Function-Versatile Thermo-Optic Switch Using Silicon Nitride Waveguide in Polymer

Abstract

A function-versatile thermo-optic switch is proposed and experimentally demonstrated using silicon nitride waveguides embedded in polymer cladding. The device consists of a 1 × 2 input splitter, 2 single-mode waveguides for phase shifting, and a thermally controlled 2 × 2 output coupler to give another degree of freedom in achieving phase-matching conditions. Combining the high waveguide birefringence of the thin silicon nitride waveguide and the excellent thermo-optic property of the polymer material, this device can realize multiple functions by applying different micro-heater powers, i.e., polarization-independent path switching, beam splitting, and polarization beam splitting. For the polarization-independent path switching, the fabricated device has shown a crosstalk suppression better than 10 dB for the TE mode and over 20 dB for the TM mode in the wavelength range from 1500 nm to 1620 nm. For the polarization beam splitting function, the device can reach a polarization extinction ratio greater than 10 dB at selected bands. This simple yet scalable device may find applications in polarization-multiplexed optical communication technology and complex photonic computing networks.

1. Introduction

Photonic integrated circuits (PICs) use photons rather than electrons to perform a wide variety of functions [1,2]. Because of their compact size, robustness, high performance, and low power consumption, PICs become ideal candidates for many applications, including high-speed data transmission for telecommunication [3,4], on-chip optical sensing [5], parallel photonic computing [6,7], light detection and ranging (LiDAR) systems for autonomous vehicles [8,9], etc. Under the PIC technology, beam splitters (BS), path switches (PS) and polarization beam splitters (PBS) are some of the key building blocks. For example, in modern optical communication networks, orthogonal polarizations of light have been used as independent data carriers and multiplexed to transmit in the same physical channel, thereby doubling the transmission capacity. A PBS becomes an essential component.

The BS and PS switches can be easily realized with Mach-Zehnder interferometers (MZIs) [10,11,12,13], and microring resonators [14,15]. However, the working polarization is commonly fixed in the silicon optical switches [12,13], because of their high birefringence and high insertion loss compared with the silica and polymer-based devices [16]. Although the silica [10] and polymer platforms [11] can certainly avoid the polarization-dependence, it is difficult to integrate polarization-sensitive elements on the same platform.

Though the BS and PS functions are relatively simple to achieve, the design of waveguide-based PBS requires some niche effort, using, e.g., asymmetrical directional couplers (DC) [17], grating couplers [18], multimode interference devices (MMIs) [19], Mach-Zehnder interferometers (MZIs) [20,21], and structures from arithmetic inverse design [22]. The realizations have been made on a variety of material platforms, such as silicon-on-insulator (SOI) [23,24], silicon [17,20], silicon nitride (SiNx) [25,26], and indium phosphide (InP) [27,28].

The first step to designing a waveguide-based PBS is to break the waveguide symmetry so that TE and TM modes exhibit different effective refractive indices and eigenmode profiles. Different polarization modes can then be coupled or guided into separate channels by a proper coupler or interferometer design. For example, a directional coupler can guide the polarization modes into different output ports at a propagation distance equal to multiple times their respective coupling lengths. However, this would require precise control of the fabrication process to simultaneously match the coupling lengths for both polarizations. The inverse-design-based PBS also demands precise fabrication technology to pinpoint the sub-wavelength structures at nanometer accuracy [22,29,30].

Furthermore, in most of the demonstrated PBS devices so far, the TE and TM modes are usually guided to fixed ports, without the possibility to actively select the output port for a given polarization [17,18,19,20,21,22]. Moreover, the device function itself, PS or PBS, cannot be “switched” easily, thereby limiting the control of light flow on a chip for a fully flexible architecture, i.e., it is difficult to integrate the polarization-dependence, polarization-independence, and switchable output ports on a single integrated device.

In this work, we come up with a design for a function-versatile optical switch using a combination of phase-shiftable MZI and MMI structures. The heaters in the MZI single-mode waveguide arms as well as in the first part of the 2 × 2 MMI region work together to reach the required phase difference between the polarization modes under thermal tuning. The remaining part of the 2 × 2 MMI will guide light into different ports, depending on their polarization and phase conditions. Under different heater power settings, the device can provide versatile functions for polarization-independent BS, PS, as well as PBS. We believe this work can provide some inspiration for the development of an on-chip polarization control device to be used in optical switch networks and photonic computing technology.

