1. Introduction

Space-division multiplexing (SDM) and mode-division multiplexing (MDM) technologies are now gaining popularity among both the academia and the industry. Fiber modes or waveguide modes are used as another degree of freedom in addition to the traditional multiplexing technologies to increase the transmission capacity of communication networks or on-chip systems. Due to a large number of available spatial and mode channels, the spectral efficiency of the experimental optical transmission system has been increased to a record level of 1000 bit/s/Hz^[1] with the aid of SDM/MDM technology, two orders of magnitude higher than that of the single-mode channel. The future deployment of SDM/MDM technology in long-distance telecom will be a viable solution with the maturation of few-mode devices such as few-mode transceivers, amplifiers, routers, etc. In cost-insensitive long-distance applications, the few-mode devices can be manufactured in many material platforms, including III/V, Si, glass, lithium-niobate, etc. While for short-reach application, such as datacom and on-chip interconnects, the vast number of communication channels and access points place considerable pressure on components cost. Cost-effective and technologically-simpler solutions are more preferred when introducing new technologies to short-reach applications in addition to the currently adopted wavelength division multiplexing (WDM) and multilevel coding (such as PAM-4) technologies. On-chip SDM/MDM technologies in III/V or Si platforms provide a promising solution, which can be attributed to the channel characteristics of short-range applications and the volume-production characteristic of the semiconductor industry. For systems, in short-reach applications, especially the on-chip systems, the influences of dispersion, polarization perturbations, and intermodal interference are much less than that of the long-distance system, significantly reducing the need for complex multiple-input-multiple-output (MIMO) digital signal processing (DSP). The demodulation and detection of signals from SDM/MDM channels can be much simpler and cheaper than that of the long-distance network. Simple intensity-modulation direct-detection (IM/DD) technology can still be used in short-reach datacom and on-chip systems after introducing the SDM/MDM technology. For devices, only the semiconductor integration platforms can meet the demands for large volume, low-cost production of functional few-mode devices in SDM/MDM systems, especially the few-mode converters/multiplexers/demultiplexers (C/M/D) and transmitters.

Silicon photonics and InP photonics are two major material systems for few-mode devices. Silicon photonics provides a high-level of flexibility and processing maturity for complex and large-scale integrated passive few-mode C/M/D^[2–8]. However, the difficulty in efficient lasing and amplification in the silicon platform remain as a problem for a silicon-based fully integrated few-mode transmitter. While the InP platform, with the advantage in the active and passive components integration, can offer the integration possibility of not only the passive few-mode C/M/Ds^[9–16] but also the lasers and amplifiers with the passive few-mode components. Fully functional integrated few-mode transmitters can be readily fabricated based on current technology^[17]. It is a promising solution to considerably reduce the complexity, footprint, power consumption and cost of short-reach applications.

Silicon-based few-mode devices have been widely studied and reviewed^[18], but there is still no systematic review on the InP-based few-mode devices. The differences in material platforms, design considerations, structures, performance and applications in the two-material system still necessitate an overview of the InP-based few-mode devices. In this paper, we will review the recent progress in InP-based few-mode devices, including few-mode converters, multiplexers, demultiplexers, and transmitters. The working principle, structures, and performance of the InP-based few-mode devices are discussed.

2. Working principle and structures

Modes generation, conversion, multiplexing, and demultiplexing are indispensable to an MDM system. Fig. 1 shows the typical configurations of a few-mode transmitter and a few-mode receiver. The few-mode transmitter usually consists of one or several laser sources, modulators, mode converters, and mode multiplexers. The laser sources are normally single (transverse) mode lasers, serving as optical carriers. The modulators can be either external modulators (e.g., Mach–Zehnder modulators or electronic-absorption modulators) or directly modulated lasers (DML). If a directly modulated laser is used, it can be viewed as a multifunctional device, capable of generating optical carrier and being modulated. The modulators transfer the information into different mode (to be converted) channels. The mode converters convert the single (transverse) modes into desired higher-order (transverse) modes. The mode multiplexers combine all mode channels into a single channel, ready for transmission into few-mode fibers or waveguides. To demodulate the signal at the receiver, the multiplexed modes should be first demultiplexed into different channels and then converted into a single mode, and finally detected by the photodetectors in each channel. In the receiver, mode conversion is optional if the photodetectors are accessed by multimode waveguides or fibers, i.e., no mode conversion is needed in the case of direct detecting higher-order modes.

