Post-Release Deformation and Motion Control of Photonic Waveguide Beams by Tuneable Electrothermal Actuators in Thick SiO2

Photonic packaging, which includes high-precision assembly of photonic sub-systems, is currently a bottleneck in the development of commercially-available integrated photonic products. In the pursuit of a fully-automated, high-precision, and cost-effective photonic alignment scheme for two multi-channel photonic chips, this paper explores different designs of the on-chip electrothermal actuators for positioning mechanically-flexible waveguide structures. The final alignment goal is ∼100 nm waveguide to waveguide. The on-chip actuators, particularly for out-of-plane actuation, are built in a 16 μm-thick SiO2 photonic-material stack with 5 μm-thick poly-Si as an electrothermal element. A major challenge of out-of-plane positioning is a 6 μm height difference of the waveguides to be aligned, due to different built-up material stacks, together with a misalignment tolerance of 1 μm–2 μm from the pre-assembly (flip-chip) process. Therefore, the bimorph-actuator design needs to compensate this height difference, and provide sufficient motion to align the waveguides. We propose to exploit the post-release deformation of so-called short-loop bimorph actuator designs to meet these joint demands. We explore different design variants based on the heater location and the integration of actuator beams with waveguide beams. The actuator design (with 30 μm poly-Si and 900 μm SiO2 in length) has ∼8 μm out-of-plane deflection and is able to generate ∼4 μm motion, which meets the design goal.


Introduction
Photonic integration technology, i.e., the design and microfabrication of on-chip optical functions, is key to establishing advanced applications in, e.g., data communication and sensing. Several material platforms are available to create Photonic Integrated Circuits (PICs), including indium phosphide (InP) [1] and silicon dioxide/silicon nitride (SiO 2 /Si 3 N 4 ) [2]. Today, PICs with complex functionality can be designed and fabricated to contain both passive (waveguides, splitters) and active (lasers, detectors) optical functions.
Generic foundry-based processes in the photonic domain have brought PIC costs within the scope of many applications, i.e., ∼e 10-100 per chip [1,3,4]. A key bottleneck for market entrance is volume-compatible integration, i.e., assembling one or more PICs together with other optical and electrical components into a single housing [5]. Packaging requires the establishment of opto-mechanical connections, electronic interconnections and thermal management. The standardization of packaging and the automation of assembly processes for photonic products are still in their infancy. The current photonic packaging and assembly technologies are mainly based on custom-engineered solutions, motion of the bimorph actuator. Additionally, we relate the experimental results to the analytical and numerical models.
Firstly, the proposed short-loop bimorph actuators with the main design variants are introduced in Section 2. Then, Section 3 shows the experimental characterization work, together with analytical simulation. In Section 4, the results are discussed and Section 5 provides conclusions.

Design Variants
The core of the design consists of a set of waveguide (WG) beams and actuator beams. As far as the SiO 2 stack is concerned, these beams have the same dimensions (length, cross-section). The actuator beams are provided with a section of poly-Si (Figures 4 and 5). They interact with the WG array through a crossbar which connects the beams at their free ends. This design preserves the lithographically-defined pitch between the WG end-facets, and supports the joint motion of all WG beams by a limited number of actuators. The mechanical coupling between actuator and WG beams obviously leads to cross-sensitivities during operation.
We explore two different designs of the poly-Si heater, based on its location. The poly-Si heater either forms a loop across two adjacent beams, or is placed on the chip body at the base of the suspended structure while straight poly-Si strips conduct the heat to the rest of the poly-Si track. These designs are referred to as bridge design (BD) and non-bridge design (NBD), respectively. Their motion performance will be different. The heaters can be powered until a maximum allowable temperature is reached. In the NBD (Figures 4b and 5b), some of the heat will be conducted into the Si chip, and will therefore not effectively contribute to the bimorph's motion. The BD variant (Figures 4a and 5a) heats up more effectively along the entire heater structure. Modeling and experimental results presented in later sections confirm this expectation.
Another design variation concerns the integration of actuators in the overall design. The actuator beams are either placed adjacent to the WG beam array, or are integrated with the WG beams. These configurations are referred to as separate bimorph actuator (SBA) configuration or integrated bimorph actuator (IBA) configuration, respectively. Figure 4 presents both BD and NBD with the SBA configuration, which has a (mechanically) more compliant crossbar. As a result, this can allow more degree of rotational adjustment. Figure 5 shows both BD and NBD with the IBA configuration. In Appendix A, Table A1 summarizes the main design parameters and values for the MEMS design.

