As it is well known, silicon photonics has become very promising because of CMOS compatibility, low waveguide loss, and high integration density. Numerous silicon photonic devices have been demonstrated successfully for data transmission [5–9]. In particular, for PON systems, lots of special photonic components are needed [10–12], such as diplexers, triplexers [13–15], and quadplexers [16]. Among them, diplexers and triplexers have been developed successfully with silicon photonic technology in recent years. In contrast, it is still a big challenge to develop on-chip quadplexers due to the special requirements for the wavelengths as well as the bandwidths of the four channels. To the best of our knowledge, no on-chip quadplexer has been reported yet.

As shown in Figs. 1(a)–1(c), a quadplexer is proposed to realize the co-existence of GPON and 10G GPON by combining the two channels of 1270/1310 nm for downloading and the two channels of 1490/1577 nm for uploading. The corresponding bandwidths for these four channels are 20, 40, 20, and 5 nm, respectively. There have been several typical on-chip WDMs using for example, arrayed-waveguide gratings (AWGs) [17,18], etched diffraction gratings (EDGs) [19], microring resonators (MRRs) [20], and Mach–Zehnder interferometers (MZIs) [21,22]. According to the working principle, AWGs and EDGs are suitable for multichannel WDM systems with uniform channel spacing in a wavelength range limited by the free spectral range (FSR). Even though a novel cross-order silica AWG triplexer with three channels of 1310/1490/1550 nm was proposed to break the FSR limitation [18], it is still inflexible for AWGs/EDGs to achieve nonuniform bandwidths as desired for triplexers as well as the quadplexers considered in this paper. Similarly, WDMs based on cascaded MZIs do not have nonuniform channel spacing as desired and the wavelength range is also limited by the FSR. Furthermore, it usually requires critical control of the phase shifts and the coupling ratios for MZIs, which causes fabrication challenges. When using MRRs, designers can make the channel wavelengths flexible and the channel bandwidth is also flexible. However, the bandwidth is usually 1–2 nm or less and the FSR is only several tens of nanometers, according to the minimum bending radius. As a result, MRR-based WDMs also do not provide a channel bandwidth as wide as 5–40 nm and an operation wavelength range as wide as $\sim 300\mathrm{nm}$. In summary, it is difficult to realize such a quadplexer with highly nonuniform broad bandwidths (i.e., 5–40 nm) and very sparse channels in a 300 nm wide wavelength range from 1270 nm to 1577 nm by using regular photonic filters, such as AWGs/EDGs, MRRs, and MZIs.

In this paper, we propose and demonstrate what we believe, to the best of our knowledge, is the first on-chip quadplexer by using four cascaded multimode waveguide grating (MWG) filters. Previously, we realized a four-channel coarse wavelength division (de)multiplexer based on cascaded MWGs that works well in the 80 nm wide O-band covering the channels of 1271/1291/1311/1331 nm to satisfy the demands in short-distance optical interconnects (e.g., data centers) [23]. Recently, a silicon-based on-chip triplexer was also realized with three cascaded MWGs for the three channels of 1310/1490/1550 nm in fiber-to-the-home (FTTH) systems [24]. For the realized triplexer, the challenge is to design the MWGs carefully with optimized apodization strengths, corrugation depths, and periods, to achieve low losses and low cross-talk as well as the desired bandwidths of 100, 20, and 10 nm for the channels of 1310, 1490, and 1550 nm, respectively. In contrast, the design for the MWGs for the present on-chip quadplexer is even more challenging because of the ultrawide operation wavelength range covering the O-, S- and L-bands (i.e., 1270/1310/1490/1577 nm) and their nonuniform bandwidths of 20/40/20/5 nm. It means that one should carefully take the waveguide dispersion and the grating effects for more than one mode into account. Fortunately, four MWG-based filters were successfully designed with satisfactory performances. For the fabricated on-chip quadplexer, flat-top responses have been achieved with low excess losses of $<0.5\mathrm{dB}$ and low cross-talk of $<-24\mathrm{dB}$ for all channels. Such a compact silicon-based, on-chip quadplexer with excellent performances can be easily integrated with other silicon photonic devices and thus will be useful for PON systems.