Many types of mode C/M/D have been demonstrated in both silicon and InP platforms. Typical structures include asymmetrical directional couplers (ADC)^{[4, 19]}, tapered-waveguide structures^{[20, 21]}, multimode interference (MMI) couplers^[9–14], Y–junctions^[22–25], etc. Fig. 2(a) shows a schematic ADC mode converter where a single-mode access waveguide and a multimode bus waveguide are coupled through an evanescent coupling region. By properly choosing the waveguide width, gap width, and coupling-region length, efficient mode conversion can be realized for modes with the same effective refractive indices, e.g., the fundamental mode in the narrower single mode waveguide and the first-order mode in the wider multimode waveguide. For modes with different effective refractive indices, mode coupling is very weak. Thus, mode conversion between desired modes can be realized with low crosstalk. Fig. 2(b) shows a tapered-waveguide-based mode converter, where a single-mode waveguide and a multimode waveguide are connected by an adiabatic tapered waveguide. Mode conversion is realized in the adiabatic taper due to mode hybridization^[18]. The mode hybridization usually happens at some specific waveguide width (w_h) where some eigenmodes of specific orders (e.g., TM₀, TE₁) are indistinguishable. If the end width of the taper (w₁, w₂) satisfies w₁ < w_h < w₂, efficient mode conversion will be achieved after the launched optical mode propagates through the taper. ADC and tapered-structure based mode converters usually require precise control of the coupling gap, waveguide width on a submicron level. The two structures are mostly reported in silicon platforms but rarely realized in the InP platform due to the difference in processing the precision control level.

In the InP platform, MMI and Y-junctions are two major structures for mode converters. The MMI-based mode converters can normally realize mode conversion between the fundamental TE₀ mode and the first-order TE₁ mode. When an MMI is used as a TE₀ –> TE₁ mode converter, the MMI should be a general-MMI. The input arm (width w₀) should be placed at the MMI edge, and the output arm needs to be a merged waveguide (width w₁ = 2w₀) of two adjacent output ports, as shown in Fig. 3(a). The basic idea is to combine two images of the input TE₀ mode into the form of TE₁ mode at the output waveguide. Two factors have to be taken into consideration^[9].

1) The position of the input waveguide should be chosen so that two adjacent output waveguides of the MMI touch each other to form a wider waveguide supporting a first-order mode. At the same time, the portion of energy distributed in the two adjacent output waveguides should be equal to guarantee an optimum excitation of the first-order mode. The condition will be satisfied for an N × N general-MMI with a position parameter being half of the rib width, as shown in Fig. 3(a).

2) Secondly, the phase relationship between the output fields at the two adjacent waveguides should be out-of-phase, so that the combined field has the same phase distribution as that of a first-order mode. For an N × N general-MMI, if the optical mode is launched at the first input port with port number i = 1，a phase-shift of π will be obtained between the output ports with port number j = 2m and j = 2m + 1, with m being an integer in the range of [1, N – 1]^[26]. For example, the output port 2 and 3 will be out of phase for a 3 × 3 MMI and a 4 × 4 MMI when the light is launched at input port 1, as shown in Figs. 3(a) and 3(b).

The phase relationship Δφ_ij between the input and output ports for a general N × N MMIs can be summarized as Eq. (1)^[9]:

$$\begin{gathered} i + j\,\,{\rm{ even}}: {\rm{ }}\Delta {\phi _{ij}} = {\phi _0} - \pi - \frac{\pi }{{4N}}(j - i)(2N - j + i), \\ i + j\,\,{\rm{ odd}}: {\rm{ }}\Delta {\phi _{ij}} = {\phi _0} - \frac{\pi }{{4N}}(j + i - 1)(2N - j - i + 1),\end{gathered}$$

(1)

where φ₀ is a constant phase shift depending on the MMI length. The phase relationships for a general 2 × 2 MMI, 3 × 3 MMI and 4 × 4 MMI are listed in Table 1 for reference.