Fabrication
For the purpose of this paper, it is relevant to understand the material layers which constitute the WG and actuator beams. The complete fabrication process flow is summarized in Appendix B.
Both the WG beams and the actuator beams are made of the TriPleX-material, which in itself is a stack of various SiO 2 layers with a total thickness of about 16 µm. The WG beams contain a Si 3 N 4 core. This core is only ∼200 nm in thickness, and plays no role of significance in the mechanical behavior of the beams.
To create an actuator structure, a 5 µm-thick poly-Si layer is deposited. The SiO 2 /poly-Si structures are patterned from the front side, and then are released from the bulk Si chip to form free-standing structures. During this release step, the poly-Si layer is protected by a 2 µm-thick (t pox ) layer of SiO 2 by plasma-enhanced chemical vapor deposition (PECVD). As the PECVD SiO 2 stack causes additional post-release deflection, it is removed from the SiO 2 -only sections, i.e., beyond the poly-Si section of the beams.

Basic Models
Estimation of the vertical post-release-position of a beam's end-facet is done in two steps. First, the curvature of the trilayer section is computed using a multilayer-beam model [23,24]. The model describes the curvature of a multilayer beam depending on dimensions and material properties (Young's modulus, coefficient of thermal expansion) of the layers, as well as strains that result both from thermal and residual/intrinsic stress. The thermal stress results from a considered temperature difference (∆T). To find the post-release deformation, the difference between the deposition temperature of each layer and the ambient temperature needs to be considered. For estimating the motion range, the difference between the ambient temperature and the operation temperature needs to be used. Details of the model can be found in Appendix C. The computed curvature applies to the trilayer section of the beams. To compute the end-facet position of this section, the following equation is used: where L poly is the length of poly-Si on the suspended beam structure, and ρ is the radius of curvature of the trilayer stacks ( Figure 6). Second, the remaining SiO 2 stack is considered straight by approximation; the small curvature in the SiO 2 -only sections resulting from the intrinsic stress gradient can be ignored. Hence, the end-position of the entire beam can be approximated by where L wg is the length of the entire suspended cantilever. If only considering the thermal effects from the fabrication, there are two major temperature cycles: the epitaxial poly-Si growth at 1050 • C and later the PECVD SiO 2 deposition at 400 • C. By substituting these temperature values into the formulas in Appendix C, the post-release deformation can be computed. For simulating the actuation, the same model can be used as well, and a possible poly-Si operation temperature (e.g., 400 • C, which is far below the poly-Si recrystallization temperature ∼600 • C [25,26]) is considered. Figure 7 shows the simulation results for post-release deformation and motion, and suggests that L poly of 30 µm-75 µm, with a total SiO 2 beam length of 900 µm, can achieve the design goal. The total beam length of 900 µm was chosen, since this gave a good fabrication yield and low stiffness for the entire WG beam and actuator array.

Experimental Results
This section summarizes the characterization of the various bimorph actuator designs. First, the post-release deformation and motion range of actuator beam arrays are characterized for varying L poly on post-release deformation (Section 3.1). Then, the motion behavior of the bridge design (BD) and non-bridge design (NBD) actuator is compared (Section 3.2). Moreover, the integrated bimorph actuator (IBA) and separate bimorph actuator (SBA) designs are compared, focusing on their post-release deformation (Section 3.3) and rotational actuation (Section 3.4).
The post-release deformation was characterized by a white-light interferometer, Contour GT-K 3D optical profilometer (Bruker Corporation, Billerica, MA, USA) with ∼10 nm-50 nm resolution. The motion range was measured using a Polytec MSA-400 vibrometer (Polytec GmbH, Waldbronn, Germany) with ∼50 nm-100 nm resolution in low frequencies. The Keithley 2611 source meter (Keithley Instruments, Inc., Cleveland, OH, USA) was used to drive the actuators.

