2.1.

Alignment Concept

The alignment device consists of eight V-grooves to mount the eight fibers and three V-grooves for the alignment. The width of all V-grooves was chosen to be the same. Essentially, the device could be small ($<10{\mathrm{mm}}^2$), but for ease of handling we chose the size of the device to be $10\mathrm{mm}\times 5\mathrm{mm}$. On the corresponding waveguide board, depicted in Fig. 1(b), there are three alignment structures and eight waveguides arranged matching to the V-grooves on the silicon device.

When the silicon device, which holds the fibers, and the waveguide board are assembled together, the silicon surface is in contact with the bottom cladding of the waveguide board. This provides accurate vertical alignment. By choosing the right width of the V-groove, the height of the fiber in front of the waveguide can be adjusted. For optimal coupling, the center of the fiber has to be matched to the center of the waveguide (compare Fig. 2).

The lateral position of the fiber is adjusted by alignment structures, which define the position of the alignment grooves. To achieve maximum position precision and accuracy, the alignment structures are fabricated in the same processing step as the waveguide cores. A top-cladding structure was added to enable precentering during assembly (see left side of Fig. 2).

In the axial position, the distance between the fiber facet and the waveguide facet has to be minimized to achieve maximum coupling efficiency. The fiber facets and the waveguide facets act as a mechanical stop. By pushing the silicon device and the board against each other, a minimal gap between the fiber and waveguide facets can be achieved.

2.2.

Simulation of the Fiber-Coupling Characteristics

To determine the necessary position accuracy, the coupling characteristics of a $9/125\mu \mathrm{m}$ single-mode fiber (e.g., SMF-28) to an integrated polymer waveguide have been investigated using beam propagation method simulations. The integrated waveguides exhibit a refractive index difference $\mathrm{\Delta }n=0.006$, a width of $5\mu \mathrm{m}$, and a height of $5\mu \mathrm{m}$. The simulations reveal that a transition loss of 0.18 dB has to be expected due to the mismatch of the modal fields of fiber and waveguide.

Assuming a displacement loss of up to 0.5 dB as acceptable, a placement error of up to $1.5\mu \mathrm{m}$ can be tolerated, according to Fig. 3. The investigation of angular errors revealed that a tilt angle of up to 1.2 deg can be tolerated. Because there is a large overlap between the PLC board and the silicon device, the tilt errors are expected to be small. Considering an overlap length of 4 mm, a tilt angle of 1 deg would result in an offset of $70\mu \mathrm{m}$. This is large compared to the necessary position accuracy of $1.5\mu \mathrm{m}$. Thus, tilt errors can be neglected.

Fig. 3

Displacement dependency of the fiber-to-waveguide coupling efficiency.

3.1.

Design and Fabrication of the Silicon Device

To fabricate the silicon device, we used standard MEMS fabrication processes based on photolithography and potassium hydroxide (KOH) etching. The V-grooves were formed by {111} surfaces, allowing the precise fabrication of V-grooves with a desired width.⁹ Assuming a waveguide height $H$ of $5\mu \mathrm{m}$ and a fiber diameter $F$ of $125\mu \mathrm{m}$, the center of the fiber has to be positioned $2.5\mu \mathrm{m}$ above the bottom cladding level (see Fig. 4). Considering the angles of the {111}-type surfaces, a V-groove width $V$ of $156.7\mu \mathrm{m}$ results. To obtain V-grooves with an accurate width, all etching processes have to be tightly controlled to avoid over etching. In preceding tests, the occurring undercut, which widens the V-grooves, has been measured and subsequently compensated for on the photomask.

Fig. 4

Cross section of the coupling system consisting of the silicon V-groove and the fiber; the waveguide is indicated as a gray square.

3.2.

