substrate or absorbing material in contact with the superstrate.
Three variations of SiN waveguides are explored in this work, and one wafer was characterized
for each, which are referred to as Wafer 1, Wafer 2, and Wafer 3. The SiN type and the thicknesses
of the waveguide layer, top cladding, and bottom cladding are summarized in Table 1.
Table 1. Variations of SiN waveguides fabricated
SiN Type
t
SiN
t
clad,top
t
clad,bot
Wafer 1
PECVD
200 nm
1.2
μ
m
1.48
μ
m
Wafer 2
PECVD
135 nm
1.55
μ
m
1.48
μ
m
Wafer 3
LPCVD
200 nm
1.2
μ
m
1.48
μ
m
The waveguides and devices were fabricated on 200 mm diameter Si wafers. The SiO
2
bottom
cladding and SiN waveguide layer were deposited first. Fully-etched SiN waveguides were formed
by DUV lithography and reactive-ion etching (RIE). The SiO
2
top cladding was deposited and
deep trenches were etched to form edge couplers. Chemical mechanical planarization (CMP)
was used to planarize layers. Finally, backgrinding was used to thin the wafers to
≈
100
μ
m
similar to the procedure in [11]. Additional wafers were fabricated with
≈
50
μ
m thicknesses,
however, at the time of writing, they have not been fully characterized. The whole-wafer thinning
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37402
Fig. 1.
(a) Schematic of the SiN waveguide platform.
t
SiN
,
t
clad
,
bot
, and
t
clad
,
top
are
the SiN, bottom SiO
2
cladding, and top SiO
2
cladding thicknesses, respectively. Wafer
backgrinding is used to thin the wafers for the neurophotonic probe application. (b) Cross-
section transmission electron micrograph (X-TEM) of a single-mode waveguide from Wafer
1. (c) Simulated mode profiles at wavelengths (
λ
) 488 nm and 633 nm of waveguides from
Wafers 1 and 2. The electric field magnitudes
|
E
x
|
and
|
E
y
|
are shown for the TE and TM
polarizations, respectively.
is optional but was carried out for the purpose of testing the fabrication process of implantable
neuroprobes for optogenetic stimulation and functional optical imaging [12,13]. This etching
before grinding technique (auto-dicing) also separated the dies on the grinding tape since test dies
were surrounded on all sides by deep trenches. A cross-section transmission electron micrograph
(X-TEM) of a fabricated 270 nm wide SiN waveguide from Wafer 1 is shown in Fig. 1(b).
X-TEMs and cross-section scanning electron micrographs of a small number of dies from Wafers
1-3 confirmed the SiN thicknesses.
The measured PECVD and LPCVD SiN refractive indices monotonically decreased from
1.82-1.79 and 2.02-1.98, respectively, over a wavelength range of 450-648 nm. Mode profiles
and single-mode cutoff widths were calculated using a finite difference eigenmode solver with
the nominal SiN thicknesses and measured refractive indices. The widths for the single-mode
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37403
condition are summarized in Table 2. We defined the cutoff condition as the largest waveguide
width (to the nearest 10 nm increment) where the calculated effective indices of all higher order
modes were
<
10
−
4
above the cladding index.
Table 2. SiN refractive indices and single-mode cutoff waveguide widths
Refractive index
λ
=
450 nm
λ
=
488 nm
λ
=
532 nm
λ
=
633 nm
λ
=
450
−
648 nm
Wafer 1
1.82 - 1.79
340 nm
390 nm
440 nm
580 nm
Wafer 2
1.82 - 1.79
410 nm
470 nm
540 nm
>
700 nm
Wafer 3
2.02 - 1.98
240 nm
270 nm
320 nm
410 nm
3. Measurement results
To measure waveguide loss and device loss spectra across the visible spectrum, the measurement
setup in Fig. 2(a) was used. A 20 W supercontinuum laser (Fianium WhiteLase SC480-20) was
coupled to a narrowband tunable optical filter (Photon Etc. LLTF Contrast VIS HP20). The
full-width-at-half-maximum (FWHM) linewidth of the filter was
<
2 nm over a wavelength
range from 430 to 648 nm. The free-space output of the filter was passed through a polarizer
and coupled to a polarization-maintaining (PM) fiber (Nufern PM460-HP). The axis of the
polarizer was aligned to the slow-axis of the PM fiber, and the opposite end of the PM fiber was
cleaved for edge-coupling to the photonic chip. This end of the fiber was mounted in a 5-axis
micromanipulator with a fiber rotation mount for aligning the fiber to on-chip edge couplers
[Fig. 2(b)] and aligning the slow-axis of the fiber with the principal axes of the chip for either
transverse-electric (TE) or transverse-magnetic (TM) light injection. Prior to chip measurements,
a free-space polarizer was placed in front of the fiber facet to verify the polarization extinction
ratio was
>
20 dB over a wavelength range of 430-648 nm and to identify the angle of the fiber
slow axis within the rotation mount. A cleaved single-mode (SM) fiber (Nufern 460-HP) was
coupled to edge couplers on the output facet of the chip, Fig. 2(c), and the fiber was connected to
an optical detector for detection of output light from the chip.
