3.1.

Bragg Grating Sensor PIC

A Bragg grating implemented in a waveguide reflects a band of wavelengths centered around the so-called Bragg wavelength ${\lambda }_{\mathrm{B}}$, defined by ${\lambda }_{\mathrm{B}}=2n_{\mathrm{eff}}\mathrm{\Lambda }$ (Bragg equation), where $\mathrm{\Lambda }$ is the grating pitch and $n_{\mathrm{eff}}$ is the effective index of the guided mode in the grating section.¹⁰ The effective index was simulated using Ansys Lumerical Mode solutions taking into account the standard transverse electric (TE) Si waveguide thickness (215 nm) and width (450 nm) for the ISIPP50G technology targeted. The Bragg grating was implemented by modulating the waveguide with $\pm 10\mathrm{nm}$ sidewall corrugations, as illustrated in Fig. 3. This results in a change of the effective index, i.e., 2.3271 and 2.3699, for the 440 and 460 nm wide sections, respectively. Considering the expected effective index, grating pitches of 324, 326, and 328 nm were implemented to yield Bragg wavelengths within the C-band (1530 to 1565 nm). Increasing the pitch or effective index results in a Bragg wavelength red-shift according to the Bragg equation.

Fig. 3

(a) Sensor layout: top view and cross-sectional view illustrating layer thicknesses and material refractive indices (RI). (b) Simulated reflection spectrum of a phase-shifted Bragg grating in a TE silicon waveguide having 200 periods (328 nm pitch) of $\pm 10\mathrm{nm}$ sidewall corrugations, a $\pi$-shift section, and then another 200 periods.

The influence of the design parameters (corrugation width, number of periods, pitch, effective index, and fill factor) on the spectral response of the Bragg grating can be studied analytically¹⁷ or using the eigenmode expansion propagation solver (Ansys Lumerical) as detailed in Refs. 13 and 14. In summary, increasing the fill factor has a minor influence on the Bragg wavelength, increasing the number of periods increases peak reflectivity, and increasing the corrugation width increases the bandwidth. Nevertheless, even Bragg gratings with only $\pm 10\mathrm{nm}$ sidewall corrugations, being on the limit of fabrication capabilities, yield a broad peak in the reflection spectrum (bandwidth of several nanometers). Therefore, phase-shifted gratings having a $\pi$-shift section in the center were considered. Figure 3 shows the sensor layout and the simulated reflection spectrum of such a grating having 200 periods of $\pm 10\mathrm{nm}$ sidewall corrugations (328 nm pitch), a $\pi$-shift section, and then another 200 periods. As can be observed, there is a very narrow dip in the middle of the Bragg grating spectrum, enabling more accurate tracking of the central resonance wavelength compared with the broader spectrum envelope. According to this design, the Bragg wavelength is 1544.8 nm and the grating effective index of 2.354.

To interface the Si waveguides having Bragg grating sensors, TE grating couplers were used. The ISIPP50G technology by default offers “regular” and “enhanced performance” TE grating couplers. The latter has improved coupling efficiency for interfacing with a single-mode fiber owing to the improved upward directionality. However, this makes it unsuitable for the current packaging process because of the very bad downward directionality. Therefore, a regular TE grating coupler (nominal design: 625 nm pitch, 50% duty cycle, and 70 nm etch depth) was used; it was optimized to interface with a $10\mu \mathrm{m}$ mode field diameter (MFD) and was found to exhibit a good downward directionality.

The test structures that were implemented on the PIC design consisted of short (e.g., 3 mm) Si waveguide sections having a TE grating coupler on both sides and a phase-shifted Bragg grating sensor toward the middle of the waveguide sections.

