In the last decades, the photonic integrated circuit (PIC) has attracted much research interest, and enables integration of active and passive optical components on a single chip in a scalable manner. Numerous PIC platforms, such as silicon-on-insulator (SOI) [1], silicon nitride (SiN) [2], and indium phosphate (InP) [3], have been widely explored for photonic integration. Lithium niobate (LN), due to its wide transparent window, strong electro-optic (EO) effect, large nonlinear optical coefficient, and chemical stability, has been widely used in making various kinds of active optical components. Traditional waveguides on bulk LN are typically fabricated using technologies of titanium diffusion or proton exchange, which leads to an indistinct waveguide boundary and small index contrast between the waveguide core and its cladding regions [4]. Therefore, optical components fabricated on bulk LN are usually large, and high density integration is considered not feasible on this material platform. Recently, thin-film LN (TFLN) [5] has attracted a lot of attention due to its capability to support sub-micrometer-sized waveguides on LN, which can then lead to compact optical components as compared to those on bulk LN. Ultra-low-loss TFLN waveguides using a dry etching technique [6] have already been demonstrated. Therefore, TFLN is considered as one of the most promising PIC platforms for future optical communications.

Due to the aforementioned large EO coefficient of LN, high-speed EO modulators on a TFLN platform, such as phase modulators [7], Mach–Zehnder modulators [8,9], Michelson interferometer modulators [10], and in-phase/quadrature modulators [11], have shown excellent performance as compared to their counterparts on other PIC platforms, such as SOI. As TFLN can simultaneously confine optical and acoustic modes, waveguide-based acousto-optic modulators working at sub-gigahertz acoustic frequencies have been demonstrated [12] based on suspended TFLN waveguides. Due to the large nonlinear optical coefficient and low linear waveguide losses, ultra-high-efficiency wavelength conversion with a normalized conversion efficiency of $4600\%/(\mathrm{W}·{\mathrm{cm}}^2)$ was achieved in a periodically poled TFLN waveguide [13].

Although a number of studies on active TFLN devices have been reported, there are still many passive devices worth exploring, such as wavelength-division multiplexing (WDM) devices. Combining WDM technology and high-performance EO modulators [14], multi-channel, multi-wavelength transmitter chips, which are important components for data-center optical communications, can then be built on TFLN [15,16]. As LN is an anisotropic material, TFLN waveguides exhibit unique mode properties as compared to waveguides based on conventional isotropic materials [17]. It has been shown that significant polarization coupling related to mode hybridization through a waveguide bend can be observed. This has become the dominant factor limiting the compactness of on-chip routing for TFLN waveguides. Therefore, WDM devices using conventional structures such as array-waveguide gratings (AWGs) [18], planar concave gratings (PCGs) [19], and cascading Mach–Zehnder interferometers (CMZIs) [20] are difficult to design on the TFLN platform. Recently, a CWDM device using a multimode waveguide grating structure [21] on a hybrid platform consisting of TFLN and silicon rich nitride has been demonstrated at C-band. However, this structure is incompatible with state-of-the-art high-speed EO modulators on the monolithic TFLN platform discussed here. In this paper, we demonstrate a coarse WDM (CWDM) device based on an angled multimode interferometer (MMI) structure on the TFLN platform. The key part of the proposed device is essentially a straight multimode waveguide without involving any bending structures. Although CWDM devices using an angled MMI structure have been demonstrated on SOI and SiN platforms [22,23], implementations of such a structure, as well as efficient CWDM structures in general, on the TFLN platform are still missing. The proposed device here can support ultra-low insertion losses and cross talks for four supported wavelength channels, and can be adopted for CWDM transmitters at O-band.

Figure 1(a) shows the schematic layout of the proposed four-channel CWDM device on TFLN. The basic structure consists of a multimode dispersive waveguide, i.e., the angled MMI of width $W_{\mathrm{MMI}}$ and lengths $L_i$ ($i=1$, 2, 3, 4) for the four output channels, and the corresponding input and output waveguides of width $W_a$ and tilted angle $\theta$ with respect to the MMI structure. The input and output waveguides are all tapered to a relatively narrow width of $W_{\mathrm{IO}}$ to ensure single-mode input before entering the MMI. The transverse-electrical (TE) mode is considered throughout the structure. The input waveguide is put at the beginning of the MMI section, and the output waveguides are placed along the opposing edge of the MMI section according to the inverted self-imaging condition [22], as shown in Fig. 1(a). Due to the dispersion of the angled MMI section and the tilted input, the input signals are separated into different wavelength channels that are imaged at different axial positions $L_i$ with respect to the input waveguide. The imaging length $L_i$ can be defined using Eq. (1), which depends on the effective refractive index of the fundamental TE mode $n_{\mathrm{eff}}$ of the MMI section, the MMI width $W_{\mathrm{MMI}}$, and the wavelength of the corresponding output channel ${\lambda }_i$: $$L_i=\frac{4n_{\mathrm{eff}}{W}_{\mathrm{MMI}}^2}{{\lambda }_i}·$$(1)

