Open Access
Add to BibSonomy
Abstract
We present a 1 × 3 optical switch based on a translational microelectromechanical system (MEMS) platform with integrated silicon nitride (SiN) photonic waveguides. The fabricated devices demonstrate efficient optical signal transmission between fixed and suspended movable waveguides. We report a minimum average insertion loss of 4.64 dB and a maximum average insertion loss of 5.83 dB in different switching positions over a wavelength range of 1530 nm to 1580 nm. The unique gap closing mechanism reduces the average insertion loss across two air gaps by a maximum of 7.89 dB. The optical switch was fabricated using a custom microfabrication process developed by AEPONYX Inc. This microfabrication process integrates SiN waveguides with silicon-on-insulator based MEMS devices with minimal stress related deformation of the MEMS platform.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Photonics switching technologies were developed using different kind of microelectromechanical actuators [1–3]. These actuators use piezoelectric [4], electrothermal [5], electromagnetic [6] and electrostatic [7] actuation-based methods in both planar 2-D [8,9] & out-of-plane 3-D [10,11] configurations. Piezoelectric actuators provide fast switching but largely rely upon out-of-plane actuation for optical switching [12,13]. Electrothermal actuators have also been designed for out-of-plane switching but are relatively slow and operate at high power [14,15]. Electrostatic actuators can provide both out-of-plane and planar switching solutions through comb drives or parallel plate-based actuation with low power consumption and fast switching operation [16–19]. Optical switching technologies based upon electrostatic actuators have been widely designed for silicon (Si) based photonics [20–23] whereas only a few optical switches exist that integrate optical switching with silicon-nitride (SiN) photonics [24,25]. Due to their lower refractive index contrast, SiN waveguides are less sensitive to width variations in comparison to Si waveguides [26]. Also, SiN waveguides can have lower scattering losses from sidewall roughness in comparison to Si waveguides [27,28].
In this work, we present a planar 1 × 3 optical switch based on a translational microelectromechanical system (MEMS) platform. The MEMS platform is designed in the device layer of a silicon-on-insulator (SOI) wafer and integrated with SiN channel waveguides that carry the optical signal across the MEMS platform. The MEMS platform presented here is a significant improvement upon our previous work [29] where only the in-plane translational motion of the actuator was verified since no waveguides were fabricated on it. The PiezoMUMPs microfabrication process available through MEMSCAP [30] was then used to demonstrate the actuator. The improved MEMS presented in this article is designed to manage the residual stress caused by the silicon dioxide (SiO2) cladding of the SiN waveguides [31] upon integration with the suspended MEMS layer. A schematic of the 1 × 3 optical switch along with its operating principle is presented in section 2. The trade-offs made to accommodate the residual thin film stress while minimizing the switching actuation voltage are also discussed in section 2. A brief overview of the microfabrication process flow and high-resolution scanning electron microscope (SEM) images of the fabricated devices are given in section 3. Mechanical and optical characterization results are reported in section 4. A discussion of the characterization results is presented in section 5, followed by concluding remarks and the envisioned future work in section 6.
2. Design methodology
2.1 1 × 3 Optical switch
The basic MEMS structure of our 1 × 3 optical switch is inspired from our previous work on electrostatic actuators [29]. Electrostatic actuation relies upon two isolated actuator plates fabricated in the same Si device layer. One of the actuator plates is fixed to the substrate and the movable plate is anchored through supporting spring system. The air gap between the actuator plates acts like the dielectric medium, and overall, the actuator is electrically equivalent to a capacitor. One of the actuator plates is grounded and the other plate is kept at a potential difference. This creates an electrostatic force of attraction due to charge accumulation of opposite polarity on the two actuator plates. As shown in Fig. 1, our device consists of a central waveguide platform connected through serpentine spring structures to a support beam. Left and right parallel plate switching actuators are located on opposite ends of the support beam. Each switching actuator takes advantage of the electrostatic pull-in phenomenon and is designed to provide a lateral displacement of 4 µm. This displacement is limited to 4 µm through mechanical stoppers located on the two opposite corners of each actuator. The gap closing actuator works following a similar principle as the switching actuators. Upon actuation, the waveguide platform is pulled closing the two air gaps between the suspended waveguides on the platform and the fixed waveguides on the substrate. This is required to reduce the optical coupling loss between suspended and fixed waveguides. The interfaces between the suspended and fixed waveguides at the two air gaps act as mechanical stoppers and limit again the displacement to 4 µm. The serpentine spring structures are designed to provide low mechanical stiffness to minimize the required actuation voltage during both switching and gap closing actuation.
Fig. 1. (a) Schematic of the 1 × 3 optical switch showing the MEMS platform with switching and gap closing actuators along with the waveguide layout. The insets show design dimensions for the (b) spring, (c) switching actuator and (d) gap closing actuator.
A top view of the SiN channel waveguides in a 1 × 3 switching configuration is shown in Fig. 1. The single input waveguide is located over a non-movable substrate portion. It is aligned to the center switching waveguide on the suspended MEMS platform when no switching voltage is applied. The waveguide platform can be moved in the left and right directions using the left and right switching actuators, respectively. Once a switching voltage is applied, the left or right output waveguide is aligned to the input waveguide depending upon the actuator used. The gap closing actuator can be used in all three switching positions (center, left and right) to minimize optical losses over the two air gaps. The waveguide platform size is 162.5 µm x 520 µm to accommodate all three waveguides. SiN channel waveguides with 3.2 µm top and bottom SiO2 cladding, and a core of 435 nm x 435 nm are used to implement the optical paths. The side cladding of the input and output waveguides over the fixed sections of silicon is 12.25 µm wide on each side. The three output waveguides are spaced on a 127 µm pitch to match that of our test setup optical I/O fiber array. Each waveguide has four 90° bends. Two of these bends are on the suspended platform to minimize the size of the platform. The other two bends are over the fixed silicon to separate the three output waveguides as per the aforementioned 127 µm pitch. Each bend has a 100 µm radius to minimize optical losses due to bending of the waveguides. Our previous work [29] presented 2.5 Finite Difference Time Domain (FDTD) simulations to estimate the crosstalk between adjacent waveguides for different separations, and Eigen Mode Expansion analysis for the edge-coupling between the waveguides on the fixed substrate and the movable platform with inverted tapers. Based on this work, inverted tapers with a tip-width of 400 nm and a length of 20 µm were included at the edge of the platform and fixed silicon to improve the coupling efficiency with minimum crosstalk. The tip-width was limited to 400 nm to comply with the minimum feature size of the optical lithography process used to fabricate the switch. The dimensions for the cladding and the space between waveguides are modified near the air gaps and on the suspended MEMS platform. Detailed dimensions of the layout of the waveguides are shown in Fig. 2.