After receiving the PICs from the MPW fabrication service, the Bragg grating sensors were characterized by reading them out in reflection, similar to what is done by default for FBGs. Therefore, an superluminescent light emitting diode (Exalos EXS1510-1110) was used as the source and was connected to port 1 of a circulator (Thorlabs 6015-3-APC), and an optical spectrum analyzer (OSA; Agilent 86142B) was connected to port 3. A cleaved single-mode interfacing fiber was connected to port 2 and was actively aligned above the grating coupler on the PIC top surface. Index matching liquid was used to avoid spurious reflections at the fiber–air interface. Figure 4 shows typical recorded reflection spectra for the three different grating pitches that were implemented, clearly illustrating the dip in the reflection spectrum owing to the phase shift section. However, the dip in the reflection spectrum was not exactly at the simulated Bragg wavelength, most likely caused by a different actual effective index. Furthermore, significant differences in the effective index were observed between different PICs (when measured at a constant temperature).

Fig. 4

Typical measured reflection spectra of the phase-shifted Bragg gratings in TE silicon waveguides at three different pitches. The total reflectivity includes grating reflectivity and grating coupler efficiency (two passes). Differences in absolute reflectivity for the different graphs are caused by the spectral response of the TE grating coupler.

In an MPW run, the reticle containing the PIC designs from all customers is imaged at several positions on the wafer using a stepper, so each customer receives PICs fabricated at different positions on the wafer. Assuming the lithographic process is very accurate, and therefore the pitch of the gratings can be considered identical throughout the whole wafer, shifts in Bragg wavelength can only be explained by changes in the effective index. Therefore, the measured Bragg wavelength (at constant temperature) can be used to assess the effective index uniformity over the wafer. Figure 5 shows the measured (at 22°C) Bragg wavelength for several 328 nm pitch phase-shifted Bragg gratings as a function of their position on the wafer. The Bragg wavelength, and hence the effective index, is the highest in the center of the wafer and decreases toward the edge. For a given grating pitch, the difference between the longest (1550.9 nm) and shortest (1535.9 nm) Bragg wavelength measured was 15 nm, which corresponds to a significant difference in the effective index of 0.023.

Fig. 5

(a) Measured Bragg wavelength (in nm) as a function of the designed grating pitch, for several PICs fabricated on different positions of the same wafer. (b) Measured Bragg wavelength (in nm) for 328 nm pitch phase-shifted Bragg gratings as a function of position on the wafer. The PICs positioned at uncolored squares were not measured.

Grating coupler spectra are also sensitive to changes in the effective index. Although clear spectra were obtained (cfr. below), it was noticed that on average all grating coupler spectra were redshifted compared with the default behavior in addition to variations in the center wavelength that were observed over different PICs. Therefore, in the following, larger coupling angles (between 12.5 deg and 15 deg off vertical incidence) than the default 10 deg were used to compensate for this by blue shifting the spectra to better overlap the Bragg grating reflection spectra to achieve higher peak reflectivity values. It was verified that this increase in coupling angle mainly resulted in a blue shift without significantly affecting the coupling efficiency.

3.2.

Optical Interface

A ball lens was used to image the $10.4\mu \mathrm{m}$ (nominally) MFD of the single-mode optical fiber (at back side) onto the TE grating coupler (at the device side). Additionally, the ball lens was laterally shifted with respect to the optical fiber axis to achieve the desired coupling angle for the grating coupler (between 12.5 deg and 15 deg), as illustrated in Fig. 6. A fused silica holder was created to assemble the ball lens at the precise distance with respect to both the fiber tip and PIC back side. Detailed optical interface design considerations and possible variants were published in Ref. 9. The results in this paper were achieved using a $300\mu \mathrm{m}$ sapphire-based ball lens interface and a configuration as illustrated in Fig. 6. Although these grating couplers are designed for top side coupling, they also show acceptable performance for back side coupling, as shown below.

Fig. 6

Schematic illustration of the optical interface for accessing the grating coupler from the back side of the PIC.

