The mid-infrared (MIR) spectral region used for biomedical and sensing applications is of great importance, because most molecules display stronger fundamental vibrational absorptions in this region ^[1–3]. Not only that, the pursuit of high-capacity optical communications in the MIR spectral region has intensified in recent years. The realization of a low-loss hollow core photonic bandgap fiber (HC-PBGF) waveband offers a great opportunity to realize a wide and low-loss 2 μm transmission window ^[4]. Furthermore, thulium-doped fiber amplifiers (TDFAs) provide an exceptionally wide (100 nm) bandwidth in this wavelength range, and thus long-distance communication in the 2 μm transmission window is considered to be feasible ^[5]. From then on, several high-speed and high-capacity optical communication systems have been demonstrated experimentally by using the proposed HC-PBGF or other fibers ^[6–8]. 100 Gbit/s wavelength-division multiplexing (WDM) transmission in the 2 μm waveband was successfully demonstrated over 1.15 km of low-loss HC-PBGF and over 1 km of solid core fiber (SCF) ^[6]. After that, an externally modulated 4×10Gbit/s non-return-to-zero (NRZ) on–off keying (OOK) WDM signal transmitted through 1.15 km of low-loss HC-PBGF, employing an InP-based Mach–Zehnder modulator (MZM) for the first time ^[7], to the best of our knowledge. Recently, 80 Gbit/s data transmission using the direct-detection optical filter bank multicarrier (FBMC) modulation technique is achieved, which is the highest single-channel bit rate through a 100-m-long SCF designed for single-mode transmission at 2 μm ^[8].

To reach the potential of WDM and coherent communications in the 2 μm waveband, a suite of active and passive functional integrated components should be designed to improve the performance of the transmission system. As known, the matured complementary metal–oxide–semiconductor (CMOS) technology has been developed for integrated silicon photonics, which has the distinct advantages of small footprint, high density, low loss, and reduced power consumption with improved stability ^[9–17]. Until now, many key components in 2 μm optical communication systems have been implemented on InP, silicon, and other semiconductor materials systems, such as MIR lasers ^[18–21], MIR modulators ^[22–25], and high-speed MIR photodetectors (PDs) ^[26–29]. Besides, much research on basic passive functional MIR devices has been reported, such as MIR fiber-to-chip grating couplers ^[30], multimode interferometer (MMI) ^[31,32], Mach–Zehnder interferometer (MZI) couplers ^[33], arrayed waveguide grating (AWG) ^[34–38], microring resonators (MRRs) ^[39,40], and polarization control devices ^[41]. In addition, the mode-division multiplexing (MDM) technique provides another dimension to increase the optical transmission capacity ^[42–46]. It is believed that the on-chip mode (de)multiplexer at 2 μm has a great potential for chip-scale high-speed optical interconnect applications.

In this paper, we design and fabricate on-chip two-mode and four-mode division multiplexing photonic circuits at the wavelength of 2 μm by using tapered asymmetrical directional couplers. The mode (de)multiplexers and vertical grating coupler are fabricated on a silicon-on-insulator (SOI) platform. In the experiment, the coupling loss of the grating coupler and the average crosstalk of four channels are measured over a wide wavelength range from 1950 to 2020 nm. Moreover, stable data transmission with 5 Gbit/s NRZ-OOK through the fabricated mode (de)multiplexers is further demonstrated. The experimental results including clear eye diagrams and bit error rate (BER) curves are obtained, and the penalties at a BER of 3.8×10−3 are less than 2.5 dB for all the channels.

As shown in Fig. 1, the four-mode division multiplexing structure is composed of a mode multiplexer and a mode demultiplexer, which both have three tapered directional couplers and four input/output ports. Compared to a conventional asymmetrical normal directional coupler, the tapered structure has greater fabrication tolerance and wider working bandwidth ^[43]. As shown in Fig. 1, inset is the zoom-in view of the tapered asymmetric directional coupler, which parallel-couples a narrow silicon waveguide with width w1 to a wide tapered waveguide (from w2a to w2b with a center width of w2) with a coupling length L and gap g, respectively. The adiabatic couplings from the fundamental mode to higher-order modes (TE0−TE1, TE0−TE2, TE0−TE3) rely on the phase matching between the waveguides, i.e., the effective refractive index of the fundamental TE0 mode of the narrow waveguide equal that of the high-order mode in the wide waveguide at the center wavelength of 2 μm. The conversion process can be described as follows: from input ports, the fundamental mode TE0 is first used for the excitation of high-order modes via the tapered directional couplers. After a propagation distance, these four modes are then converted back into the fundamental modes for detection by the corresponding asymmetric directional coupler.

