Given the ever-increasing internet traffic, scaling the per-fiber transmission capacity is highly desired^[1]. Several multiplexing schemes, including wavelength-division multiplexing (WDM), polarization-division multiplexing (PDM), and mode-division multiplexing (MDM), are exploited to increase the communication capacity. Among them, MDM provides an attractive option to further expand the transmission capacity by introducing multiple modes for each wavelength carrier^{[2, 3]}.

Silicon optical switches based on a silicon-on-insulator platform offer the potential to enable high-speed optical networks due to their compatibility with a commercial complementary metal–oxide–semiconductor (CMOS) process^[4-6]. When leveraging multiple physical dimensions, a two-dimensional or three-dimensional switch can be realized to achieve scaled capacity. An on-chip multimode optical switch was reported based on quasi-phase-matching and a staged coupling method with the capability of three-wavelength and two-mode multiplexing^[7]. Recently, a WDM-compatible multimode optical switch working for three wavelengths and two modes was demonstrated^[8]. A silicon wavelength and mode switch-and-selector architecture with two wavelengths and two modes was reported with improved flexibility of the network^[9]. In contrast, mode and polarization dimensions own the merit of multiplexing signals with one laser source. Recently, silicon optical switches carrying four-mode and two-mode dual-polarization were demonstrated using Mach–Zehnder interferometers (MZIs), respectively^{[10, 11]}. To further scale the capacity, we presented a switch with three hybrid dimensions including mode, polarization and wavelength^[12]. However, previous multi-dimensional switches were based on thermo-optic tuning with a slow response time of several µs. Dynamic and high-speed multi-dimensional optical switching is highly desired.

In this paper, we report a 2 × 2 high-speed mode and polarization switch-and-selector architecture (HMPSA) based on MZI switches with PN junction-based phase shifters. Compared to thermo-optic switches, the electro-optic (EO) HMPSA exhibits a fast switching time of 11 ns benefitting from the free-carrier plasma dispersion effect. Moreover, the numbers of the MZIs are reduced by half in this configuration. The measured overall insertion losses (ILs) are below 9.03 dB. The inter-modal and intra-modal crosstalk (XT) of the HMPSA are lower than –15.86 and –7.32 dB for all the channels at 1550 nm, respectively.

The architecture of the silicon-integrated HMPSA for two modes and dual polarizations is shown in Fig. 1(a). Similarly to a wavelength switch-and-selector architecture, the HMPSA can route any optical signal from one input port to an arbitrary output port.

The proposed switches consist of mode polarization beam splitters (PBSs), mode multiplexers (MMUXs), polarization rotators (PRs) and crossings. The mode multiplexers are based on asymmetric directional couplers to multiplex the TE₁ and TM₁ modes^[13]. The crossings are implemented by 90°-crossed multi-mode interferometer (MMIs)^[14]. The polarization beam splitters and rotators are the building blocks of the process design kit (PDK) from Advanced Micro Foundry (AMF)^[15]. The device has two input ports (I₁, I₂) and two output ports (O₁, O₂). Firstly, the four mode- and polarization-multiplexed channels with TE₀, TE₁, TM₀ and TM₁ modes from an input port I₁ or I₂ are converted to the fundamental modes by the corresponding mode de-multiplexers (MDEMUXs) and the PBSs. After rotating the TM fundamental mode to the TE fundamental mode by PRs, signals are switched to the output waveguides by the follow-on MZI-based EO switches. After the TM channels are recovered by the PRs, the switched signals are multiplexed by the MMUXs and the polarization beam combiners (PBCs) and finally routed to output port O₁ or O₂.

Various structures have been used as MMUX, such as Y-splitters^{[16, 17]}, MMI coupler^[18] and asymmetrical directional couplers^{[19, 20]}. Here we choose asymmetric directional couplers to achieve mode (de-)multiplexing for their better compactness in this design. As shown in Fig. 1(b), the widths of the waveguides carrying the two modes and dual polarization are chosen to be 0.4, 0.845, 0.4 and 1.045 µm, respectively. We use 10-µm-length adiabatic tapers to connect the multimode waveguides of different widths. For TE₁ and TM₁ modes, the optimized gaps between the access waveguides and the multimode waveguides are 0.2 and 0.3 µm, respectively. The coupling lengths are designed as 18.5 and 8.25 µm, respectively. The coupling between a few-mode fiber and a silicon multimode waveguide is still challenging^[21-23]. Consequently, two MMUX-As and two MDEMUX-As outside the 2 × 2 HMPSA are used to couple high-order modes.

