As a fundamental element in microwave photonic systems, microwave photonic links (MPLs) could find diverse applications, including RF over fiber wireless communication^[1], radio astronomy^[2], and optical delay lines^[3]. The MPL technology has achieved huge developments, featuring high gain, low-noise floor, and high linearity within constrained bandwidths. However, the widespread adoption of commercial MPLs faces challenges, primarily due to noise over the useful bandwidth, including thermal noise, relative intensity noise (RIN), interference, and cross talk^[4–7], which originate from the optical components like tunable lasers, erbium-doped optical fiber amplifiers (EDFAs), modulators, etc. Among these components, the modulator plays a particularly crucial role. Its characteristics, including half-wave voltage (Vπ), bandwidth, extinction ratio, and insertion loss, significantly impact the overall performance of MPLs^[8,9]. Given that this noise introduces instability into the system, the development of low Vπ and insertion loss, and broad bandwidth modulators become necessary. Moreover, by utilizing dual-output modulators in conjunction with a balanced photodetector (BPD)^[10], it is possible to suppress the noise introduced by various components in the link, thereby enhancing the overall performance of MPLs.

As a newly proposed platform, thin-film lithium niobate (TFLN) exhibits a significant refractive index contrast between the lithium niobate (LN) waveguide and the cladding. It inherits nearly all the exceptional material characteristics of bulk LN but reduces the device footprint, becoming a promising candidate to replace bulk LN modulators in various applications. In recent years, low half-wave voltage^[11–14], broad bandwidth^[15–20], and low insertion loss^[21–25] TFLN modulators have been successfully developed^[26]. However, there are still few reports on the packaged dual-output Mach–Zehnder modulator (DO-MZM) based on a pure TFLN platform for the application in MPLs^[27–30]. Moreover, integrating larger-sized spot size converters (SSCs) on the TFLN platform is necessary to achieve low-loss coupling with a polarization-maintaining fiber (YOFC PM1017E, 6.5 µm mode field diameter) and reduce packaging loss.

In this work, a TFLN dual-output electro-optic (EO) modulator chip is fabricated, which consists of a Mach–Zehnder interference (MZI) structure, coplanar ground–signal–ground (GSG) electrodes, 6.5 µm spot-size converters, and a thermal-optical (TO) bias electrode. A symmetric 2×2 multimode interference (MMI) splitter is placed at the output of the chip to yield two outputs with intensities inversely proportional to each other. After packaging, the device exhibits a packaging loss of 4.2 dB with a 3 dB EO bandwidth of 48 GHz, a half-wave voltage Vπ of 2.5 V, and an extinction ratio of ≥25dB. When utilizing a BPD at the receiver end, 18 dB common-mode rejection ratio (CMRR) with 6.2 dB signal enhancement has been observed.

Figure 1(a) illustrates the designed dual-output modulator based on 500 nm thick X-cut TFLN, bonded on 4.7 µm thermally grown silicon dioxide with a Si substrate (NanoLN). The device includes an EO-MZI modulation region, a thermo-optic (TO) heater, 2×2 MMI, and SSCs. In the EO modulation region, a 9-mm-long configuration with a coplanar waveguide (CPW) traveling-wave electrode is utilized for the transmission of radio frequency (RF) signals. A 200-µm-long, 7-µm-width Ti-heater is employed to adjust the bias points of the MZM structure through the TO effect of LN, which is positioned above the LN waveguide and separated by a 1 µm thick SiO2. For the input and output sections, 6.5 µm mode field diameter SSCs are utilized to maintain low coupling loss.

Figure 1(b) illustrates the cross-sectional diagram of the MZM modulation region, where the optical signal is equally split into two paths and modulated by coplanar GSG electrodes. The TE₀ mode and electric field distribution in the modulation region are shown in Fig. 1(b). The LN waveguide is designed as a ridge waveguide with an etching depth of 260 nm and is broadened to 4 µm through an adiabatic taper waveguide. The gap between electrodes is set as 7.4 µm according to the trade-off among Vπ, bandwidth, and the optical loss. Then the width of the signal electrode is set as 20 µm with an impedance close to 43 Ω; this value was chosen slightly lower than the characteristic impedance of the electrode (actual value of 47 Ω), which helps reduce the drop in the low-frequency S₂₁ curve to achieve higher bandwidth. Figure 1(c) shows the simulation results of bandwidth; the 3 dB bandwidth is 67 GHz. 1×2 and 2×2 MMI couplers are utilized as splitter and combiner for the MZM configuration, respectively. The design of MMI is primarily based on the waveguide’s self-imaging effect. To conserve space and reduce losses of the 2×2 MMI, a paired interference-type of MMI structure is employed^[31]. In this configuration, the input optical field is distributed at G=WMMI/3 around, where G represents the gap between two waveguides and W_MMI represents the width of the MMI. Figure 2(a) illustrate the top view of the device, the optimized parameters are also marked on this figure. The power is divided equally between the two outputs. The optical field distribution of the device is shown in Fig. 2(b); the splitting ratio at both ends is 49%.

