Thin-film lithium niobate (TFLN) photonics is rapidly emerging as a versatile platform for high-speed electro-optics and nonlinear optics applications [1]. The material properties of lithium niobate (LN) lie at the center of this breakthrough: high refractive index, large χ⁽²⁾ nonlinearity, wide transparency window, and efficient piezo-electric response [2]. Recently, with commercialized TFLN wafers and advanced nanofabrication techniques based on high-quality dry etching [3], the exploration of lithium niobate photonics has been transformed from traditional bulk-scale components into integrated photonic chips, enabling both better performance and novel functionalities [1]. For example, integrated TFLN photonic devices, featuring strong nonlinear interaction, low propagation loss, and small form factors, enable significant advances in electro-optic (EO) modulation [4–6], EO frequency conversion [7], EO frequency comb generation [8–10], Kerr frequency comb generation [11–17], synthetic crystal generation [18–20], transduction [21–25], and engineered all-optical nonlinearities via periodic poling [26–30]. These components represent fundamental building blocks for advanced large-scale photonic systems, boosting applications in optical communications [31], optical computation [32], precision measurements [33], microwave signal processing [34,35], and quantum information science [36]. To unlock the full potential of the TFLN platform, propagation loss arises as a fundamental challenge. At the device level, enhancing light–matter interaction in various applications necessitates ultra-low propagation loss, as the ratio between coupling and loss determines the strength of the desired interaction. At the system level, the requirement of integrating numerous devices to form large-scale photonic circuits is ultimately limited by loss.

Substantial research efforts have been made to minimize the propagation loss of TFLN waveguides, that is maximize the quality (Q) factor of TFLN resonators. Using different resonator designs [37,38], researchers have explored different etching methods to optimize fabrication quality, such as dry (ion-plasma) etching, wet etching, and chemical mechanical etching. Dry etching, utilizing reactive ion plasma, is the predominant method of etching LN [3,39–43] as it allows for precise control of feature sizes. Wet etching, though simpler, encounters the challenges of nonuniform etching along different crystal directions [44,45]. While chemical mechanical polishing yields an ultra-smooth surface, it has difficulty in preserving narrow features, thus making it hard to realize, e.g., optical waveguide coupled resonators within a single TFLN layer [46–48]. The highest Q factor measured in TFLN resonators realized using dry etching stands at 12 million [3,49], though even higher Q factors have been achieved using the CMP process [46]. However, these achievements are below the Q factors observed in other low-loss photonics platforms like 442 million on the silicon nitride platform [50] and the theoretical upper limit of 163 million inherent to the TFLN platform [49]. Furthermore, most breakthrough devices exhibit average Q factors limited to a few million [7–9], restricting the overall device performance. Consequently, the Q factor remains a critical bottleneck within the TFLN platform, requiring a further push on the upper limit.

Here, we demonstrate monolithic microring racetrack resonators on the TFLN platform with a record-high intrinsic Q factor of 29.3 million, and a corresponding ultra-low propagation loss <1.3dB/m, realized using the dry reactive ion etching. We achieve this by optimizing the fabrication process and tailoring the resonator widths and lengths to achieve the largest Q values. We evaluate the device fabrication quality using scanning electron microscopy (SEM) and atomic force microscopy (AFM). We also analyze different mode families observed in the transmission spectra of resonators, investigate the statistical relationship between the Q factor and resonator geometries, and calibrate the resonator linewidth using a radio frequency (RF) modulated laser.

Our microresonators are based on a racetrack geometry. As depicted in Fig. 1(a), the waveguide on the left can couple light in and out, and the cavity on the right is where light circulates for multiple roundtrips, experiencing the resonant enhancement. We chose the width of the racetrack section to be large, aiming for a reduced overlap between the light and sidewall. Figure 1(b) illustrates the anticipated spectrum of a ring resonator featuring a Lorentzian-shaped resonance at the specific resonant wavelength. The Q factor can be extracted from the linewidth and wavelength mathematically. A high Q factor implies a narrow resonance linewidth and minimal propagation loss.

We fabricate high-Q resonators with an optimized fabrication process. The fabrication begins with a chip cleaved from a TFLN wafer with 600 nm thickness of LN and 4.7 μm buried silicon dioxide. The device is first patterned by electron beam lithography using 700 nm thick hydrogen silsesquioxane (HSQ) electron beam resist. We found the optimal dose to be 1600μC/cm2. The multipass writing technique is implemented to ensure high writing quality. Subsequently, approximately 325 nm of LN is etched using an inductively coupled plasma reactive ion etching tool with argon gas. 325 nm is determined to be the optimal etch depth as we observed that deeper or shallower etches result in lower Q. The dry etch process is split into two separate runs in order to check the etch rate in between (approximately 0.6 nm/s) and to reach the target etch depth. Following these steps, the device undergoes a chemical cleaning (hot potassium hydroxide, hot SC1, diluted HF, and piranha) and annealing at 520°C in oxygen gas. For wet processing we place glassware in water baths and for temperature treatment we use an LN-only annealer to avoid possible cross-contamination. This optimized fabrication method exhibits stability and repeatability in consistently yielding functional ultra-high-Q devices.

