Kerr frequency comb (microcomb) generation based on parametric four-wave mixing (FWM) in monolithic microresonators with high quality (Q) factors [1], emerging as an alternative scheme, has attracted considerable interest due to miniaturization, chip-scale integration, repetition rate in the microwave and terahertz regime, and spectral coverage from visible to mid-infrared [2–5]. The recent demonstration of dissipative Kerr solitons (DKSs) [6,7], with a double balance between Kerr nonlinearity and dispersion as well as loss and parametric gain in the optical resonator, has provided a route to a fully coherent microcomb with a smooth spectral envelope and broadened width due to soliton-induced Cherenkov radiation [8,9]. The soliton microcomb has ubiquitous commercial potential for applications in dual-comb spectroscopy [10], coherent communications [11], frequency synthesizer [12], long-distance measurements [13], and quantum key distribution [14].

However, accessing soliton states in microresonators is still challenging since the DKS generation requires keeping the pump in an effective red-detuned regime, where the thermo-optic instability in the microresonator causes complex behavior [6]. Solitons have been demonstrated in MgF2 [6], silica [15], Si3N4 [8,16–21], and AlN-based [22] systems. Ultrahigh nonlinearity AlGaAs microresonators with low loss were developed for the stable generation of DKSs at a cryogenic temperature [23]. Sundry techniques were adopted to demonstrate the above-mentioned solitons, including different frequency tuning schemes such as rapid forward or/and backward pump laser scans [17–19], pump modulation [21,22], and fast control of the resonator temperature [16], as well as power kicking [8]. However, the complicated control or extra equipment needed will increase the cost and limit the microcomb applications. Recently, LiNbO3 has been demonstrated for accessing the perfect soliton crystal or single DKS, on-demand, due to its intrinsic photorefractive effect [24–26].

An alternative method to manipulate the power coupled into the resonance is using an auxiliary laser to pump another cavity mode, which can suppress thermal dragging dynamics during soliton formation and improve the soliton stability and access window [27–31]. However, generally for Si3N4 and AlN microresonators, both pump and auxiliary lasers need to be amplified to high powers with two fiber amplifiers. Moreover, a nearby auxiliary mode with cross-polarization has been proved useful to stabilize the thermal dynamics in the Si3N4 resonator, enabling stable access to an octave-spanning single DKS with a soliton step width of ∼12pm [18].

AlN, with a similar optical refractive index to Si3N4 and a large transparency window, is an ideal crystalline material for integrated photonics applications. AlN microresonators have been used to realize broadband microcombs successfully [32–34] and achieve the single Kerr soliton using the scheme of dual-pumping [30] and single-sideband modulation [22]. In addition, the strong Pockels nonlinear effect was exploited for harmonics, bi-chromatic microcomb, and Pockels quadratic soliton generation [35–37]. Nevertheless, stably accessing the Kerr soliton state with a single pump under direct tuning has not been reported for this material.

In this work, we reveal that an adjacent mode near (e.g.,  35 pm longer than) the pump mode can help to mitigate the cavity thermo-optical effects, thereby allowing stable access to DKSs in an AlN microresonator via slow laser tuning. By pumping the fundamental TE (TE00) mode under appropriate power, the long-lived (10.4-GHz-wide soliton step) soliton microcombs beyond an octave-spanning spectral bandwidth can be deterministically attained at a pump tuning speed of 1 nm/s. The broad soliton step allows us to directly produce the DKS with high probability by simply switching the pump wavelength from an off- to an on-resonance state. A measured 30 fs pulse trace confirms the high coherence of the soliton microcomb.

Figure 1 illustrates the device characterization and the approach taken in this work. Using standard photolithography and inductively coupled plasma etching processes [39], we fabricated the devices in a quarter of a 2 inch wafer, as shown in Fig. 1(b), that consists of a 1.2-μm-thick epitaxial single-crystal AlN film and a sapphire substrate [40]. We employed a microring resonator with a radius of 60 μm and a cross section of 2.29μm×1.2μm (fully etched). The integrated bus waveguide, 500 nm apart from the resonator, is tapered from 0.91 to ∼4μm at both ends to reduce the fiber-waveguide coupling loss, which is measured to be 3.2 dB per facet. Figure 1(d) shows the measured transmission, featuring two sets of TE polarization modes, TE00 and TE10, which are easily identified due to the different free spectral ranges (FSRs), ∼374 and ∼366GHz. The resonance wavelengths of the TE00 and TE10 modes are 1550.582 and 1550.617 nm, respectively, with a separation of only 35 pm. The significantly changed extinction ratio near 1553.6 nm indicates a weak mode coupling between the two transverse modes. As fitted in Fig. 1(e), the loaded Q(QL) of the two nearby fundamental and first-order modes is 4.8×105 and 2.8×105, respectively. The QL and intrinsic Q(Qi) of the TE00 mode resonant at 1562.675 nm, which is free from mode interactions, are extracted to be 6.1×105 and 1.4×106 considering a slightly over-coupling condition. The target geometry was chosen meticulously to ensure that the TE00 resonance frequency is several GHz higher than that of TE10 mode. For the same resonator geometry, we have observed similar behavior (two adjacent TE modes in C-band) in multiple devices, indicating that the design is easy to replicate in fabrication.