To assess the suitability of the ball lens interface, the recorded Bragg grating reflection spectra obtained using a bare fiber (coupling angle 10.6 deg and index matching liquid) and using the ball lens interface (coupling angle about 15 deg), both aligned above the PIC from the device side [see Fig. 7(a)], were compared. It should be noted that the coupling angle was different in both cases because the coupling medium was different (index matching liquid compared to air). As shown in Fig. 7(b), the spectra recorded using both configurations are very similar, indicating that the ball lens interface is well-suited for reading out Bragg grating sensors on a PIC. It is noteworthy that, without index matching liquid, spurious reflections at the cleaved fiber tip heavily distort the reflection spectrum, whereas for the ball lens, which is by default operated in air (no index matching liquid used), this was not observed because the index matching glue and the lateral shift of the ball lens avoid direct back reflections into the fiber. However, the glue interface between the fiber and ball lens is critical because delamination would result in a Fabry–Pérot cavity formed by a thin air gap between the fiber and glue layer, which also results in spurious ripples in the reflection spectrum.

Fig. 7

(a) Configuration under test: a fiber with a ball lens interface mounted on a ferrule was aligned above the TE grating coupler. (b) Comparison of the reflection spectra recorded using a bare fiber and ball lens interface. Red curve: device side coupling using bare fiber + index matching liquid (coupling angle 10.6 deg). Blue curve: device side coupling using ball lens interface (coupling angle 15 deg).

Figure 8 shows the effect of coupling from the back side (using ball lens) compared with coupling from the device side. Because coupling from the back side required polishing, a different PIC was used, which explains the different observed Bragg wavelengths caused by a different effective index, as discussed above. It can be seen that the reflectivity is lower when coupling from the back side because of the lower back side efficiency of standard TE grating couplers and because for this measurement no AR coating was applied on the polished back side of the PIC. As explained below, when applying an AR coating, the peak reflectivity would improve by 1.5 dB per interface, or in total by 3 dB, because the Si–air interface at the back side is passed twice when reading out the sensor in reflection.

Fig. 8

Comparison of reflection spectra when coupling from the top side using bare fiber and index matching liquid (red curve, coupling angle 10.6 deg) and from the back side using ball lens interface (black curve, coupling angle 15 deg). Two different PICs were used for the experiment, explaining the different Bragg wavelengths.

3.3.

PIC Postprocessing

PICs require polishing when interfaced from the bottom side. This was achieved by mounting individual MPW PICs with their device side facing down on a glass carrier using a temporary wax. Several PICs can be mounted simultaneously in a square pattern and preferably as closely spaced as possible. After mounting the PICs, their back side was polished on a Logitech PM5 lapping and polishing workstation using a polishing slurry with ${\mathrm{Al}}_2{\mathrm{O}}_3$ particles having a size of $0.3\mu \mathrm{m}$. After polishing, the roughness was checked using a Veeco NT3300 interferometer in the phase-shifting interferometry mode using monochromatic light. Although after polishing a high-quality optical interface is obtained (nanometer roughness), the high refractive index contrast silicon–air interface yields significant reflections. These can be avoided by applying an AR coating. Here, a single 200 nm silicon nitride layer (refractive index $\sim 2$) was deposited (using plasma-enhanced chemical vapor deposition at 120°C) on the silicon back side to minimize reflections in the C-band.

Finally, the PICs were diced to the proper dimensions (here $970\mu \mathrm{m}\times 970\mu \mathrm{m}$). This was done by mounting individual PICs on UV-release dicing tape, which has a higher tackiness than regular dicing tape. This was needed to ensure sufficient adhesion of the very small (and thick) pieces after the dicing process. Upon UV illumination from the back side, this tape significantly reduces tackiness, making it possible to release the pieces without damage.

