Optical filters are extensively employed in wavelength division multiplexing (WDM) systems^[1–3], microwave photonics^[4–6], external cavity lasers^[7,8], and optical differentiators^[9,10]. The implementation of these filters relies on various photonic integrated circuits (PICs), such as a long-period grating (LPG)^[11,12], Bragg grating^[13], grating-assisted directional coupler^[14,15], Mach-Zehnder interferometer (MZI)^[16,17], micro-ring resonator (MRR)^[2,10,18,19], and MRR-MZI^[1,20–22]. To process flexibly the optical signals in the next-generation reconfigurable WDM network, optical filters with bandwidth and wavelength tunability simultaneously are highly desirable and have attracted ever-increasing attention^[2,12]. A scheme based on grating-assisted contra-directional couplers has been proposed in Ref. [23], which shows a wide bandwidth tunability but a small wavelength tuning range. Compared with the grating filters, MRR filters have the potential to achieve a larger bandwidth and wider wavelength tuning ranges simultaneously^[19]. However, the filters based on a single MRR offer a limited bandwidth tuning range. To enlarge the wavelength and the bandwidth tuning range, the optical filters based on MRR-cascaded^[18,19], MRR-assisted-MZI^[1,20,21], and MZI-assisted-MRR^[22] structures have been proposed. Most prior MZI-assisted-MRR filters are realized on mature material platforms such as silicon on insulator (SOI) and silicon nitride platforms.

As an emerging platform for PIC devices, lithium niobate (LN) on insulator (LNOI) is a low-loss material and capable of realizing high-index-contrast waveguides and hence compact PIC devices^[24,25]. Moreover, the LNOI platform offers the opportunity to implement simultaneously electro-optic (EO) and thermo-optic (TO) PIC devices with a moderate power consumption and a low driving voltage, as it features not only TO characteristics but also a strong EO response (r33=30pm/V)^[24,25]. Up to now, various TO^[19] or EO^{[11,12,15–17,26–30]} devices or both^[31] on LNOI, including modulators^[31,32], optical filters^{[11,12,19,27]}, interleavers^[16,17], and switches^[28,29], have been demonstrated experimentally.

In this Letter, an MZI-assisted-MRR filter is proposed and demonstrated experimentally on the LNOI platform. The proposed filter combines the advantages of the MZI and the MRR structures, and the asymmetric MZI (AMZI) and MRR used in this work facilitate a periodically variable free spectral range (FSR) of the transmission spectrum at the through and drop ports, and hence have a tunable bandwidth. Therefore, the proposed filter can achieve EO and TO tuning for the bandwidth and the dip wavelength simultaneously. Our typically fabricated filter shows that the minimum and maximum 3 dB bandwidths at the through port are 27.86 and 31.74 GHz, respectively, while at the drop port, these values are 14.68 and 30.69 GHz. Meanwhile, the TO and EO tuning rates of the dip wavelength are approximately −6.5pm/mW and −65.09pm/V, respectively. This flexible tunability makes the proposed filter have the potential to be used in optical communication and optical information processing systems to achieve multifunctional filtering operation.

Our proposed filter is formed on an X-cut LNOI wafer. Its three-dimensional view incorporating waveguides and electrodes and the top view of the waveguide layout are shown schematically in Figs. 1(a) and 1(b), respectively. The filter consists of a straight waveguide and an MZI-assisted MRR formed by bending one arm (Arm1) of a 2×2 AMZI and connecting its two input and output waveguides. The MRR is coupled to the straight waveguide via a two-mode interferometer (TMI1). Here, the use of TMI couplers instead of directional couplers (DCs) eliminates the need for fabricating submicron-sized coupling gaps, thereby reducing the difficulty of the MRR fabrication, but the resulting MRR performance and dimensions also remain comparable to those achieved using DCs. The radius of the MRR is r. In addition, two TMIs (TMI2 and TMI3) are used to split and combine light signals in the aforementioned AMZI. These three TMIs have the same widths wT but different lengths li (i=1,2,3). Except for the three TMIs, all other waveguides are single-mode waveguides (SMWs) with identical widths of wS. Considering that the splitting ratio of the three TMIs and the length difference Δl (=l5−l4) between Arm1 and Arm2 will affect the 3 dB bandwidth (W3-dB), extinction ratio (ER), and FSR of the transmission spectra, four electrode heaters are placed on three TMIs and Arm1 to provide TO tuning. In addition, a pair of tuning electrodes is placed on both sides of Arm2 to provide EO tuning.