For the first MWG-based filter operating with the channel of 1270 nm, the ${\mathrm{TE}}_0$ mode carrying the data is received at the input port, propagates forward, and is then converted to the backward ${\mathrm{TE}}_1$ mode through the MWG. Finally, it is coupled to the ${\mathrm{TE}}_0$ mode at output port ${\mathrm{O}}_1$ through a ${\mathrm{TE}}_0/{\mathrm{TE}}_1$ mode (de)multiplexer. For the 1310 nm channel, the data carried by the ${\mathrm{TE}}_0$ mode are also launched at the input port and go through the first MWG-based filter with low excess loss and low cross-talk. Similarly, the data for the 1310 nm channel finally drop at port ${\mathrm{O}}_2$ by the second MWG-based filter. For the 1490 nm and 1577 nm channels, data are launched from ports ${\mathrm{O}}_3$ and ${\mathrm{O}}_4$, respectively. For the 1490 nm channel, the launched ${\mathrm{TE}}_0$ mode from port ${\mathrm{O}}_3$ is first converted to the ${\mathrm{TE}}_1$ mode through the ${\mathrm{TE}}_0/{\mathrm{TE}}_1$ mode (de)multiplexer and then is converted to the reflected ${\mathrm{TE}}_0$ mode through the third MWG filter. This 1490 nm ${\mathrm{TE}}_0$ mode goes through the second and first MWG filters successively with low excess loss and low cross-talk and finally exits from the input port. Similarly, the launched data of 1577 nm from port ${\mathrm{O}}_4$ are manipulated by the fourth MWG-based filter and also finally exit at the input port. In particular, bent waveguides are introduced between any two filters to radiate the residual higher-order mode (${\mathrm{TE}}_1$) and greatly reduce the undesired Fabry–Perot resonance.

Regarding the ultrawide working wavelength range for the quadplexer, two broadband ${\mathrm{TE}}_0/{\mathrm{TE}}_1$ mode (de)multiplexers are designed, respectively, for the wavelength channels of 1270/1310 nm and 1490/1577 nm. For the wavelength channels of 1270/1310 nm (i.e., $i=1$, 2), the core widths at the input/output ends of waveguides A and B for the adiabatic coupler are chosen as $(w_{ia1},w_{ia2})=(450,550)\mathrm{nm}$, and $(w_{ib1},w_{ib2})=(250,120)\mathrm{nm}$, respectively. We chose the gap widths $(w_{ig1},w_{ig2},w_{ig3})=(1.2,0.18,0.5)\mathrm{\mu m}$ and the taper lengths $(L_{i01},L_{i12},L_{i23})=(25,35,5)\mathrm{\mu m}$. Figure 3(b) shows the calculated wavelength dependence of the transmissions from the ${\mathrm{TE}}_1$ mode in waveguide A and to the ${\mathrm{TE}}_0$ mode in waveguide B. From this figure, it can be seen that the designed ${\mathrm{TE}}_0/{\mathrm{TE}}_1$ mode (de)multiplexer has a low excess loss ($<0.05\mathrm{dB}$) and a low cross-talk of $<-32\mathrm{dB}$ in the desired wavelength bands around 1270 nm and 1310 nm. For the mode (de)multiplexers for the wavelength channels of 1490 nm and 1577 nm, the parameters are chosen as follows: $(w_{ia1},w_{ia2})=(480,720)\mathrm{nm}$, $(w_{ib1},w_{ib2})=(260,120)\mathrm{nm}$, $(w_{ig1},w_{ig2},w_{ig3})=(1.2,0.18,0.5)\mathrm{\mu m}$, and $(L_{i01},L_{i12},L_{i23})=(20,35,5)\mathrm{\mu m}$. The calculated transmission of the designed ${\mathrm{TE}}_0/{\mathrm{TE}}_1$ mode (de)multiplexer is shown in Fig. 3(c). It shows that the low excess loss is $<0.1\mathrm{dB}$ and the cross-talk is $<-30\mathrm{dB}$ for the mode (de)multiplexer in the desired wavelength bands around 1490 nm and 1577 nm. The simulated light propagation in the designed mode (de)multiplexers is also shown in the insets when operating at the wavelengths of 1270, 1310, 1490, and 1577 nm, respectively, which verifies the waveguide structure design.

