MMI devices are required in order to separate the
lobes. With this in mind, 50 µm wide MMIs were
studied in this work. Since these devices behavior is
highly dependent on width and femtosecond laser
direct writing has a finite resolution, it is to be
expected that while doing N writings with a given
waveguide separation S the total width is larger than
NxS. To compensate the final devices were written
with 47.5 m.
Another variable that can change the device
performance is the refractive index distribution
inside the multimode waveguide. The BPM
simulations made in this work to support the device
fabrication were obtained with a uniform refractive
index distribution, and, as such, the separation
between fabricated waveguides was controlled in
order to obtain, as close as possible, the simulations
refractive index distribution. In figure 4 longitudinal
and transversal MMI devices with a waveguide
separation of 1 µm are displayed. From the figure it
is possible to observe that with a waveguide
separation of 1 µm the modification is not uniform
since individual writing tracks are still observable.
Due to this fact the separation between fabricated
waveguides was reduced to 0.5 µm.
Figure 4: Dark field microscopy image of a longitudinal
(a) and transversal (b) MMI device fabricated with a
waveguide separation of 1 µm.
To fabricate power splitters with multimode
waveguides three values are required, namely
multimode waveguide width, length, and working
wavelength (device refractive index is also
important but not critical). The width was set to 50
µm, working wavelength to 1550 nm, but device
length is an unknown for the different splitting
ratios. To determine the device length BPM
simulations were made with Rsoft (see figure 5 (a)
to (d)) using a 2D model and a 7x10
-3
refractive
index difference. In this work 1x4, 1x3 and 1x2
power splitters were investigated and the device
length was found to be, through simulation, roughly
690, 900 and 1350 µm, respectively. With the
fabrication parameters set above, several MMI
devices were fabricated using the three designs
specified in section 2. All these designs were tested
in an effort to improve the device symmetry and
decrease stresses originated by the fabrication
process. The first longitudinal design and the
transversal design showed very similar results in
terms of modal distribution (first longitudinal design
results are displayed in figure 5 (e) to (g)), while the
second longitudinal design did not show the
simulated behavior but rather random distributions.
It should be noted that the CCD from which the
modal distributions were measured does not have a
uniform sensitivity over its area, explaining why
unequal distribution is observed.
Figure 5: Figure showing the simulated behaviour and the
results obtained with the first longitudinal design for 50
µm width devices. First the 1:1 MMI behaviour is
simulated (a) and the 1:4 (b), 1:3 (c) and 1:2 (d)
simulations obtained for a length of 690, 900 and 1350 µm
respectively. From these simulated lengths the 1:4 (e), 1:3
(f) and 1:2 (g) behaviour was obtained experimentally.
These results can be explained by the microscope
images in figure 6. From the top view images, in
bright and dark field, it is possible to see that the
second longitudinal design is not as uniform as the
others. In the cross section view this becomes
obvious since the guiding region is much more
irregular. Apart from this, it is also interesting to
notice that the other two designs have problems. In
the first longitudinal design all MMI devices
fabricated had a crack in the corner of border B (last
waveguide to be written). This probably happens
due to stress accumulation on this corner, resulting
from the fact that waveguides were written from
border A to B. On the transversal design some
problems can also be identified by the dark field
image. This design was implemented since it avoids
the stress build-up but another problem arises due to
hardware communication times. At border B more
light is visible than at border A due to the laser being