The cross-sectional view of the LNOI waveguide is shown in Fig. 1(c). In our design, the etch depth he is set to 150 nm and the sidewall angle θ is 70°. With these data, the dispersion characteristics of the first three modes at 1550 nm wavelength are calculated with a commercial mode solver (COMSOL), and based on the calculated results, wS and wT are set to 1.3 and 2.6 µm to support only the fundamental mode and the first two modes, respectively. Aluminum (Al) heaters are used in this work; their thickness and width are set to 400 nm and 6.0 µm, respectively. To balance the absorption loss induced by the Al heaters and the TO tuning efficiency, a 900-nm-thick silica (SiO2) film is used as a buffer layer to isolate the LNOI waveguides and the heaters. Meanwhile, the gap of the EO tuning electrodes is set to 6.0 µm to achieve a negligible absorption loss.

When the light signal is launched into the input port, it couples into the MRR at TMI1 and propagates in two arms of the AMZI, subsequently interfering in TMI3 and TMI1, resulting in complementary transmission spectra measured at the through and drop ports. The normalized amplitudes of the two ports can be expressed as E2E1=t1−[t2t3ei(β−α/2)(l4+πr)−k2k3e(β−α/2)(l5+πr)]1−t1[t2t3ei(β−α/2)(l4+πr)−k2k3ei(β−α/2)(l5+πr)],(1)E3E1=k1[t2k3ei(β−α/2)(l4+πr)+k2t3e(β−α/2)(l5+πr)]1−t1[t2t3ei(β−α/2)(l4+πr)−k2k3ei(β−α/2)(l5+πr)],(2)where β (=2πNeff/λ) is the propagation constant of the fundamental mode of the SMW, λ is the wavelength, and Neff is the effective refractive index of the fundamental mode of the SMW. ti=cos(Δβili/2), ki=sin(Δβili/2) (i=1,2,3) are both related to the splitting ratio of TMIi. Δβi is the difference between the propagation constants of the fundamental and first-order modes of TMIi. α is the round-trip amplitude transmission coefficient. Additionally, the integrated coupling coefficient, determined by the AMZI structure, can be expressed as^[9]keff=k22(1−k32)(αarm1+αarm2+2αarm1αarm2cosφ),(3)where αarm1 and αarm2 are the transmission coefficients of Arm1 and Arm2, and φ=βΔl is the phase difference between the two arms.

With the above waveguide dimensions, the normalized transmission spectra at the through and drop ports are calculated using Eqs. (1) and (2) and shown in Fig. 2(a), assuming r=259μm, Δl=1634μm, l1=39μm (corresponding to k1=0.629), l2=18μm (k2=0.875), and l3=18μm (k3=0.875). Note that the spectra exhibit complementary characteristics and two FSRs. One is the FSR of the MRR, varying from ∼0.4 to 0.7 nm; another is the FSR of the MZI, fixed at ∼8.5nm. For ease of differentiation, the former is denoted as FSRMRR and the latter FSRMZI.

Further, the spectra after tuning the k1, k2, and the refractive index of Arm2 are also investigated and shown in Figs. 2(b)–2(d), respectively, with each figure incorporating an inset showing the tuning heater position with a yellow mark. From Figs. 2(b) and 2(c), tuning k1 results in slight impacts on the ER, the dip wavelength, and the W3-dB, but tuning k2 results in significant impacts on the three parameters above. Additionally, from Fig. 2(d), tuning the refractive index of Arm2 also results in significant impacts on the above three parameters. Based on the EO effect of LN, ΔNeff is calculated to be ∼1×10−5 when the tuning voltage changes 1 V. Thus, a tuning rate of ∼−93pm/V can be theoretically predicted.