Silicon nitride PIC technologies [31] provide a compact footprint, low-loss waveguides (<1 dB/m), high birefringence, and a wide wavelength range (400–2350 nm), as well as a compact footprint and a build-a-block complimentary to SOI and III–V PICs. Though the thermo-optic of SiNx is relatively low [32] (~10⁻⁵/K), the SiNx fabrication technologies are well compatible with the polymer material, which has a large thermo-optic coefficient [33] (~−10⁻⁴/K). Though the thermo-optic effect suffers from a large responding time (~ms) [34] compared with the electro-optic effect (~ns) [35], the fabrication of a thermo-optic polymer waveguide is simple and flexible, and the propagation loss is low (<1 dB/cm). Therefore, we choose to work on a thin SiNx waveguide buried in polymer cladding [36] for the following design. This type of waveguide is known to exhibit a strong birefringent effect, i.e., the TE and TM modes show very different effective indices and mode profiles due to a large aspect ratio of the core. The fabrication technology is simple, tolerant, and low-cost. The waveguide propagation loss for either polarization can be as low as 1 dB/cm [36]. The coupling loss between the waveguide and the standard single-mode fiber (SSMF) can be as low as 1~3 dB/facet [36]. It can be further reduced by using reduced-core fiber (RCF) with a reduced mode-field diameter and a taper structure on the waveguide side [37]. The SiNx waveguide embedded in polymer proves to be robust and can keep its loss properties upon heating. Due to the low mode confinement factor, a large percentage of the mode intensity lies on the polymer cladding, and the waveguides can be thermally tuned effectively [36,38,39].

The following sections introduce the waveguide design, the device design, and the simulation results, followed by a description of the fabrication technology and the measurement setup. The device is first qualitatively characterized by the polarization-selective examination of the output waveguide near-field profiles through an imaging system. The device is then further verified by the fiber-waveguide-fiber transmission measurement. The data are analyzed, and the difference from the design expectations is discussed. The prospects for further development and applications are summarized at the end.

2. Design and Simulation

The cross-section of the SiNx waveguide buried in polymer is shown in Figure 1a. In this study, the thickness and the width of the SiNx waveguide are chosen to be 150 nm and 2 μm to satisfy the single-mode condition. The refractive indexes of the SiNx and polymer cladding layers (ZPU12-450 from ChemOptics Inc, Korea) are 1.949 and 1.450 at 1550 nm, respectively, measured by an ellipsometer (SE-VM-L, Wuhan Eoptics Technology Co., Ltd., Wuhan, China). The bottom cladding thickness is set to 15 μm to satisfy the thermal balance between the device and the silicon heatsink. The top cladding thickness is set to 6 μm for efficient thermal tuning of the waveguide. The waveguide is single mode above the wavelength of 1495 nm. The polymer materials feature high thermo-optic coefficients (−1.14 × 10⁻⁴/K) and low thermal conductivity. As most of the light field resides in the cladding, the thermo-optic effect of the polymer material dominates [36,38]. Microheaters with 7 μm width and 100 nm thickness are designed to provide efficient thermal tuning according to our previous work [40].

The schematic of the proposed function-versatile thermo-optic switch is depicted in Figure 1c. The device is constructed by a 1 × 2 MMI splitter (460 × 30 μm²), connected to two single-mode waveguides as phase-shifting arms, and subsequently a 2 × 2 MMI (1875 × 30 μm²) for path switching. In total, four microheaters are applied, two of which are used to tune the phase on the waveguide arms (H₂, H₃, 1000 μm length), while the other two are used to tune the phase on the 2 × 2 MMI (H₁, H₄, 100 μm length). The upper output port is defined as port A, and the lower is port B. Taper waveguides (width increases from 2 μm to 7 μm in 200 μm length) are added to reduce the coupling loss between the single-mode waveguide and MMI coupler. A gap of 500 μm is set to suppress the thermal crosstalk between the two phase-shifting arms. Figure 1d shows the simulated results of the polarization insensitive 1 × 2 MMI splitter and the 2 × 2 MMI coupler for the TE and TM polarizations without a heater.