Fig. 2. (a) Schematic of the 1 × 3 optical switch. The insets show the waveguide spacing and side cladding dimensions for (b) waveguides on the suspended platform, (c) the input waveguide interface, (d) the output waveguides interface, and (e) the output waveguides over fixed silicon. Cross-sectional view of (f) waveguide core and cladding dimensions. Top view of (g) inverted taper design near air gaps with dimensions.
2.2 Optical material stack residual mechanical stress
In our previous work, we designed a translational switching MEMS platform without optical waveguides using a 10 µm thick silicon device layer. However, previous research has shown that the thin films used to build the optical waveguides cause significant residual mechanical stress. This can lead to deformation of the silicon platform [31]. In this work, we integrated 435 nm thick SiN waveguides with 3.2 µm thick top and bottom SiO2 cladding layers with the MEMS for optical switching. Thus, before fabricating the optical switch, a finite element modelling (FEM) analysis was performed to assess the impact of mechanical stress from the integration of the optical material stack (i.e., cladding and waveguide). Stress related deformation of our MEMS platform was simulated using the COMSOL Multiphysics software (version 5.5, COMSOL, Inc., Burlington, MA, USA). FEM simulation results shown in Fig. 3 demonstrate that using a thicker silicon device layer significantly reduce stress related deformation across the entire waveguide platform. Since our waveguide couplers are located at the corners of the platform, it is important to minimize the out-of-plane displacement of the platform near these regions. Therefore, in this work we used a thicker Si device layer of 59 µm. As seen in Fig. 3, a device layer with a thickness of 59 µm shows only 10 nm to 18 nm of displacement along the vertical direction (i.e., the z-axis in Fig. 3) at the corners of platform. In comparison, a device layer of 10 µm shows a vertical displacement of 425 nm to 555 nm at the corners of the platform. The inner corner of the optical stack shows higher displacement in comparison to the outer corner for both device layer models. Clearly, increasing the thickness of the silicon device layer helps minimizing vertical misalignment between the waveguides on the suspended platform and those on the fixed substrate. It should be noted that the minute deformation of 10 nm to 18 nm of the suspended waveguide platform will not have any impact on the gap closing actuator mechanism. This is because the electrostatic force generated by the gap closing actuator relies upon the 59 µm silicon device layer thickness, which is significantly larger than the deformation of the platform. Also, the left and right switching actuators were not affected by the residual stress in the simulation model shown in Fig. 3(a). Thus, the switching actuation mechanics are not impacted by the integration of the waveguides. Optical simulation results for vertical misalignment between the suspended and fixed waveguides performed in our previous work also show that a 10 nm to 18 nm misalignment has negligible effect on the optical performance of the switch [29].
Fig. 3. FEM simulation results of vertical (z-axis) displacement of the MEMS structure due to the mechanical stress caused by the optical waveguides with: (a) a 10 µm thick silicon device layer; and (b) a 59 µm thick silicon device layer used in this work.
2.3 Serpentine springs
As demonstrated in section 2.2, increasing the thickness of the device layer from 10 µm to 59 µm helps minimize the impact of the residual stress on the vertical optical alignment. However, increasing the thickness of the device layer creates other challenges. In our previous work [29], translational MEMS actuators with a single beam spring required a switching voltage of 65 V. Also, the use of springs made of a single beam anchored at opposite ends provided enough planar vertical stiffness to avoid rotation of the platform during its displacement. The increase in the thickness of device layer dramatically increases the stiffness of the supporting spring structure. Static structural analysis of the actuator design based upon [29] with a 59 µm device layer was performed using ANSYS (version 19.0, ANSYS Inc., Canonsburg, PA, USA). Devices with a single beam spring of 4 µm x 250 µm for the switching actuator showed electrostatic pull-in at a voltage of 165 V. This is significantly higher than the 65 V pull-in voltage for the switching actuator with a 10 µm device layer [29].
Similar electrostatic simulations for the gap closing actuator were performed with ANSYS for both 59 µm and 10 µm device layer thicknesses. The gap closing actuator showed electrostatic pull-in at 62 V with a 59 µm device layer in comparison to 50 V for a 10 µm device layer [29]. The increase in pull-in voltage for the gap closing actuator is significantly less than the increase in pull-in voltage for the switching actuator with the increase in device layer thickness. The serpentine spring supporting the gap closing actuator has a lower stiffness of 127 N/m in comparison to 171 N/m for the single beam spring supporting the switching actuator. Thus, static structural analyses were performed for MEMS devices with serpentine spring structures for both the switching and gap closing actuators.
The simulated structure was similar to the illustration presented earlier in Fig. 1. Each silicon beam in the serpentine spring model of the switching actuator was designed to be 6 µm x 215 µm. The switching actuator gap was kept at 4.5 µm. The serpentine spring dimensions for the gap closing actuator were designed to be 4 µm x 256 µm for each silicon beam. The gap closing actuator gap was kept at 6 µm. It should be noted that electrostatic pull-in in a parallel plate actuator occurs when the movable actuator plate is displaced by 1/3rd of the initial gap between the fixed and movable actuator plates. Thus, electrostatic pull-in is expected after 1.5 µm and 2 µm of displacement for the switching and gap closing actuators, respectively. The simulation results shown in Fig. 4 present the clear advantage of the serpentine springs since they require a switching pull-in voltage of 130 V in comparison to 165 V for the single beam springs. The simulated gap closing actuation voltage for electrostatic pull-in is 62 V. The electrostatic pull-in points are present at the end of each curve presented in Fig. 4(a).
