A polarization-insensitive thermo-optic switch is proposed and demonstrated on the silicon-on-insulator platform with a 220 nm silicon core layer. The present device is based on the Mach–Zehnder interferometer structure, consisting of polarization-insensitive power splitters and polarization-insensitive phase shifters (PIPSs). The polarization-insensitive power splitter has been realized by employing an adiabatic directional coupler, which utilizes the fast adiabatic mode evolution by introducing cubic Bézier curves on outer contours, providing broadband 3-dB power splitting for TE and TM polarization modes with only 70 µm coupling length. For the novel PIPSs, the ridge waveguide with large aspect ratio, based on the mode hybridness property, could obtain the same power consumption (Pπ) for an optical switch working at TE and TM polarizations. Experimental results indicate that the measured insertion losses are less than 2 dB and the extinction ratios are larger than 15 dB over a 40 nm wavelength band (covering the C-band).

Silicon-on-insulator (SOI) is an attractive platform for realizing low fabrication cost photonic integrated circuits (PICs) due to its complementary metal–oxide–semiconductor (CMOS) compatibility. The high refractive index contrast between core (nSi = 3.46) and cladding (nSiO2 = 1.44) could provide a strong confinement of light wave, which is of benefit to high-density integration. However, a high refractive index contrast will also bring a large polarization dispersion, especially for devices with large aspect ratios. The optical switch is an important fundamental block in optical neural networks1,2 and programmable photonic circuits.3,4 Due to the strong polarization dependence, the devices usually work in the single TE polarization state, which limits the potential to make full use of the polarization dimension of light to increase the capacity of optical interconnects and computations. In addition, the optical switch shows its application prospect in beam steering.5,6 Generally, the tuning efficiency is limited for one specific polarization state in the wavelength tuning dimension.7 The introduction of polarization-insensitive optical switches can also provide the possibility of doubling the scanning range.

To realize polarization-insensitive optical switches, there are mainly two options. The first one is the polarization diversity scheme. Using polarization diversity devices, such as polarization splitter-rotators (PSRs), two orthogonal polarization states are transmitted within two independent optical switches or switch arrays,8 which will almost double the number of optical devices and electrical connections. The polarization beam splitter (PBS) can also be introduced to phase shifters to reduce complexity.9 However, the optical switch still requires individual heaters to control two polarization states. Another common option is using polarization-insensitive devices: splitters and phase shifters. This scheme has been implemented on a 340 nm-thick silicon core layer by designing square waveguides and optimizing the width of multimode interference (MMI) couplers,10,11 which is incompatible with the commonly used 220 nm-SOI platform. For the polarization-insensitive power splitters based on the 220 nm-SOI platform, the MMI width with the same beat length for both polarizations is not wide enough to support a sufficient number of modes in the multimode section, which will induce a relatively large insertion loss (IL).12 For the phase shifters, the mode field of the square waveguide (220 × 220 nm2) is mainly concentrated in the outer boundary of the waveguide. The transmission loss and power consumption are greatly increased compared to those of the common single-mode waveguide.13,14

In this work, we demonstrate a polarization-insensitive thermo-optic switch consisting of novel polarization-insensitive adiabatic direction couplers (PIADCs) and polarization-insensitive phase shifters (PIPSs), for the first time to the best of our knowledge. The PIADC employs the fast adiabatic mode evolution method by introducing cubic Bézier curves on outer contours, achieving a broadband polarization-insensitive power splitting. The PIPSs, with a low transmission loss for the TE0 mode and a high mode conversion efficiency for TM0–TE1–TM0, could provide the same Pπ for TE0 and TM0 modes on the 220 nm-SOI platform. The experimental results show that the measured Pπ for TE and TM polarizations is 24.5 and 24.6 mW, respectively, the insertion losses are less than 2 dB, and the extinction ratios (ERs) are larger than 15 dB in the wavelength range from 1525 to 1565 nm (covering the C-band).