For a mode converter using a 3 × 3 MMI, the output first-order mode contains 66% of the input energy from input port 1. It is usually referred to as a 66%-MMI-converter. Similarly, a structure using a 4 × 4 MMI is called a 50%-MMI-converter. To realize a mode converter with 100% conversion efficiency, a more sophisticated structure should be adopted, as shown in Fig. 3(b). It consists of a symmetric 1 × 2 MMI as the optical splitter, a 4 × 4 MMI as the mode converter. A π-phase-shifter is introduced into one of the output arms of the 1 × 2 MMI so that the outputs at the port 1 and 4 of the 4 × 4 MMI cancel each other, while the outputs at port 2 and 3 interfere constructively.

The MMI-based mode converters can be further used as mode-converter-multiplexers. By placing another input waveguide at an appropriate position as the fundamental mode channel in conjunction with the previously described first-order mode channel, it is possible to obtain a multiplexed output combining the fundamental mode and the first-order mode at the same output port. For a 66%-MMI-converter-multiplexer, as shown in Fig. 4(a) the fundamental mode channel A should be placed at W_eq/3 from the top of the MMI if the first-order mode channel B is placed at the bottom, where W_eq is the width of the MMI. For a 100%-MMI-converter-multiplexer, an extra access waveguide can be introduced at W_eq/2 of the 50%-MMI-converter as the fundamental mode channel. However, such an access waveguide will result in a waveguide crossing in-between the 1 × 2 MMI and 4 × 4 MMI. An alternative solution is to replace the 1 × 2 MMI with a 2 × 2 MMI. The fundamental-mode channel is placed at W_eq/2 of the 2 × 2 MMI, and the first-order-mode channel is placed at the edge. Considering there is already a π/2 phase-shift between the upper and lower input arms of the 50%-MMI-converter, only an extra π/2-phase-shifter is needed. The mode-converter-multiplexers can also be used as mode converter-demultiplexers by simply reversing the input and output ports.

Y-junctions are also widely used as mode C/M/Ds. The symmetrical Y-junction consists of two single-mode branching waveguides and a multimode stem waveguide, as shown in Fig. 5(a). When two in-phase modes are launched from the single-mode arms, they will excite an even mode at the junction, and evolve into the fundamental mode in the multimode waveguide^[27]. When two out-of-phase modes are launched, they will excite an odd mode at the junction, and evolve into the first-order mode in the multimode waveguide. When two sets of in-phase and out-of-phase modes are launched simultaneously, mode conversion and multiplexing are realized. Asymmetrical Y-junctions are usually used as a mode converter-demultiplexer. For an asymmetrical Y-junction as shown in Fig. 5(b), when a fundamental mode is launched from the stem multimode waveguide, it will exit from the wider branching arm as a fundamental mode due to the matching of effective modal indices between the stem and the wider waveguide. When a first-order mode is launched from the stem waveguide, it will exit from the narrower branching arm as a fundamental mode since the effective modal index of the first-order mode matches that of the narrower waveguide^[24].

3. Recent advances in InP-based few-mode devices

In the following section, we will review several typical InP-based few-mode converters, multiplexers, and demultiplexers based on their structure, including single-MMI-based, cascaded-MMI-based, MMI-Y-junction-based, and reconfigurable mode C/M/Ds. Finally, a fully-integrated few-mode transmitter is discussed.

The basic working principles of mode manipulation structures have been well established in the 1980’s and 1990’s. In the 1990’s both Y-junction-based^[28] and MMI-based mode converters^[9] have been demonstrated to realize polarization splitter, all-optical switches, and wavelength-converters. The main functions of higher-order modes were intended for control instead of as information channels. The real functional devices aiming at MDM only appear recently with the development of MDM technology in both telecom, on-chip interconnects, and datacom. As mentioned in the introduction, most integrated mode manipulation devices are still passive, using silicon as the material platform. Silicon-based few-mode components can be further integrated to a large scale level to perform more complexed functions. It is suitable for telecom requiring the combination of several multiplexing methods, including WDM, PDM (polarization division multiplexing), MDM, etc. as well as higher-order modulation format. However, for datacom and short-reach applications, a low-cost solution with simple structures is highly preferred. The InP platform provides an excellent option for such applications due to its capability to integrate lasers, modulators, and mode C/M/Ds.