Critical to the function are the phase values from the end-plane of the 1 × 2 MMI (X1) to the end-plane of the heater-region in the 2 × 2 MMI (X2), in relation to the up and down paths, as well as the TE and TM polarizations. The phases at plane X2 can be expressed as:

$${\phi }_{X2-TE-upper/lower}={\phi }_{X1-TE-upper/lower}+\mathsf{\Delta }{\phi }_{Arm-TE-upper/lower}+\mathsf{\Delta }{\phi }_{MMI-TE-upper/lower}$$

(1)

$${\phi }_{X2-TM-upper/lower}={\phi }_{X1-TM-upper/lower}+\mathsf{\Delta }{\phi }_{Arm-TM-upper/lower}+\mathsf{\Delta }{\phi }_{MMI-TM-upper/lower}$$

(2)

where Δφ denotes the imposed phase change in the single-mode waveguide arm or in the MMI region by the heater. The phase difference between the upper and lower paths for the TE and TM modes at X2 can be written as:

$$\mathsf{\Delta }{\phi }_{TE-X2}=\left(\mathsf{\Delta }{\phi }_{Arm-TE-upper}-\mathsf{\Delta }{\phi }_{Arm-TE-lower}\right)+\left(\mathsf{\Delta }{\phi }_{MMI-TE-upper}-\mathsf{\Delta }{\phi }_{MMI-TE-lower}\right)$$

(3)

$$\mathsf{\Delta }{\phi }_{TM-X2}=\left(\mathsf{\Delta }{\phi }_{Arm-TM-upper}-\mathsf{\Delta }{\phi }_{Arm-TM-lower}\right)+\left(\mathsf{\Delta }{\phi }_{MMI-TM-upper}-\mathsf{\Delta }{\phi }_{MMI-TM-lower}\right)$$

(4)

The net phase difference between the TE and the TM paths at X2 is defined as:

$$\psi =\mathsf{\Delta }{\phi }_{TE-X2}-\mathsf{\Delta }{\phi }_{TM-X2}$$

(5)

Table 1. The phase configuration for different device function: polarization independent beam splitter (BS), path switch (PS), and polarization beam splitter (PBS). p and k are both numbers from the integer set Z.

Function	Output	ψ	ΔφTE-X2
I-Beam Splitter (BS)	Port A and Port B	kπ, k∈Z	pπ, p∈Z
II-Path Switch (PS)	Port A	2kπ, k∈Z	−π/2 + 2pπ, p∈Z
	Port B	2kπ, k∈Z	π/2 + 2pπ, p∈Z
III-Polarization Beam Splitter (PBS)	TE → Port A TM → Port B	π + 2kπ, k∈Z	−π/2 + 2pπ, p∈Z
	TM → Port A TE → Port B	π + 2kπ, k∈Z	π/2 + 2pπ, p∈Z

summarizes the conditions to reach different device functions under specific phase settings.

In practice, it is difficult to satisfy the phase conditions for the polarization beam splitting (PBS) function listed in Table 1 because the imposed phase change must be able to differentiate the polarization modes precisely. One can adopt the polarization-dependent nonlinear effect in a crystal, e.g., quadratic electro-optic effect, to construct an MZI-based PBS where the applied electric field will lead to a prominent index change only for a selected polarization direction [41]. However, the technology involved is still challenging.

Silicon nitride waveguides in polymer are easier to fabricate and exhibit strong birefringence [36]. Though the thermo-optic effect itself is mostly isotropic, the difference in the fundamental TE/TM mode distributions can still give an effective differential phase tuning. However, we have encountered difficulty reaching the target functions by tuning the phases on the two single-mode waveguide arms alone. Basically, for the PS and PBS functions, we fail to find a reasonable thermal setting on the two arms to match the phase conditions required for each function simultaneously.

Inspired by our previous work, where a multiport all-logic optical switch is developed in a compact polymer MMI [42], we find that when light diverges in the multimode region (higher-order modes join in), the total mode-field area/intensity distribution of the TM polarization is significantly larger than that of the TE component, subject to more influence from the temperature gradient induced by the thermal electrodes in the polymer cladding (negative thermo-optic effect, −1.14 × 10⁻⁴/K [36]) as well as in the SiNx core (positive thermo-optic effect, ~10⁻⁵/K [32]). Tuning the phases in the first part of the 2 × 2 MMI region gives us an extra degree of freedom in the design. In this way, the relative phase difference between the TE and TM modes, in addition to their respective phase differences between the upper and lower paths, can be matched more easily. Once the input phase is set (at plane X2), the remaining part of the MMI behaves similar to a 2 × 2 coupler and results in constructive interference for light at the desired output ports. Hence, the switch function is complete.