To achieve efficient edge coupling between the TFLN waveguide and the polarization-maintaining fiber, an SSC with a mode field diameter (MFD) of 6.5 µm was designed. The structure of SSC is illustrated in Figs. 2(c) and 2(d), which was achieved by modifying our previous work of 3.2 µm SSC^[32]. The 6.5 µm SSC consists of three components: the LN taper, the SiON cladding waveguide (SiON CLWG), and the cladding silicon oxide. The refractive index of the SiON was tuned to be 1.56, between the LN and silicon oxide. By optimizing the size of the LN taper and the thickness of the SiON, TE₀ mode could propagate stably in SSC and highly match with the fundamental mode field in 6.5 µm MFD polarization-maintaining (PM) fiber. Furthermore, to achieve an efficient transition of TE₀ modes between the edge-coupler and the LN taper and prevent excitation of higher-order modes, the LN taper should be located at the center of the SiON cladding waveguide. Thus, an additional etching of 3 µm depth is applied to the silicon oxide layer underneath the LN strip waveguide. A coupling loss of 0.04 dB per facet is obtained for the transverse electric (TE) mode in simulations after optimization, as shown in Fig. 1(e), which includes the mode conversion loss from cross section C1 to C4. Figure 1(f) shows the simulated optical field distribution in the SSC.

The DO-MZM chip was fabricated on a commercial TFLN wafer (NANOLN). The chip has a compact footprint of 16mm×1.5mm. The thicknesses of the LN and buried oxide layers are 500 nm and 4.7 µm, respectively. The fabrication process of the optical component is detailed in the following: electron beam lithography (EBL) was first used to define the ridge waveguide structures on the AR-P 6200 resist. Second, the patterns were transferred to the top LN layer with an etching depth of 260 nm by inductively coupled plasma (ICP) dry etching. Then, the strip waveguides for the edge coupler were defined on the LN layer with an etching depth of 3.24 µm using EBL and ICP dry etching, which include a 0.24-µm LN taper and a 3-µm buried oxide taper. Finally, a SiON layer with a thickness of 3.5 µm was deposited on the wafer as the cladding waveguide by plasma-enhanced chemical vapor deposition (PECVD). More details can be found in Appendix A.

Microscope images of the fabricated sample are shown in Figs. 3(a)–3(c). The DO-MZM chip was fabricated on a commercial TFLN wafer (NANOLN), which has a compact footprint of 16mm×1.5mm. Following fabrication, optical and electrical packaging work was carried out on the device. In terms of optical packaging, the PM single-mode fiber (YOFC PM1017E) was chosen to achieve optimal mode-field matching, considering the MFD of the device’s SSC at 1550 nm is about 6.5 µm. The fiber arrays (FAs) were aligned with the SSCs and fixed by UV-curable glue. The LN chip within a custom-built metal tube shell can be seen in Fig. 3(e). As for electrical packaging, the RF connection and direct current (DC) bias connection parts are involved. The DC pad is wire-bonded to the DC pins on the tube shell. Additionally, the input RF pad is wire-bonded to the ceramic transmission line and connected to the 1.85-mm-long RF coaxial connectors. The ceramic transmission line, which consists of alumina ceramic with metal coplanar waveguide electrodes, has a thickness of 254 µm. Thus, the RF signal can be transmitted from the microwave source to the chip. To achieve impedance matching, a 43 Ω load resistor is attached at the end of the CPW traveling-wave electrode. Figure 3(d) illustrates that a group of gold wires are employed for electrical interconnection between the electrode and external load resistance.

After packaging, the optical characteristics were tested using a tunable laser (Santec TSL-550), a power meter (YOKOGAWA A02211), and a DC power source (GWINSTEK GPD-4303s). The DC transmission with TO phase shifters was initially measured. Figure 4(a) displays the TO transmission curve in relation to the applied voltage. The optical power output of the two ports is represented by the black and red lines, while the blue line represents the total power of both ports. Consequently, the fiber-to-fiber loss of the device amounts to −4.2dB, accompanied by an extinction ratio (ER) of 25 dB. The results indicate that intensities of both ports are inversely proportional to each other. Notably, the length of the TO phase shifter is merely 200 µm, with a resistance value of 350 Ω. The required voltages for achieving null bias points adjacent to each other are determined as 5 and 8 V, respectively, corresponding to Pπ (power required for π-phase shift) of 55 mW. Additionally, the loss of 2×2 MMI is −0.14dB, and the spectrum of the MMI can be observed in the inset of Fig. 4(b); further experimental details are available in our previous work^[33]. The total loss comprises several factors: two SSC ports contribute −3dB (−1.5dB/facet), two MMIs account for −0.5dB (with the 1×2 MMI at −0.35dB and the 2×2 MMI at −0.14dB), waveguide contributes to a loss of −0.3dB (with propagation loss of −0.2dB/cm). The remaining loss of −0.4dB is attributed to the modulation region, which may be caused by misalignment between different layers during EBL exposure, leading to an increase in metal absorption loss within the waveguide region. Figure 4(b) shows the half-wave voltage Vπ measurements for the DO-MZM with a 100 kHz triangular voltage sweep. The Vπ value for the 9 mm long device is 2.5 V, corresponding to a Vπ·L of 2.25V·cm.