Figure 1 shows the schematic of the proposed polarization-insensitive thermo-optic switch, which consists of two PIADCs and two PIPSs. The SOI platform with a 220-nm silicon core layer is considered in this work. The buried oxide layer and the cladding oxide layer are 2 and 1.8 µm, respectively. The heater of 3 µm width for the PIPS is located upon the cladding oxide layer. As shown in Fig. 1, the launched TE0 or TM0 mode is split into two parts by the PIADC. For the TE0 mode, it propagates through the PIPSs without mode conversion. For the TM0 mode, it first converts to TE1 mode by the mode converter and passes through the phase-shifting section as a TE1 mode. Then, the TE1 mode converts back to TM0 mode by another mode converter. The two arms with identical waveguide structures and the heaters are arranged above. The TE0 or TM0 modes in the two arms are recombined by the PIADC. The switching states depend on the phase-shifting Δφ introduced by the thermo-optic phase shifters.

FIG. 1.

3D schematic of the proposed polarization-insensitive thermo-optic switch and cross section of PIPS (inside the black dashed box).

FIG. 1.

3D schematic of the proposed polarization-insensitive thermo-optic switch and cross section of PIPS (inside the black dashed box).

Close modal

Polarization-insensitive 2 × 2 3-dB coupler is one of the basic building blocks of the proposed optical switch, which could be obtained using a bent directional coupler (bent-DC),15 subwavelength grating (SWG) structures,16 and adiabaticity directional couplers (ADCs).17 Among them, ADC is a preferred scheme to achieve broadband and fabrication-tolerant power splitting. The schematic of the proposed PIADC is shown in Fig. 2, which consists of three parts: an input section, a coupling section, and an output section. In the input section, the gap between two waveguides W1 and W2 is gradually reduced from g1 to g0 using S-bend waveguides with length Lin. In the coupling section, the gap between the two waveguides is g0. A small g0 can improve the coupling strength and reduce the coupling length but also increases the fabrication difficulty. g0 is chosen to be 0.18 µm. The waveguide widths in the input port of the coupling section are W1 = 0.3 µm and W2 = 0.5 µm, gradually changing to W0 = 0.4 µm through the coupling section. In the output section, the two waveguides with widths of 0.4 µm drifted apart until separated by a gap of g2 by using two S-bend waveguides. The gaps g1 and g2 are selected to be 1.5 µm to ensure sufficiently low coupling strength between the two waveguides. In order to illustrate the mode evolution process clearly, the TE polarization state is taken as an example. The mode profiles in Fig. 2 are the first-order (red box) and second-order (green box) eigenmodes of the double-waveguide system with input/output ports of each section. The light input from the wide waveguide at the input port evolves into the first-order mode in the coupling section and then outputs at the two output ports with the same intensity and phase. The light input from the narrow waveguide evolves into the second-order mode in the coupling section. The light at the two output ports has the same intensity and the π phase difference. The Lin and Lout should be long enough to ensure a high mode evolution efficiency, which are set to be 40 and 20 µm, respectively, according to calculation.

FIG. 2.

Schematic of the proposed PIADC. The mode profiles show the simulated eigenmodes at each section’s input or output ports for TE polarization (first-order eigenmodes in red boxes and second-order eigenmodes in green boxes). The schematic of the coupling section is shown in the black dashed box, cubic Bézier curves are indicated by the red lines, and the red solid circles represent the end points or curve points.

FIG. 2.

Schematic of the proposed PIADC. The mode profiles show the simulated eigenmodes at each section’s input or output ports for TE polarization (first-order eigenmodes in red boxes and second-order eigenmodes in green boxes). The schematic of the coupling section is shown in the black dashed box, cubic Bézier curves are indicated by the red lines, and the red solid circles represent the end points or curve points.