Fig. 4. (a) Simulation based displacement vs actuation voltage results for switching and gap closing actuators of (b) single beam spring and (c) serpentine spring-based MEMS structures with 59 µm device layer thickness. The dotted lines are a polynomial fit to the simulated data with the electrostatic pull-in point represented at the end of each curve.
3. Microfabrication
A proprietary fabrication process developed by AEPONYX inc. in a commercial foundry was used to integrate SiN waveguides with Si MEMS on SOI wafers with predefined cavities underneath the device layer. These cavities simplify the release of the MEMS structure. Initially, a 3.2 µm thick SiO2 bottom cladding layer was deposited through Tetra EthOxy Silane (TEOS) low pressure chemical vapor deposition (LPCVD) over a SOI wafer with a 59 µm Si device layer. A 435 nm thick SiN waveguide core layer was then deposited through LPCVD on top of the bottom cladding layer. Photolithography with a stepper tool and dry etching was used to pattern the SiN waveguides. A 3.2 µm thick SiO2 top cladding layer was then deposited through TEOS plasma enhanced chemical vapor deposition over the SiN waveguides. Openings were patterned and etched into the optical stack to expose the silicon layer underneath. A 250 nm thick film of aluminum copper (AlCu) alloy was then sputtered and patterned in the exposed Si regions to form the bonding pads required to actuate the MEMS electrostatically. The optical stack next to waveguides and over the MEMS region was removed through dry etching to minimize residual stress in the final device. The MEMS actuator structures were patterned in a hard mask before being transferred into the Si through deep reactive ion etching. This also released the MEMS since they are fabricated over empty cavities. A cross sectional view of the optical switch is shown in Fig. 5.
Following microfabrication, high resolution SEM imaging was performed. An analysis of the SEM micrographs obtained showed variations in the fabricated dimensions of the actuator and stopper gaps. The images shown in Fig. 6 are from one of the three samples tested. The mechanical stopper gap for switching was designed to be 4 µm and the switching actuator gap was designed to be 4.5 µm. As shown in Fig. 6(b) and 6(c), these dimensions changed to 4.74 µm and 5.86 µm, respectively. As shown in Fig. 6(d), the air gap closing actuator dimension also changed to 7.19 µm after fabrication (compared to 6 µm in the design). The mechanical stopper / air gap closing interface between waveguides is shown in Fig. 6(e) and 6(f) with dimensions and the slope etch angle. The etch profile of the SiO2 cladding and SiN waveguide clearly shows that the slope of the etch is not vertical. This slope creates a variation in the air gap across the height of the optical stack. The top layer of SiO2 shows an air gap of 5.04 µm at the waveguide interface region. However, when measured at the level of the silicon layer underneath the optical stack the gap is 3.68 µm. These variations in the geometry of the fabricated devices were expected since AEPONYX’s process is optimized for specific etch loading and aspect ratios that differ from our experimental devices. The impact of these variations on the actuation voltage and the optical performance of the devices are presented in the next section.
Fig. 6. High resolution SEM micrographs with measurements of the fabricated 1 × 3 optical switch sample 1: (a) fabricated device; (b) mechanical stopper gap; (c) switching actuator gap; (d) air gap of the gap closing actuator; (e) air gap closing interface; and (f) etch profile of the optical stack.
Similar variations were observed during SEM imaging for the other two samples. The mechanical stopper gap for switching varies between 4.58 µm – 4.87 µm across samples. The switching actuator gap varies between 5.55 µm – 6.01 µm. The mechanical stopper gap and actuator gap for the gap closing actuator also vary between 3.62 µm – 3.85 µm and 6.69 µm – 7.41 µm, respectively. Variations in the width of the gap around the mechanical stoppers of the gap closing actuators will not impact the optical performance as the device relies upon the pull-in phenomena for closing the air gap between suspended and fixed waveguides. However, variation in the mechanical stopper gaps for the switching actuator can lead to misalignment between waveguides during switching. Experimental optical measurement results showing the impact of device fabrication variations upon transmission during switching are presented in section 4.2. The critical dimensions of the three samples tested are shown in Table 1.
Table 1. Fabrication Variations Across Samples
4. Results
4.1 MEMS characterization
MEMS characterization was done using a Wentworth microprobes station. Two high voltage DC sources were connected to the switching and gap closing actuators through microprobes, as shown in the test circuit in Fig. 7. While one of the high voltage sources was connected to a switching actuator, the second voltage source was connected to the gap closing actuator for bi-axial motion of the central waveguide platform. Both actuators were grounded through a 100 kΩ resistor to prevent device damage in case the grounded moveable actuator plate comes in contact with the high voltage static actuator plate.
A high magnification lens system was used along with a high-resolution camera to image the devices during actuation. Microscope images of the device under actuation in different switching positions are shown in Fig. 8. Images of the actuator in different switching positions at different voltages for one of the samples were analyzed using the ImageJ (version 1.53a) software to extract the displacement vs voltage curves and pull-in voltage. Experimental results for both the left and right switching actuators showed electrostatic pull-in at 170 V. This was higher than the 130 V expected from the simulation results presented in section 2.3. Similarly, the experimental pull-in voltage for the gap closing actuator was 80 V. This was also higher than the 62 V pull-in voltage predicted by simulation. This increase in pull-in voltage can be explained by the larger gap dimensions in the fabricated devices in comparison to the designed dimensions presented in section 3. Simulations based on the fabricated dimensions are compared with the experimental results for one of the samples in Fig. 9. We can see that the simulation results with modified dimensions are closer to the experimental results, although the simulated pull-in voltages are slightly higher than the experimental values. The difference between simulation and experimental results could be due to variations in the spring beam width along the length of silicon beams and potential device layer sidewall topography. It should be noted that no cross sensitivity was observed between the switching and gap closing actuators in the three switching positions. This is due to the high electrostatic force generated during pull-in in left and right switching positions that maintain the desired platform position while the gap closing actuator is enabled. The center switching position demonstrated no cross sensitivity as only the gap closing actuator is required for efficient transmission in this position. The symmetry in the design of the device mitigates cross sensitivity in this switching position.