Besides the high Q, another basic requirement to yield broadband soliton microcomb is near-zero anomalous dispersion, which was also accommodated during design. The integrated dispersion Dint [6] was calculated and plotted in Fig. 1(f), where the center frequencies of the TE00 and TE10 modes are 193.292 and 193.289 THz, respectively. Their D2/2π are extracted to be 4.8 and 36.8 MHz. Both modes are in anomalous dispersion, while the TE00 mode has near-zero dispersion and the Dint=0 is obtained around 255.4 THz (∼1170nm). As the schematics illustrated in Figs. 1(g) and 1(h), the TE10 mode can compensate for the intracavity power change and mitigate the thermal requirements to access stable soliton states.

As the experimental setup depicted in Fig. 2, the light source (Santec TSL-710) is amplified by an erbium-doped fiber amplifier (EDFA) and injected into the waveguide through a lensed fiber. The output including transmitted pump light and generated comb lines is divided into two branches, one of which is connected with two optical spectrum analyzers (OSAs) for recording the spectra. The other branch is injected into a fiber Bragg grating (FBG) filter (bandwidth 0.4 nm) to differentiate the pump transmission (bandpass, BP) and comb lines (band rejection, BR) separately. The BR part is used for comb coherence characterization with the electrical spectrum analyzer (ESA) and autocorrelator (AC) after a photodiode (PD) and a fiber polarization controller (FPC), respectively. The BP part is sent to an oscilloscope or a powermeter for the pump transmission measurement. The powermeter with two synchronized channels enables the transmission measurement of all comb lines and the filtered pump at the same time.

Figure 3(a) shows the measured all and pump transmissions at 350 mW on-chip power and a pump tuning speed of 1 nm/s, as well as the calculated difference between them. A pronounced 63-pm-wide step-like structure characteristic of soliton formation is formed. The low Q factor of TE10 mode results in a high extinction ratio in the pump transmission, implying a strong power absorption in the cavity. By sweeping the laser wavelength to different detuning positions, we record the comb spectra and plot the evolution map in Fig. 3(b). One can observe that the spectrum changes from primary comb to modulation instability [MI, dashed line (i)] comb and solitons [dashed lines (ii) and (iii)]. The solitons exist stably when the laser wavelength ends between ∼1550.78 and ∼1550.84nm, consistent with the soliton regime marked in Fig. 3(a).

The top stack of Fig. 3(c) is the generated MI comb spectrum (i), which ranges from 140 to 260 THz with a dispersive wave (DW) bump around 255 THz. Typical spectra of solitons 1 [region (ii) of Fig. 3(c)] and 2 [region (iii) of Fig. 3(c)] are presented in Fig. 3(c) subsequently, which cover an octave-spanning range from 130 to 273 THz (1100–2300 nm). The measured soliton spectra are significantly extended toward shorter wavelengths due to the emission of the DW via soliton-induced Cherenkov radiation at ∼264THz (∼1136nm). Upon further tuning the pump toward the red-sided regime before dropping out of the cavity resonance, we can only observe a few weak sidebands (state iv) generated from the TE10 mode due to the lower Q factor and larger dispersion. Therefore, our geometry design is ideal for generating the DKSs without any concern about the nonlinear competition between the two transverse modes.

We also observed the shiny green light emission during the soliton formation [see insets in Fig. 3(c)], corresponding to the third-harmonic generation (THG) at ∼580THz. We believe the other weak spectrum bump occurrence around 290 THz arises from the THG and the optical parametric oscillation [37]. The soliton centroids are 5.2 and 7.4 THz lower than the pump because of the Raman self-frequency shift effect and the DW recoil [41–43]. By fitting the spectral with sech2 function, we can find the full widths at half-maximum (FWHMs) of solitons 1 and 2 are 15.2 and 12.3 THz, corresponding to 40 and 33 modes, respectively. Their minimum transform-limited pulse width τ is estimated to be 21 and 26 fs, using 0.315 × the pulse FWHM where the pulse shape is sech2 [6].

The transition from the MI comb to the soliton comb is verified by the drastic reduction of low-frequency intensity noise [see Fig. 3(d)]. To further confirm the single DKS states, the temporal characteristics were also carried out through the second-harmonic generation (SHG)-based autocorrelation measurement. Figure 3(e) shows the measured soliton autocorrelation trace, which has a poor signal-to-noise ratio (SNR) since the soliton comb power is at the limit that can be detected by the AC. However, the pulses are clearly separated by ∼2.66ps, inversely proportional to the FSR of ∼374GHz. Within a narrow measurement range, an extremely narrow pulse [see Fig. 3(f)] with ∼30fs width and improved SNR can be obtained based on the sech2 fitting. The measured pulse width is larger than that estimated from the spectrum (21 fs) due to the limited responsivity beyond ∼1600nm of the AC.