Close modal
The coupling section of the ADC usually requires a long evolution length to avoid unexpected coupling and obtain uniform 3-dB power splitting.18 We propose a Bézier curve optimized ADC, of which shape is convenient and intuitive to control,19 to dramatically shorten the evolution length. The configuration is shown in the black dashed box of Fig. 2, The outer contour of the waveguide is cubic Bézier curve B(t), including two end points and two curve points, which determine the shape of the curve. Three constraints are set to reduce the optimization complexity and ensure a smooth connection: (1) A1 and A4 locate on the edge of the waveguide; (2) A1 and A2 have the same Y coordinate; and (3) A3 and A4 have the same Y coordinate. We define L1 = a*Lc and L2 = b*Lc. Taking A1 as the origin, the cubic Bézier curve equation is expressed as
Bt=A11t3+3A21t2t+3A31tt2+A4t3,t0,1,
(1)
where A1, A2, A3, and A4 are (0, 0), (a*Lc, 0), [(1 − b) *Lc, ΔW], and (Lc, ΔW), respectively. ΔW = W0W1 = W2W0. The two waveguide widths along the X direction are expressed as W1 + B(t) and W2B(t). As mentioned above, the shape of ADC is determined by the curve parameters a and b and the coupling length Lc. Eigenmode expansion (EME) is a frequency-domain algorithm, which scales exceptionally well with propagation distance and is an ideal method for simulating long structures. We first simulate the minimum length required for the mode evolution efficiency of TE (Lth_TE) and TM (Lth_TM) polarizations to reach 99.9% under different curve parameters as shown in Figs. 3(a) and 3(b). Then, a larger value is selected as the Lth of the ADC [shown in Fig. 3(c)], that is, Lth = max (Lth_TE, Lth_TM). With the selected curve parameters, the evolution efficiency of the coupling section is equal to or greater than 99.9% with the corresponding length Lth for both polarizations. From Fig. 3(c), one can find that the minimum length of the coupling section is 58 µm with the curve parameters of a = 0.6 and b = 1. When the curve parameters are a = 0 and b = 0, the two waveguide widths are linearly tapered and the corresponding Lth is 299 µm. Hence, the length of the coupling section Lc could be shortened five times by using the Bézier curve optimized ADC. With high robustness, the coupling length Lc is chosen as 70 µm. Here, we denote the splitting ratio (SR) between the outputs as SR = TO1/(TO1 + TO2), where TO1 and TO2 are the transmittances at the O1 and O2 ports. The three-dimensional finite-difference time-domain (3D-FDTD) is used to calculate the insertion loss and splitting ratio of the entire PIADC. As shown in Figs. 3(d)3(f), the calculated splitting ratios are within 0.47–0.53 for both polarizations. The insertion losses are below 0.02 dB for TE0 and 0.15 dB for TM0 within a 100 nm bandwidth. Figures 3(g) and 3(h) show the calculated SR spectra of PIADC with a waveguide width deviation of ΔW = 0, ±20 nm for TE and TM polarizations, respectively. The SR spectra are all within 0.47–0.53, indicating a large fabrication tolerance.
FIG. 3.

Calculated (a) Lth_TE, (b) Lth_TM, and (c) Lth with parameters a and b varying from 0 to 1; calculated transmission spectra at two output ports for (d) TE polarization and (e) TM polarization (the insertions are the corresponding propagation mode field at 1550 nm); (f) calculated insertion loss spectra of TE and TM polarizations in the wavelength range of 1500–1600 nm; calculated SR spectra of PIADC with a waveguide width deviation of ΔW = 0, ±20 nm for (g) TE and (h) TM polarizations.

FIG. 3.

Calculated (a) Lth_TE, (b) Lth_TM, and (c) Lth with parameters a and b varying from 0 to 1; calculated transmission spectra at two output ports for (d) TE polarization and (e) TM polarization (the insertions are the corresponding propagation mode field at 1550 nm); (f) calculated insertion loss spectra of TE and TM polarizations in the wavelength range of 1500–1600 nm; calculated SR spectra of PIADC with a waveguide width deviation of ΔW = 0, ±20 nm for (g) TE and (h) TM polarizations.

