Harnessing the enhanced light-matter interaction inside the micro-nano optical cavities, frequency combs have triggered the development of multi-channel source generation via the nonlinearity-driven four wave mixing (FWM) [1,2]. By balancing group velocity dispersion (GVD) and inherent Kerr nonlinearity, the dissipative Kerr soliton (DKS)—mode-locked pulses by high-power pumping—offers low noise, high coherence, and static comb spectra [3,4]. These features have advanced applications in optical telecommunication [5,6], microwave engineering [7,8], and LiDAR [9,10]. Rapid deployment of large-volume manufacturing has made chip-integrated frequency combs accessible across various thin-film materials, including silica [11,12], ${\mathrm{Si}}_3{\mathrm{N}}_4$ [13,14], AlGaAs [15,16], ${\text{LiNbO}}_3$ [17,18], and ${\text{LiTaO}}_3$ [19]. More importantly, multi-layer heterogeneous integration and efficient planar microfabrication enable the integration of multiple optical devices onto a single chip, revolutionizing the architectures of photonic integrated circuits (PICs) and frequency microcombs. Among these novel materials, ${\text{LiNbO}}_3$ and ${\text{LiTaO}}_3$, categorized as ferroelectric crystals, emerge as promising candidates for versatile photonic platforms due to their high-linear EO modulation, plentiful nonlinear optical effects, and ultralow propagation loss [19–24].

Given the excellent nonlinear features within the ferroelectrics, it naturally raises the conceptional combination of a high-speed EO modulator and a broadband Kerr comb source to establish a fully stabilized microcomb. To date, the Kerr soliton microcomb has been demonstrated in Z-cut ${\text{LiNbO}}_3$ [18,25–27], particularly with a bi-started characteristic [28] and mode-locked state self-stability [29]. However, this thin-film cut type physically limits access to the maximum Pockels coefficient ($r_{33}$), impeding the integration of high-performance EO modulators with Kerr combs. Although seeding Kerr combs is feasible in X-cut ${\text{LiTaO}}_3$ [19] or ${\text{LiNbO}}_3$ [30], achieving soliton frequency combs with long equilibration times remains a significant challenge. This is due to the slow blue shift caused by the photorefractive effect and the rapid red shift caused by the thermo-optic (TO) effect, which together create a dynamic disequilibrium that hampers the stabilization of mode-locked states. Only recently was the rapid single sideband sweep investigated for the initial addressing of soliton states in ferroelectric cavities [19], but the triggered soliton state trends to disappear within a few seconds, since the detuning condition on which the soliton generation depends will vary with the refractive index change on a time scale of seconds. Although this resonance excursion can be tracked by the power-kicking [13] or Pound–Drever–Hall (PDH) laser locking techniques [31], the electronic devices are sophisticated and not suitable for PICs with miniaturization requirements. Thus, for the realization of a Kerr soliton microcomb with long equilibration time in X-cut ferroelectrics, it is essential to eliminate the complex photorefractive disruptions affecting micrometer-sized waveguides. This approach could further simplify the search for Kerr soliton states, and support the development of low-voltage-driven modulators in a monolithic way. While techniques such as annealing and cladding removal have been employed to mitigate the PR effect [32,33], their effectiveness in stabilizing soliton microcombs under high-optical-power conditions has not been sufficiently studied yet.

In this paper, we realize the photorefractive-free resonators by simply heating up the ${\text{LiTaO}}_3$-based microresonator. Without the need for an additional electronic feedback loop to dynamically stabilize the laser-resonance offset, we demonstrate Kerr soliton generation using the dual-suppressed strategy, which involves mitigating photorefractive effects through thermally controlled carrier dynamics and minimizing TO effects via an auxiliary-assisted approach. This comb generation has proven to be highly reliable, with only manual or piezo-module tuning of the optical instruments. Notably, the deterministic generation and prolonged lifetime of mode-locked states represent significant advantages of this technique, overcoming the limitations posed by time-dependent photorefractive phase modulation and the huge thermal dragging response in X-cut ferroelectric photonic platforms.