To concretize the design, we have used commercial software (Lumerical) for the thermal simulation and subsequent light field propagation simulation based on the eigenmode expansion method. The thermal simulation of the single-mode waveguide (WG1) under 50 mW/mm heat power is shown in Figure 2a. The maximal temperature is 240 °C, well below the degradation temperature (300 °C) of the chosen polymer material. Figure 2a also shows the simulated thermal crosstalk between two single-mode waveguides (WG1 and WG2). It is shown that it needs at least a 100 μm gap for WG2 to keep a room temperature (25 °C) when WG1 is heated by a maximal power of 50 mW/mm, indicating that the gap between the two waveguides should be larger than 200 μm when 50 mW/mm heat power is applied to both WG1 and WG2. Both TE and TM modes undergo a negative shift of the effective index upon heating, but at different rates/slopes, as demonstrated in Figure 2b, in which the difference between the TE and TM effective indexes versus heater power is plotted as the line with triangle markers. The relative index change, and in turn the phase difference, between the TE and TM modes is the key to developing an interferometer-based PBS. The maximal difference of the phase change between the TE and TM modes in the single-mode waveguide can be calculated by:

$$\mathsf{\Delta }{\phi }_{TE-TM}=\frac{2\pi }{\lambda }\left(\mathsf{\Delta }nef{f}_{\max }-\mathsf{\Delta }nef{f}_{\min }\right)L$$

(6)

where Δneff_max and Δneff_min are the maximal and minimal difference of the effective refractive indices between the TE and the TM modes in the range of 0–50 mW/mm heat powers. L is the length of the microheater on the single-mode waveguide. To achieve 2π relative phase shift, the electrode length should be at least 419 μm long, as obtained from Equation (6), and is set to 1000 μm in our design to reduce the local heat power density and temperature for long-term practical use.

Table 2. Function list of the versatile thermo-optic switch at the required phase conditions, heat power configurations and simulated light field propagation results. p and k are both numbers from the integer set Z.

Function	Heat Powers (mW)	Output	Simulation Results
			TE Polarization	TM Polarization
I-BS	ψ = kπ, k∈Z ΔφTE-X2 = pπ, p∈Z
	H1 = 0, H2 = 0, H3 = 0, H4 = 0
II-PS	ψ = 2kπ, k∈Z ΔφTE-X2 = −π/2 + 2pπ, p∈Z
	H1 = 0, H2 = 0, H3 = 0, H4 = 2.4
	ψ = 2kπ, k∈Z ΔφTE-X2 = π/2 + 2pπ, p∈Z
	H1 = 2.4, H2 = 0, H3 = 0, H4 = 0
III-PBS	ψ = π + 2kπ, k∈Z ΔφTE-X2 = −π/2 + 2pπ, p∈Z
	H1 = 0, H2 = 5, H3 = 16.5, H4 = 3.8
	ψ = π + 2kπ, k∈Z ΔφTE-X2 = π/2 + 2pπ, p∈Z
	H1 = 3.8, H2 = 16.5, H3 = 5, H4 = 0

lists the functions of the device, the required heater power, and the simulated light field propagation results. I-BS shows the phase configurations for the 3-dB beam splitting (BS) function, in which the input light is equally divided into two parts, irrelevant to their polarizations. II-PS section shows the function of polarization-insensitive path switching (PS). Both the TE and TM lights can be guided into the same physical output channel at the same time. Section III-PBS shows the function of a polarization beam splitter (PBS). The input TE and TM lights can be selectively guided into opposite output ports. It is worth mentioning that the heater powers of each function can be set with different parameter combinations as long as the phase conditions are satisfied. Table 2 shows one of the simulated symmetrical power configurations (H₁ and H₄, H₂ and H₃) for the switchable PS and PBS functions as an example.

