Thin-film lithium niobate (TFLN) modulators have rapidly advanced in recent years, driven by the material’s strong electro-optic (EO) effects, the commercialization of high-quality films, and significant progress in etched-waveguide techniques. These advancements have enabled high modulation efficiency, large bandwidth, low loss, and compact modulator^[1,2]. Various modulator types have been demonstrated, including the Mach–Zehnder interferometer (MZI)-based traveling-wave (TW) modulators^[3,4], slow-light modulators^[5,6], ring resonators^[7,8], Bragg grating modulators^[9,10], and Fabry–Perot modulators^[11,12]. Among these, the MZI-type modulator, despite its longer length, is particularly favored in the industry due to its low loss and broad operating wavelength range. Furthermore, the bandwidth and modulation efficiency limits of TW modulators have been broken using periodic capacitively loaded TW (CLTW) electrodes^[13–16]. To maintain index matching between microwave and optical waves, quartz or undercut-etched silicon substrates are typically used. Notably, a high-performance folded TFLN modulator has been realized, exhibiting a large EO bandwidth of 45 GHz and a low half-wave voltage of 0.7 V, thereby supporting driverless single-lane 100 GBaud PAM-4 modulation^[17].

Another significant challenge in the application of TFLN modulators is fiber-to-chip coupling. Two primary methods are employed: edge couplers and grating couplers. Edge couplers offer the advantages such as broadband operation, polarization insensitivity, and low loss^[18–21], with coupling losses reported as low as 0.54 dB/facet using lensed fibers^[20]. However, edge couplers often necessitate facet polishing, which is not conducive to wafer-scale testing, and coupling with single-mode fibers (SMFs) typically results in higher losses^[22,23]. Conversely, grating couplers, which are compatible with SMFs having large mode-field diameters (MFDs), provide flexible layout options and are better suited for preliminary testing of integrated photonic devices. Nevertheless, the traditionally high loss and narrow bandwidth of grating couplers have constrained their broader application, leading to a preference for edge couplers in final packaging. Although there have been efforts to improve the coupling efficiency and bandwidth of grating couplers, these often involve special or complex processes that are incompatible with standard device fabrication. In our previous work, we proposed a high-performance grating coupler that incorporated a reflector mirror and a one-port cavity, which achieved a coupling loss of 0.89 dB and a perfectly vertical coupling configuration, being superior to those of other gratings^[24].

In this paper, we present a high-performance modulator that integrates the fabrication processes for both the reflective mirrors of perfectly vertical grating couplers and CLTW electrodes, resulting in a device compatible with perfectly vertical coupling and packaging. Initially, the grating mirrors and electrodes were fabricated on the surface of a TFLN on a silicon substrate. The entire device was then flip-bonded onto a quartz substrate using UV adhesive. This approach eliminates the need for etching openings for the fiber or metal mirror on the back side of the quartz when both the grating mirror and the modulator electrodes are directly fabricated on TFLN with a quartz substrate. It also avoids mechanically unstable suspended structures on silicon substrates for high-performance modulators. By combining the advantages of a CLTW modulator with the cavity-assisted grating structure on a quartz substrate, we achieved a device that offers a large bandwidth, a low half-wave voltage, low coupling losses with SMFs, and a broad operating wavelength range. Furthermore, the perfectly vertical coupling simplifies packaging and enhances robustness, allowing for scalability to parallel multichannel and wafer-level production.

Figure 1 shows the schematic structure of the proposed device. The modulator is based on an MZI, with optical waveguides formed from the 400-nm-thick half-etched TFLN. The modulator’s electrodes employ a CLTW design, featuring a wide signal electrode to minimize microwave loss and T-shaped periodic structures to maintain a narrow electrode gap for efficient modulation^[13,15]. The grating design embeds the diffractive grating (with period p2 and groove width w2) within a one-port cavity formed by partially and perfectly reflecting Bragg gratings (with period p1 and groove width w1), along with a metal mirror at the bottom to enhance light emission directionality and coupling efficiency. Further details can be found in Ref. [24]. The grating mirror and CLTW are positioned beneath the LN waveguide, separated by a SiO2 isolation layer. The entire device is bonded to a quartz substrate using UV adhesive (NOA61), which has a refractive index of approximately 1.541 at 1550 nm, a dielectric constant of 4.04 at 1 MHz, and a dielectric loss tangent of 0.045. Figures 1(b)–1(d) depict the schematic diagrams of the modulation region and grating cross section, with the corresponding parameters listed in Table 1. It should be noted that in this design, the light transmission direction in the grating is along the z axis, while in Ref. [24], it is along the y axis. Consequently, the optimized grating parameters are different.