Close modal
To realize the polarization-insensitive optical switch, the PIPSs should be designed to have the same Pπ for TE and TM polarization modes. We use the HEAT solver in DEVICE Solutions to simulate the thermal distribution of the heaters. The thermal conductivities of silicon and silica are 148 and 1.38 W/(m K), respectively, and the heat convection coefficient between air and silica is 5 W/(m2 K). The thermal distribution is then imported into MODE Solutions to calculate the change in effective refractive index caused by heating. Pπ can be obtained by the following equation:
Pπ=Pλ2LΔneff,
(2)
where P is the power applied to the heater, L is the effective length, and λ is the working wavelength. For strip waveguides, Pπ as a function of waveguide width is calculated for TE0 and TM0 modes based on 340 nm-SOI and 220 nm-SOI platforms, respectively, as shown in Figs. 4(a) and 4(b). For the one based on the 340 nm-SOI platform, there are two intersection points of the two modes, located near 0.34 and 0.38 µm. The Pπ of TE0 and TM0 is the same at these two waveguide widths, which refers to the width of the phase shifter used in Refs. 10 and 11. Figure 4(b) shows that the Pπ curves of the TE0 and TM0 modes gradually progress with the increase in the waveguide width, but there is no intersection point. Therefore, the same Pπ cannot be achieved by only changing the waveguide width. The mode converters based on the mode hybridness property could achieve low-loss transmission for the launched TE0 mode and efficient mode conversion for TM0 to TE1.20 The introduction of mode converters paves a new way for implementing PIPSs, which can obtain the same Pπ for TE0 and TE1 modes. The Pπ of TE1 mode is also calculated as shown in Fig. 4(b). Fortunately, there is an intersection point of TE0 and TE1 modes with a waveguide width of 0.65 µm. However, the Pπ curve of TE1 mode is quite steep. About 6.1% Pπ difference will be introduced with only a 20 nm waveguide width variation, which means a quite high fabrication precision is required. In addition, the TE1 mode has a high field amplitude distribution along the side wall of the waveguide, which is not conducive to achieving low waveguide loss. Thus, the ridge type waveguide is considered. Here, the height of the slab layer is selected as 0.15 µm (0.07 µm shallow etched depth) to meet the standard process provided by the foundry. Figure 4(c) shows the effective refractive index of TE0, TM0, and TE1 modes with the slab layer width varying from 0.8 to 5 µm when the waveguide width is 0.8 µm. The effective refractive index of the three modes weakly changes when the slab width is larger than 2 µm. Therefore, the slab width is chosen to be Ws2 = 2 µm. The calculated Pπ curves for the TE0, TM0, and TE1 modes of ridge waveguide with different waveguide widths are shown in Fig. 4(d). The curve of TM0 mode is separated from the other two by a large distance. The TE0 and TE1 modes have the same Pπ with a waveguide width of Wwg = 0.85 µm. The Pπ differences with a Wwg of 0.7 and 1.0 µm are 0.55 and 0.34 mW, respectively, that is, a variation in the waveguide width of 150 nm induces only 2.5% difference in Pπ. Figure 4(e) shows the calculated Pπ of ridge waveguide with slab height deviation for TE0 and TE1 modes when Wwg = 0.85 µm, where the Pπ difference is less than 0.3 mW (1.4%). Thus, the ridge waveguide has a superior performance, which is selected for the phase shifter.
FIG. 4.

Calculated Pπ (at 1550 nm wavelength) of strip waveguide with different waveguide widths based on (a) 340 nm-SOI and (b) 220 nm-SOI platforms; (c) calculated effective index of TE0, TM0, and TE1 modes with different slab widths; (d) calculated Pπ of ridge waveguide for TE0, TM0, and TE1 modes with different waveguide widths; and (e) calculated Pπ of ridge waveguide with a slab height deviation of ΔH = 0, ±10, ±20, ±30, and ±40 nm for TE0 and TE1 modes when Wwg = 0.85 µm.

FIG. 4.