3. Fabrication and Measurement Setup

The device is fabricated on a 4-inch silicon wafer (substrate), and the flow is shown in Figure 3a. Commercial polymer material (ZPU12-450 series from ChemOptics Inc, Daejeon, Republic of Korea), which has a refractive index of 1.45 at 1550 nm, was adopted as cladding. Firstly, the bottom polymer cladding is spin-coated, cured under UV illumination, and baked at 200 °C for 30 min. Then, the SiNx core was directly deposited on the cured bottom polymer cladding by plasma enhanced chemical vapor deposition (PECVD) using SiH₄ and NH₄ gases. After that, the SiNx thin film was characterized by an ellipsometer and found to be 1.949 at 1550 nm. Following the deposition, the waveguide core was patterned by conventional contact lithography and subsequent inductively coupled plasma (ICP) etching. The same polymer material is then spin-coated and cured as the top cladding. The micro-heaters are made with a second lithography, metal evaporation, and lift-off process. Finally, the wafer is diced into bars/chips for measurement.

The characterization of the device follows two steps, as adopted in our previous work [37,42]. Firstly, a polarization-selective imaging system is built to examine the near-field light profiles at the output facet of the chip. The heater power configuration is recorded for each of the switch functions. The experimental setup is shown in Figure 3b. A tunable laser (EXFO T100S-HP) covering 1550 nm to 1620 nm is used as the light source. The light is coupled to the chip with a standard single-mode fiber. The heater pads are contacted by needle probes connected to a series of current sources (Keithley 2400). The imaging system, consisting of a collimation lens, an adjustable aperture, a rotatable polarizer, and a focusing lens, is placed at the output of the chip. The polarization state of the input light needs to be adjusted and determined by the three-wheel fiber polarization controller, the rotatable polarizer, and the infrared CCD camera. After that, we use the imaging system to simultaneously display the two output spots images and get a suitable heat power configuration.

Secondly, fiber-chip-fiber alignment and transmission measurement are carried out by switching the imaging system with a standard single-mode fiber at the output facet (Figure 3c). The output fiber is connected to a detector system (EXFO CT440-PDL). The transmission spectra are obtained by scanning the wavelengths through a home-made LabVIEW program.

4. Results and Discussion

The near-field images at the device output facet (at 1550 nm) and the transmission spectra are summarized in Figure 4 with respect to each function. All the transmission data are normalized with the straight single-mode waveguide included on the same chip as a reference. The corresponding heat power configurations are recorded in

Table 3. Heat power configurations for the I-BS, II-PS, and III-PBS functions in the experiment.

Function	Output	H2 (mW)	H3 (mW)	H4 (mW)
I-BS	Port A and Port B	40.7	12.2	5.2
II-PS	Port A	41.6	10.0	5.3
	Port B	14.5	42.9	3.4
III-PBS	TE → Port A TM → Port B	18.6	8.8	3.9
	TM → Port A TE → Port B	9.8	10.4	2.5

. The thermal simulation shows that the maximal required temperature (212 °C for 42.9 mW) is much smaller than the degradation temperature (300 °C) of the polymer material. The local heater temperature can be further decreased by increasing the microheater’s length. Previous work [43] shows that the polymer-based device can work steadily over several months. Since the device suffers from a fabrication error (e.g., the original state of the device is not an ideal 3 dB beam splitter), the values of H₂, H₃, H₄ are different from the simulated values to compensate for the fabrication error. H₁ and H₄ are symmetrical microheaters, their heating powers can be exchanged with each other if the output port is needed to be reversed, such as in the configurations of the BS and PBS functions in Table 2.

For the BS function as shown in Figure 4a, light should be divided into two identical parts at the two output ports without applying any heaters. However, we observed quite a large imbalance for the pure passive measurement and had to adjust heaters H₂, H₃ and H₄ for more balanced light spots on the camera. This indicates that the structure is not perfectly symmetric and an initial phase delay exists between the upper path (A) and lower path (B) owing to the fabrication inaccuracies. After thermal adjustment, the device can function as a polarization-insensitive beam splitter across a broad wavelength range. In particular, the transmission measurement shows a flat response for the TM mode from 1540 nm to 1580 nm with a power imbalance lower than 1 dB. The transmission loss is higher in the TE mode. Nevertheless, an imbalance lower than 1 dB is also achievable from 1500 nm to 1550 nm. Fine tuning of heater powers can improve the balance in other wavelength bands.