To explore the dependence of soliton existence range on the pump power, we characterized the all and pump transmissions, under various powers, and we plot their difference in Fig. 4. A striking soliton step with 10.4 GHz (83 pm) width is observed at 335 mW, which is the widest soliton existence window as far as we know. When further increasing the power, the soliton step declines nearly linearly (∼−0.1GHz/mW) and disappears at 440 mW due to the residual thermal effects. It should be noted that the soliton existence range can be further increased to ∼90pm when we use a simple, slow laser piezo tuning (several GHz/s, by hand). The soliton states are sustained for 2 h at an on-chip power of ∼410mW.

One highlight of our system is the direct creation of the DKS state by simply red tuning the pump light with one step, thereby eliminating the requirement for complex pump tuning or power kicking techniques. As the schematic depicted in Fig. 5(a), the laser wavelength was switched 50 times between an off- (λstart) and an on-resonance (λstop) state. The duration of each wavelength is 2 s.

The possibility of accessible soliton versus the λstart and λstop was recorded and depicted in Fig. 5(b), as confirmed by monitoring the microcomb spectrum and pump transmission with both OSA and real-time oscilloscope. We can directly access the soliton with high probability, over 80% when the λstop is 1550.720 or 1550.730 nm and the step values locate between 0.09 and 0.12 nm. Compared with the soliton existence regime (1550.781–1550.844 nm) marked in Fig. 3(a), we note that tuning the pump wavelength with step mode will reduce the thermal effects within the microresonator, thus decreasing the detuning required for soliton generation. Figure 5(c) plot the measured pump transmission when the λstart and λstop are 1550.620 and 1550.720 nm, 1550.630 and 1550.730 nm, and 1550.640 and 1550.740 nm, respectively, with a fixed step of 0.1 nm. The successful and failed (accessible and inaccessible to the soliton states) attempts are marked by green rectangles and red circles. The success rate is as high as 94%, 92%, and even 100%, respectively.

As investigated in Refs. [18,20], with an auxiliary mode, the intracavity power is modified such that a part of the soliton regime becomes thermally stable. Therefore, the soliton behavior is affected by the mode separation and a thermally induced cavity redshift, which is related to the pump power, the mode Q factors of the two adjacent modes, and thermal conductance of the microresonator. The influences of these parameters on the soliton dynamics in our single-crystalline AlN microresonators need further study in both numerical and experimental investigations.

To increase the overall yield for ensuring the pump and auxiliary modes are close, we will design and fabricate microresonators with minor changes around the target parameters by taking the fabrication variation into account. A patterned metal contact can be deposited near the devices for attempting to fine-tune the mode separation and coupling. This provides another dimension, together with optimized pump tuning speed and power, to control the thermal effect for accessing soliton states more easily. We can estimate that the pump power required for soliton generation can be decreased to tens of mW assuming the intrinsic Q factor was increased to 10 million, which is needed to directly generate the soliton by pumping the passive microresonator with a laser diode chip. Moreover, instead of the mixed polarization of pump and auxiliary modes [18], the same polarization scheme eliminates the need for an in-line polarization controller for realizing the vision of hybrid integration. By designing different radii and resonator widths, one can also expect the soliton microcombs with different repetition rates in different materials by pumping at other wavelength windows such as 1300 and 1064 nm.

A simple route is proposed and demonstrated to stably achieve the octave-spanning DKS in an AlN microresonator, in which a nearby auxiliary mode is slightly red-detuned from the pump mode. The auxiliary resonance can compensate for the intracavity power change and balance the thermal effects in the resonator, thus producing and broadening the soliton step significantly. A comparison of different nonlinear material platforms for single soliton generation is shown in Table 1. In this work, an octave-spanning soliton microcomb ranging from 130 to 273 THz, with a repetition rate of ∼374GHz, is first demonstrated in AlN with a single pump under direct wavelength tuning. The spectral bandwidth is at a similar level to that realized with the state-of-the-art Si3N4 technology, while more comb lines are expected in our AlN microresonators due to the smaller repetition rate. Moreover, the demonstrated 10.4 GHz-wide single soliton step is, to the best of our knowledge, the widest soliton accessing window so far, which is enhanced by ∼7 times compared with the soliton step demonstrated in Si3N4 microresonator [18]. Benefiting from this, we can directly reach the soliton states with high probability by controlling the pump wavelength with step mode, which will relax the strict requirements in controlling the pump or the cavity temperature. The free-noise at low frequency and 30 fs soliton pulse trace confirm the high coherence of the generated solitons. We believe such a deterministic and simplified approach for accessing soliton states is critical for chip-based self-referenced microcombs and their potential applications outside the laboratory.Table 1.

Comparison of Single Kerr Solitons Generated with Various Chip-Integrated Microresonators

Acknowledgment. The authors would like to thank Prof. Liam Barry for the technical assistance.