The grating structures are designed according to the phase-matching condition between the ${\mathrm{TE}}_0$ and ${\mathrm{TE}}_1$ modes so that the launched ${\mathrm{TE}}_0$ mode at the input port can be converted to the reflected ${\mathrm{TE}}_1$ mode. Here, the phase-matching condition is given as $n_{\mathrm{eff}1}+n_{\mathrm{eff}2}={\lambda }_B/\mathrm{\Lambda }$, where $n_{\mathrm{eff}0}$ and $n_{\mathrm{eff}1}$ are, respectively, the effective indices of the ${\mathrm{TE}}_0$ and ${\mathrm{TE}}_1$ modes in the MWG, $\mathrm{\Lambda }$ is the grating period, and ${\lambda }_B$ is the Bragg wavelength. Here, grating tapers are introduced to connect the grating section and the input/output section, as shown in Fig. 2(b), to suppress the undesired reflection loss at the front/back ends of the grating section. To improve the sidelobe suppression ratios (SLSRs) and reduce the cross-talk, the grating is apodized longitudinally. For example, the superposition of the gratings is modulated with a Gaussian function of the position $z$ in the propagation direction, as shown in Fig. 2(d) [i.e.,$\mathrm{\Delta }s=\exp [-b(z-L/2)/L^2]L/2$], where $\mathrm{\Delta }s$ is the longitudinal shift, $b$ is the apodization strength, and $L$ is the length of the Bragg grating.

Figures 4(c) and 4(d) show the calculated spectral responses for the designed MWG-based filters for the channels of 1490 nm and 1577 nm, respectively. Their parameters are chosen as follows: $W_3=W_4=1100\mathrm{nm}$, $(b_3,b_4)=(15,5)$, $({\delta }_3,{\delta }_4)=(240,160)\mathrm{nm}$, $({\mathrm{\Lambda }}_3,{\mathrm{\Lambda }}_4)=(306,320)\mathrm{nm}$, $(N_3,N_4)=(200,300)$, and $N_{3\mathrm{tp}}=N_{4\mathrm{tp}}=20$. Here, the apodization strengths and the corrugation depths for these two MWGs are chosen quite differently because these two channels are very separate and the bandwidths are very different. In particular, weak apodization strength is chosen for the 1577 nm MWG to achieve a narrow bandwidth of 5 nm. From Figs. 4(c) and 4(d), the designed MWGs have 1 dB bandwidths of 20 nm and 5 nm as desired, respectively. Both have box-like responses with low excess losses of $<0.05\mathrm{dB}$. When the quadplexer is used as a (de)multiplexer in ONUs, the crosstalk between these two wavelength-channels is $-42–-38\mathrm{dB}$.

In conclusion, in this paper we have proposed and realized a silicon-based, on-chip quadplexer for the first time, to the best of our knowledge. The quadplexer is designed with four cascaded MWG-based filters. In the present design, apodization gratings, grating tapers, and bent waveguides have been introduced to greatly reduce the cross-talk. All channels for the designed quadplexer have box-like responses with low excess losses ($\sim 0.1\mathrm{dB}$) and low crosstalk ($<-30\mathrm{dB}$) in theory. For the fabricated silicon quadplexer, the bandwidths for the wavelength-channels of 1270, 1310, 1490, and 1577 nm are about 16, 38, 19, and 6 nm, respectively. The cross-talk is as low as $-30–-20\mathrm{dB}$, $-32–-28\mathrm{dB}$, $\sim -28\mathrm{dB}$, and $\sim -36\mathrm{dB}$ for these four channels, respectively. One should notice that the present quadplexer only works for TE polarization. A polarization-insensitive quadplexer can be achieved by introducing an on-chip polarization splitter rotator [26]. In particular, in the present proof-of-concept demonstration, we used two types of grating couplers in the characterization for convenience. On the other hand, edge couplers can be easily used for light coupling between a fiber and a silicon strip waveguide to achieve broadband and polarization-independent coupling. More importantly, the present on-chip quadplexer will be very useful to further develop monolithically integrated transceivers on silicon, which has great potential in reducing the module size and power consumption as well as the package complexity and cost. Furthermore, it is also possible to combine the video signals (1550–1560 nm) and/or optical time-domain reflectometer signals (1625–1675 nm) by cascading two more MWG filters designed with the corresponding central wavelengths and bandwidths. Such a scalable quadplexer with high performances could be used widely to develop next-generation PON and WDM systems.