We start with a simple band model for illustrating the photorefractive mechanism in the ferroelectric material, as displayed in Fig. 1(a). Defect levels between the valence band and the conduction band are ubiquitous in congruent ferroelectric materials and are thought to originate from vacancies and antisite defects inside the crystal lattice [34], as well as defects introduced during bonding and etching processes. These defects often match the photon energy, allowing them to interact with the incident light. Depending on the laser intensity and photon frequency, photoexcited carriers are promoted to the conduction band, where they transport before either recombining with other carriers or returning to the defect levels. This electron process can produce a refractive index change in ferroelectric materials, known as the photorefractive effect [35]. In waveguide components, the electric fields of confined optical modes direct the movement of these photo-excited carriers. For instance, in transverse-electric (TE) modes, the horizontal potential difference created by the optical field leads to a photovoltaic electric field and hence to the Pockels modulation, as carriers accumulate along the direction opposite to the electric field component of the light mode [Fig. 1(b)(i)]. In comparison with the transient light-matter interaction, the electronic reaction has a longer relaxation time, which stands for a longer period of the carrier migration and results in the emergence of a slowly developing space-charge field. Based on this effect, a Z-cut soliton can be accessed via manually tuning the laser frequency into the cavity resonance [25,29]. However, the circumstances of in-plane optical mode pumping in X-cut ${\text{LiNbO}}_3$ or ${\text{LiTaO}}_3$ resonators are markedly distinct from the Z-cut variant, primarily owing to its larger thermo-optic (TO) coefficient [36]. Although the rapid TO shift locks the pump laser within the thermal bistability regime [4], it also narrows the soliton existence step and renders the soliton seeding operations ineffective [37]. Therefore, the opposite phase shift produced by the photorefractive effect might over-compensate for the thermal resonance drift to a certain extent, and the distinction between the above two reactions is also manifested in the timescale, whereby the postponed photorefractive behavior exacerbates the drawback of locking the laser-resonance offset to an accessible DKS state. Figure 1(c) reflects that the pump-resonance offset jitters when the pump laser is fixed in the red-detuned resonance region. A dramatic blue shift happens slowly once tuning the laser into the cavity resonance dip from the red-detuned side, and vice versa. At the same time, the swift TO shift towards the long wavelength regime would result in the pump exceeding the cavity resonance, which has appeared in previous works [17,38,39], betraying the competitive relationship between two effects and its irresistible prohibition to an emerging mode-locked comb state. In the optical microresonator, the standing wave characteristic shapes the spatial distribution of the built-in electric field [Fig. 1(b)(ii)], creating the refractive-index-modulated pattern in analog to the grating with the frequency selective feature [40]. With high-power pump frequency tuning over the resonance, the formed mode splitting exhibits an apparent evolution [Fig. 1(d)] that can also account for the enhanced charge-discharge-type electric field originating from the production and the neutralization of carriers at room temperature [41,42]. This introduces the instability into the nonlinear frequency conversion process, such as the deviation in local integrated dispersion and the reduction in the extinction ratio of the pumped resonance [43]. Furthermore, besides the racetrack microresonator, the configuration of various devices has the capacity to influence the generation of a photorefractive-induced grating and differentiate the mode splitting value, primarily depending on the effective EO coefficient of the designed waveguide. As a result, a more complicated pumping mechanism is required to enable the soliton formation.

Heating is a plausible technique for degenerating the photorefractive effect in the confined optical waveguide, which has been employed for nonlinear harmonic generation [44,45] and EO modulation [46]. According to the classical photorefractive principle [35], the carrier lifetime $\tau$ is inversely proportional to the ion conductivity $\sigma$. Based on the Arrhenius-type relationship [47,48], $\sigma (T)\propto e^{-\frac{1}{T}}$, it is prone to enhance the ion conductivity and subsequently reduce the lifetime $\tau$ via elevated temperature operation. This behavior is verified by the I-V measurement of a coplanar waveguide (CPW) device on 200 nm LT thin film, using the probe station aligned with a heating function to thermally control on-chip temperature. The I-V curve is traced by a power device analyzer (Keysight B1505A) through contacting the probes in the ground and signal electrodes of the CPW. As shown in Fig. 2(a), an exponential growth in current is observed, consistent with the conductivity enhancement as the temperature rises from 25°C to 125°C. For a more thorough illustration, we set up a modulator/demodulator system [40] to conduct a photorefractive grating measurement based on our X-cut ${\text{LiTaO}}_3$ microresonators by planar manufacturing, which is described in detail in Section 4. The grating-induced separation between clockwise (CW) and counter-clockwise (CCW) modes, as inferred from the probe transmission in the inset of Fig. 2(b), is defined as the splitting value. To measure the temporal change of the pumped optical resonance, the mode splittings periodically scanned by a low-power laser are recorded at a sampling interval of 0.3 s for 3 min. With the increase in temperature from 30°C to 90°C, the mode splitting curves display a numerical decline and a shorter period of oscillation at higher temperature [Fig. 2(b)]. A possible explanation of the above phenonmenon could be that the attenuated grating behavior is a consequence of a reduced existence time of the heated carriers, and that the diffusion and recombination of accelerated charge carriers could be responsible for a more obvious charge-discharge cycle for a photovoltaic electric field. Nevertheless, this experiment does not fully demonstrate the ideal suppression of the photorefractive grating, primarily ascribed to the fact that the coupling of low-power probe light and measurement of the small mode splitting value are more susceptible to the fiber jitter, which is induced by the prominent thermal airflow at higher temperature. Correspondingly, Fig. 2(c) plots the photorefractive grating appearance (upper panel) and the maximum resonance splitting degrees versus temperature (lower panel), enabling access to on-demand manipulating the photorefractive effect inside a high-Q microcavity under the temperature-assisted operation.