Varying Poly-Si Length versus Post-Release Deformation
The post-release deformation (δ pr ) is obtained by measuring the position of the crossbar with respect to the top layer of the SiO 2 stack. Since the PECVD oxide is locally removed from the beam, the thickness (t pox = 2 µm) of that layer needs to be added. Thus, δ pr is the sum of δ mpr and the thickness of PECVD SiO 2 ( Figure 6): where δ mpr is the distance from the top surface of the base to the top surface of the crossbar (free end). The post-release deformation of a series of bimorph actuator designs, with various of poly-Si lengths, including 1-pair, 2-pair and 3-pair of NBD-actuator (50 µm pitch) variants ( Figure 8), was measured. L poly varies from 20 µm to 200 µm. The results are plotted in Figure 9. Each measurement point is the average result of three measurements on the free-end of the crossbar. A single-beam analytical model is included in the graph, both with and without considering intrinsic stress. Values for the intrinsic stress were estimated and summarized, and can be found in Table A1.
The analytical model that does consider the intrinsic stress is in a good agreement with the measurements. For a given L poly , the post-release deformation should be the same for each of the actuators. The minimum and maximum difference between all designs at the same value for L poly is in the order of ∼1 µm-2 µm. As described in Section 1, for the specific PHASTFlex case, an offset of roughly 6 µm needs to be compensated. The results indicate that this is possible with a poly-Si range of 20 µm-50 µm.

Varying Poly-Si Length versus Motion
The design of 1-pair bimorph actuator was chosen to analyze the relationship between L poly and motion. Note that the number of actuator pairs does not influence the generated motion (Section 4.4). Figures 10 and 11 show the motion measurements of a series of 1-pair BD and NBD bimorph actuators with varying L poly , respectively. Each measurement point is the average result of five measurements on the free-end of the crossbar, and their standard deviation is included. Both plots show the generated vertical motion as a linear function of its power consumption, and the motion increases with a longer L poly . Out-of-plane motion (µm) L poly =20µm L poly =30µm L poly =50µm L poly =60µm L poly =70µm L poly =80µm L poly =90µm L poly =100µm L poly =150µm L poly =200µm Figure 11. Motion characterization with standard deviation of the 1-pair NBD bimorph actuator with different poly-Si lengths. Figures 12 and 13 show the post-release deflection and motion range for the NBD and the BD actuator, respectively. These graphs can be used to design WG arrays for selected specifications. The solid line points out the 6 µm nominal offset between the WG layers in the chip in the PHASTFlex case. The flip-chip assembly error in vertical direction is estimated to be ∼1 µm-2 µm (chip to chip). This implies that, in the worst case, the initial WG offset needs to be ∼8 µm (6 µm nominal WG layer offset plus 2 µm assembly error). The WG array needs to be designed for this situation, i.e., it needs to have an ∼8 µm initial offset. If the assembly error is in the other direction, the offset between the WG layers after assembly will be 4 µm (6 µm nominal WG layer offset minus 2 µm assembly error). Hence, the motion range must be such that this position is achieved, i.e., a motion range of 4 µm (8 µm targeted initial offset minus 4 µm) is needed. For these design specifications, designs with ∼30 µm poly-Si lengths are suitable.
There is a lower limit to the manufacturable poly-Si length, which is 40 µm for the BD variant. This is due to the need for perforations in the SiO 2 structure for successful release that requires a perforation with a suitable dimension to realize the suspended beams. The perforation has the minimum dimension of ∼38 µm (determined by the dimension of the SiO 2 beam). From this perspective, the NBD actuator shows more design freedom on choosing L poly .    Figure 14 shows the height profile of an array of suspended beams of a NBD-IBA configuration (W wgp = 50 µm, L poly = 200 µm). The inner and outer WGs show the same post-release deflection, which is confirmed by the measured horizontally-flat crossbar, see Figure 14c,d). The post-release deflection of the beams is ∼35 µm, which matches the analytical result shown in Figure 9. The curvature of the poly-Si section was measured to be ∼230 /m, and the SiO 2 -only section remains almost straight (Figure 14a,b). For the SBA configuration (W wgp = 50 µm, L poly = 80 µm), with more suspended beams and a longer crossbar structure, the post-release deflection of inner beams (the inner WG and actuator beams) is still the same. The outer actuator beam is about 800 nm lower than the others (Figure 15a,b). This could be due to some remaining underneath the SiO 2 . We did notice some Si remaining underneath the SiO 2 layer along the crossbar, due to differential etching rates. This could lead to the slightly curved crossbar (Figure 15c,d). Furthermore, the structures do have a gradient intrinsic stress. For very long structures, such as the crossbar, this might result in some curvature.
For a small-footprint configuration, such as the IBA with the 50 µm pitch, all suspended beams (with the same stacks and geometry) have the same free-end position. In addition, this configuration tends to have robuster manufacturing yield. For a large-footprint configuration with more suspended beams (e.g., the SBA configuration), the crossbar is not entirely flat. However, the WG beams that are located in the middle of the crossbar, and the relative post-release offset of their end-facets is still small. Hence, this configuration might still be useful for an alignment task, despite the deformation of the crossbar.