The proposed device was fabricated with our in-house microfabrication facilities. The LNOI wafer (NANOLN) with 400-nm-thick X-cut LN film formed on a 4.7-μm-thick silica layer was used to fabricate the device. First, the negative pattern of the waveguides was defined on the LNOI wafer by the standard photolithography process. The 60-nm-thick chromium (Cr) was then deposited on the sample, and the waveguide pattern was formed with Cr by the lift-off process and further transferred into the LNOI by the proton-exchange-process-assisted dry etching method^[33]. After the residual Cr film was removed with a dechroming solution, the SiO2 buffer layer was deposited on the chip. Subsequently, Al electrodes were formed in the tuning regions by the thermal evaporation and lift-off processes. Finally, the input and output end faces were formed by cleaving the sample.

Considering the inevitable fabrication errors and to compare the performance of the samples with different parameters, we first fabricated several chips without SiO2 cladding, each containing four filters. Through testing, we selected three chips, each containing one optimally performing filter. We designated them as S1, S2, and S3. Next, a 100-nm-thick SiO2 buffer layer was further deposited on S1, which was then renamed as S4. Meanwhile, a 900-nm-thick SiO2 buffer layer was deposited on S2, and subsequently, the EO and TO tuning electrodes were fabricated on it, and the sample was then renamed as S5. Table 1 summarizes the typical parameters of these five samples. Figures 3(a) and 3(b) show, respectively, a photograph of the chip including S5 and the microscope image of the tuning electrodes of S5. Figures 3(c) and 3(d) show, respectively, the scanning electron microscope (SEM) images of the waveguides and the TMI of S2.

To characterize the fabricated filters, light from an amplified spontaneous emission (ASE) source (1530–1600 nm) was launched into the input port of the filter under test via a lensed single-mode fiber (SMF). The filter was placed on a metal platform with a thermoelectric cooler stuck underneath. The polarization state of the light was controlled with an inline fiber polarizer and a polarization controller (PC). The position of the fiber was adjusted carefully to launch the light exactly at the center of the LNOI core to excite only the fundamental mode. The output light from the through port or drop port was collected with another lensed SMF and monitored with an optical spectrum analyzer (OSA) (Anristu, MS97740A). With this method, the transmission spectra of all five samples are measured sequentially at a temperature of 22°C, and shown in Figs. 4(a)–4(e). To ensure clarity, only the spectra within the 1542–1558 nm wavelength range are presented here, with each figure further incorporating an inset showing a narrower spectral segment. Note that Figs. 4(a)–4(e) also present the load quality factors (Q factors), the W3-dB, and the ERs obtained from the spectra shown in each inset (see QL, W3-dB, and ER in these five figures). In addition, the transmission spectra of S5 across the 1530–1600 nm wavelength range are also shown in Fig. 4(f), indicating a large bandwidth of the filter. Note that here all spectra have been normalized to their respective maximum values for ease of comparison.

From Figs. 4(a) and 4(b), S1 and S2 show nearly identical performance characteristics, and especially, both achieve quite large Q factors at the through port, although the Q factor of S1 is slightly higher, reaching 31942. From Fig. 4(c), S3 achieves a significantly different ER, Q factor, and especially, a maximum bandwidth difference of 16.9 GHz between its through and drop ports. Additionally, a comparison of the transmission spectra of S1 [see Fig. 4(a)] and S4 [see Fig. 4(d)] demonstrates that the deposition of a 100-nm-thick SiO2 film enhances the ER and Q factor of the drop port, respectively, from 9.2 to 21.3 dB and from 12493 to 14384, but constricts the W3-dB at this port to 25.6 GHz near 1551 nm. Further, as shown in Fig. 4(e), when increasing SiO2 thickness to 900 nm, the ER at the drop port stabilizes within 12–15 dB. However, the Q factors decline to 18599 and 12859 at the through and drop ports, respectively. These experimental results can be attributed to the changes in the transmission loss of light in the MMR and MZI waveguides as well as the changes in the TMI splitting ratio, caused by different structural parameters.

