a The EO comb is generated by coupling a continuous-wave (cw) pump laser into a LN resonator under microwave modulation. b Spectrum of the EO comb when the pump laser is above the Raman lasing threshold. The EO combs spread around the pump laser and Raman laser share the same f_rep but have different f_ceo (unlocked state). With the help of four-wave mixing (FWM), these two combs expand their bandwidths and overlap. The beating frequency of the two combs in the overlapped modes is transferred to the Raman laser mode and finally locks the Raman lasing frequency being integer multiples of the microwave frequency from the pump frequency (locked state).

Fortunately, the collaboration of these nonlinearities provides a route to self-adaptively eliminate the δ and merge these combs into a single broader microcomb. The mechanism is organized as follows: (1) Anomalous dispersion in the microresonator enhances FWM, creating overlapping spectral lines between the primary and Raman EO combs. In the overlap region, there are multiple comb lines in a single mode, with frequencies differed by δ; (2) The Kerr nonlinearity transfers this frequency difference δ to the Raman gain region, producing a sideband around the Raman laser. The frequency of this sideband differs to the pump laser by integer multiples of the microwave frequency; (3) This sideband competes in the Raman scattering process, potentially serving as a seed for frequency locking. Proper tuning of microwave frequency and pump power can reduce δ, leading to the frequency synchronization of the Raman EO comb lines with the primary EO comb and the alignment of the f_ceo, as shown in the inset of Fig. 1b. This mechanism exhibits partial similarities to laser injection locking¹⁸. However, unlike external locking approaches, the frequency stabilization here originates from coherent nonlinear interactions within a single device³⁷, mediated by the coupling between the broadened comb spectrum and the Raman gain profile, which collectively enforce phase locking.

Experimental setup and microcomb characterization

The experimental setup is shown in Fig. 2a. The broadband REO microcomb is experimentally demonstrated using an x-cut thin film lithium niobate (TFLN) photonic chip, as shown in Fig. 2b. Figure 2c presents the microscope image of the racetrack microresonator with the FSR of about 26.03 GHz. Detailed information about the device fabrication can be found in the Methods section. The pump laser (Toptica CTL 1550) is amplified and polarization-controlled to match the quasi-TE mode of the on-chip waveguide, and then coupled into the waveguide supporting the fundamental TE mode through a lensed fiber. The transmitted light is collected by another lensed fiber and sent into subsequent characterization instruments. The coupling loss is approximately 6 dB per facet. An RF signal, which matches the FSR of the microresonator, is amplified and applied to drive the on-chip electrode. Figure 2d shows the pumped mode resonates around the wavelength of 1571.04 nm with a loaded Q-factor of 1.1 × 10⁶. The dispersion of the microresonator around 1571 nm, shown in Fig. 2e, is analyzed using the integrated dispersion formula ????=12?2?2+16?3?3+?, yielding D₂/2π = 3.92 kHz indicating anomalous and flat dispersion, which is crucial for expanding the spectral width and locking the microcomb.

Fig. 2: Experimental setup and device characterization.

a Schematic of the experimental setup. EDFA: erdium doped fiber amplifer; FPC: fiber polarization controller; OSA: optical spectrum analyzer; DWDM: dense wavelength division multiplexing; NF: notch filter; ESA: electrical spectrum analyzer. b Photograph of the thin film lithium niobate (TFLN) photonic chip. c The microscope image of the device consisting of a LN racetrack microresonator and electrodes. d The transmission spectrum of a typical fundamental TE mode with a fitted loaded Q-factors of 1.1 × 10⁶. e Red dots are the dispersion of the microresonator around 1571 nm measured by the Mach–Zehnder interferometer (MZI). Blue curve is the fitted curve, with the fitted value of D₂/2π being 3.92 kHz. f The optical spectra of the hybrid microcomb under different on-chip optical pump power. The driven RF power is 30 dBm. The intensity of the yellow curve for 80mW (Unlocked) is vertically offset by  −5 dB to mitigate overlap with other curves. The inset on the upper left corner shows a flat spectrum between 1520 nm and 1540 nm. g The RF spectra of the output comb field. The disappear of the beating frequency indicates the whole output field being in a phase-locked state.

Figure 2f displays the optical spectra of microcombs under different pump conditions. The comb initiation scheme is the same as the convenitional EO combs (see Supplementary Information). At a low on-chip optical pump power of 2 mW, the EO effect predominates, producing a typical EO comb profile with the spectral width of 80 nm with the modulation frequency around 26.03 GHz. When the pump power exceeds the Raman laser threshold, two additional sets of EO combs emerge on both sides of the primary comb due to the combined effects of Raman scattering and FWM, as shown by the yellow curve in Fig. 2f. These comb structures are referred to as the Stokes comb and the anti-Stokes comb. The spectral width of the hybrid comb becomes significantly wider than that of the pure EO comb at low pump power. After filtering out the pump laser, the output comb field is injected to the photodetector (PD) and analyzed by the electrical spectrum analyzer (ESA). A beat signal in the upper panel of the Fig. 2g, indicates a frequency difference δ of about 48 MHz between the primay EO comb line and the Raman EO comb line within the same resonance mode, implying distinct frequency offsets f_ceo and preventing treatment as a single phase-locked comb. This beating frequency depends on the pump and microwave driving conditions as well as environment noise. Through adjusting the microwave modulation frequency to reduce the δ, an abrupt disappearance of the beat signal reveals that these combs is merged into a single one (lower panel of the Fig. 2g). The locked REO microcomb shows a broad spectral width and obvious flatness along the wings (blue curve in Fig. 2f), which is consistent with numerical simulations in Supplementary Information. With further increaseing optical pump power to 180 mW, more significant Kerr effect can expand the spectral width further to over 300 nm. It is worth noting that the sideband slope of the REO microcomb in the region around the pump is comparable to that of a conventional EO comb (about 1 dB/nm), while the sideband slope in the flattest region (inset of Fig. 2f) is only about 0.1 dB/nm, which is an order of magnitude smaller.