Transmission spectra of on-chip waveguides and devices were collected under computer control
by stepping the center wavelength of the tunable optical filter in 2 nm steps and measuring the
fiber-coupled output power of the chip for each wavelength. Simultaneously, a second detector
measured the tapped input power to the chip to verify the input optical power did not drift, as
shown in Fig. 2(a). The wavelength range of measurements was fixed to 430-648 nm. The lower
end was limited by the single-mode cut-off wavelengths of the fibers (specified as 410
±
40 nm
and 430
±
20 nm for PM460-HP and 460-HP, respectively). Single-mode operation of the fibers
between 430-450 nm is not guaranteed, however, reasonable fits and reproducibility in cutback
measurements indicates multimode behaviour was not significant.
Tapered edge couplers were used for fiber-to-chip coupling on all devices in the platform.
The waveguide width was 5.2
μ
m at the chip facet and narrowed over a 400
μ
m length to a
single-mode waveguide width. Coupling efficiencies for the edge couplers were measured for
Wafers 1-3 by measuring the transmission spectrum of a straight waveguide with edge couplers at
each facet and normalizing to the measured power at the input fiber facet. The loss of the 1.798
mm long straight waveguide between the edge couplers was not de-embedded from the edge
coupler loss. The measured coupling efficiencies are shown in Fig. 3. For the TE polarization,
the edge coupler coupling efficiency was
−
7.9 to
−
9.8 dB/facet,
−
8.0 to
−
8.9 dB/facet, and
−
8.5
to
−
11.3 dB/facet for Wafers 1, 2, and 3, respectively, over a 430-648 nm wavelength range. For
the TM polarization, the edge coupler coupling efficiency was
−
7.5 to
−
8.9 dB/facet,
−
6.9 to
−
7.9 dB/facet, and
−
8.1 to
−
10.2 dB/facet for Wafers 1, 2, and 3, respectively, over a 430-648 nm
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37404
Fig. 2.
(a) Schematic of measurement setup. Filtered and polarized light from a supercon-
tinuum laser is coupled into a polarization-maintaining (PM) fiber and edge-coupled to the
photonic chip. Output light is collected by a single-mode (SM) fiber edge-coupled to the
opposite facet of the chip and detected. (b) Optical micrograph of one of the test chips with
waveguide loss cutback structures showing the input and output fibers edge-coupled to the
chip. (c) Output facet of the chip in (b) with different wavelength settings of the optical filter.
wavelength range. The TM coupling efficiency was higher than the TE efficiency for all wafers
due to the lower optical confinement of the fundamental TM mode and better mode match to
the
≈
3.5
μ
m mode field diameter of the fiber. The higher efficiency of the Wafers 1 and 2 edge
couplers was due to the lower refractive index and resulting reduced optical confinement of the
PECVD SiN compared to the LPCVD SiN of Wafer 3. The use of two different fiber types for the
input and output coupling in the edge coupler measurements is justified by the similar mode field
diameters of the fibers. To confirm the accuracy of the measurements, additional measurements
were performed with only 460-HP fiber used for both input and output coupling and optical input
from a 473 nm diode laser. The edge coupler efficiencies from these measurements and those in
Fig. 3 agree to within 0.2, 0.1, and 0.7 dB/facet for Wafers 1-3, respectively. The error in these
measurements are limited by fiber alignment to
<
0.2 dB.