The transmission spectra of S4 and S5 measured for the TE polarization at different ambient temperatures are shown in Fig. 5. Note that from Figs. 5(a) and 5(b), for S4, as the ambient temperature increases from 22°C to 28°C, the dip wavelength at the through port (drop port) blueshifts from 1549.76 (1549.88) to 1549.54 (1549.67) nm. Meanwhile, the ER at the through port (drop port) has a fluctuation of less than 1.2 (1.0) dB, while for S5, as the ambient temperature increases from 24°C to 41°C, the dip wavelengths at the through (drop) port redshifts from 1556.26 (1556.32) to 1556.40 (1556.42) nm, as shown in Figs. 5(c) and 5(d). Meanwhile, the ER at the through (drop) port has a fluctuation of less than 0.9 (2.0) dB. The above results indicate that the thickness of SiO2 cladding will change the drift direction of the dip wavelength, and as the thickness increases, the drifted velocity decreases and the temperature stability of the filter improves. Further, the variations of the dip wavelength with temperature for S4 and S5 are plotted in Figs. 5(e) and 5(f), respectively. For S4 (S5), the linear tuning rates are ∼33.4 (∼11.7) and ∼34.9pm/°C (∼9.7pm/°C) at the through and drop ports, respectively.

The transmission spectra of S5 after fabricating Al electrodes are measured and shown in Fig. 6(a). Further, the TO tuning performance of S5 is measured and shown in Figs. 6(b)–6(f). As shown in Figs. 6(a), 6(c), and 6(d), the periodic variations of the FSRMRR reveal two distinct spectral types. The spectrum near 1545.3 nm wavelength is named as type A, featuring its spectrum at the drop port with a small dip between two large dips [see Fig. 6(c)], while the spectrum near 1549.2 nm wavelength is named as type B, featuring no such small dip [see Fig. 6(d)]. The appearance of type A and type B transmission spectra can be explained as follows. The proposed MZI-assisted MRR can be viewed as an embedded two-ring resonator formed by the two arms of the AMZI and a shared bent waveguide. The two rings exhibit different FSRs, causing their spectra to exhibit distinct superposition characteristics at different wavelengths due to the Vernier effect. Figure 6(b) shows that as the heating power applied to TMI1 increases from 0 to 65 mW, the dip wavelengths and the W3-dB of the two output ports are basically unchanged. However, the ERs at the drop port fluctuate between 11.3 and 14.9 dB. Figures 6(c)–6(f) show, respectively, the spectral evolution (point lines) of type A and type B when tuning TMI2 and TMI3, together with the fitting curves (solid lines). From Figs. 6(c) and 6(d), as the tuning power applied to TMI2 increases from 0 to 126 mW, at the through port, the dip wavelengths blueshift from 1545.16 (1548.95) to 1545.11 (1548.89) nm for type A (B). The ERs are nearly unchanged, but the W3-dB changes ∼2.25(∼2.75)GHz for type A(B), while at the drop port, the dip wavelengths blueshift from 1544.96 (1548.88) to 1544.92 (1548.85) nm for type A (B). The ER has a fluctuation of less than ∼5.4(∼6.3)dB for type A (B). The W3-dB changes ∼1.00(∼2.00)GHz for type A (B). From Figs. 6(e) and 6(f), the spectral evolution when tuning TMI3 exhibits similar behavior. As the power applied to TMI3 increases from 0 to 120 mW, the dip wavelengths blueshift ∼0.06nm at the through port and ∼0.04nm at the drop port, accompanied by ER fluctuations less than ∼1.54dB. The W3-dB at different heating powers applied to TMI2 and TMI3 is summarized in Table 2. The bandwidth ratio, defined as the drop port bandwidth divided by the through port bandwidth, are plotted in Fig. 6(g), indicating an overall increasing trend with increasing tuning power.