Fig. 8. Zoomed in microscope images of the fabricated 1 × 3 optical switch device at the input and output waveguide interface during actuation in: (a, b) left switching position; (c, d) center switching positions; (e, f) right switching position.
Fig. 9. Experimental and simulation actuation results for the gap closing and switching actuators with the pull-in points. The dotted lines are a polynomial fit to the simulated data based on the fabricated dimensions and the solid lines are a polynomial fit to the experimental measurements.
4.2 Optical characterization
The transmission of optical signals was successfully characterized for all switching positions for three samples. Each sample was wire bonded to a custom printed circuit board (PCB) designed to control the actuation of the MEMS. A tunable laser (T100S-HP) and optical component tester (CT440) from EXFO were used for these measurements. The output of the tunable laser was connected to the input optical fiber array with a 30° polish angle through the optical component tester. The fiber array could be optimally aligned with the surface grating couplers (SGCs) at the input and output ports of the chip on each sample at a vertical distance of approximately 50 µm. Three detector ports on the optical component tester were connected to the output optical fibers from the fiber array. Polarization maintaining fibers were used for the optical connections between the tunable laser, optical component tester and the fiber array to maintain the transverse electric (TE) mode during these tests since the alignment between the fiber array and SGCs used to demonstrate these prototypes were optimized for this polarization. The pitch between the SGCs was designed to match the pitch of the optical fiber array, which was fixed at 127 µm. A microposition controller was used to align the sample and the optical fiber array with 1 µm precision. The sample could be moved in-plane (X & Y direction) and the fiber array could be moved vertically (Z direction) for this purpose. A detailed image of the test setup used for optical characterization is shown in Fig. 10.
Fig. 10. (a) Test setup for optical characterization of the 1 × 3 optical switch; (b) cross-sectional camera view of the optical fiber array aligned to the sample; (c) a zoomed in image of the sample wire bonded to the PCB during measurements.
We fabricated reference waveguide structures on all our samples. The length of the reference waveguide was the same as the total length of the center channel in our device. The number and position of bends was kept the same for both the reference waveguide and the center channel waveguide. Our reference waveguide structure also consisted of adjacent output waveguides similar in position and length to the left and right channels. This would help us replicate crosstalk, if any, between the waveguides present on our optical switch. It should be noted that only the center channel had the input port in the reference waveguide structure for optical measurements. The transmission data obtained for the reference waveguide was used to normalize experimental results for the 1 × 3 optical switch in all switching positions. This was done by subtracting the transmission data of the reference waveguide from transmission data of each switching position at each wavelength. The wavelength scan resolution used was 10 pm. The propagation loss for the TE mode in the C-band and L-band is 3.3 dB/cm in the fabrication process used. Therefore, for the center switching waveguide, which as a length of 7.618 mm, the total propagation loss was 2.48 dB. Each switching position (left, center and right) was tested with and without activating the gap closing actuator for all three samples. The average insertion loss obtained for the three samples in all switching positions at wavelengths between 1530 nm to 1580 nm is shown in Fig. 11. The measured transmission spectra for the three switching positions of the three samples are presented in the appendix. The experimental results presented focus mainly on the C-band and the beginning of the L-band of the telecommunication spectrum because the SGCs were optimized for this wavelength range. The transmission data of the optical switch in Fig. 11 showed small undulations for the entire wavelength scan. These undulations increased outside the wavelength range presented for all three samples. Since we observe the same undulations in our reference waveguide structures, it is safe to assume that the switch does not cause these undulations. They are probably caused by unwanted reflections between the SGCs and the fiber array.
Fig. 11. Average transmission of three 1 × 3 optical switch samples in all three switching positions with/without gap closer (GC) actuator across the 1530 nm – 1580 nm wavelength. Crosstalk center (CTC) represents the optical signal transmission in the left and right channels when the center switching position and GC actuator are ON.
The average insertion loss over the entire wavelength scan for the three samples was 4.64 dB when the gap closing actuator is ON and the switch is aligned to the center position (solid green curve in Fig. 11). This is less than the average insertion loss observed for the left and right switching positions with the gap closer actuator ON, which were 5.38 dB (solid orange curve in Fig. 11) and 5.83 dB (solid blue curve in Fig. 11), respectively. This can be explained by the fabrication variations in the mechanical stopper gaps of the switching actuators observed during SEM imaging. While the mechanical stopper gap for switching actuator was designed to be 4 µm, the fabricated gap varied between 4.67 µm and 4.87 µm, as discussed in section 3. Since the switch is digital and relies upon the ON and OFF state of the switching actuators, this 670 nm to 870 nm gap variation creates a misalignment between waveguides when switching to the left and right positions. Similar average insertion loss variations at the different switching positions were observed in all three samples. The average insertion loss for each sample is shown in Table 2. Once the gap is closed, the crosstalk between adjacent waveguides is less than -30 dB.
Table 2. Average Insertion Loss Comparison Between Samples
The etch profile shown in section 3 (see Fig. 6) creates up to a 1 µm gap between the waveguides on the fixed substrate and the ones on the suspended platform (see Fig. 12(b)). Optical losses due to this residual air gap cannot be compensated with the gap closing actuator. The simulated and experimental average insertion losses for all samples over the wavelength range of 1530 nm to 1580 nm are compared in Fig. 12. The simulation results were obtained through FDTD simulations using Lumerical (version 2020 R2) software for edge coupling between waveguides with slanted sidewalls (see Fig. 12(b)). The simulation also includes the scenario for perfectly aligned waveguides and waveguides misaligned laterally by 740 nm, which corresponds to the shift observed in sample 1. Both cases include a 1 µm residual air gap at the center of the waveguide core due to etch profile of the optical stack. The waveguide dimensions were kept at 435 nm by 435 nm except for 20 µm long inverted tapers with a 400 nm tip width near the air gap interface. Simulation results compare well with the experimental one. However, as explained earlier the SGCs cause undulations in the experimental results in Fig. 12(a). Our simulation model did not include the SGCs due to computational limitations. The simulation and experimental results in Fig. 12(a) also show reduced optical losses with increase in wavelength. This is expected due to lower confinement and hence, higher coupling efficiency at the air gaps for higher wavelengths.