Calculated Pπ (at 1550 nm wavelength) of strip waveguide with different waveguide widths based on (a) 340 nm-SOI and (b) 220 nm-SOI platforms; (c) calculated effective index of TE0, TM0, and TE1 modes with different slab widths; (d) calculated Pπ of ridge waveguide for TE0, TM0, and TE1 modes with different waveguide widths; and (e) calculated Pπ of ridge waveguide with a slab height deviation of ΔH = 0, ±10, ±20, ±30, and ±40 nm for TE0 and TE1 modes when Wwg = 0.85 µm.

Close modal

The mode converters contain two taper structures with vertical asymmetry, as shown in Fig. 5. There is an offset Loffset = 40 µm between the heater and the mode converter to avoid heating of the mode converter. The mode conversion occurs in the first segment, where the waveguide width (W3) remains unchanged and the slab width Wslab is linearly tapered from W3 to Ws1. The calculated mode refractive index of the ridge waveguide is shown in Fig. 6(a). One can find that the mode hybridness happens when the slab width Wslab is around 0.7 µm, where the dispersion curves for the TM0 and TE1 modes are close to each other. In order to achieve efficient mode conversion, we choose W3 = 0.45 µm and Ws1 = 0.9 µm. The second taper with a length of Lt2 = 20 µm is used to connect the first taper to the ridge waveguide under the heater. The TM0–TE1 mode conversion efficiency is calculated by using EME solutions. The TM0–TE1 mode conversion efficiency can be greater than 99.9% when Lt1 > 40 µm as shown in Fig. 6(b). The Lt1 is set to be 50 µm for high conversion efficiency and large fabrication tolerance. The 3D-FDTD is used to verify the overall performance of the PIPSs. With the selected parameters, the transmission spectra and propagation mode fields for the TE0 and TM0 modes are shown in Figs. 6(c) and 6(d), respectively. The calculated losses are below 0.02 dB for the TE0 mode and 0.1 dB for the TM0 mode. The mode crosstalk for the TM0 mode is less than −19 dB within a 100 nm wavelength band.

FIG. 5.

Schematic of the proposed mode converter connected with a phase shifter.

FIG. 5.

Schematic of the proposed mode converter connected with a phase shifter.

Close modal
FIG. 6.

(a) Calculated mode effective index of the ridge waveguide; (b) calculated TM0–TE1 mode conversion efficiency with Lt1 varying from 0 to 80 µm; the simulated transmission spectra in the wavelength range of 1500–1600 nm and propagation mode fields at 1550 nm for the launched (c) TE0 and (d) TM0 modes.

FIG. 6.

(a) Calculated mode effective index of the ridge waveguide; (b) calculated TM0–TE1 mode conversion efficiency with Lt1 varying from 0 to 80 µm; the simulated transmission spectra in the wavelength range of 1500–1600 nm and propagation mode fields at 1550 nm for the launched (c) TE0 and (d) TM0 modes.

Close modal

The proposed device is fabricated with the E-beam lithography (EBL) followed by an inductively coupled plasma reactive ion etching (ICP-RIE) process. Figure 7 shows the detailed images of the fabricated device. The polarization division-multiplexings (PDMs), consisting of a PBS,21 a TE-type grating coupler, and a TM-type grating coupler, are added at the input/output ports of the switch. The amplified spontaneous emission (ASE) source, source meter, and optical spectrum analyzer (OSA) are used to characterize the fabricated device. The TE-polarized or TM-polarized light is launched to the devices through the input port, and the OSA is applied to record the transmissions at the cross or bar ports with the source meter applying different voltages on the heater. Two PDMs connected with a straight waveguide are fabricated on the same chip for power normalization. Figures 8(a) and 8(b) show the measured transmissions at 1550 nm wavelength with the applied power sweeping from 0 to 90 mW for TE and TM polarizations. The Pπ for TE0 and TM0 is 24.5 and 24.6 mW, and the Pπ difference is only 0.1 mW (0.4%), which indicates the superior performance of the proposed PIPSs. The minima of transmissions for two polarizations do not coincide completely, and the initial state has some deviations. Those are attributed to the random deviation of the phase delay in the Mach–Zehnder interferometer (MZI) arms and could be minimized by optimizing the fabrication processes and introducing a low random phase error design. With the same applied voltages, the measured transmissions at the bar and cross ports for TE and TM polarizations, when the switch state is on and off, are shown in Figs. 8(c) and 8(d), respectively. The extinction ratios are larger than 15 dB, and the insertion losses below 2 dB in the wavelength range from 1525 to 1565 nm for both TE and TM polarizations. The polarization-dependent loss (PDL) is less than 0.8 dB over a 50 nm wavelength band. Figures 8(e) and 8(f) show the measured response waveforms by applying the 1-kHz square-wave pulse train. The rise time (10%–90%) and fall time (90%–10%) are 6 µs (10 µs) and 12 µs (14 µs), respectively, for TE (TM) polarization. The thermo-optic modulation bandwidths are calculated to be 39 kHz for TE polarization and 29 kHz for TM polarization by using the first-order approximation. In Table I, we summarize the performances of several reported polarization-insensitive thermo-optic switches. The proposed optical switch, with a low insertion loss, a large extinction ratio, and a large bandwidth, could provide a practical solution for the 220 nm-SOI platform.