3.1. Waveguide loss
Waveguides losses were measured using the cutback method. Microscope images of a subset
of the cutback structures are shown in Figs. 2(b) and 2(c). 5 cutback structures were used for
each waveguide loss measurement with lengths 0, 1.5, 3, 5.4, and 6.24 mm or 0, 1.5, 3, 6,
and 7.2 mm, relative to the shortest structure. The cutback structures used large 80
μ
m radius
bends. Waveguides of different widths, both single-mode and multimode over the majority of the
visible spectrum, were measured on 4 dies from both Wafers 1 and 3 and 3 dies from Wafer 2.
Both TE and TM polarization losses were measured by rotating the input PM fiber to select the
input polarization. The cutback structures for the multimode waveguides had a short length of
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37405
Fig. 3.
Measured coupling efficiency per facet for edge couplers from (a) Wafer 1, (b) Wafer
2, (c) Wafer 3 for TE and TM polarizations.
single-mode waveguide following the input edge coupler to strip off higher order modes injected
by the input fiber.
Figure 4(a) shows an example of the measured optical power spectrum from a set of cutback
structures on Wafer 1. Since this plot is the raw fiber-coupled output power from the chip,
much of the spectral dependence of the power is due to the spectral power distribution of the
supercontinuum laser. Figure 4(d) shows linear fitting of the losses of the cutback structures versus
relative waveguide length at multiple wavelengths; the goodness of fit is high with
R
2
>
0.97.
Waveguide loss spectra extracted via the cutback method are shown in Figs. 4–6 for the TE
and TM polarizations. Waveguide widths were measured using X-TEM and X-SEM imaging.
Single-mode waveguides at red, green, and blue wavelengths as well as multimode waveguide
were measured. For Wafer 1, the simulated single-mode cutoff wavelengths are
<
430 nm and
610 nm for the 280 nm and 540 nm wide waveguides, respectively. For Wafer 2, the single-mode
cutoff wavelengths are
<
430 nm,
<
430 nm, and 525 nm for the 270 nm, 340 nm, and 520 nm
wide waveguides, respectively. Finally, for Wafer 3, the single-mode cutoff wavelengths are 470
nm, 515 nm, and
>
650 nm for the 250 nm, 290 nm, and 520 nm wide waveguides, respectively.
Ripple occurs throughout the 520-540 nm wide waveguide loss spectra and at short wavelengths
close to and below the single-mode cutoff wavelength for the narrower waveguides. As explained
in the Appendix, this is most likely due to excitation and interference of higher order modes. This
ripple and fiber alignment error are the primary sources of error in the cutback measurements and
are quantified in the standard errors listed below. To reduce alignment error, the fiber alignment
process was automated. For Wafer 2, spectra were terminated at wavelengths above which the
optical confinement became sufficiently low for bend losses and substrate absorption to prevent
an accurate linear fit of the cutback structure losses.
The waveguide loss measurements are summarized in Table 3. For all wafers, the losses
decreased with increasing waveguide width due to reduced modal overlap with the etched
sidewalls. Wafer 2 showed the lowest losses at blue and green wavelengths between 430-550
nm likely due to the thinner SiN compared to Wafers 1 and 3, which results in a lower modal
overlap with the etched sidewalls. For Wafer 2, the average waveguide losses for the single-mode
340 nm wide waveguide from 466-500 nm were
≤
4.1 dB/cm and
≤
2.8 dB/cm for the TE and
TM polarizations, respectively. From 502-550 nm, the average waveguide losses were
≤
3.3
dB/cm and
≤
2.3 dB/cm for the TE and TM polarizations, respectively. The blue-green (466-550
nm) single-mode waveguide losses (280 nm width) on Wafer 1 were
≤
8.8 dB/cm. Wafer 3
generally exhibited the highest losses at blue and green wavelengths. This may be due to the
higher refractive index of the LPCVD SiN causing increased sidewall scattering. This may also
be due to the absorption mechanism noted in [9] at wavelengths close to and below 400 nm.
However, since the exact roughness of each wafer was not quantified, variations in roughness
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37406
Fig. 4.