The channel spacing $\mathrm{\Delta }\lambda$ of the adjacent channels for the proposed device can also be derived from Eq. (1) by taking a derivative with respect to $\lambda$ on both sides, which gives $$|\mathrm{\Delta }\lambda |=\frac{1}{4n_{\mathrm{eff}}}{\left(\frac{\lambda }{W_{\mathrm{MMI}}}\right)}^2\frac{W_a+\mathrm{\Delta }x}{\sin \theta },$$(2)where $\mathrm{\Delta }x$ is the gap between the adjacent output waveguides and usually defines the resolution requirement of the proposed structure. Practically, $\mathrm{\Delta }x$ should be sufficiently large depending on the patterning tool used. In the proposed design, the whole structure is made on an $x$-cut TFLN layer, which is typically used for modulators and other devices on TFLN, with the optical axis, i.e., the $z$ axis, of the LN material perpendicular to the MMI section as shown in Fig. 1(a). The thickness of the TFLN is 400 nm, and all the patterns shown in Fig. 1(a) are partially etched for 200 nm. The bottom and top cladding layers of the structure are all silicon oxide. Considering the wavelength channels (CWDM standard at O-band) and the resolution requirement for fabrication, we simulate the filter performance using a bidirectional eigenmode expansion (EME) algorithm (Lumerical MODE solution), and the structural parameters of the proposed device are carefully designed, and are summarized in Table 1. Figures 1(b) and 1(c) show the calculated effective refractive index $n_{\mathrm{eff}}$ at different wavelengths and MMI widths for TE polarized modes. The MMI width $W_{\mathrm{MMI}}$ is first set to 17.54 μm, which gives a resaonable overall length of the device around 2 mm, as well as low insertion losses. The light propagation in the MMI structure with a tilted input is shown in Fig. 1(d).Table 1.

Optimized Structural Parameters for the Four-Channel CWDM Device on TFLN

One can find that the minimal line width of the proposed design is $\mathrm{\Delta }x=0.88\mathrm{\mu m}$, which is within the resolution of typical patterning tools, such as a deep ultraviolet stepper or an electron beam lithographer (EBL). The input and output tapered waveguides have a length of 100 μm, which are long enough and present negligible losses. Here, due to the non-vertical sidewalls resulting from LN etching (typically $\phi =60°$), all the waveguide widths mentioned above refer to the top widths of the TFLN structures. The simulated wavelength responses of the four output channels in the optimized design are presented in Fig. 2(a). The resulting insertion losses for all four channels are $<0.6\mathrm{dB}$ at the corresponding CWDM channels with 3 dB bandwidths ${\mathrm{BW}}_{3\mathrm{dB}}$ of 12.4 nm.

The fabrication tolerance of the proposed CWDM device is also studied. The spectral responses of channel #1 with different $L_1$, $W_a$, $W_{\mathrm{MMI}}$, etching depth $t$, and total thickness of TFLN $h$ are investigated. As shown in Fig. 4, the peak wavelength can be affected by variations of all the above structural parameters, but the spectral shape is barely changed. The sensitivity of the peak wavelength to changes in $L_1$ is only 0.5 pm/nm, while the variation in $W_{\mathrm{MMI}}$ results in a shift of the peak wavelength at a relatively large rate of 105 pm/nm. This is typical for an MMI structure. Variations in the widths of the input and output waveguides can also affect the peak wavelength position as shown in Figs. 4(c) and 4(d), but the effects are less than those induced by changes in $W_{\mathrm{MMI}}$. Nevertheless, modern patterning tools, such as EBL, can ensure an accuracy of better than 10 nm, and therefore can support the fabrication of the proposed design. Moreover, tolerances to variations in the etching depth $t$ and the total thickness of TFLN $h$ also need to be considered. The peak wavelength shifts at a large rate of 200 pm/nm for variations in $t$ as shown in Fig. 4(e). This effect can be released by tight control of the etching process using, e.g., an in situ film thickness measurement tool. Better than 5 nm etching depth accuracy can be achieved. As shown in Fig. 4(f), the sensitivity of peak wavelength to the total thickness is the largest (410 pm/nm) among all the structural parameters studied here. This indicates that the initial film thickness from the wafer vendor is critical for the proposed structure.

The designed angled MMI based four-channel CWDM device with parameters shown in Table 1 was then fabricated on a wafer (NanoLN) consisting of a 400 nm thick $x$-cut TFLN on a 3 μm buried ${\mathrm{SiO}}_2$ layer. The structure was patterned using an EBL system (Raith 150 II), and a 200 nm thick LN layer was etched using inductively coupled plasma reactive ion etching technology. The device was then covered with an 800 nm thick ${\mathrm{SiO}}_2$ layer deposited by plasma enhanced chemical vapor deposition. The 800 nm thick ${\mathrm{SiO}}_2$ cladding layer was chosen here not only to protect the device, but also stay consistent with that used on an EO modulator on TFLN to achieve efficient velocity matching [24]. Figure 5 shows some images of the fabricated device. The input and output TFLN waveguides are coupled to fibers using grating couplers (GCs) for measurement purposes.

In conclusion, we have demonstrated a four-channel CWDM device on the TFLN platform using an angled MMI structure. The key part of the proposed device is simply a straight multimode waveguide. This avoids using bent waveguides needed in other types of CWDM structures, whose radius is extraordinarily large in anisotropic planar structures such as TFLN [17]. The overall footprint of the proposed device is about 2.1 mm × 0.02 mm. Ultra-low insertion losses of $<0.72\mathrm{dB}$ and averaged cross talks of 18 dB have been realized experimentally. As compared to the proposed device, the footprint of an MMI structure is much larger (15 mm) on the traditional bulk LN platform [28], and mode leakage losses will become obvious if the tilted angle is larger due to the low refractive index difference. This makes it hard to achieve an efficient CWDM device using the proposed structure based on traditional LN waveguides. We believe that further integration of the proposed device with high-speed EO modulators, and probably lasers [29], would lead to a fully integrated CWDM transmitter chip on TFLN.