FIG. 7.

(a) Optical microscope images of the proposed polarization-insensitive MZI; enlarged view of (b) PIPSs, (c) PDM, and (d) PIADC; scanning electron microscope (SEM) image of the (e) input and (f) output ports of the coupling section; and (g) SEM image of the ridge waveguide of the phase shifter.

FIG. 7.

(a) Optical microscope images of the proposed polarization-insensitive MZI; enlarged view of (b) PIPSs, (c) PDM, and (d) PIADC; scanning electron microscope (SEM) image of the (e) input and (f) output ports of the coupling section; and (g) SEM image of the ridge waveguide of the phase shifter.

Close modal
FIG. 8.

Transmissions at 1550 nm with heater power sweeping from 0 to 90 mW for (a) TE and (b) TM polarizations; measured transmissions at (c) bar and (d) cross ports when switch state is on and off for TE and TM polarizations; measured response waveforms for (e) TE and (f) TM polarizations by applying the 1-kHz square-wave pulse train.

FIG. 8.

Transmissions at 1550 nm with heater power sweeping from 0 to 90 mW for (a) TE and (b) TM polarizations; measured transmissions at (c) bar and (d) cross ports when switch state is on and off for TE and TM polarizations; measured response waveforms for (e) TE and (f) TM polarizations by applying the 1-kHz square-wave pulse train.

Close modal
TABLE I.

Summary of the polarization-insensitive thermo-optic switch on the SOI platform.

ReferenceCore height (nm)Pπ (mW)aBandwidth (nm)bER (dB)IL (dB)
22  1500 ∼40 ⋯ ∼25 ∼0.5 
10  340 15.8 55 20 <4 
11  340 14.57 for TE 35 ∼15 ∼2.5 
14.85 for TM 
14  220 152.78 35 14 <4.3 
This work 220 24.5 for TE >40 15 <2 
24.6 for TM 
ReferenceCore height (nm)Pπ (mW)aBandwidth (nm)bER (dB)IL (dB)
22  1500 ∼40 ⋯ ∼25 ∼0.5 
10  340 15.8 55 20 <4 
11  340 14.57 for TE 35 ∼15 ∼2.5 
14.85 for TM 
14  220 152.78 35 14 <4.3 
This work 220 24.5 for TE >40 15 <2 
24.6 for TM 
a

The measured Pπ for TE and TM polarizations is indicated separately if given in references.

b

Bandwidth is under the present ER.

In summary, we propose and experimentally demonstrate a polarization-insensitive thermo-optic switch on a 220 nm-SOI platform, which consists of two identical PIADCs and PIPSs. A novel Bézier curve optimized ADC is proposed to obtain broadband polarization-insensitive 3 dB power splitting. With the optimized curve parameters, the coupling length could be dramatically shortened five times. The PIPSs are achieved by introducing mode converters and manipulating TE0 and TE1 modes with the same Pπ. The experimental results indicate that the insertion losses are below 2 dB, and the extinction ratios are larger than 15 dB for two polarizations over a 40 nm wavelength band (covering the C-band). In addition, the Pπ difference between TE and TM polarization modes is only 0.1 mW (0.4%). The proposed device has the potential for the large-scale optical switch array and optical phase array.