Figure 2 presents the simulation results of the modulator’s performance. To ensure high bonding yield, the UV adhesive layer cannot be excessively thin. Given that the dielectric loss tangent of the UV adhesive is 0.045, microwave transmission loss becomes significant. Consequently, the modulator must be sufficiently separated from the UV adhesive to minimize the microwave loss. We simulated the microwave loss and EO S21 for different thicknesses (hc2) of the SiO2 isolation layer between the modulator and the UV adhesive, as shown in Figs. 2(a) and 2(b). The results show that as hc2 increases, the microwave loss decreases and the EO bandwidth increases, with both approaching convergence at hc2=5μm. Therefore, we selected hc2=5μm as the optimal thickness. The simulated half-wave voltage is 2.3V·cm, and the characteristic impedance is designed to be approximately 42 Ω to achieve a better match between microwave and optical waves, while still maintaining a low radio frequency (RF) loss, as shown in Figs. 2(a) and 2(d). Under this configuration, the refractive indices of the microwave and optical waves are well matched, as shown in Fig. 2(d).

The fabrication process flow for the proposed TFLN modulator is shown in Fig. 3. The LN waveguides and electrodes/mirrors were fabricated on a commercially available x-cut LN-on-insulator (LNOI) wafer (NanoLN, China), which consists of a 400-nm x-cut LN thin film and a 3-µm buried SiO2 layer on a silicon substrate. The LN waveguides were patterned using electron beam lithography (EBL) followed by argon-plasma-based dry etching. Next, the entire structure was covered with a 700-nm-thick SiO2 layer via plasma-enhanced chemical vapor deposition. The CLTW electrodes and metal mirrors were then patterned using contact i-line photolithography, followed by evaporation and a lift-off process. Up to this point, these steps adhere to a standard modulator fabrication process. Subsequently, an additional SiO2 layer was deposited to isolate the modulator from the UV adhesive. The entire structure was then flipped and bonded onto a quartz substrate using UV adhesive. To remove bubbles during the bonding process, the sample was placed in a vacuum system; stress can be applied to expel them. The silicon substrate of the LNOI wafer was removed by mechanical grinding and chemical etching. At this stage, the CLTW electrodes and metal mirrors are embedded beneath the LN thin film and the SiO2 layer. Finally, dielectric openings were created to access the metal pads. Figure 4 shows images of a completed sample, including the grating coupler with a mirror and the modulation section. Although die-to-die bonding was utilized in this example, the process can be easily scaled to full-wafer bonding.

The optical transmission was measured using SMFs (SMF-28) mounted perfectly vertical to the chip, as shown in Fig. 5(a). The laser source was normalized out. The proposed modulator exhibited a minimum fiber-to-fiber insertion loss of approximately 6.5 dB. Comparing the modulator’s response to that of a reference straight waveguide with two grating couplers revealed an additional 2 dB loss attributable to the modulator. The coupling loss is higher than the simulation (0.9 dB per grating), and the spectrum shifts to shorter wavelengths, primarily due to deviations in grating dimensions and overcladding thickness. The variations of 15% in hc1, w1, and w2 may lead to this result. Further details on fabrication tolerance can be found in Ref. [24]. The deviation from the optimized results is due to the lack of dimension compensation during the fabrication of this chip. However, fabrication errors within approximately 15% can be controlled using current exposure and etching techniques. The additional loss in the modulator is mainly attributed to the multimode interference (MMI) coupler and the waveguide transmission loss in the modulation region. With precise process control, the total insertion loss of the modulator could be reduced to less than 3 dB (<1dB from the modulator^[15] and <2dB from the grating coupling loss^[24]). Despite this, the result demonstrates a low fiber-to-fiber insertion loss using the perfectly vertical coupling method, which is not achievable with conventional grating couplers. The modulation efficiency of this structure was further evaluated using a 100-kHz triangular-wave voltage drive. A typical sinusoidal response of the output optical signal was observed, as shown in Fig. 5(b). The half-wave voltage (Vπ) was determined to be 3.8 V, corresponding to a VπL of 2.66V·cm, which is slightly larger than the theoretical value. This discrepancy may be due to variations in overcladding thickness and misalignment of metal patterns resulting from manual mask alignment. The extinction ratio (ER) was measured at 17 dB from the direct current (DC)-driven responses, as shown in Fig. 5(c).