Wafer 1 waveguide loss cutback measurements. (a), (d) Example measurement
showing (a) output power spectra of the cutback structures (L
0
-L
4
) for 280 nm wide
waveguides from Die 4 (TM polarization) and (d) the corresponding linear fits of the output
power versus length showing extraction of the waveguide losses at 488 nm, 532 nm, and
633 nm wavelengths. (b), (e) Waveguide losses of 280 nm wide waveguides for (b) TE-
and (e) TM-polarized light. (c), (f) Losses of 540 nm wide waveguides for (c) TE- and (f)
TM-polarized light. The red waveguide loss traces are averages (Avg.) of Dies 1-4. The
median (90th percentile) of the standard errors of the Avg. traces over the measurement
wavelength range are (b), (e) 0.7-0.8 (1.3-1.4) dB/cm and (c), (f) 0.4-0.5 (0.7-0.9) dB/cm.
between the wafers may also contribute to this observation. At yellow and red wavelengths
>
552
nm, the differences in waveguide loss between the wafers was marginal. The best performing
single-mode waveguides from 552-648 nm were the 520 nm wide Wafer 2 waveguides with TE
polarization losses
≤
1.9 dB/cm; closely followed by the Wafer 3 - 290 nm wide and Wafer 1 - 280
nm wide waveguides with TM polarization losses
≤
3.4 dB/cm and
≤
4.4 dB/cm, respectively.
The thin PECVD SiN on Wafer 2 appears to provide a significant waveguide loss advantage
over Wafers 1 and 3 at blue and green wavelengths. However, this comes at the expense of
reduced optical confinement. Designs requiring maximum optical confinement may benefit from
the parameters of Wafers 1 or 3 and mitigate waveguide loss by using only short lengths of
single-mode waveguides and long lengths of multimode waveguides.
The waveguide loss standard errors averaged over the measurement wavelength range for each
individual waveguide loss spectrum in Figs. 4–6 ranged from 0.3 - 1.3 dB/cm, 0.3 - 0.9 dB/cm, and
0.3 - 1.5 dB/cm for Wafers 1-3, respectively. The average values are stated because the cutback
structure transmission spectra ripple discussed in the Appendix leads to wavelength-dependent
standard errors. The ripple is not correlated across dies and the average waveguide loss traces have
less wavelength-dependent standard errors, which is evident in the median and 90th percentile
(over wavelength) standard errors listed in the captions of Figs. 4–6.
Bend losses were also measured with cutback structures with increasing numbers of 90
◦
waveguide bends. The extracted loss per 90
◦
bend included the bend mode radiation losses, the
mode overlap losses between the bend mode and input/output straight waveguide modes, and
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37407
Fig. 5.
Wafer 2 waveguide loss cutback measurements for (a),(d) 270 nm, (b),(e) 340 nm,
and (c),(f) 520 nm wide waveguides. The top row (a)-(c) is for the TE polarization, and the
bottom row (d)-(f) is for the TM polarization. The median (90th percentile) of the standard
errors of the Avg. traces over the measurement wavelength range are 0.5 (0.6-0.8) dB/cm.
Fig. 6.
Wafer 3 waveguide loss cutback measurements for (a),(d) 250 nm, (b),(e) 290 nm,
and (c),(f) 520 nm wide waveguides. The top row (a)-(c) is for the TE polarization, and the
bottom row (d)-(f) is for the TM polarization. The median (90th percentile) of the standard
errors of the Avg. traces are (a),(b),(d),(e) 0.5-0.7 (1.3) dB/cm, (c),(f) 0.6-0.7 (0.9) dB/cm.
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37408
Table 3. Summary of average waveguide (Wg.) losses for TE and TM polarizations (Pol.) over
various wavelength (
λ
) ranges
a
Wavelength span truncated to 502-540 nm
b
Wavelength span truncated to 552-570 nm
Research Article
Vol. 27, No. 26 /23 December 2019 /
Optics Express
37409
waveguide propagation losses. For Wafer 1, 20
μ
m radius bends had
≤
0.05 dB/bend for 280 nm
wide waveguides over the wavelength range 430-592 nm for TE and TM polarizations. Above
this wavelength range, the bend losses increased significantly due to reduced optical confinement.