Both the imaging and transmission results have shown a higher loss for the TE mode. To investigate this, we have performed a standard cutback measurement using 3 sets of straight waveguides with different lengths (0.31 cm, 0.82 cm, and 1.5 cm) to extract the propagation loss (slope) and the coupling loss with the standard single mode fiber (intercept). It is found that the fiber-chip coupling loss is 4.0 dB/facet for the TM mode and 8.5 dB/facet for the TE mode at 1550 nm, respectively. The propagation loss is 1.90 dB/cm for the TM mode and 5.56 dB/cm for the TE mode, respectively. Compared with the previous work [36,38], the SiNx layer appears to have a much higher intrinsic loss, and the PECVD technology adopted in this work leaves much room for improvement.

For the PS function, as shown in Figure 4b, light should be guided into either of the two output ports regardless of its polarization state. The transmission measurement of the crosstalk for the PS function (to A) is shown as an example, and the results for PS to B are similar. The PS function proves to be broadband, i.e., over 10 dB of crosstalk suppression can be achieved for the TE mode, and over 20 dB can be achieved for the TM mode nearly for the entire range from 1500 nm to 1620 nm.

For the PBS function, as shown in Figure 4c, the TE and TM modes should be guided into the opposite output ports. It shows one of the conditions under which the TE mode and TM mode are guided into Port B and Port B, respectively. It shows that over 10 dB polarization extinction ratio (PER) can be achieved for the wavelength range from 1500 nm to 1560 nm. The TM modes behave better as the propagation loss is lower and a larger part of the light field resides in the cladding for more efficient thermal tuning. Over 20 dB of crosstalk suppression can be achieved between the two ports for the TM modes.

It is noted under which the transmission spectra show some periodic bumps, especially for the TE mode in the PBS function. This is caused by the instable phase tuning, as manual current adjustment via needle contact is adopted in this experiment. The phase condition may not be satisfied ideally. The phase error makes the output intensity change more rapidly during the wavelength scanning, and as the TE mode has a higher change of effective index when increasing the heating power, the TE polarization has a more frequent bump along the measured spectra. This can be improved by proper packaging design with stable wire bonds and dedicated driver circuits to regulate the phase automatically and precisely.

The crosstalk for the TE modes appears to be wavelength sensitive, and the highest crosstalk suppression occurs at 1523.1 nm with 13.1 dB. We believe this is again attributed to the nonoptimal PECVD technology, which leaves a large concentration of residual N-H bonds in the layer that gives rise to high material absorption [44]. This is particularly severe for the TE mode because, compared with its TM counterpart, a larger ratio of the light field stays in the core. Upon heating, not only the real part of the refractive index is changed but also the imaginary part undergoes a shift, resulting in extra loss and a different phase tuning than expected. Nevertheless, the fabricated device demonstrates the basic ability of switching between different functions by varying the thermal settings as proof of concept. A plan for improvement will be made as part of our future work.

5. Summary and Conclusions

To summarize, a function-versatile thermo-optic switch using silicon nitride waveguides embedded in polymer cladding is proposed, fabricated, and measured. Micro-heaters are added on the single-mode waveguide arms for phase-shifting but also on the front part of the 2 × 2 coupler region to facilitate path/polarization switching. Under different phase settings, the device can be “switched” to provide different functions, including polarization-independent beam splitting, path switching, and polarization beam splitting. The theory and simulation results are first presented, followed by device fabrication and experimental verification from output facet imaging and fiber-chip-fiber transmission measurement.

In future work, the SiNx deposition technology must be improved to reduce the waveguide loss and allow more accurate phase tuning. Other transparent materials, such as TiO₂ and Al₂O₃, can also be adopted as the core layer by evaporation or sputter technology. Instead of cumbersome needle probe contacts, we plan to develop a dedicated circuit to drive the electrodes automatically and more precisely. Moreover, the microheater length of the single-mode waveguide can be further increased to reduce the local heat power density and temperature for long-term use.

This work could lead to the construction of a low-cost M × N switch network with versatile functions, including path switching and polarization multiplexing, to be used in the optical communication or computation systems. It may also lead to the realization of a compact, chip-based analyzer to determine the state of polarization in a fiber.