This work was supported by the National Major Research and Development Program (Grant No. 2021YFB2801703); the National Natural Science Foundation of China (Grant Nos. 62135011, 62105286, and 62335001); and the “Pioneer” and “Leading Goose” R&D Program of Zhejiang (Grant No. 2022C01103), the Fundamental Research Funds for the Central Universities.

The authors have no conflicts to disclose.

Shi Zhao: Conceptualization (equal); Data curation (lead); Formal analysis (equal); Investigation (lead); Methodology (equal); Visualization (lead); Writing – original draft (lead). Jingye Chen: Formal analysis (equal); Methodology (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Daoxin Dai: Funding acquisition (supporting); Supervision (equal); Writing – review & editing (supporting). Yaocheng Shi: Conceptualization (equal); Funding acquisition (lead); Methodology (equal); Supervision (lead); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
W.
Bogaerts
,
D.
Perez
,
J.
Capmany
,
D. A. B.
Miller
,
J.
Poon
,
D.
Englund
,
F.
Morichetti
, and
A.
Melloni
, “
Programmable photonic circuits
,”
Nature
586
,
207
216
(
2020
).
2.
Y.
Shen
,
N. C.
Harris
,
S.
Skirlo
,
M.
Prabhu
,
T.
Baehr-Jones
,
M.
Hochberg
,
X.
Sun
,
S.
Zhao
,
H.
Larochelle
,
D.
Englund
, and
M.
Soljačić
, “
Deep learning with coherent nanophotonic circuits
,”
Nat. Photonics
11
,
441
446
(
2017
).
3.
K.
Suzuki
,
R.
Konoike
,
H.
Matsuura
,
R.
Matsumoto
,
T.
Inoue
,
S.
Namiki
,
H.
Kawashima
, and
K.
Ikeda
, “
Polarization and wavelength performance improvement in large-scale silicon photonics switches
,”
J. Lightwave Technol.
41
,
841
847
(
2023
).
4.
L.
Song
,
T.
Chen
,
W.
Liu
,
H.
Liu
,
Y.
Peng
,
Z.
Yu
,
H.
Li
,
Y.
Shi
, and
D.
Dai
, “
Toward calibration-free Mach–Zehnder switches for next-generation silicon photonics
,”
Photonics Res.
10
,
793
801
(
2022
).
5.
S.
Zhao
,
J.
Chen
, and
Y.
Shi
, “
Dual polarization and bi-directional silicon-photonic optical phased array with large scanning range
,”
IEEE Photonics J.
14
,
6620905
(
2022
).
6.
C.
Li
,
X.
Cao
,
K.
Wu
,
G.
Qiu
,
M.
Cai
,
G.
Zhang
,
X.
Li
, and
J.
Chen
, “
Blind zone-suppressed hybrid beam steering for solid-state Lidar
,”
Photonics Res.
9
,
1871
1880
(
2021
).
7.
J. K.
Doylend
,
M. J. R.
Heck
,
J. T.
Bovington
,
J. D.
Peters
,
L. A.
Coldren
, and
J. E.
Bowers
, “
Two-dimensional free-space beam steering with an optical phased array on silicon-on-insulator
,”
Opt. Express
19
,
21595
21604
(
2011
).
8.
K.
Suzuki
,
R.
Konoike
,
S.
Suda
,
H.
Matsuura
,
S.
Namiki
,
H.
Kawashima
, and
K.
Ikeda
, “
Low-loss, low-crosstalk, and large-scale optical switch based on silicon photonics
,”
J. Lightwave Technol.
38
,
233