Figure 9 shows an interferometric scan of the back side of a PIC after polishing, indicating the nanometer roughness as required. Figure 10 shows coupling efficiencies measured on 3 mm long straight waveguide sections with phase-shifted Bragg grating and two TE grating couplers, when coupling with a ball lens interface (coupling angle 12.5 deg) from the device side and back side respectively and illustrates the improved back side coupling efficiency after applying an AR coating. In all cases, the other grating coupler was interfaced from the device side using a regular single-mode fiber (same coupling angle, i.e., 12.5 deg). It can be seen that the coupling efficiency (for one grating coupler interface) reduces by about 3.7 dB when moving the ball lens interface from the device side to the polished back side. However, after applying the nitride AR coating, the back side coupling efficiency improves by 1.5 dB, so the reduction in efficiency when coupling from the back side is effectively about 2.2 dB. Judging from the improvement in efficiency, the AR coating virtually eliminates all reflections at the silicon–air interface. Because this interface is passed twice for reading out Bragg grating sensors in reflection from the back side, the AR coating results in a total coupling efficiency improvement of 3 dB. Although the AR coating can eliminate the reflections at the back side of the substrate, there are still reflections at the silicon-BOX (buried oxide) interface that cannot be influenced by postprocessing, limiting the back side coupling efficiency. Furthermore, the grating couplers used had limited directionality. We previously showed that adding metal reflectors on top of a grating coupler can drastically improve directionality and further boost back side coupling efficiency.¹⁸

Fig. 9

Interferometric scan of a polished PIC back side illustrating the obtained nanometer roughness.

Fig. 10

Coupling efficiency measurements on 3 mm long straight waveguide sections with a grating coupler on either side and a phase-shifted Bragg grating with Bragg wavelength around 1531 nm. The input grating coupler was interfaced using a single-mode fiber from the device side (coupling angle 12.5 deg) in all cases. (a, red curve) Output grating coupler interfaced from the device side using ball lens interface. (b, blue curve) Output grating coupler interfaced from the back side, without AR coating applied, (c, green) Output grating coupler interfaced from the back side after application of the AR coating. The coupling angle at the output grating coupler was maintained at 12.5 deg in all cases.

Figure 10 also shows the typical signature in transmission of the phase-shifted Bragg grating (Bragg wavelength around 1531 nm), superimposed on the grating coupler spectra.

3.4.

Holder Fabrication

The optical interface was implemented by realizing a precision (according to the design) three-dimensional (3D) holder for the ball lens. The holders were fabricated in fused silica using a femtosecond laser direct-write inscription method followed by an etching step in potassium hydroxide (KOH).¹⁶ The technique involves irradiating the contours of the desired 3D features, so the irradiated fused silica becomes much more susceptible to etching in KOH than the nonirradiated regions. An ytterbium-doped fiber laser having a pulse length $<400\mathrm{fs}$, repetition rate 500 kHz, circular polarization, and frequency-doubled operating wavelength of 515 nm (Satsuma, Amplitude) was used. The beam was focused in nominally $500\mu \mathrm{m}$ thick fused silica substrates (supplier: Siegert Wafer) using a 0.55 numerical aperture aspheric lens (Newport 5722-A-H). After laser irradiation, etching occurred in a 30% KOH solution heated at 85°C.

The holder can be considered a 3D version of V-grooves, which we previously demonstrated using the same laser technology.¹⁹ Both inverted conical and square pyramidal shapes were fabricated to position the ball lens in three dimensions. As shown in Fig. 11(a), the squared pyramid showed cracks at the sharp corners due to excessive internal stress concentrations during the laser inscription step.^20,21 Therefore, the smoother conical holder design was selected. Figure 11(b) shows the glass part that was lifted out after the contour lines were etched, nicely illustrating the negative shape of the holder.

Fig. 11

Ball lens holder: (a) top view of the conical and pyramidal design; (b) glass part lifted out after etching the contours, and (c) side view with mounted $300\mu \mathrm{m}$ ball lens.

The position of the holder (in the depth direction) defines the eventual separation between the fiber tip and the ball lens and between the ball lens and the back side of the PIC. Ideally, a fused silica substrate with a thickness equal to the distance between the fiber and PIC in which the holder is monolithically implemented at the proper depth should be used. Unfortunately, not all fused silica thicknesses are easily available. Therefore, because here a $500\mu \mathrm{m}$ thick fused silica substrate was used, an additional spacer ($\sim 130\mu \mathrm{m}$ thick) was used to obtain the correct total fiber-to-PIC distance. To fabricate a circular opening (diameter $500\mu \mathrm{m}$) in the spacer, the same femtosecond laser direct-write inscription method was used. A substrate with a matrix of several identical holders was realized to accelerate subsequent packaging processes.