The device under test for soliton generation has an average intrinsic loss (${\kappa }_0$) of approximately 53 MHz, revealing a mean intrinsic Q-factor of near $3.6\times {10}^6$. To generate a DKS soliton relying on the foregoing finding, a dual-suppressed strategy is performed using our experimental setup, as shown schematically in Fig. 3(c). First, to counteract the unwanted photorefractive modulation while pumping the cavity with more than one hundred milliwatts order of magnitude, a stronger heating operation via TEC is considered in the experiment to meet the demand of no additional resonances drift. Second, as for the cavity temperature stabilization under a huge TO effect, the thermal dragging dynamics that exists in soliton formation is bypassed using the dual-laser-driven method [49]. With a total insertion loss of approximately 10 dB (5 dB per facet from fiber-to-chip coupling via a lens fiber and negligible loss from the optical chip), two high-power sources are employed to inject into the cavity from two opposite directions after the fiber optical system including power amplifying, polarization controlling, and edge coupling. Following a conventional tuning trace from the blue regime to red regime, a 1559 nm transverse-magnetic (TM) polarized optical source with 185 mW on-chip power, serving as the auxiliary laser, would level off to the resonance dip in advance. In the meantime, the TEC-controlled temperature of the chip holder should be adjusted finely till the time-varying resonance excursion vanishes and thermal resonance triangle dominates. The final settled temperature is set to 230°C to fully compensate for the carrier relaxation time in the photorefractive effect. As such, the auxiliary laser can be adjusted close to the blue-detuned regime of the TM mode resonance. The low quality factor and overwhelming anomalous dispersion of the TM mode group raise the microcomb generation threshold [Fig. 3(b)], which is preferable to suppress the unfavorable beatnote noise arising from different microcombs generated by two high-power sources [50]. To get soliton microcomb seeding started, a 1564 nm tunable external-cavity diode laser, with 97 mW on-chip power, TE polarization, and 0.75 GHz/ms piezo-controlled laser sweeping, is required to tune near the high-Q TE mode resonance. Throughout the soliton formation process, the abrupt alteration in pump power, which corresponds to the transition of microcomb states, creates a passive tuning of auxiliary laser-resonance offset due to the TO effect. This equivalent change in intracavity auxiliary power in turn maintains the total intracavity power to guarantee the thermal balance of all cavity resonances.

Figure 3(d) shows the generated comb power and optical transmission; the stable comb power profile betrays a simultaneous suppression of the photorefractive effect and the thermal effect. When the pump laser is biased from the short wavelength regime to the long wavelength regime with respect to the pumped resonance, the cavity power exhibits oscillating optical intensity followed by multistage flattened steps. This signal delineates a specific progression from the modulation instability (MI) state to the low-noise mode-locked states. By setting the pump laser rigidly in the effective blue-detuning steps, different types of temporally dissipative soliton waveforms can be witnessed in Fig. 3(e), including the MI comb (state i), breather soliton comb (state ii), multi-soliton comb (state iii), two-soliton comb (state iv), and single-soliton comb (state v), with RF noise spectra confirming their coherence [Fig. 3(f)]. The available extent of the microcomb spectra can be optimized through the elaborate dispersion design and higher-power pumping, so as to establish the octave spanning microcomb catching up to the state-of-the-art performance in the Z-cut ferroelectric platform [18]. As a proof of concept in our work, this method similarly provides a universal method for accessing a DKS comb on the X-cut thin film ${\text{LiNbO}}_3$ photonic platform [30].