239
(
2020
).
9.
C.
Sun
,
Y.
Lai
,
Y.
Yu
,
S.
Fu
,
P. P.
Shum
, and
X.
Zhang
, “
Crossing-free on-chip 2 × 2 polarization-transparent switch with signals regrouping function
,”
Opt. Lett.
43
,
4009
4012
(
2018
).
10.
S.
Wang
and
D.
Dai
, “
Polarization-insensitive 2 × 2 thermo-optic Mach–Zehnder switch on silicon
,”
Opt. Lett.
43
,
2531
2534
(
2018
).
11.
H.
Yang
,
Y.
Kuan
,
T.
Xiang
,
Y.
Zhu
,
X.
Cai
, and
L.
Liu
, “
Broadband polarization-insensitive optical switch on silicon-on-insulator platform
,”
Opt. Express
26
,
14340
14345
(
2018
).
12.
J.
Chen
and
Y.
Shi
, “
Polarization-insensitive silicon waveguide crossing based on multimode interference couplers
,”
Opt. Lett.
43
,
5961
5964
(
2018
).
13.
L.
Zhang
,
S.
Hong
,
Y.
Wang
,
H.
Yan
,
Y.
Xie
,
T.
Chen
,
M.
Zhang
,
Z.
Yu
,
Y.
Shi
,
L.
Liu
, and
D.
Dai
, “
Ultralow‐loss silicon photonics beyond the singlemode regime
,”
Laser Photonics Rev.
16
,
2100292
(
2022
).
14.
W.
Li
,
L.
Xu
,
J.
Zhang
,
D.
Mao
,
Y.
D'Mello
,
Z.
Wei
, and
D. V.
Plant
, “
Broadband polarization-insensitive thermo-optic switches on a 220-nm silicon-on-insulator platform
,”
IEEE Photonics J.
14
,
6658807
(
2022
).
15.
X.
Chen
,
W.
Liu
,
Y.
Zhang
, and
Y.
Shi
, “
Polarization-insensitive broadband 2 × 2 3 dB power splitter based on silicon-bent directional couplers
,”
Opt. Lett.
42
,
3738
3740
(
2017
).
16.
S.
Han
,
W.
Liu
, and
Y.
Shi
, “
Ultra-broadband dual-polarization power splitter based on silicon subwavelength gratings
,”
IEEE Photonics Technol. Lett.
33
,
765
768
(
2021
).
17.
Y. L.
Wu
,
F. C.
Liang
,
H. C.
Chung
, and
S. Y.
Tseng
, “
Adiabaticity engineering in optical waveguides
,”
Opt. Express
28
,
30117
30129
(
2020
).
18.
X.
Sun
,
H. C.
Liu
, and
A.
Yariv
, “
Adiabaticity criterion and the shortest adiabatic mode transformer in a coupled-waveguide system
,”
Opt. Lett.
34
,
280
282
(
2009
).
19.
S.
Zhao
,
S.
Han
,
M.
Zhu
,
J.
Chen
, and
Y.
Shi
, “
Ultra-broadband dual-polarization and arbitrary ratio power splitters based on Bezier curve optimized multimode interference
,”
Opt. Lett.
48
,
1331
1334
(
2023
).
20.
D.
Chen
,
M.
Liu
,
Y.
Zhang
,
L.
Wang
,
X.
Hu
,
P.
Feng
,
X.
Xiao
, and
S.
Yu
,
C + L band polarization rotator-splitter based on a compact S-bend waveguide mode demultiplexer
,”
Opt. Express
29
,
10949
10957
(
2021
).
21.
H.
Wu
,
Y.
Tan
, and
D.
Dai
, “
Ultra-broadband high-performance polarizing beam splitter on silicon
,”
Opt. Express
25
,
6069
6075
(
2017
).
22.
S.
Nakamura
,
S.
Yanagimachi
,
H.
Takeshita
,
A.
Tajima
,
T.
Hino
, and
K.
Fukuchi
, “
Optical switches based on silicon photonics for ROADM application
,”
IEEE J. Sel. Top. Quantum Electron.
22
,
185
193
(
2016
).