By using the finite difference eigenmode (FDE) method, we first calculate the dispersion relationship about the mode effective refractive index and width of the silicon waveguide with height h=0.22μm at the wavelength of 2 μm. As shown in Fig. 2(a), width w=0.65μm for the TE0 mode of the narrow waveguide, central width w=1.32μm for the TE1 mode, central width w=2.02μm for the TE2 mode, and central width w=2.705μm for the TE3 mode of the wide waveguide are chosen since their effective indices have almost the same value neff=2.18. In order to relax the fabrication limitations, we then design these sloped tapers with Δw=w2b−w2a=0.1μm, as shown in Fig. 2(a). By using the three-dimensional (3D) finite-difference time-domain (FDTD) method, we then simulate the whole transmission process for three high-order modes. In Figs. 2(b)–2(d), it can be observed that the fundamental mode is converted into the desired high-order modes by the asymmetric directional coupler. The mode transmission and crosstalk (less than −18dB) are calculated and presented in Figs. 2(b)–2(d).

The optimal widths for all the waveguides of the designed four asymmetric directional couplers are summarized, as shown in Table 1. For the asymmetric directional couplers, the width of the narrow waveguide and the gap are chosen to be 0.65 and 0.2 μm, respectively. According to the phase matching condition shown in Fig. 2, we simulate the whole evolution processes and obtain the optimized coupling lengths of 69.8, 91.8, and 94.8 μm for the TE1, TE2, and TE3 modes.

The measured microphotograph of the fabricated four-channel MDM device is shown in Fig. 3(a), which consists of a four-mode multiplexer and demultiplexer. The designed silicon photonic device is fabricated on a standard SOI wafer with a 220-nm-thick top silicon layer and a 2-μm-thick buried oxide (SiO2) layer. The proposed structure is fabricated by the 248-nm-deep ultraviolet lithography and inductively coupled plasma etching. Waveguide outlines are etched fully down 220 nm to the buried oxide, while the grating couplers are shallow etched down nominally 70 nm. The size of the fabricated device is ∼1.8mm×0.23mm, and the length of four-mode multiplexing is 200 μm. The microscope image in Fig. 3(a) shows the whole view of the device. Details of three asymmetric directional couplers and the vertical grating coupler are shown in Figs. 3(b)–3(d). The width of the narrow waveguide is 660 nm, the wide waveguide is tapered from 2.678 to 2.785 μm, and the coupling gap is 203 nm, as shown in Fig. 3(d). As a comparison, we characterize the performance of the fabricated two-mode (TE0, TE1) multiplexing structure first, which has the same structure parameters as shown in Fig. 3(b).

In the experiment, a supercontinuum source (OYSL SC-5-FC) and a long wavelength optical spectrum analyzer (Yokogawa AQ6375B) are used to measure the transmission spectra from the output ports. As shown in Fig. 4(a), the discrepancy between the measured insertion loss of the vertical grating couplers and the simulated coupling efficiency is probably due to an over-etching error that occurred in the fabrication. Most insertion loss of a single straight waveguide comes from the vertical grating coupler, including ∼20dB coupling loss from the fiber-chip/chip-fiber vertical coupling system for TE polarization around the wavelength of 2 μm. The transmission spectra and the crosstalk of two channels are measured by coupling the broadband optical source into the input ports one by one. By detaching the grating loss and the laser’s power fluctuation, the transmission spectra and the crosstalk of two channels are depicted in Figs. 4(b) and 4(c). Obviously, low crosstalk of less than −18dB between the TE0 and TE1 modes can be obtained, owing to the successful mode coupling process.

After that, the four-mode multiplexing device is measured. With the transmission of a straight waveguide as a reference, the present four-mode (de)multiplexer has an average insertion loss (around 2 μm) of about 1.3, 2.6, 4.8, and 5 dB for the TE0, TE1, TE2, and TE3 mode channels, respectively. The excess loss is mainly caused by the scattering loss and incomplete mode coupling in the asymmetric directional couplers due to the practical fabrication deviations. The average crosstalk of four channels is less than −18dB with a wavelength range of ∼70nm (from 1950 to 2020 nm). Among the measured curves, the curves of four colors (red, blue, green, and black) correspond to the detected powers of the TE0, TE1, TE2, and TE3 modes, respectively. As shown in Fig. 5(a), the measured TE0 mode has a relatively higher power than other three modes. In particular, for the TE2 and TE3 modes in Figs. 5(c) and 5(d), the transmission curves tend to descend when coming to a longer wavelength, which is mainly caused by the fabrication deviation. As shown in Fig. 3(d), the widths of the fabricated narrow and wide tapers are both relatively larger than the designed values in Table 1, which leads to the shift of the central wavelength. Meanwhile, high-order modes suffer from relatively larger scattering loss due to the sidewall roughness of the fabricated waveguide.