Importantly, to showcase the potential of our dual-suppressed technique, we test the long-term performance of the multi-soliton state via fixing a TE-polarized pump laser (1564.240 nm) and a TM-polarized auxiliary laser (1559.147 nm) engaged by an elevated temperature to heat up the cavity. The absence of the resonance drift caused by the photorefractive-induced charge field enables the mode-locked comb existence exceeding 3 min, as depicted in the time-wavelength map in Fig. 4(a). In Fig. 4(b), sliced spectral frames at the fixation time of 20 s and 160 s provide further details that the representative multi-soliton-shaped envelope is well preserved over long equilibration periods in our experiment. Note that the power fluctuations of the depicted comb intensity are ascribed to the temperature instability of the TEC and the variations of fiber-to-chip coupling caused by the thermal airflow. These disruptions do not impair the mode-locked nature of our generated microcomb, providing compelling advantages for photonic applications requiring stable comb sources.

Apart from the redundancies of complex electronic devices to access and stabilize the mode-locked soliton, the heat-assisted approach faces limitations due to the large device volume of TEC-controlled equipment and unavoidable heating-air perturbation affecting the edge coupling system. To address these issues, we fabricated the integrated platinum (Pt) heater (see Section 4) with a spiral channel structure designed to form and localize the heating field, enabling a fully integrated soliton generation prototype [Fig. 4(c)]. This greatly reduces the footprint of the soliton generator. By applying an 80 V DC voltage to an electrode with approximately 2162 Ω resistance, we can get the heat distribution map of the etched surface using a temperature measurement microscope system (QFI/InfraScope-TM-HS). The localized thermal domain near the waveguide reached temperatures exceeding 200°C [Fig. 4(d)], which has the same elevated temperature as the TEC-controlled heater, confirming the effectiveness of the all-on-chip demonstration of our dual-suppressed method. Based on the same dual-laser-driven configuration, we reproduced the microcomb generation via sweeping the pump laser. As inferred from the normalized comb power [Fig. 4(e)], the step-like comb power confirms the triggering of different mode-locked states similar to the aforementioned results. However, the use of the localized thermal field reduces device volume and minimizes coupling issues, demonstrating the advantages of the integrated approach.

In summary, the intracavity photorefractive effect in the X-cut ${\text{LiTaO}}_3$ thin film can be readily restrained at high temperature, as thermally improving carrier conductivity breaks the non-static balance between internal charge field and intracavity electromagnetic waves. Via the dual-laser-driven scheme, we obtain various soliton states in a low-loss ${\text{LiTaO}}_3$ cavity, including soliton crystals and a single soliton. Our results about the robustness of a low-noise state with considerable equilibration lifetime in experiments and all-on-chip demonstration are core to applications involving fast reconfigurable soliton microcombs [51,52] based on the X-cut ferroelectric platform.

Nevertheless, there are several issues requiring further attention. The Pt-crafted heater, like other metals such as titanium and chromium, cannot endure prolonged exposure to high voltages, which may lead to severe electrical breakdown. Similarly, the air-exposed Joule heating is unattainable with long-term reliable operation due to the penetrated air attempts to oxidize the metal at elevated temperatures. Moreover, the experimental results for DKS steps are not ideal, shown in the blue-shaded regions in Fig. 4(e). This can be attributed to residual photorefractive effects and coupling fluctuations, leading to a mode splitting and the ongoing resonance instability. To address the existing problems, some solutions relevant to this work are promising. For example, the thermal management by advanced device structure can be utilized to optimize heat distribution [53] and reduce the heating impact on the coupling stability. Modified microheater microstructures, especially in the realm of microelectronics [54,55] and micro-electromechanical systems (MEMS) [56,57], might enable a loop in a more reliable thermal field control. Tricky ways, such as rising soliton steps in larger FSR resonators [16] to weaken the thermal response, can reduce the need of auxiliary lasers and further miniaturize the chip footprint.