3.1 Single-MMI-based mode C/M/D

In 2016, we revisited Leuthold’s scheme using MMI as a few-mode C/M/D^[9]. The purpose was to verify their characteristics of broadband operation and higher-order fiber-modes excitation. MMI-based mode converters working in the O-band^[11] and C-band^[12] were realized and demonstrated. Fig. 6(a) shows the epitaxial structure of the O-band mode-converter. It consisted of a 500-nm InP buffer, a 300-nm InGaAsP core layer with a 1.15-μm bandgap, and a 1.7-μm InP cladding layer. Fig. 6(b) shows the schematic structures of the 50%- and 66%-MMI-converter. The two devices were fabricated with a length of 371 and 495 μm, respectively. Mode conversion from TE₀ to TE₁ was successfully demonstrated in the wavelength range of 1280–1320 nm. Fig. 6(c) shows the mode pattern obtained from the manufactured 50%- and 66%-MMI-converter. Similar results were also obtained for mode converters in the wavelength range from 1520–1565 nm. The generated TE₀ to TE₁ were coupled into a two-mode fiber, LP₀₁ and LP₁₁ modes were successfully excited. Table 2 shows the first-order TE₁ modes from a 66%-MMI-converter and LP₁₁ modes excited by the TE₁ modes at various wavelengths.

For the MMI-based mode converters/multiplexers, the operational mode order is typically limited to two modes. To realize higher-order mode conversion, additional structure can be introduced. By using a modified 3 × 3 MMI and an enlarged butterfly-shaped phase shifter, a mode converter capable of converting TE₀ to TE₂ has been demonstrated in SOI platform^[8], which can also be applied to the InP platform to further scale the mode conversion order.

3.2 Cascaded-MMI-based mode C/M/D

To realize a 100%-MMI-C/M/D as shown in Fig. 3(b) or Fig. 4(b), a phase-shifter (PS) is required. The phase-shifter can be realized through bent waveguides with different radius^[9], butterfly-shaped tapered structure^[3] or tilted-joint connection^[29]. However, these schemes will either result in a long PS section or a strict fabrication tolerance; care should also be taken to minimize the loss. To achieve more precise control and increase the processing tolerance, we proposed a PS structure with a thickness-detuned core layer^[10], as shown in Fig. 7(a). By tuning the core-layer thickness of the PS section, the phase difference between the upper and lower input arms of MMI2 can be controlled precisely since the thickness error of the epitaxial layers is only a few nanometers, resulting in a precise control of the effective refractive index difference (Δn) between the PS arm and the other waveguide arm. Δn in this scheme will be on the order of ~10⁻². Considering the phase difference of Δφ = 2π/λ(ΔnL), where λ is the wavelength, L is the PS length, to achieve a π/2 phase shift, L will be on the order of 3–25λ, which is an appropriate length for InP processing control while still maintaining a small PS size. On the other hand, Δn will be small enough to make the structure rather insensitive to the PS length and increase the fabrication tolerance. Fig. 7(b) shows the calculated shortest π/2-PS length when changing the core-layer thickness of the PS section. During the simulation, the core layer of other waveguides was fixed at 300 nm. Taking into account the growth quality of epitaxial materials, a core layer thickness of 400 nm for the PS section was selected, with the corresponding PS section length being 14.7 μm. The epitaxial structure of the 100%-MMI-converter-multiplexer was similar to that shown in Fig. 6(a), with the exception that the core layer of the PS section was 100-nm thicker than that of other waveguides. The influence of the PS length on the insertion loss is shown in Fig. 7(c). Within ±8 μm from the selected PS length of 14.7 μm, the simulated insertion losses are lower than 1 dB for both the TE₀ mode and TE₁ mode. The crosstalk of the TE₀ mode and the TE₁ mode are below −21.8 and −36.3 dB, respectively. Fig. 7(d) shows the scanning electron microscopy (SEM) image of the fabricated device. In the enlarged inset, a visible elevated region can be found in the PS section due to the thicker core layer. The device was tested as a mode-converter by injecting a fundamental mode into Port3 and Port4, respectively. TE₀ and TE₁ modes were successfully obtained, as shown in Fig. 7(e). Broadband operation from 1520 to 1600 nm was also realized.