Fig. 12. (a) Simulated and experimental average insertion losses comparison for all samples in three switching positions with the gap closing actuator ON, (b) measured etch profile and residual air gap between center of the SiN waveguides, and (c) dotted lines showing 740 nm lateral misalignment scenario used for FDTD simulations of edge coupling waveguides.
The effect of increasing incrementally the voltage across the gap closing actuator on the insertion loss of the device was also characterized. The voltage applied to the gap closer was increased in 10 V increments up to 70 V and in 5 V increments afterwards until the optical losses in the center switching waveguide were minimized. The voltage for the lowest optical losses was found to be 80 V for all three samples. This corresponds to the electrostatic pull-in voltage for the gap closing actuator presented earlier in section 4.1. The gap closing actuator provides a large reduction in the insertion loss as shown by the loss vs applied voltage curves in Fig. 13(a). Results from the three samples are shown demonstrating the repeatability of the performance of the GC actuator.
Fig. 13. Optical characterization results at a wavelength of 1600 nm showing (a) a reduction in the optical loss across the center switching position as the gap closer actuation voltage increases for three samples from different wafers and (b) the impact of left and right switching actuation voltage upon optical signal transmission in the center and switching waveguides for sample 1. The ends of curves show the experimental data points for loss reduction and transmission after electrostatic pull-in of the gap closing actuator.
The optical switch is designed such that the center waveguide is aligned to the input when no voltage is applied. When a voltage is applied to a switching actuator (left or right), the optical signal is transmitted through the corresponding waveguide. This switching effect was characterized for the left and right switching channels and the effect of switching upon the center waveguide was measured simultaneously. A voltage was applied to the two switching actuators in 50 V increments up to 100 V and in 10 V increments afterwards up to 170 V. The results for sample 1 can be seen in Fig. 13(b). After applying the switching voltage of 170 V, the gap closing actuator was also used to further reduce the optical losses. This explains why the signals transmitted through the left and right waveguide have lower losses than the initial value in the center waveguide. The optical signal transmission through the center waveguide was monitored to clearly show the digital switching behavior of the MEMS. These measurements were performed at a wavelength of 1600 nm with 10 dBm of input power during our initial alignment tests. Similar results can be expected for the wavelength range presented in Fig. 10 with minor offset in transmission due to the response of the SGCs.
5. Discussion
The translational MEMS platform for planar optical switching [29] was modified to accommodate SiN channel waveguides in a 1 × 3 optical switch configuration. The thickness of the silicon device layer was increased to sustain the residual mechanical stress caused by the optical material stack used to create waveguides on the suspended MEMS structure. The spring mechanism for the switching actuator was improved to minimize the actuation voltage. Multiple devices were fabricated and tested. The maximum displacement of the switching actuator was 4.87 µm after electrostatic pull-in at 170 V. Similarly, the gap closing actuator closed the two air gaps between the suspended and fixed waveguides by moving the platform by up to 3.85 µm after electrostatic pull-in at 80 V.
The optical characterization of the 1 × 3 switch presented in this work showed the lowest average insertion loss of 4.64 dB for the center switching channel for wavelengths between 1530 nm and 1580 nm. The left and right channels showed a minimum average insertion loss of 5.38 dB and 5.83 dB, respectively, across the same wavelength range. The difference in insertion loss is due to the misalignment variation between waveguides in the left and right switching channels across samples. The residual stress simulations presented in section 2.2 showed a slight variation of the height of the suspended platform across the width of the waveguides. The inner corners of the waveguide array showed lower vertical displacement in comparison to the outer corners along the edge of the platform. The left switching channel was fabricated closer to the inner corner of the array and the right switching channel near the outer corner. The placement of waveguides on the platform therefore creates a variation in the vertical misalignment of left and right switching channels. In addition to this variation, the slight difference in total optical path length of the switching waveguides can contribute towards variation in optical losses measured as well. The center switching channel has an optical path length of 7.618 mm. The left switching channel has the shortest optical path length of 7.471 mm while the right switching channel has the longest optical path length of 7.765 mm. This length difference along with the variation in the height of the suspended waveguides can explain the small 0.45 dB variation in insertion loss between the left and right channels. Also, fabrication variations in the left and right switching stopper gaps across samples (see Table 1) cause optical loss variations across samples during switching (see Table 2).
The optical switch presented in this work could in theory completely close the air gap between the suspended and fixed waveguides. However, in practice the slope at the edge of the waveguides (see Fig. 12(b)) creates a residual air gap of up to 1 µm when measured between the center of the core of the waveguides. Nevertheless, this is an improvement over our previous MEMS silicon nitride waveguide based optical switch [25], which only closes the air gap to a minimum of 500 nm by design in addition to any residual air gap due to etch profiles, across each coupling region. The reduction in the air gap presented in this work helps lowering the average insertion loss by 7.89 dB, 7.46 dB and 7.75 dB for center, left and right switching channels, respectively.
The fabrication variation in the switching stopper gap dimension of sample 1 leads to a misalignment of 740 nm and 750 nm between the suspended and fixed waveguides during left and right channel switching, respectively (see Table 1). This variation increases the optical insertion losses for these channels (see Table 2). The 740 nm change in stopper gap increases the average optical loss by 0.74 dB for the left switching channel. The 750 nm variation for the right switching channel increases the optical loss by 1.19 dB. The slight difference in the fabricated stopper gaps along with the difference in length of switching waveguides discussed earlier in this section can contribute towards this observed variation in optical losses. FDTD simulation results showed a 1.51 dB variation in insertion loss due to a 740 nm misalignment between suspended and fixed waveguides with a 1 µm residual air gap. These optical switching losses can be successfully minimized in a future implementation by introducing a fabrication bias in the photolithography mask. The mask can be designed for a smaller stopper gap dimension to compensate for the over etching observed during fabrication. This can help minimize optical losses by reducing the misalignment between waveguides during switching. However, even with this misalignment the device successfully manages to switch the optical signal between different SiN channel waveguides with a crosstalk level below -30 dB. The grating coupler could also be optimized further to reduce reflections across the wavelength range.