Mounting of the Fibers onto the Silicon Device

To finalize the silicon device, the fibers (Corning $\mathrm{SMF}\text{-}28\mathrm{e}+$) have to be mounted into the respective V-grooves (see Fig. 5). For low-loss coupling of multiple fibers, all fiber facets have to be aligned coplanar. One way to achieve this is to cleave the fibers and mount the fibers with an accurate $z$-position onto the silicon device. A disadvantage of this approach is that the position of each fiber has to be adjusted individually. Additionally, a very precise amount of adhesive has to be applied to bond the fibers to the silicon device without voids while avoiding wetting of the fiber facets. To avoid these issues, we chose to adjust the lengths of the fibers after the adhesive bonding using a wafer dicer. By cutting a notch [see Fig. 1(a)], the resulting fiber facets are arranged in one single line and protruding adhesive is removed. The resulting fiber facets provide the mechanical stop for the facet of the waveguide board; therefore, the actual position of the notch does not need to be very accurate as long as it is perpendicular to the V-grooves.

While one fiber after another was fed into the V-grooves, a UV transparent glass lid was used to clamp the fibers down into the grooves. To attach the fibers to the silicon device, a UV curable adhesive was applied into the strain-relief trench. By capillary forces, the adhesive was drawn into the fiber equipped V-grooves, effectively filling up the gaps between the fibers and the silicon device. After the curing step with UV-light, the fibers were cut to length. Thus, all eight fiber facets were precisely arranged in one line and excess adhesive was removed. The applied cutting process produced fiber facets with an RMS roughness of 75 to 100 nm.

3.3.

Fabrication of the Waveguide Board

The waveguides used in our device were fabricated similar to Ref. 8. As a first step, a bottom cladding layer was deposited onto a glass substrate and cured by a UV flood exposure. As a second step, a layer of core material was applied. To structure the core layer, the material was cured using a laser direct imaging machine followed by removing the uncured material by a development step. Subsequently, a layer of cladding material was applied as top cladding. Using the laser direct writing setup equipped with a second laser, featuring a beam diameter of $45\mu \mathrm{m}$, and a subsequent developer rinse, all waveguides were covered with a line of cladding material. Additionally, a support structure was added to the alignment structures (see Fig. 6).

Fig. 5

Photography of the silicon device equipped with eight fibers.

Fig. 6

Bottom illuminated microscopy picture of a cross section of an alignment structure.

3.4.

Assembly of the Coupling System

To achieve optical coupling between the fibers on the silicon device and the integrated waveguides, the two parts have to be joined. When the joining is done manually, there is a tactile feedback when the silicon device snaps correctly into place. Additionally, interference fringes can be observed when viewing through the glass substrate. This confirms close proximity between the silicon device and lower cladding. To attach the silicon device to the PLC board, the two parts were pressed together and UV curable adhesive, which allows a quick bonding, was applied. Additionally, the adhesive acts as index-matching agent in the gap between waveguide and fiber facet. In Fig. 7, a photograph of an assembled system is shown.

Fig. 7

Assembled connection of eight fibers to eight waveguides.

4.1.

Microscopy Analysis of the Assembled System

In the first step, a microscopy analysis was done to characterize the joining process and the passive alignment of the two devices. The microscopy analysis on cross sections using bottom illumination confirmed the proper alignment of the silicon device and the waveguide (see Fig. 8).

Fig. 8

Cross section of an alignment structure fitting into a V-groove.

The lateral deviation of the silicon device to the alignment structures is in the range of about $1\mu \mathrm{m}$ or less. One can also observe that the device is closely attached to the waveguide board without a gap between the bottom cladding and the devices surface (Fig. 8). Figure 9 shows that the fiber fits tightly to the V-groove and therefore has the desired linear support.

Fig. 9

Cross section of a silicon device with mounted fiber.

Microscopy analysis also allowed the measurement of the waveguide geometry and the width of the V-groove. The waveguides have a height of $5\mu \mathrm{m}$ and a width of $5\mu \mathrm{m}$. The V-grooves are about $157\mu \mathrm{m}$ wide.

4.2.