At longer wavelengths, wider waveguides closer to the single-mode cutoff are more appropriate
for small bends, e.g., 20
μ
m radius bend losses with 340 nm wide waveguides were
≤
0.05
dB/bend for wavelengths between 450-648 nm and 450-620 nm for the TE and TM polarizations,
respectively. For Wafer 2, 20
μ
m radius bend losses for 340 nm wide waveguides were
≤
0.06
dB/bend from 430-522 nm for the TE polarization and
<
0.3 dB/bend from 430-490 nm for the
TM polarization. At the same waveguide width, 80
μ
m radius bend losses were
≤
0.06 dB/bend
from 430-648 nm and 430-600 nm for the TE and TM polarizations, respectively. For Wafer 3,
20
μ
m radius bend losses were
≤
0.05 dB/bend for a 290 nm waveguide width from 450-648 nm
for TE and TM polarizations. The standard errors on the measurements were
<
0.02 dB/bend for
Wafers 1 and 2 and
<
0.03 dB/bend for Wafer 3 over the measurement wavelength range.
3.2. Waveguide crossing
We used the waveguides to realize several devices that are useful for photonic circuits in the
visible spectrum. The first is a waveguide crossing using multimode interference (MMI). In-plane
MMI waveguide crossings [14] were designed using finite difference time domain (FDTD)
simulations. Crossings were fabricated and measured on each wafer, and an optical microscope
image of one of the crossings is shown in Fig. 7(c). An example simulated top-down electric
field profile is shown in Fig. 7(b) showing the operation of the device; interference between
TE0/TM0 and TE2/TM2 modes excited in the multimode section reduces optical overlap with
the discontinuity created by the intersecting waveguides. The same design was used for all
wafers with the input and output waveguide widths
w
IO
=
300 nm, taper length
L
taper
=
3
μ
m,
multimode section length
L
MM
=
6.4
μ
m, and multimode section width
w
MM
=
1
μ
m. The
crossing design dimensions are defined in Fig. 7(a).
Crossing losses were measured using cutback structures with 1, 15, 30, and 45 crossings.
Figure 7(d) shows example linear fits of the cutback structure output power versus the number of
crossings at multiple wavelengths.
R
2
>
0.9 for all crossing cutback measurements over the full
measurement wavelength range. The standard error of all linear fits was
≤
0.02 dB/crossing for
all measurements. Crossing loss spectra are shown in Fig. 8. The minimum loss per crossing
ranged from 0.07-0.09 dB/crossing, the loss was
<
0.1 dB/crossing for at least a 58 nm bandwidth
in all cases, and the loss was
<
0.2 dB/crossing for at least a 138 nm bandwidth in all cases.
The optical crosstalk of waveguide crossings was measured by sending light into Input 1 in
Fig. 7(c), measuring the power at the crosstalk port (Output 2), and normalizing to the measured
power at the thru port (Output 1). The analagous procedure was performed for Input 2 generating
two crosstalk measurements for each polarization as shown in Fig. 9. The crosstalk was
<
−
20
dB in all cases over the full measurement bandwidth. Over the 0.1 dB-bandwidth of the crossings,
the maximum measured crosstalk was
−
26(49) dB,
−
21(47) dB, and
−
27(32) dB for Wafers 1-3,
respectively, for the TE(TM) polarization.
3.3. MMI
1
×
2
power divider
A second related device is a 1
×
2 MMI power divider [15]. The devices were designed using
FDTD simulations, fabricated, and measured on all three wafers. An optical micrograph of one of
the MMI power dividers is shown in Fig. 10(a). Figure 10(b) shows an example top-down electric
field intensity profile of light propagating through the MMI divider. The design parameters are:
aperture width
w
ap
, output aperture spacing
d
ap
, multimode section length
L
MMI
, and multimode
section width
w
MMI
. Three designs were tested. Design 1 (
w
ap
=
360 nm,
d
ap
=
760 nm,
L
MMI
=
4.74
μ
m,
w
MMI
=
1.52
μ
m) is a blue wavelength design for Wafers 1 and 3, Design
2 (
w
ap
=
360 nm,
d
ap
=
760 nm,
L
MMI
=
4.14
μ
m,
w
MMI
=
1.52
μ
m) is a yellow wavelength