During our experiments we also found that some of the devices shorted during switching unlike our previous work, which involved testing only the MEMS actuators for translational motion [29]. This was due to the critically low difference between the stopper gap and the actuator gap dimensions. The minimum gap in the Si device layer allowed by our fabrication process is 4 µm to release the MEMS structure. The switching mechanical stopper gap was thus kept at 4 µm while the switching actuator gap was kept at 4.5 µm, creating a difference of 500 nm in the stopper and actuator gap designs sent for fabrication (see Fig. 1(c)). This 500 nm gap variation was detrimental towards successful testing of the switching actuator as some of the devices would short during electrostatic pull-in. The switching actuator gap was not increased beyond 4.5 µm to minimize the actuation voltage for switching. It should be noted that the gap closing actuator had a 2 µm difference between the actuator gap and the mechanical stopper gap (see Fig. 1(d)). No shorting was observed while closing the gap in any of the devices. Thus, a larger difference in the actuator and mechanical stopper gap can easily mitigate this issue in future implementation and increase device reliability.
Most of the SiN based optical switching solutions reported in the literature rely upon thermal tuning of optical components, such as Mach-Zehnder interferometers (MZIs), to switch light between waveguides. Although a large number of ports can be achieved through this method, such structures consume a lot of power due to the DC current required to heat the MZI arm for optical switching [32,33]. Alternatively, a few MEMS based phase shifting / switching solutions exist that work with SiN based optical filters. A low power electrostatic actuation-based solution that can filter only 3 wavelengths in the C-band through digital control of a suspended aluminum-based MEMS bridge over a silicon nitride ring resonator was demonstrated in [24]. Another low power piezoelectric actuation-based solution that can filter multiple wavelengths by straining a silicon nitride (Si3N4) ring resonator using an integrated Lead Zirconate Titanate (PZT) actuator arm was reported in [34]. However, this switch can operate over a wavelength range of only 10 nm and has high losses due to the PZT fabrication and release processes. Also, PZT based piezoelectric actuation comes with environmental concerns due to the lead contained in the material. Similarly, aluminum nitride (AlN) based piezoelectric actuation has been recently used to tune Si3N4 based ring resonators [35]. However, the tuning range is limited to only 20 pm. A rotational SOI based MEMS actuator has been integrated with SiN waveguides in a crossbar switching configuration in recent years [25]. This rotational switch design had 12.2 dB – 14.8 dB total insertion loss at an operational wavelength of 1550 nm. The air gaps between the suspended and fixed SiN waveguides could not be completely closed in this approach due to design limitations. In this work, the air gaps can be closed to reduce optical losses, and simple parallel plate actuators allow for simple digital control of the switch. Also, the optical switch in this work provides broadband operation with an insertion loss of 4.45 dB – 6.64 dB over a 50 nm wavelength range across multiple samples. We obtained the switching time for the switching and gap closing actuators by performing time domain simulations with COMSOL Multiphysics. The switching actuator requires 50 µs to close completely whereas the gap closing actuator needs 100 µs as per our simulation results. Since the center switching channel relies only upon gap closing actuation for efficient coupling, the switching time for this channel will be 100 µs. The left and right switching channels can operate at 50 µs without the gap closing actuator. Combining gap closing actuation with either the left or right switching actuators will result in efficient optical coupling after 150 µs. Also, simulations including the effects of gravity were performed on our device model in COMSOL Multiphysics. This showed no impact of gravity upon the device structure. The high stiffness of our device due to the 59 µm Si thick device layer mitigates any impact of gravity upon our device. In addition to this, the electrostatic force generated during pull-in makes the structure very stable to g-forces and vibrations. The direct contact of the movable structure and continuous use of the pull-in voltage to keep the switch in the required switching state anchor the structure as opposed to a free-standing structure that can be sensitive to vibrations. However, the optical performance of a switch can be sensitive to vibrations as they can impact the alignment between the fiber array and the grating couplers. We measured less than a 0.15 dB optical loss variation across 10 measurements in all switching positions for Sample 1. Furthermore, the 15 µm and 35 µm mechanical stoppers (see Fig. 1(c) and 1(d)) ensure minimal device contact during actuation. We did not notice any stiction issues during our testing for any of the samples. A comparison with the state-of-the-art SiN based optical switching solutions is presented in Table 3.
Table 3. Comparison of SiN Based Optical Switches
6. Conclusion
A novel 1 × 3 planar MEMS optical switch with integrated SiN channel waveguides was presented. The challenges of integrating SiN waveguides with residual mechanical stress onto a MEMS switching platform were mitigated by using a thick silicon device layer. The design of serpentine springs for both the switching and gap closing actuators to minimize actuation voltages was also presented. The 1 × 3 optical switch was fabricated successfully and required a switching voltage of 170 V and a gap closing voltage of 80 V. No actuation is required to couple light into the center switching waveguide and it demonstrates an average optical loss of 4.64 dB across the 1530 nm to 1580 nm wavelength range with only the gap closing actuator activated. The left and right channels required actuation of the relevant switching actuator and the gap closing actuator and achieve an average optical loss of 5.38 dB and 5.83 dB across the same wavelength range.
The wide bandwidth of the proposed optical switch makes it a versatile device to enable reconfigurable SiN photonic circuits. For example, it can be integrated with high performance passive SiN optical filters (e.g., ring resonators, Bragg gratings and/or micro-disk resonators) to select wavelength channels or even an entire wavelength band inside an optical telecommunication network. The key advantage over existing SiN based wavelength channel selection systems will be low power operation, no thermal tuning, and a simple fabrication process that allows us to control dimensions for each output waveguide and optical filter. Furthermore, since the switch and filters can be manufactured at low cost at high volumes, these systems can make possible the deployment of wavelength selective devices, such as receivers, at large scale in access networks to increase their capacity. For instance, access networks based on the NG-PON2 standard require receivers able to select among a few channels (4 to 8) that could be implemented with this technology [36]. The experimental results presented in this work focused on the C-band of the telecommunication spectrum. This was because of the bandwidth limitation our SGCs. However, the switch is broadband and can work across the whole wavelength range for which the waveguides are single mode and transparent. For the waveguide configuration used in the prototypes, this goes from 1100 nm up to 2200 nm. The demonstrated wavelength range of the switch in this work can be useful towards realization of a MEMS based tunable transceiver operational in the C-band of the telecommunication spectrum. Such a system can provide viable solutions towards building cost effective and energy efficient passive optical networks [37].