3.5.

Sensor Assembly

The multistage process for assembling the sensor required building a semi-automatic assembly station (Fig. 12) ready for sub-micrometer positioning accuracy with the possibility of continuous output signal monitoring. This station supports the epoxy gluing processes, as well as the positioning feedback given by external cameras, detectors, and a spectrum analyzer.

Fig. 12

(a) Schematic view and (b) photo of the implemented assembly station.

The sensor assembly process consists of four steps, as described below.

3.5.1.

Manual insertion of ball lens in holder

This step involves using a substrate with a number of conical fused silica holders, each of which must be filled with epoxy. A spherical lens is then positioned in the center of each holder, after which the epoxy is ready for UV curing (cfr. Fig. 13). After fixing the spherical lenses, the substrate is singulated into individual holders ($970\mu \mathrm{m}\times 970\mu \mathrm{m}$), ready for the next process step.

Fig. 13

(a) Top and (b) side view of the $300\mu \mathrm{m}$ ball lens glued in the holder.

3.5.2.

Distance plate assembly (spacer)

To match the proper distance between the ball lens surface and the PIC back side (see design shown in Fig. 6), an additional spacer was attached to the top side of the holder using UV-curable adhesive. Figure 14(a) shows a top view of this $130\mu \mathrm{m}$ thick glass spacer, and Fig. 14(b) shows a top view of the lens holder with the spacer attached. The slight intentional decentering of the ball lens with respect to the holder outlines can be seen.

Fig. 14

(a) Top view of diced spacer consisting of a $970\mu \mathrm{m}\times 970\mu \mathrm{m}\times 130\mu \mathrm{m}$ glass piece with $500\mu \mathrm{m}$ etched hole in the center. (b) Top view of lens holder with spacer attached.

3.5.3.

Bonding of the holder on top of the optical fiber (inside a ceramic ferrule)

On top of the fiber ferrule, the holder is placed at the correct laterally shifted position relative to the optical axis, according to the design (Fig. 6). Subsequently, the holder is bonded to the ferrule using UV-curable adhesive. When correctly positioned, the ball lens emits at a well-defined angle to interface the grating coupler. This was visually and qualitatively verified by coupling the red light in the fiber and observing the angle of the emitted beam [Fig. 15(b)].

Fig. 15

(a), (b) Side views of the ball lens holder being assembled on top of the fiber ferrule. (c) Evaluation of the emitted beam angle using red light.

3.5.4.

Bonding of the PIC on top of the holder

Finally, the correct position of the PIC relative to the lens holder [Fig. 16(a)] is found using feedback from the OSA, which captures the reflection spectrum of the Bragg grating sensor on the PIC. Once the correct position is found, the PIC is bonded to the holder using UV curable epoxy [Fig. 16(b)]. To do this, small droplets of epoxy are applied to the four sides and then UV cured.

Fig. 16

(a) Assembling the PIC on top of the holder (side view). (b) View on the assembly after gluing the PIC to the holder. (c) Glue dispensing for the sensor encapsulation based on a metal tube.

3.5.5.

Final sensor encapsulation

Proper encapsulation (to withstand high temperature, humidity, and pressure) of the sensor requires placing the sensor into a protective rigid tube (sheath). It also protects the sensor from mechanical impact, e.g., during insertion of the sensor probe for process monitoring. Therefore, the tube is placed over the sensor assembly [Fig. 16(c)], thermally curable glue is dispensed, and finally curing is done at a high temperature. After this process, the sensor is ready for further use. Figure 17 shows the sensor probe packaged in a protective metal tube with an interfacing fiber leading out.

Fig. 17

(a), (b) Photo of the assembled sensor, including protective tube and interfacing fiber. (c) Filled areas around the sensor seen from the top side.