We have conducted characterizations on our racetrack resonator devices to verify the fabrication quality. Figure 1(c) presents the optical microscope image of a racetrack resonator with length of 500 μm and bending radius of 200 μm. Figures 1(d) and 1(e) are the SEM images of a coupling region of 3 μm width racetrack device with a coupling gap of 0.5 μm. Taken by an AFM image, Fig. 1(f) is consistent with Fig. 1(e), demonstrating the well-defined waveguide and smooth sidewalls, suggesting high fabrication quality. Additionally, Fig. 1(g), obtained from Lumerical simulation, illustrates the fundamental mode supported by a 3 μm width racetrack, showing the reduced overlap between the light and sidewall enabled by the ultra-wide waveguide. We note that wide racetrack waveguides may decrease the coupling efficiency between the bus waveguide and the resonator. Indeed, by characterizing devices with different coupling gaps, we confirmed that all resonances presented in this paper operated in the under-coupled regime. This can be attributed to the reduced overlap between the modes propagating in the coupling waveguide and resonator waveguide. To improve the coupling efficiency, as well as mitigate the excitation of higher-order modes, resonator width could be tapered down in the coupling region, or the directional or pulley coupler can be utilized [51–53].

We measure our device using a tunable laser (TSL-570) and a low-noise detector (Newport 1811). Figure 2(a) demonstrates an exemplary spectrum obtained from our measurements, where the x-axis represents the wavelengths generated through laser sweep, and the y-axis denotes the signal received by the detector. The spectrum presents diverse resonance shapes featuring different linewidths and extinction ratios. This is attributed to the fact that wide racetracks support both fundamental and higher-order modes. We assigned mode family to each resonance based on their shape, extinction ratio, and free spectral range (FSR). Fundamental modes are selected by finding the mode family with the highest FSR [depicted by the red dots in Fig. 2(a)], corresponding to the minimum group index predicted by the Lumerical eigenmode simulation. We observe that the fundamental mode exhibits the highest Q factor among all the mode families (referred to as “high-Q mode family” later), which is consistent with our design as the fundamental mode typically has the smallest amount of overlap with waveguide sidewall roughness.

Many measured resonances feature ultra-high Q factors. The highest intrinsic Q factor is 29.32 million, with corresponding loaded Q factor of 19.56 million, corresponding to a device with parameters of 4.5 μm width, 10 mm length, and 0.6 μm coupling gap [Fig. 2(b)]. The intrinsic and loaded linewidths are 6.4 MHz and 9.6 MHz, respectively. The equivalent propagation loss can be calculated as α=10log10(e−κingL/c)=1.3dB/m, where κi=2π×6.4MHz is the intrinsic linewidth, ng=2.3 is the group refractive index, L=1m is the length, and c is the speed of light. We note that this is the upper bound for the propagation loss, since κi includes bending losses as well. Other resonances within the same mode family exhibit ultra-high Q values exceeding 20 million, as shown in Figs. 2(c) and 2(d). Resonances from the high-Q mode family may feature single or split peaks, which could have resulted from the mode splitting between the clockwise (CW) and counter-clockwise (CCW) modes. The typical CW and CCW mode coupling is small; however, in this ultra-high-Q regime, the linewidth could be comparable to this coupling, leading to the observation of mode splitting. In addition, the potential overlap with resonances from other mode families may also generate similar doublet features.

To verify that the direct laser sweep approach can characterize the actual spectrum precisely, we further evaluate the measurement using RF modulation as a calibration. Figure 4(a) depicts a phase modulator driven by an RF source that is added before the resonator to generate sidebands. The distance between the sideband and the original signal is used as a ruler to calibrate the x-axis. Figures 4(b)–4(d) demonstrate the calibration process. Initially, we measure resonance without RF modulation. Immediately after, we turn on the RF modulation with an RF frequency of 100 MHz and observe the generation of sidebands. With sidebands as a reference, we calibrate the x-axis, extract the linewidth, and calculate Q values. It is worth noting that although this approach allows one to calibrate the linewidth using a precise RF frequency reference, it underestimates the extinction ratio: sidebands generated by RF signal have nonzero transmission when the laser signal is on resonance with the cavity. The fitting, using simple Lorentzian, then underestimates effective “coupling loss” and thus underestimates intrinsic Q. Care needs to be taken to compensate for this effect. Taking this into account, we evaluate the perceived RF-calibrated loaded Q of 18.39 million, which is comparable to the original loaded Q of 16.07 million.

We demonstrate microresonators with a record-high intrinsic quality factor of 29 million on monolithic TFLN. We performed statistical analysis of the measured quality factor over different device designs and calibrated the measurement with an RF modulated laser. The optimized design and fabrication process of monolithic TFLN microresonators enable our realization of ultra-high-Q devices. Moving forward, the enhanced high-Q devices on TFLN could substantially improve the device performances of electro-optics and nonlinear photonics, transitioning devices into a new parameter space. This advancement holds great potential to catalyze system-level applications, thus facilitating applications in microwave photonics [54], quantum computing [55,56], and nonlinear optics [57,58]. Our efforts further push the state-of-the-art, showcasing the potential of the TFLN platform and paving the way for future innovative explorations in integrated photonics.

Acknowledgment. S. L. acknowledges fellowship from Agency for Science, Technology and Research (A-STAR); H. W. acknowledges fellowship from National Science Foundation; L. M. acknowledges Capes-Fulbright and Behring Foundation fellowships; A. C. acknowledges Rubicon postdoctoral fellowship from the Netherlands Organization for Scientific Research (NWO).