Fig. 1(c) depicts the schematic configuration of the 2 × 2 high-speed MZI switch. A 2 × 2 MMI structure is used as the 3 dB coupler for its compact size and broad-band response. One arm of the MZI contains an EO phase shifter based on a lateral 50-μm-long p–n diode to induce the π phase shift for high-speed switching operations.

The fabrication of the HMPSA chip is carried out by ultra-violet lithography on a silicon-on-insulator wafer with a 220-nm-thick silicon layer on 2 µm buried dioxide layer using CMOS processes in AMF, Singapore. Fig. 2(a) shows the micrograph of the fabricated 2 × 2 HMPSA with a footprint of 2.2 × 0.9 mm². Fig. 2(b)–2(e) depict magnified micrographs of a PBS, a MMUX, a PR, a 2 × 2 high-speed MZI switch and waveguide crossings, respectively.

In the experiment setup, a tunable light source (Keysight 81960A) and an optical power meter (Keysight N7744A) are utilized to measure the spectral responses of the HMPSA. Light is coupled into and out of the HMPSA by grating couplers with a shallow etching depth of 70 nm and periods of 630 and 980 nm for supporting TE and TM polarization, respectively. The fiber-to-chip coupling losses are 5.3 and 6.4 dB/facet at 1550 nm, respectively. Fig. 3 shows the measured transmission spectra, which have been normalized by the grating couplers and the MMUX on the same chip. Table 1 summarizes the ILs performance of the building blocks. Take the input port I₁ of TE₀ channel as an example. When the powers supplied to the heater are 6.23 and 29.03 mW, the signals are switched to output port O₁ and O₂, respectively. By tuning the powers applied to the corresponding MZI switches, optical signal from each input port is capable of being routed to all available output ports for all the channels. Here we manually adjusted the switching power to allow the maximum output optical power, while it is possible to achieve switch control and calibration with built-in power monitors and a feedback loop^{[24, 25]}. For all the channels, the ILs are below 9.03 dB at 1550 nm, which are mainly caused by the manufacturing imperfection, incomplete coupling in the MMUXs and the PBSs. The ILs can be reduced by using dual-core adiabatic tapers^[26] or higher silicon layer thickness^[27] with improved fabrication tolerance. The inter-modal XT performance of the HMPSA is characterized by transmitting an optical signal from a defined input port and then measuring the transmission spectrum at all output ports sequentially. All MZI heating powers are manually adjusted to enable maximum output at the same port. For example, when we measure the inter-modal XT from port I₁-TE₀ to port O₂-TE₁, the MZI switches for the port I₁-TE₀ and I₁-TE₁ are both set to output at the port O₂. As shown in Fig. 3, the measured inter-modal XT values of the fabricated device are lower than –15.86 dB at 1550 nm for all the channels.

Fig. 4 shows the measured intra-modal XT introduced by the high-speed MZI switches. For the port I₁-TE₀, the measured intra-modal XT are below –10.13 dB at 1550 nm. For all the inputs, the overall intra-modal XT are lower than –7.32 dB at 1550 nm. The relatively large crosstalk introduces significant, but tolerable, impairments for a quadrature phase-shift keying (QPSK) format by using an on-chip self-homodyne coherent detection scheme^[28]. Further intra-modal XT reduction can be realized via two MZI switches^[29] or using the variable coupler^[30].

We then measure the dynamic routing performance of the switch by applying a 1 MHz square-wave voltage signal to the device. The peak-to-peak drive voltage is 1.1 V biased at a direct current voltage at 0.8 V. Fig. 5 shows the measured response for the switch. The measured 10%–90% switching time upon electrical tuning are 11 and 10 ns for the rising and falling edges, respectively.

In conclusion, a 2 × 2 HMPSA is experimentally demonstrated based on EO MZIs. The ILs of the switch are 3.55–9.03 dB at 1550 nm. The measured inter-modal and intra-modal XT values are better than –15.86 and –7.32 dB at 1550 nm, respectively. The switching time (10%–90%) for the rising and falling edges are 11 and 10 ns, respectively. The demonstrated silicon 2 × 2 HMPSA has promising potential for future high-speed optical networks with switching time of only nanoseconds. Furthermore, this scheme can be extended to higher-order modes by employing cascaded subwavelength-grating-based directional couplers^[31].

This work was supported in part by the National Key Research and Development Program of China under Grant 2019YFB2203600, the National Natural Science Foundation of China (NSFC) under Grant 61975115/61835008/62035016, and the Science and Technology Commission of Shanghai Municipality under Grant 2017SHZDZX03.