Unfortunately, when we subsequently attempted to measure the transmission spectra of S5 under the condition that Arm2 was tuned by TO, the tuning electrodes were damaged, necessitating the re-fabrication of the electrode. Figure 7(a) presents the transmission spectra of S5 in the wavelength range of 1542–1558 nm after the electrodes re-fabrication. Compared with Fig. 6(a), the spectra in Fig. 7(a) show certain changes in dip wavelength and ER while still maintaining the two spectral characteristics described above, i.e., types A and B. To clearly show the spectral variations during TO tuning of Arm2, the evolutions of the type A spectrum near 1557.5 nm and type B spectrum near 1553 nm with the applied tuning power are presented in Figs. 7(b) and 7(c), respectively. It can be seen that as the tuning power applied to Arm2 increases from 104 to 558 mW, at the through port, the dip wavelengths blueshift from 1557.56 (1552.67) to 1553.19 (1549.62) nm for type A (B). Meanwhile, the ER has a fluctuation of less than ∼3.8(∼4.8)dB for type A(B), and the W3-dB changes ∼3.07(∼3.78)GHz for type A (B); at the drop port, the dip wavelengths blueshift from 1557.35 (1552.70) to 1552.88 (1549.55) nm for type A (B). The ER has a fluctuation of less than ∼3.17(∼4.23)dB for type A(B), and the W3-dB changes ∼2.30(∼4.56)GHz for type A(B). The dependence of the bandwidth ratio on the heating power exhibits an upward trend, as depicted in Fig. 7(e). Additionally, both ports exhibit nearly identical blueshift rates of ∼6.5pm/mW, as shown in Fig. 7(f). The W3-dB at different heating powers applied to Arm2 is also summarized in Table 2.

Further, the evolution of the type B spectrum at the drop port with the EO tuning voltages applied to Arm2 is presented in Fig. 7(d). As the tuning voltage increases from 0 to 37 V, the W3-dB changes from ∼0.13nm(∼16.21GHz) to ∼0.17nm(∼21.27GHz) with a tuning rate of 1.1 pm/V, and the dip wavelength changes from 1550.95 to 1548.49 nm with a tuning rate of ∼65.09pm/V, as shown in Figs. 7(g) and 7(h).

Finally, when TO tuning was applied to Arm1, the transmission spectra of S5 at the drop port were also measured and shown in Fig. 8(a), which indicates the FSRMZI is ∼9nm. As the tuning power increases from 0 to 180 mW, near the wavelength of 1550 nm, the spectral type changes from type B to type A. However, when the power was further increased to 312 mW, the spectral type reverts to type B. In the meantime, as shown in Fig. 8(b), the dip wavelength redshifts from 1549.24 to 1558.10 nm with a tuning rate of ∼28.7pm/mW, exceeding those achieved when TO tuning TMI2 or TMI3.

We have designed and demonstrated experimentally a TO and EO tuning filter based on an MZI-assisted-MRR structure on the LNOI platform. Our typical fabricated filter achieves a maximum load Q factor of 31942 and a maximum ER of 21.3 dB, together with the minimum and maximum 3 dB bandwidths of 27.86 and 31.74 GHz at the through port, respectively, and 14.68 and 30.69 GHz at the drop port. Additionally, by fabricating tuning electrodes in critical regions of the device, including the three TMIs and the two arms, we investigated the TO and EO tuning characteristics of the device. The results indicate that the proposed filter can achieve EO and TO tuning for the bandwidth and the dip wavelength simultaneously. Moreover, the results also reveal significant variations in TO tuning efficiency across different regions, with higher efficiency when tuning the two arms. Similarly, EO tuning of the arm also achieves substantial tuning efficiency. Our proposed filter has the potential to be used in optical communication and optical information processing systems to achieve multifunctional filtering characteristics.