Motion of Separate Bimorph Actuator (SBA) and Integrated Bimorph Actuator (IBA)
The MEMS designs also allow rotational adjustment of the WG beam array along the optical axis by differentially powering the individual bimorph actuators. This is needed to compensate the chip pre-assembly error in that direction. This section compares the rotational adjustment of the SBA and IBA configuration, respectively.
In this experiment, a safe power was applied to drive the actuators to prevent any permanent change in the poly-Si resistance. The single power source we used provided a maximum total power of 120 mW was given to drive different number of actuators, and this power was distributed evenly to each activated pair of driven actuator beams. The vertical position of each of the WG end-facets was measured, by actuating a subset of the bimorph actuators. Each measurement point is the average result of five measurements on the free-end of the crossbar.
A typical example of an IBA design (W wgp = 250 µm, L poly = 50 µm) is shown in Figure 16a, including 5-pair actuator beams (WG beams). Figure 16a depicts the measurement points observed with the vibrometer. Figure 16b shows the measurement results when actuating different combinations of actuator beams. Moreover, the standard deviation is included. As expected, by differential powering of the actuators, the angle of the cross-bar can be adjusted. Generally, with less pairs of actuator activated, the crossbar deforms more significantly. For example, using the chip body as reference plane, the crossbar has a rotational angle of 40 • with 2-pair activated actuators, while the crossbar rotates 30 • with 3-pair activated actuators. When all actuator pairs are driven with the same power, the crossbar keeps horizontally flat, confirming the outcome of former experiments. In this situation, the vertical offset among each WG end-facets are within ∼100 nm. Figure 17a shows a SBA design (W wgp = 50 µm, L poly = 80 µm): three pairs of bimorph actuators are placed to each side of the WG array. Besides the end position of the WG beams, the end position of the bimorph actuators are also measured, to reflect the deformation of the entire crossbar. The measurements include either actuating both sets of actuators or only one set, and the results are shown in Figure 17b. When only one set of actuators are activated, the crossbar has the largest rotational angle, and this angle is measured to be the same (∼33 • ). The crossbar remains straight, particularly where the WGs are located.

Motion Limit
The maximum motion range that can be achieved is determined by the maximum temperature the actuators can withstand. With an increased power input, more resistive heating is generated. The actuator will burn out when temperature reaches a certain point.
To experimentally find the limit, a 1-pair BD-bimorph actuator (50 µm pitch, L poly = 90 µm) was measured, and its experimental results are presented in Figure 18. Each motion measurement point is the average result of five measurements on the free-end of the crossbar. The maximum deflection was measured to be 9.34 µm at 88.56 mW dissipated power. Figure 18a shows a linear relationship between the supplied power and the vertical deflection, and all the motion measurements have less than 50 nm standard deviation. The measured electrical resistance of poly-Si firstly shows a slight decrease and then a rapid raise, and eventually drops after reaching a peak (Figure 18b). During this experiment, the actuator started to emit visible light after applying 7 V. Further increase of the voltage increased light intensity until the actuator burnt out. This indicates that the poly-Si lattice is recrystallized when the temperature is elevated, which allows more excited electrons passing through. After the experiment, the actuator is electrically defective; mechanically, the overall actuator was still intact, although some cracks and delamination have occurred ( Figure 19).