Then the EO responses of the fabricated device were characterized under the RF signal. As shown in Fig. 4(c), the on-chip 3-dB EO bandwidth (S₂₁ parameter) of the DO-MZM is 59 GHz, and the input return losses (S₁₁ parameter) is less than −15dB at up to 58 GHz. After packaging, the EO bandwidth became 48 GHz, as shown in Fig. 4(d). The S₁₁ is less than −10dB at up to 47 GHz but increases beyond 47 GHz. The reasons are as follows. First, the RF input section is composed of RF connectors and RF pins, and both have specific working frequencies, which will cause an increase in reflections beyond 40 GHz. Second, there are discontinuities in the RF links between RF connectors and pins that will further degrade reflection performance. This issue can be solved by optimizing the connections between RF connectors and RF pins.

When paired with a BPD, the DO-MZM can restrain DC components, suppress common-mode noise, and enlarge the dynamic range of the MPLs. The responsivity of the two photodetectors in the BPD is η1, η2, and the noise is n1(t), n2(t). Hence, the response current of the photodetectors is obtained in Eqs. (1) and (2), i1=η1E1E1*=η1[A2+B22+ABcos(dωt+dφ−π2)]+n1(t),(1)i2=η2E2E2*=η2[A2+B22+ABcos(dωt+dφ+π2)]+n2(t),(2)where E1 and E2 represent the two output ports of the DO-MZM. dω and dφ stand for the angular frequency and phase difference between the two ports. A and B are amplitudes of E1 and E2, respectively, which are complementary to each other. According to Eqs. (1) and (2), the total output current of the BPD iout is given asiout=i1−i2=(η1−η2)A2+B22+(η1+η2)⁢ABcos(dωt+dφ−π2)+n1(t)−n2(t).(3)

Equation (3) illustrates that when a dual-output modulator is paired with a BPD, according to n1(t)−n2(t), common-mode noise can be suppressed to a lower level.

To test the DO-MZM’s ability of the noise floor suppression in MPLs, we have constructed a balanced detection structure, as shown in Fig. 5(a). In this measurement, a continuous-wave light source was provided by a DFB laser with a center wavelength of 1550 nm (Santec TSL-510). The light passed through a variable optical attenuator (VOA) and an optical power amplifier (EDFA, Amonics AEDFA-IL-23-B-FA) before being directly coupled into the sample (DO-MZM). The bias point was then set at the linear operating point. Simultaneously, a microwave source (ROHDE & SCHWARZ SMB 100 A) generated a 1 GHz signal at a power level of −25dBm, which was applied to the DO-MZM. Subsequently, the two optical output signals of the device went through the optical time delay lines (OTDLs) and VOAs to ensure consistent delay (dφ) and power (A & B) levels. Finally, both signals were simultaneously input into the BPD.

To conduct a comparative analysis on the disparity in noise floor between the single output and dual output received by the BPD, we individually connected the device’s two ports, 01 and 02, to the BPD, respectively. The resulting output data are displayed in Figs. 5(b) and 5(c). The signal powers were measured at −49dBm (port 01) and −46dBm (port 02), with corresponding background noise levels of −65.04dBm (range from −65.04 to −68.63dBm at 10 GHz) and −65.44dBm (range from −65.44 to −68.15dBm at 10 GHz). Subsequently, both output ports were simultaneously connected to the BPD, as illustrated in Fig. 5(d). The signal power increased to −41dBm (both ports), while the signal’s background noise decreased to −83.78dBm (range from −83.78 to −86.53dBm at 10 GHz). Through this experiment, an 18 dB CMRR was measured. Additionally, by balancing the output power of ports 01 and 02, a signal enhancement of 6.2 dB was calculated. The CMRR and signal enhancement achieved in this study are comparable to those of commercial bulk LN devices, but with a more compact size and larger useful bandwidth.

In this work, a low-loss, broadband TFLN EO-MZM with two opposite outputs is designed, fabricated, and characterized. The device shows an insertion loss of 4.2 dB, a half-wave voltage of 2.5 V, and an on-chip 3-dB EO bandwidth of 59 GHz. A large bandwidth of 48 GHz is achieved after packaging. When paired with a BPD receiver, an 18 dB CMRR and 6.2 dB signal enhancement are measured and calculated, respectively. TFLN has shown its advance in fabrication of ultralow loss, low Vπ, and broad bandwidth modulators, and the footprint is more compact than those of modulators based on bulk LN platforms. Our result shows that the DO-MZM based on TFLN is promising for practical use in high-performance MPLs.