To evaluate the bandwidth of the proposed modulator, the electrical properties of the CLTW electrode were first assessed using a vector network analyzer (VNA). As shown in Fig. 6(a), the S21 response indicates a 6.4 dB bandwidth exceeding 67 GHz, with S11 remaining below −15dB across the entire frequency range. Key electrode parameters, derived from the S-parameters, such as microwave loss, characteristic impedance, and microwave refractive index are presented in Figs. 6(b)–6(d), respectively. The microwave refractive index is slightly lower than the designed value, likely due to variations of 1 µm in the width of the S electrode (wS) and the gap (gm3). Fortunately, this minor change in refractive index has a minimal impact on the EO bandwidth. The fitted characteristic microwave loss from Fig. 6(b) is approximately 0.6dB·cm−1·GHz−1/2, which is higher than the simulated results (0.4dB·cm−1·GHz−1/2). This increased microwave loss may be attributed to dielectric losses from the deposited oxide and UV adhesive, as well as scattering losses from rough metal surfaces. Despite this, the loss remains significantly lower than that of typical TW electrode-based TFLN modulators^[3,4].

The EO bandwidth of the modulator was further evaluated, as shown in Fig. 7(a). A large EO bandwidth with a 2 dB roll-off at 67 GHz for a 7 mm-long device was achieved, attributed to good index matching with the quartz substrate. The back-to-back (B2B) bit-error rates (BERs) for 80 and 100 Gbit/s on–off keying (OOK) signals were measured and calculated using offline digital signal-processing algorithms. As shown in Fig. 7(b), the BERs are below the KP-4 forward error correction threshold at low received optical power, with no error floor observed. Clear eye openings were observed at data rates up to 100 Gbit/s, as depicted in Figs. 7(c) and 7(d).

The initial fabrication steps of the proposed modulator, including the formation of LN thin films, optical waveguides, and electrodes, are conducted on a silicon substrate, maintaining compatibility with silicon-based complementary metal oxide semiconductor (CMOS) technology. To achieve refractive-index matching between microwave and optical waves, the entire modulator was transferred to a quartz substrate, eliminating the need for partial etching of the silicon substrate^[15–17], which could compromise mechanical stability. The fabrication process for the reflector and electrodes is also compatible with this approach. Compared to benzocyclobutene (BCB) bonding technology^[4], UV adhesive bonding offers the advantages of simpler fabrication and reduced cost. NOA61 has a greater thickness and a higher tolerance for the bonding surface roughness, allowing it to accommodate step heights exceeding 1 µm from the electrodes without the need for additional polishing. Moreover, BCB bonding requires specialized bonding equipment, while UV adhesive bonding requires only a UV lamp and an oven, with the UV adhesive itself being more affordable than BCB. Although the fabrication described here was performed on individual dies, the entire process is easily scalable to wafer-scale production. Additionally, the modulator structure is highly robust, featuring a grating coupler that utilizes a perfectly vertical coupling method, which facilitates both testing and packaging. This vertical coupling is especially beneficial for multicore fibers^[25], allowing for easy expansion into multichannel applications, such as parallel-channel optical transmitters with a perfectly vertical packaged multicore fiber interface, as illustrated in Fig. 8.

In conclusion, we have demonstrated a large-bandwidth TFLN modulator with grating couplers for perfectly vertical coupling to single-mode fibers. The modulator incorporates CLTW electrodes on a quartz substrate, offering high modulation efficiency and broad bandwidth. Coupling efficiency is further enhanced by incorporating a one-port cavity and a metal mirror beneath the structure. The fabrication process begins with the formation of the modulator’s electrodes and metal mirrors on an LNOI wafer with a silicon substrate. The structure is then bonded to a quartz wafer using UV adhesive, after which the silicon substrate is removed. The modulator achieves a low fiber-to-fiber loss of 6.5 dB, a 3-dB bandwidth exceeding 67 GHz, and a half-wave voltage of 3.8 V for a 7-mm long intensity modulator. Successful data transmission using OOK modulation at speeds up to 100 Gbit/s has also been demonstrated. The proposed modulator, characterized by high performance, structural stability, robust design, compatibility with CMOS processes, and convenient packaging, offers promising potential for high-capacity, low-loss optical communication systems.