Broadly speaking, we believe that the thermal-assisted approach has the potential of enriching the current microcomb generation for an all-optical ${\text{LiNbO}}_3$ or ${\text{LiTaO}}_3$ integrated circuit. As illustrated in Fig. 5, this platform, highly compatible with high-speed modulation [20,23] and horizontal periodically poled waveguides, offers high nonlinear conversion efficiency [58–60], positioning it as pivotal for next-generation optical applications. A notable application of this technique lies in coherent optical communications [61–63]. As well as empowering hyperscale data transmission networks with high-coherence parallelization and high-speed reconfiguration, the monolithic realization of stable DKS formation without electrical feedback and $r_{33}$-based EO modulation can also avoid the device complexity of optoelectronic systems. Combined with the periodically poled components, the versatility of this X-cut design also facilitates the frequency doubling applications from the visible to infrared band, such as selective harmonic generation [64], broadband synthetic frequency lattice [65], and efficient optical parametric amplifier [66] based on coherent multi-channel sources.

The device fabrication involves two processes: the manufacturing of thin film ${\text{LiTaO}}_3$ and the planer device fabrication. Via ion cutting and wafer bonding (commonly termed smart-cut technique), 600-nm-thick ${\text{LiTaO}}_3$ thin film is sliced from an optical-grade ${\text{LiTaO}}_3$ bulk wafer to the thin-film-on-insulator structure, including 600 nm LT thin film, 4.7 μm oxide layer, and 500 μm silicon substrate, subsequently followed with dicing into small chips with $1\mathrm{cm}\times 1.2\mathrm{cm}$ for a more flexible device demonstration. With regard to the device fabrication, the micro-nano patterns are transferred onto the diced chips using e-beam lithography (Elionix ELS-BODEN 125) with a hard mask (Ma-N 2405) and physical dry etching process with argon ions (Leuven, HAASRODE-I200). Next, an alkaline wet etching approach is employed to remove the sidewall redeposition as well as the etched resist. The well-defined microheater, comprising 150 nm Pt and 10 nm Ti, is fabricated through a dual-layer lift-off process, which employs a maskless lithography (Heidelberg DWL 66+) and e-beam metal evaporation system. Our under-test ${\text{LiTaO}}_3$ microresonator [inset of Fig. 3(c)] has an average intrinsic $Q$ value of 3.6 million and 100 GHz FSR, with 400 nm etched depth and 200 nm slab. In order to generate a robust bright soliton state, the balance between material nonlinearity and group velocity dispersion can be obtained by anomalous integrated dispersion, with a multimode waveguide width of 2 μm. Alternatively, the adopted racetrack device is intended to be orientation selective so that the Raman scattering in microresonators is inhibited to ensure the maximum conversion efficiency of the Kerr nonlinear process.

The resonance splitting induced by a cavity-enhanced photorefractive grating can be quantified by literally pumping the racetrack resonators. However, measured pump transmission features a large thermal-induced resonance distortion that influences the extraction of peak splitting value; thus the actual resonance shape should be detected by another laser with reduced optical power from the opposite edge facet. Using a piezo-driven module to tune the sweeping pump laser activates the photorefractive grating effect in a ${\text{LiTaO}}_3$ microresonator. To extract the mode splitting information, the probe laser that propagates in the backward direction is modulated by an electro-optic intensity modulator (IM). After injecting into the microresonator, the probe laser is then received by PD, and the probe signal mixed with strong backscattering pump light is recognized by the lock-in amplifier (Stanford Research Systems, SR844) based on the same applied RF signal.

For the Q-factor characterization, we use the Toptica DLC CTL 1550 to deliver the tunable continuous-wave laser into the racetrack resonator with specific polarization, which is realized by a variable optical attenuator (VOA), a polarization controller (PC), and tapered fiber coupling. To extract the actual group dispersion from the laser sweeping measurement with a nonlinear piezo-driven property, a customized fiber-based MZI interferometer, with 50 m delay line, is used to calibrate the definite location of resonances.

Acknowledgment. The sample fabrication in this work is supported by Shanghai Institute of Microsystem and Information Technology (SIMIT) material-device processing and characterization platform, ShanghaiTech Material and Device Lab (SMDL), JFS laboratory, and the USTC Center for Micro and Nanoscale Research and Fabrication.

Author Contributions.X.K., C.Y., and X.O. fabricated the ${\text{LiTaO}}_3$ wafers. J.C., C.W., and S.W. designed the devices. J.C., B.C., X.W., and Y.Z. fabricated the devices. J.L., J.C., P.W., B.C., D.S, W.X., and Z.Q. carried out the measurements of photorefractive-induced mode splitting, soliton generation, and the I-V curve. J.C. and J.L. analyzed the data. J.C., C.W., and S.W prepared the figures and wrote the manuscripts with contributions from all authors. C.W., C.D., and X.O. supervised the project.