The fabricated chip is further employed for the four-mode division multiplexing application with NRZ-OOK signals at 5 Gbit/s. Figure 6(a) shows the experimental setup. An arbitrary waveform generator (Tektronix AWG 70002) is used to generate a 5 Gbit/s electrical OOK signal, and a 2 μm directly modulated laser (DML, EOT ET-5000) is modulated. Then, the optical signal is amplified by a TDFA (AdValue Photonics AP-AMP-2000) and coupled into the waveguide with a pair of vertical gratings. At the receiver side, after transmitting through the mode (de)multiplexer, the light beam is first filtered by a tunable optical filter (TF). After attenuating by a variable optical attenuator (VOA), the optical signal is amplified by another TDFA before being sent to a 2 μm PD. At last, we obtain the electronic signal from a real-time sampling oscilloscope (Keysight DSA-Z 204A) operating at 80 GS/s with a bandwidth of 20 GHz. Figure 6(b) shows the measured spectra around 2 μm before modulation (red) and after modulation (blue). One can clearly see that the optical spectrum with a 5 Gbit/s NRZ-OOK signal (blue) has a wider bandwidth than an unmodulated optical signal (red), indicating successful signal modulation.

Figure 7 plots the measured BER performance of two-channel and four-channel data transmission as a function of the received optical signal-to-noise ratio (OSNR) (back-to-back, CH1, CH2, CH3, and CH4). The measured OSNR penalties at a BER of 3.8×10−3 [7% forward error correction (FEC) threshold] are less than 2.5 dB for four channels. The measured eye diagrams are also obtained in the experiment. Obviously, the penalties are larger than similar experiments demonstrated at 1550 nm ^[43], and the transmission rate is also limited by the total insertion loss, including the coupling loss of the grating coupler and the intrinsic transmission loss. Moreover, it can be also observed that the BER performance of higher-order modes is not as good as that of the fundamental mode.

Remarkably, the loss of the grating coupler could be reduced by optimizing the etching depth, increasing the thickness of silicon, and introducing a complex apodized period. In addition, instead of vertical coupling, the coupling efficiency could be enhanced by edge coupling via inverse tapers.

Moreover, in the present device design, a wide tapered waveguide with Δw=w2b−w2a=0.1μm and a narrow waveguide with width w1=0.65μm are chosen. With further improvement, the narrow and wide silicon waveguides might be both tapered structures with larger Δw (e.g., 0.2 μm) and longer adiabatic coupling length L. Consequently, larger fabrication tolerance and wider working bandwidth could be achieved ^[46,47].

Additionally, we also study the temperature dependence of the device. The simulated results show favorable operation performance (negligible crosstalk degradation) at different temperatures.

Although the 2 μm communication system can now be implemented by using the components mentioned above, its transmission performance is still limited by the immature 2 μm optoelectronic components. In the experiment, the optical signal at 2 μm is amplified by a TDFA. However, the additional noise introduced by the TDFAs is large, which greatly degrades the transmission performance. In addition, the used TF is not very stable, and its bandwidth is large, which is not efficient for noise suppression. With further improvement of the TDFA (e.g., low-noise TDFA) and TF (e.g., stable and narrow band filter), higher transmission speed up to tens of Gbit/s could be achieved in the experiment.

In summary, we have fabricated on-chip two-mode and four-mode MDMs and demonstrated on-chip high-speed data transmission in the 2 μm wavelength region. In the experiment, low crosstalks (less than −18dB) and low insertion losses of all the channels can be obtained. Moreover, chip-scale 5 Gbit/s OOK signal transmission around the wavelength of 2 μm has also been realized. The measured OSNR penalties at a BER of 3.8×10−3 are less than 2.5 dB for four channels, and clear eye diagrams are also obtained. In the experiment, the transmission of the 2 μm waveband is still limited by a variety of key optoelectronic devices, such as a modulator, TDFA, filter, and PD. It is believed that the performance including data transmission rate, BER, and transmission distance can be further enhanced by optimizing these essential optoelectronic devices. It is anticipated that the on-chip 2 μm data transmission through the WDM and MDM will play a key role in emerging MIR spectral applications such as on-chip optical interconnects and on-chip optical data processing.

Acknowledgment. The authors thank the Center of Micro-Fabrication and Characterization (CMFC) of WNLO and the facility support of the Center for nanoscale characterization and devices of WNLO.