Characterization of the Optical Path

The coupling losses caused by the passively coupled interface have been identified. For this reason, we compared measurements on actively aligned reference waveguides and passively aligned waveguides. To couple light into the waveguides, an $\mathrm{SMF}\text{-}28\mathrm{e}+$ fiber connected to a 1310-nm diode laser was used as the light source. The stability of the output power was monitored using a fiber beam splitter.

During the reference measurements, the output power of the polymer waveguide has to be measured to determine the transmission properties of the waveguides. A multimode fiber with a core diameter of $105\mu \mathrm{m}$ and an NA of 0.2 coupled to a Newport 818-IS integrating sphere was used as a detector [compare Fig. 10(a)]. To compensate for fiber and coupling losses in the setup, a calibration measurement was performed by measuring the input power of the feeding fiber directly with the multimode detector fiber [Fig. 10(b)].

Fig. 10

Block diagram of the measurement procedure for characterization of the optical link: (a) measurement: reference, (b) calibration: reference, (c) measurement: ribbon, and (d) calibration: ribbon.

In the second step, the passively aligned optical link was measured. During these measurements, the light source remained the same as during the reference measurements. On the output side, the passively coupled $\mathrm{SMF}\text{-}28\mathrm{e}+$ fibers were connected directly to the integrating sphere [see Fig. 10(c)]. Again, a calibration measurement was performed by measuring the waveguide input power by connecting the input fiber directly to the integrating sphere [Fig. 10(d)]. During all measurements, the gaps have been filled with index-matching gel.

The results obtained by the reference measurements (see Table 1) show that the waveguides, with a length of 25 mm, exhibit average transmissions of $-1.69\mathrm{dB}$. When the fiber ribbon is attached to the waveguides, they exhibit transmissions of about $-1.92\mathrm{dB}$. Previous cut-back measurements show that the polymer waveguides exhibit a propagation loss of $0.43\mathrm{dB}/\mathrm{cm}$. Considering the length of the polymer waveguide sample of 25 mm, a propagation loss of 1.08 dB has to be expected. Thus, coupling losses ${\mathrm{CL}}_{\text{Reference}}$ for the reference waveguides and ${\mathrm{CL}}_{\text{Ribbon}}$ for the fiber coupled waveguides can be calculated. Neglecting scattering losses due to surface irregularities, one can assume a lossless coupling between a single-mode waveguide and a multimode fiber during the reference measurement. Thus, a coupling loss of 0.23 dB between single-mode waveguide and single-mode fiber can be determined by subtracting ${\mathrm{CL}}_{\text{Reference}}$ from ${\mathrm{CL}}_{\text{Ribbon}}$.

Table 1

Measurement data of the optical characterization (transmission and coupling loss), including average and standard deviation σ.

4.3.

Climate Tests

After the initial measurements, the sample was stored at room temperature for 14 days, followed by a second transmission measurement, which was performed as reference measurement for thermal cycling. In the cycling test, our sample was heated to 70°C and cooled to $-{30}^{°}\mathrm{C}$ 10 times with a dwell time of 15 min each. The slope of the temperature ramp was $\pm 1\mathrm{K}$, while the humidity was free-floating. After the thermal cycling, a third transmission measurement was performed. Subsequently, the sample was exposed to an 85/85-test. In this testing scheme, the sample was exposed to 85°C with a relative humidity of 85% for 168 h. After finishing the test, a final transmission measurement was performed.

Table 2 shows that the transmission values remained almost unchanged during the storage period and the thermal cycling. During the 85/85-test, the transmission decreased to $-3.13\mathrm{dB}$ on average. As the standard deviation of 1.2 dB after the 85/85-test indicates, the variance of the measured values increased significantly after the 85/85-test. While two of the channels remained almost unchanged at $-1.9$ and $-2.1\mathrm{dB}$ after the 85/85-test, the other channels exhibited transmissions of down to $-5.50\mathrm{dB}$.

Table 2

Results of the aging and thermal cycling of the sample.