Besides using a thickness-detuned core layer to achieve the π/2 phase shift, a 1 × 1 MMI can also be used as a phase shifter^[14]. The main advantage of using an MMI as the PS is that the whole device will be solely based on MMI couplers, which have been standardized as the building block in the generic foundry platform for the photonic integrated circuits^[30]. This will make it compatible with standard manufacture processes for integration with other standard devices on InP substrate to form more advanced circuits. Another advantage is that by introducing the 1 × 1 MMI-PS, the upper and lower connecting arms between MMI 1 and MMI 2, as shown in Fig. 8(a), can adopt a simple straight waveguide of the same length, without the need to use bent or butterfly structures. The 1 × 1 MMI-PS will introduce a fixed phase shift of π, regardless of its length, while the phase delay of a straight waveguide increases linearly as its length. By adjusting the width and thereby the length of the 1 × 1 MMI, it is possible to obtain a phase difference of π/2 between the upper and lower arms. Fig. 8(b) shows the simulated field distribution of the 100%-MMI-converter-multiplexer based on an MMI-PS. In the design, the 1 × 1 MMI was 87 μm in length. In comparison, if a bent-waveguide were to be adopted, the length of the phase-shifter would be 200 μm with the curve radius of 1000 μm. Fig. 8(c) shows the TE₀ and TE₁ mode pattern from a fabricated device using this scheme.

3.3 MMI-Y-junction-based mode C/M/D

The 100%-MMI C/M/D using a 4 × 4 MMI as the mode converters as shown in Fig. 7(a) and Fig. 8(a) require three connecting arms between the two MMIs. The MMIs are usually designed to be wide enough to avoid crosstalk between adjacent connecting arms. The increase of MMI width will cause a dramatic increase in the length of MMI when considering the quadratic relation of MMI’s length with width. The total lengths of the 100%-MMI-C/M/D above mentioned were all longer than 1 mm. An MMI-Y-junction structure was then proposed to realize a miniature 100%-mode-C/M/D^[25]. It consisted of a 2 × 2 MMI coupler, a symmetric Y-junction, and a phase shifter with a thickness-detuned core layer, as shown in Fig. 9(a). For a general 2 × 2 MMI coupler, when a TE₀ mode is launched from Port 1, it will be split into two beams of the same amplitude but a phase shift of −π/2 between Port 4 and Port 3. While for the TE₀ mode launched from Port 2, the phase shift will be π/2. After passing through a π/2 phase-shifter, the two beams from Port 1 will be in-phase, and the beams from Port 2 will be out-of-phase. The in-phase and out-of-phase TE₀ modes will be guided out of Port 5 as the TE₀ and TE₁ mode respectively. The simulated field distribution is shown in Fig. 9(b). The modes pattern from a fabricated device is shown in Fig. 9(c). Due to the reduction of connecting arms, the device size has been reduced to less than 500 nm, much shorter than other 100%-mode-C/M/D structures in InP materials.