Locking behavior of the microcomb

To characterize the self-locked state of the REO microcomb and exclude any accidental alignment of the primary and Raman EO combs, we experimentally investigate the locking dynamics by varying the microwave frequency under fixed optical pump power (150 mW) and modulation power (30 dBm). The microwave modulation frequency starts at 26.03 GHz. As the modulation frequency is increased to better match the FSRs near the pumped mode, the spectral width of the entire microcomb broadens progressively, as illustrated in Fig. 3a. Meanwhile, the intensity peak of the comb lines gradually shifts from longer to shorter wavelengths (from Stokes to anti-Stokes comb). This shift is attributed to the anomalous dispersion of the microresonator, where the FSR increases from longer to shorter wavelengths. Figure 3b provides a detailed view of how the spectral intensity distribution varies with the modulation frequency.

Fig. 3: Locking behavior of the REO microcomb.

a, b The evolution of the optical spectra when scanning the microwave modulation frequencies. The green line shows the position of the probe laser. c, d Measured beat note signals between comb lines near the green line and a probe laser at 1541 nm for different microwave frequencies.The corresponding beat note signals captured by an ESA, showcasing three distinct type of spectra, representing self-locked and unlocked states of the hybrid microcomb.

Limited by the resolution of the OSA, the details around the overlap region are challenging to discern. Therefore, a probe laser around 1541 nm is employed to detect the comb lines in the overlap region. The probed comb lines near 1541 nm are extracted by filtering other comb lines using dense wavelength division multiplexing (DWDM). The corresponding beat signals are displayed in Fig. 3c, d. There are two distinct groups of beat signals in the RF spectra. The lower one, corresponding to the frequency difference δ between comb lines, gradually decreases and disappears with increasing the modulation frequency, and then reappears after hundreds of kHz, as indicated by the blue dashed line in Fig. 3d. The disappeared region corresponds to the locking range, which can also be confirmed by the beat signal of the probe laser with the comb lines in the nearest cavity mode, i.e., another group of beat signals at higher frequency. The central peak of this group of beat signals, marked in red dashed line in Fig. 3d, represents the comb line from the primary comb, and the slope of this red dashed line represents the mode index of the probed mode relative to the pumped mode. The peak spacing equals to the offset difference δ, which scales linearly with the microwave frequency. When the modulation frequency is outside the locking range, both the primary and Raman comb lines coexist within the cavity mode, along with their FWM generated sidebands. Otherwise, there is only one single comb line within the cavity mode. The sudden shift of the beat signals and its dependence on the microwave frequency are clear indicators of the self-locking process. The width of the corresponding locking range can reach 520 kHz, indicating the robustness of the self-locked microcomb against small variations in the pump and microwave fields as well as environment fluctuations. The self-locked microcomb exhibits a flat distribution of comb lines at its wings, which is attributed to the effect of Raman gain and Kerr effect. Furthermore, the self-locking mechanism also suppresses the linewidth and frequency jitter of the Raman comb, ensuring its alignment with the primary comb (see Supplementary Information), as widely seen in injection locking experiments¹⁸. In this way, the self-locking mechanism transforms the Raman scattering into a crucial nonlinear gain mechanism for broadband EO comb generation.

Indeed, the locking range of the microcomb offers a precise tuning method for fine-tuning f_rep across several hundred kHz. To realize extensive tuning of f_rep, it is necessary to demonstrate that the microcomb can remain locked under various microwave frequencies that are significantly different. By adjusting the pump laser frequency, the locking range can be changed, as depicted in Fig. 4a. It is observed that the locking range gradually shifts to lower frequency as the pump laser frequency is decreased, due to the diminishing frequency difference between the pump laser and the Raman laser. When the pump laser frequency is excessively high, the detuning between the pump laser and the cavity mode becomes too large, resulting in the decrease of intracavity power. Consequently, the locking range narrows until phase locking becomes unsustainable. Conversely, when the pump laser frequency is too low, the pump laser shifts to the red-detuned side of the cavity mode, causing it to exit the cavity mode. The upper and lower boundaries at different laser frequencies define a locking region, within which the locking range exceeds 500 kHz for most laser frequencies, with the largest range reaching 550 kHz.

Fig. 4: Variation of locking range at different modulation frequencies and powers.

a Measured beat note signal by a probe laser at 1541 nm for different pump laser frequencies. The comb is self-locked in the shaded region between the two lines. The locking range shift with laser frequency shift, demonstrating the self-locked comb works for various f_rep. b Relationship between the two boundaries of the locking range and the modulatioon power. c Relationship between the microwave amplitude and the locking range, which is well fitted by a linear function.

In addition to the optical tuning method, the microwave energy can heat the chip, thus changing the FSR of the microresonator via the thermal effect. This alteration impacts the f_rep of the locked state. Figure 4b illustrates how the two boundaries of the locking range shift when the microwave modulation power is adjusted from 21 dBm to 30 dBm, showing that the the f_rep of the locked state can vary within a range of 8 MHz. From the data in Fig. 4b, we also analyzed the relationship between the width of the locking range and the modulation power. Since the comb line amplitude is proportional to the microwave amplitude of the EO modulation²², and according to the Adler equation for injection locking³⁸, the locking range is proportional to the injection laser amplitude. Therefore, the locking range is proportional to the microwave amplitude, meaning a larger microwave amplitude results in a wider locking range. As depicted in the Fig. 4c, the locking range exhibits a positive relationship with the square root of microwave power (i.e., microwave amplitude), fitting well to a linear function.