In the future, we aim to minimize fabrication variations and improve the device reliability through optimization of the MEMS design prior to fabrication. We also look towards improving the MEMS design further to provide larger displacement for the switching motion at a lower actuation voltage. Larger displacement of the switching actuator will allow to integrate more waveguides on the central platform, and hence, increase the number of channels. It can also enable greater degree of control and minimize optical losses due to misalignment between waveguides during digital switching.
Appendix
The tunable laser used for optical measurements provides scans covering the wavelength range of 1500 nm to 1630 nm. The transmission power was measured in dBm using the EXFO CT440 optical component tester for the three samples. The spectral response for samples 1, 2, and 3 is shown in Fig. 14, Fig. 15, and Fig. 16, respectively. This includes the transmission power measured for the reference waveguides fabricated alongside each sample on the same die. The crosstalk in the left and right switching channels during transmission in the center channel with the gap closing actuator ON is also shown for each sample in the corresponding figure. Sample 1 and 2 show lower undulations in their wavelength response below 1600 nm whereas sample 3 shows low undulations only between 1560 nm and 1600 nm. Sample 2 also shows larger variations across switching positions, which corresponds to the measured fabrication variation explained in section 3 (see Table 1).
Fig. 14. Transmission power for sample 1 across 1500 nm – 1630 nm for in all three switching positions with and without gap closing actuator.
Fig. 15. Transmission power for sample 2 across 1500 nm – 1630 nm for in all three switching positions with and without gap closing actuator.
Funding
Natural Sciences and Engineering Research Council of Canada; AEPONYX Inc.; PRIMA Québec; Regroupement Stratégique en Microsystèmes du Québec (ReSMiQ).
Acknowledgments
The authors would like to thank AEPONYX Inc. for access to their test facilities, device fabrication and financial support. The authors also thank CMC Microsystems for providing the software tools necessary for design optimization. The authors would also like to thank their research group members Seyedfakhreddin Nabavi and Mohammad Kazemi for their help with switching time simulations and wire bonding of samples respectively.
Disclosures
SS: AEPONYX (F,P), NK: AEPONYX (F,P), JB: AEPONYX (E,I,P), FN: AEPONYX (F,P), MM: AEPONYX (F,P)
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. T. Liu, F. Pagliano, R. V. Veldhoven, V. Pogoretskiy, Y. Jiao, and A. Fiore, “Low-voltage MEMS optical phase modulators and switches on a indium phosphide membrane on silicon,” Appl. Phys. Lett. 115(25), 251104 (2019). [CrossRef]
2. H. Sattari, A. Y. Takabayashi, P. Edinger, P. Verheyen, K. B. Gylfason, W. Bogaerts, and N. Quack, “Low-Voltage Silicon Photonic MEMS Switch with Vertical Actuation,” in 2021 IEEE 34th International Conference on Micro Electro Mechanical Systems (MEMS) (2021), pp. 298–301.
3. Q. Qiao, M. S. Yazici, B. Dong, X. Liu, C. Lee, and G. Zhou, “Multifunctional mid-infrared photonic switch using a MEMS-based tunable waveguide coupler,” Opt. Lett. 45(19), 5620–5623 (2020). [CrossRef]
4. P. Ramesh, S. Krishnamoorthy, S. Rajan, and G. N. Washington, “Fabrication and characterization of a piezoelectric gallium nitride switch for optical MEMS applications,” Smart Mater. Struct. 21(9), 094003 (2012). [CrossRef]
5. Y. Gerson, S. Krylov, and B. Ilic, “Electrothermal bistability tuning in a large displacement micro actuator,” J. Micromech. Microeng. 20(11), 112001 (2010). [CrossRef]
6. S. Jia, J. Peng, J. Bian, S. Zhang, S. Xu, and B. Zhang, “Design and Fabrication of a MEMS Electromagnetic Swing-Type Actuator for Optical Switch,” Micromachines 12(2), 221 (2021). [CrossRef]
7. I. Plander and M. Stepanovsky, “MEMS optical switch: Switching time reduction,” Open Comput. Sci. 6(1), 116–125 (2016). [CrossRef]
8. E. Bulgan, Y. Kanamori, and K. Hane, “Submicron silicon waveguide optical switch driven by microelectromechanical actuator,” Appl. Phys. Lett. 92(10), 101110 (2008). [CrossRef]
9. S. Abe and K. Hane, “Variable-gap silicon photonic waveguide coupler switch with a nanolatch mechanism,” IEEE Photonics Technol. Lett. 25(7), 675–677 (2013). [CrossRef]
10. D. M. Sun, W. Dong, G. D. Wang, C. X. Liu, X. Yan, B. K. Xu, and W. Y. Chen, “Study of a 2× 2 MOEMS optical switch with electrostatic actuating,” Sens. Actuators, A 120(1), 249–256 (2005). [CrossRef]
11. M. Pallay and S. Towfighian, “A reliable MEMS switch using electrostatic levitation,” Appl. Phys. Lett. 113(21), 213102 (2018). [CrossRef]
12. S. J. Kim, Y. H. Cho, H. J. Nam, and J. U. Bu, “Piezoelectrically pushed rotational micromirrors using detached PZT actuators for wide-angle optical switch applications,” J. Micromech. Microeng. 18(12), 125022 (2008). [CrossRef]
13. W. Liao, W. Liu, Y. Zhu, Y. Tang, B. Wang, and H. Xie, “A tip-tilt-piston micromirror with symmetrical lateral-shift-free piezoelectric actuators,” IEEE Sens. J. 13(8), 2873–2881 (2013). [CrossRef]
14. Y. An, B. Sun, P. Wang, L. Xiao, H. Liu, and H. Xie, “A 1× 20 MEMS mirror array with large scan angle and low driving voltage for optical wavelength-selective switches,” Sens. Actuators, A 324, 112689 (2021). [CrossRef]
15. T. J. Peters and M. Tichem, “Electrothermal actuators for SiO2 photonic MEMS,” Micromachines 7(11), 200 (2016). [CrossRef]
16. J. Li, Q. X. Zhang, and A. Q. Liu, “Advanced fiber optical switches using deep RIE (DRIE) fabrication,” Sens. Actuators, A 102(3), 286–295 (2003). [CrossRef]
17. K. Takahashi, Y. Kanamori, Y. Kokubun, and K. Hane, “A wavelength-selective add-drop switch using silicon microring resonator with a submicron-comb electrostatic actuator,” Opt. Express 16(19), 14421–14428 (2008). [CrossRef]
18. M. Kaykisiz and E. Bulgan, “Low-loss optical switch using Brewster angle,” in 16th International Conference on Optical MEMS and Nanophotonics (2011), pp. 215–216.