3.4 Reconfigurable mode C/M/D

A reconfigurable photonic integrated mode C/M/D was proposed and demonstrated by Polytechnic University of Milan in 2016^[16]. The schematic device structure is shown in Fig. 10(a). The mode C/M/D was realized using an MMI-based balanced Mach-Zehnder interferometer (MZI) and a symmetric Y-junction. Two thermally tunable phase sections were introduced in one arm of the Mach-Zehnder interferometer (amplitude controller) and one arm (phase controller) of the Y-junction, corresponding to a phase shift of φ_A and φ_P, respectively. The MZI was biased as a 3-dB coupler with φ_A being π/2 or 3π/2. The phase difference between the two output branches of the MZI will be either a (π, 0) or (0, π) pair depending on φ_A if two fundamental modes are simultaneously launched at S1 and S2. φ_P can be chosen to be 0 or π to control which input channel (S1 or S2) shall be converted to the TE₁ mode. The structure is completely reconfigurable, and the two single-mode inputs can be mapped into any combination of the two modes of the output waveguide. The device was designed and fabricated on an InP-based generic technological platform through a Multi-Project Wafer run^[30]. The device picture is shown in Fig. 10(b). The phase control at the amplitude controller and the phase controller are realized by applying a control voltage of V_A and V_p respectively. The functionality of the mode C/M was tested by coupling a single mode laser through either S1 or S2, with a fixed MZI phase control of φ_A = π/2 (V_A = 3.1 V). Fig. 10(c) shows the mode patterns corresponding to φ_P = 0 (V_P = 0 V) and φ_P = π, (V_P = 4.8 V) respectively. Switchable mode conversion can be observed. Since the circuit is completely balanced the operational bandwidth of the mode C/M is extremely wide. Operation wavelength at 1480 and 1550 nm were both demonstrated at the same control voltages. A transmission experiment was further carried out by using the chip as the mode C/M in the transmitter and as mode C/D at the receiver, as shown in Fig. 10(d). SDM transmission in a 2-meter few-mode fiber was demonstrated. Fig. 10(e) shows the DEMUXed eye diagram of one channel and two channels. Clear opened eye with Q-factor of 5.7 (single channel transmission) and 3.8 (double channel transmission) were demonstrated.

3.5 Fully-integrated few-mode transmitter

In 2018，we demonstrated a fully-integrated InP-based few-mode transmitter^[17]. The transmitter consists of two directly modulated lasers (DMLs) and a 66%-MMI mode converter/multiplexer. The schematic structure of the few-mode transmitter is shown in Fig. 11(a). The two DMLs are distributed feedback lasers (DFB), responsible for generating two directly modulated TE₀ modes. The MMI-based mode converter/multiplexer converts the TE₀ mode from DML1 to TE₁ mode and multiplexes the TE₁ mode with the TE₀ mode from DML2 at Port3 of the MMI. The epitaxial structure of the device is shown in Fig. 11(b). The active section is a multiple-quantum-well (MQW) structure with photoluminescence (PL) wavelength of 1.5 μm. The passive section has a 0.3-μm core layer with PL wavelength of 1.2 μm. The passive and active section was integrated using butt-joint technology. Fig. 11(c) shows a picture of the fabricated device. To avoid the influence of the unused Port4, a curved waveguide was designed to guide the unused TE₀ mode power off the testing path. Fig. 11(d) shows the TE₀ mode and TE₁ mode patterns generated from the few-mode transmitter and the fundamental fiber LP₀₁ and first-order fiber LP₁₁ mode excited by the waveguide modes, respectively. The transmitter works at a wavelength around 1538 nm, with the side mode suppression ratios (SMSR) above 35 dB and −3 dB small-signal modulation bandwidth above 14 GHz. Direct modulation of the transmitter using 10-Gb/s non-return-to-zero (NRZ) on-off-key (OOK) signal and 10-Gbaud/s-NRZ 4-level pulse amplitude modulation (PAM-4) signal was demonstrated. Fig. 11(e) shows the eye-diagram of the 10-Gb/s-NRZ-OOK and 10 Gbaud/s-NRZ-PAM-4 signals corresponding to the fundamental mode and the first-order mode, respectively. Error-free receiving of the two-mode channels were also obtained in a 10 Gb/s-NRZ-OOK back-to-back bit error rate (BER) test. The power penalty between TE₀ mode and TE₁ mode was 4.3 dB at the BER of 10⁻⁹, as shown in Fig. 11(f). The directly modulated monolithic integrated few-mode transmitter may find potential applications in short-reach or on-chip MDM systems, where intensity modulation with direct detection (IM/DD) is preferred.

4. Conclusion

In summary, InP-based few-mode devices provide an attractive solution for both long-reach and short-reach optical SDM/MDM systems. Many structures have been proposed to realize mode conversion, multiplexing and demultiplexing through monolithic integration technology. MMI and Y-junction based structures are important schemes in InP platform due to their advantages in large fabrication error, relaxed processing precision control, and broadband working range. The readily available technology in active/passive integration makes the InP-based few-mode transmitter a perfect and commercially promising candidate for future SDM/MDM applications, especially in cost-sensitive, volume-production-demanding and technologically-simpler short-reach applications.