Bridge Design and Non-Bridge Design
To compare the performance of the BD and NBD, 1-pair bimorph actuator with the same poly-Si length (L poly = 50 µm) were chosen. Before any resistive heating, the resistance of the NBD was measured to be 473 Ω, whereas the resistance of the BD was measured to be 655 Ω.
Only considering the vertical motion, the BD variant deflects more than the NBD, with the same voltage ( Figure 20). The experimental results are verified by a FEM (finite element model) approach, using COMSOL Multiphysics R (Version 5.2, COMSOL, Inc., Burlington, MA, USA). In all FEM, the heat flux and the heat transfer in materials are set under the same condition. Figure 21 shows the temperature gradients of both designs, when their heaters reach the same maximum temperature (i.e., 400 • C). Since the heater is located on the base in the NBD variant, the actuator beams get heated less effectively than that of the BD.

Temperature Estimation
To avoid the high resistive temperature that can destroy poly-Si lattices permanently, we prefer to operate the actuators safely below the thermal runaway temperature. To estimate the mean actuator temperature, the same model as the post-release deformation model was built (Appendix C). The modeling result shows that a linear deflection change of 10 nm/ • C for the poly-Si section with 5 µm in thickness and 60 µm in length. Figure 22a firstly shows the motion measurement of a 1-pair BD bimorph actuator within a safe voltage range, and by substituting the measured motion into the analytical model, Figure 22b shows the measured motion as a function of the analytically simulating actuator temperature, together with the FEM result. Both simulations show almost the same trend in the curve of motion and temperature.

Bimorph Actuator Pitch and Number of Pairs
We have two WG-pitch system (W wgp ) for our photonic application, and by activating different number of pairs of actuator beams (N ba ), it is possible to fine-tune the WG end-position (Section 3.4). To compare the performance of the bimorph actuators with these two design variables, four BD bimorph actuators (L poly = 60 µm) were characterized ( Figure 23). For all experiments, each pair of bimorph actuator (with the same electrical resistance) was electrically connected in parallel. The comparison shows the motion as a function of power consumed by each pair of the actuator, and it corresponds with a linear fit. The results also show that each pair of the actuator generates almost the same vertical deflection when consuming the same power, regardless of the various W wgp and N ba . As reported in Section 3.4, the design parameters (W wgp and N ba ) do affect the crossbar deformation during the actuation, though they do not influence the vertical motion. Out-of-plane motion (µm) 1-pair, 50µm pitch 2-pair, 50µm pitch 2-pair, 250µm pitch 3-pair, 250µm pitch

Propagation and Coupling with Curved Beams
The curvature of the WG beams is small (∼230 /m and less) compared to the permissible curvature (up to ∼14,000 /m, depending on WG designs [2]) in the TriPleX material platform, and will not have any significant impact on the light propagation. The end-facets of the flexible WG beams will be under an angle with respect to the InP waveguide end-facets, for two reasons. Firstly, as part of the alignment process the flexible WG beams are bent, so the propagation direction is not perpendicular to the InP WG end-facet. However, this angle is very small and has an ignorable effect (e.g., 8 µm vertical deflection over a beam length of 900 µm, which corresponds to ∼0.5 • ). Secondly, the end-facet of the WG beams have a slope due to the etching process in which they are realized of ∼3.5 • . For small mode fields, the sensitivity of coupling loss to angular error is more tolerant. Using the analytical models from [27], and taking a mode-field diameter of ∼1 µm at 1550 nm wavelength, shows that an angular offset of 5 • leads to a coupling loss of less than 0.05 dB ( Figure 24). Experimentally, our previous work has proven that the flexible WG beams and etched end-facets offer suitable mode-field diameters, and allow good optical coupling [10,28].