19. T. J. Seok, N. Quack, S. Han, R. S. Muller, and M. C. Wu, “Highly scalable digital silicon photonic MEMS switches,” J. Lightwave Technol. 34(2), 365–371 (2016). [CrossRef]
20. Y. Akihama, Y. Kanamori, and K. Hane, “Ultra-small silicon waveguide coupler switch using gap-variable mechanism,” Opt. Express 19(24), 23658–23663 (2011). [CrossRef]
21. Y. Munemasa and K. Hane, “A compact 1× 3 silicon photonic waveguide switch based on precise investigation of coupling characteristics of variable-gap coupler,” Jpn. J. Appl. Phys. 52(6S), 06GL15 (2013). [CrossRef]
22. D. T. Fuchs, H. B. Chan, H. R. Stuart, F. Baumann, D. Greywall, M. E. Simon, and A. Wong-Foy, “Monolithic integration of MEMS-based phase shifters and optical waveguides in silicon-on-insulator,” Electron. Lett. 40(2), 142–143 (2004). [CrossRef]
23. K. Hane, H. M. Chu, and T. Sasaki, “Silicon photonic waveguide devices tunable by in-plane actuation,” in 2016 International Conference on Optical MEMS and Nanophotonics (OMN) (2016), pp. 1–2.
24. G. N. Nielson, D. Seneviratne, F. Lopez-Royo, P. T. Rakich, Y. Avrahami, M. R. Watts, H. A. Haus, H. L. Tuller, and G. Barbastathis, “Integrated wavelength-selective optical MEMS switching using ring resonator filters,” IEEE Photonics Technol. Lett. 17(6), 1190–1192 (2005). [CrossRef]
25. J. Briere, M. Y. Elsayed, M. Saidani, M. Bérard, P. O. Beaulieu, H. Rabbani-Haghighi, F. Nabki, and M. Ménard, “Rotating circular micro-platform with integrated waveguides and latching arm for reconfigurable integrated optics,” Micromachines 8(12), 354 (2017). [CrossRef]
26. R. Baets, A. Z. Subramanian, S. Clemmen, B. Kuyken, P. Bienstman, N. Le Thomas, G. Roelkens, D. V. Thourhout, P. Helin, and S. Severi, “Silicon Photonics: silicon nitride versus silicon-on-insulator,” in Optical Fiber Communication Conference (2016), pp. Th3J-1.
27. T. Barwicz and H. A. Haus, “Three-dimensional analysis of scattering losses due to sidewall roughness in microphotonic waveguides,” J. Lightwave Technol. 23(9), 2719–2732 (2005). [CrossRef]
28. J. F. Bauters, M. J. Heck, D. John, D. Dai, M. C. Tien, J. S. Barton, A. Leinse, R. G. Heideman, D. J. Blumenthal, and J. E. Bowers, “Ultra-low-loss high-aspect-ratio Si3N4 waveguides,” Opt. Express 19(4), 3163–3174 (2011). [CrossRef]
29. S. Sharma, N. Kohli, J. Brière, M. Ménard, and F. Nabki, “Translational MEMS platform for planar optical switching fabrics,” Micromachines 10(7), 435 (2019). [CrossRef]
30. A. Cowen, G. Hames, K. Glukh, and B. Hardy, PiezoMUMPs Design Handbook, rev. 1.3 (MEMSCAP Inc., 2014).
31. C. Yang and J. Pham, “Characteristic study of silicon nitride films deposited by LPCVD and PECVD,” Silicon 10(6), 2561–2567 (2018). [CrossRef]
32. J. Joo, J. Park, and G. Kim, “Cost-effective 2 ( 2 silicon nitride Mach-Zehnder Interferometric (MZI) thermo-optic switch,” IEEE Photonics Technol. Lett. 30(8), 740–743 (2018). [CrossRef]
33. X. Sun, J. Feng, L. Zhong, H. Lu, W. Han, F. Zhang, R. Akimoto, and H. Zeng, “Silicon nitride based polarization-independent 4 ( 4 optical matrix switch,” Opt. Laser Technol. 119, 105641 (2019). [CrossRef]
34. W. Jin, R. G. Polcawich, P. A. Morton, and J. E. Bowers, “Piezoelectrically tuned silicon nitride ring resonator,” Opt. Express 26(3), 3174–3187 (2018). [CrossRef]
35. B. Dong, H. Tian, M. Zervas, T. J. Kippenberg, and S. A. Bhave, “PORT: A piezoelectric optical resonance tuner,” in 2018 IEEE Micro Electro Mechanical Systems (MEMS) (2018), pp. 739–742.
36. C. Pinho, F. Rodrigues, A. M. Tavares, C. Rodrigues, C. E. Rodrigues, and A. Teixeira, “Photonic Integrated Circuits for NGPON2 ONU Transceivers (Invited),” Appl. Sci. 10(11), 4024–4043 (2020). [CrossRef]
37. J. A. Walker, “The future of MEMS in telecommunications networks,” J. Micromech. Microend. 10(3), R1–R7 (2000). [CrossRef]