Conclusions
We have proposed a photonic alignment concept, which uses high-precision flip-chip bonding and on-chip MEMS actuation for pre-aligning the PICs and fine-positioning photonic WG beams, respectively. The out-of-plane and rotational actuation of the bimorph actuator is the focus of this paper. The design challenge of the bimorph actuator is to compensate the nominal offset between the WGs that need to be aligned, and the need of providing sufficient motion. We proposed an electrothermal actuator design, with short-loop poly-Si heaters.This concept is implemented in different designs, based on the location of the heaters and the integration of the actuators with the WG beams. The initial position of the WG end-facets and the motion range are both dependent on the same design parameter, L poly . For the specific application case, the results suggest that the optimal poly-Si length is approximately 30 µm, with a SiO 2 length of 900 µm, for the PHASTFlex application. For other applications, obviously other values can be chosen.
Both IBA and SBA configurations have shown the capability of controlling the rotation of WG beams. With an evenly-assigned power on every actuator pair, the end-position of WG beams can maintain the same height. Having a more-compliant crossbar (e.g., with larger pitch in the IBA or the SBA), there can be more freedom in adjusting the WG rotation.
Moreover, an analytical single-beam model with multilayer stacks has been adjusted to estimate the post-release deformation. The model also suggests that, below 500 • C-600 • C actuator temperature, the generated motion can be more than 5 µm, agreed well with the FEM. This can be beneficial for predicting the post-release deformation, as well as the actuation temperature for the application that is thermally governed. Acknowledgments: We appreciate the great help for fabricating the MEMS by Tom Scholtes at Else Kooi Laboratory (EKL)/Delft University of Technology in the Netherlands. Furthermore, we also thank Tjitte-Jelte Peters from our department for the discussion of the models.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

Appendix A. Main Design Parameters
The dimension of the main design parameters are determined by photonic and mechanical demands, as summarized in Table A1.  Figure A1 illustrates the main steps of the MEMS fabrication on the front-side of the TriPleX wafer. The 16 µm-thick SiO 2 is actually made of a layer of 8 µm-thick thermally-grown SiO 2 as a bottom cladding on a Si substrate, and another 8 µm-thick PECVD SiO 2 as a top cladding. These two SiO 2 stacks are annealed, to obtain uniform optical and mechanical properties ( Figure A1a). The structures illustrated in Figure A1 represent a bondpad, a SiO 2 beam with poly-Si and a SiO 2 beam without poly-Si. On the front-side of the TriPleX wafer, 5 µm-thick poly-Si is firstly deposited, followed by the deposition of Al with 1 µm thickness ( Figure A1b). The Al stack is used for building electrical leads. After this, Al and poly-Si are patterned ( Figure A1c). As mentioned in Section 2.2, to protect the poly-Si layer during the subsequent Si etching steps to release the suspended structures, a layer of 2 µm PECVD SiO 2 is used to cover the entire wafer ( Figure A1d), and then the areas where the bondpad and suspended structure without the poly-Si pattern are located are opened by plasma etching (Figure A1e). Removing the PECVD SiO 2 layer on the bondpad enables the electrical probing and wire-bonding with PCBs (Printed Circuit Boards) [9]. As the PECVD SiO 2 stack causes additional post-release deflection, for the SiO 2 -only section without poly-Si, there is no need to keep this PECVD SiO 2 . This is also a major difference, between the front-side recipe that was used before [8][9][10] and the current one that needs to facilitate the post-release deformation within the PHASTFlex alignment scheme.

Appendix B. Fabrication Process
After this local removal of SiO 2 , another very thin SiO 2 (∼200 nm) is deposited ( Figure A1f). Figure A1g shows how the SiO 2 stack is patterned. The Si-reinforced release method is depicted from Steps h to j ( Figure A1), by introducing a passivation layer to reinforce the SiO 2 -Si stack when it is released from the Si substrate. Once the suspended structures are released, the passivation is removed ( Figure A1k), and the remaining Si underneath the SiO 2 stack is isotropically etched away completely ( Figure A1l). Lastly, the bondpad and beam structure without the Si pattern are opened again ( Figure A1m). A slight over-etch is proceeded to expose the bondpad.
where E i , I i , A i , b i , t i and α i are the Young's modulus, moment of inertia, cross-section area, width, thickness and coefficient of thermal expansion (CTE) of layer i, respectively. Moreover, i is strain of material i, mainly resulting from residual stress and thermal effects. ∆T is the temperature cycles that the beam undergoes during fabrication or actuation. Figure A2 describes the simplified cross section of the trilayer beam, and Table A1 includes the measured dimensions and material properties of the multi-stack beam. The end-deflection (δ tri ) of the trilayer stacks (TriPleX SiO 2 /poly-Si/PECVD SiO 2 ) can be approximated by: where L poly is the length of the poly-Si section (Table A1). As mentioned in Section 2.3, as the rest of the beam only includes SiO 2 , it can be simplified as a straight beam. Thus, the total deflection (δ pr ) of the beam can be concluded by where L wg is the length of the entire beam (Table A1).