Optical frequency combs are coherent light sources with a large number of evenly spaced frequency lines, which can be exploited in many scientific applications ^[1–4]. Compared with traditional optical frequency combs based on large-size mode-locked lasers that cannot be integrated on chip ^[1], the up-to-date Kerr-type microcombs ^[5,6] based on cascaded four-wave-mixing processes possess many significant advantages. These include a compact size, low optical loss, and feasibility of on-chip integration ^[5–8]. Since their invention, this field has garnered tremendous attention and enjoyed rapid progress. Kerr-type microcombs have been realized in many platforms ^[9–22], some of which are already integrated on chip. After successfully generating optical frequency combs in different material systems, many studies have been geared up toward potential applications, such as arbitrary waveform generation ^[19], coherent optical communication ^[23], optical synthesizer ^[24], ultrafast optical ranging ^[25,26], and astronomical spectrum calibration ^[27,28]. So far, most of the Kerr frequency microcombs and their corresponding applications have been realized in the wavelength near 1550 nm ^[9–22]. The generation of the visible microcombs that are desirable for optical atomic clocks and optical coherence tomography is still challenging due to the strong normal dispersion induced by the material and the lower optical quality factor of the microcavity in the visible wavelength region.

Recently, several approaches have been employed to generate visible Kerr combs ^[29–38], such as using second-harmonic/third-harmonic generation ^[29–33] to translate a near-infrared comb to the visible range or designing a peculiar geometry possessing large anomalous dispersion to overcome the inherent normal dispersion of the material ^[34–38]. However, all those methods suffer from certain disadvantages, such as a narrow visible bandwidth ^[34,35], large pump power needed ^{[31,33,37,38]}, limited conversion efficiency ^[29–32], non-integrated platforms ^[34,35], limited visible comb line power ^[37], and rigidities in fabrication ^[36].

In this work, we present a new method of visible comb generation. By employing a large-wedge-angle silica microdisk resonator with a thickness of 1 μm, the normal dispersion of silica material in the 780 nm regime can be completely compensated by its geometric dispersion. As a result, we obtain a resonator that produces anomalous dispersion in the visible wavelength regime without using mode hybridization, so that a visible comb can be generated directly without using any frequency conversion technique. As is shown in the next section, the microdisk possesses anomalous dispersion over the thicknesses of 800–1200 nm, which makes the fabrication process less stringent than the method by Lee et al. ^[36]. Furthermore, the visible comb obtained in our experiment covers the wavelengths ranging from 700 nm to 897 nm. To the best of our knowledge, this is the broadest wavelength range among the experimentally demonstrated visible Kerr combs.

The experiment conducted in this study uses the silica microdisk resonator shown in Fig. 1. The fabrication process is similar to that described in our previous work ^[39,40]. The disk is 1 μm thick with a diameter of 80 μm. The most intriguing feature of our disk is the large wedge angle of 56° (the definition of the wedge angle is illustrated in Fig. 2(a), and it is indicated by angle α), which is usually hard to attain if the silica microdisk is fabricated by wet etching ^[41]. It is this special geometry that allows our disk to overcome the large normal dispersion induced by the silica material and naturally exhibit anomalous dispersion in the 780 nm wavelength regime. In previous studies where buffered hydrofluoric acid (HF) was used to etch a thicker silica disk (∼1.5μm thickness), the wedge angle could not reach this level, and mode hybridization was employed to enable the small-wedge-angle disks to provide anomalous dispersion at ∼780nm ^[36]. However, this produced extra problems in the fabrication process, namely that the disk’s thickness should not deviate by more than 50 nm, as mode hybridization is very sensitive to the thickness of the disk. On the contrary, our disk displays the feature of a large wedge angle, and it naturally exhibits anomalous dispersion in the 780 nm wavelength region, independent of mode hybridization [see Figs. 2(c) and 2(d)].

Before conducting the experiment, we first simulate the dispersion characteristics of our disk using the standard finite-element method (FEM) ^[42]. Figure 2(a) schematically illustrates the mode profile of the fundamental TM10 mode (we define the first mode number according to the maximum intensity in the radial direction, while the second mode number is defined as l-m for m and l as defined in Ref. ^[43]), which we harness as the pump mode in the subsequent experiment inside our cavity. Figure 2(b) shows the dispersion relationship with respect to the wavelength of the cavity used in the experiment. As one can see, the TM10 mode shows the anomalous dispersion characteristics (D>0) at 780 nm. Figure 2(c) presents the dispersion characteristics of the TM10 mode at the 780 nm wavelength for different disk thicknesses and different wedge angles. It is clear that a wedge angle of 56° provides an anomalous dispersion over a large thickness range between 0.8 and 1.2 μm; additionally, the anomalous dispersion is insensitive to the wedge angle. These characteristics greatly reduce the constraints put on the disk thickness and the wedge angle, thus making fabrication significantly easier than the method by Lee et al. ^[36], in which the disk thickness has a tolerance of only 50 nm. Figure 2(d) depicts the effective index at different thicknesses for different mode families at a wedge angle of 56° and a wavelength of 780 nm. The smooth and continuous variation behaviors of the dispersion and effective refractive index with the disk’s thickness in Figs. 2(c) and 2(d) indicate that our disk does not exhibit the characteristics of mode hybridization ^[36].

Figure 2(e) shows the variation of zero dispersion wavelengths with different wedge angles for the silica microdisk resonators that are 1 μm in thickness and with a 80 μm diameter. It is clear that when increasing the wedge angle, the zero dispersion wavelength will shift to a shorter wavelength ^[39]. Figure 2(f) illustrates the dispersion character of the TM10 mode for a 56° wedge angle silica disk with a diameter of 2 mm. The zero dispersion wavelength can be shifted to ∼680nm, which is beneficial for visible comb generation with a shorter wavelength, as well as a lower repetition rate.

We now describe an experiment using the setup shown in Fig. 3(c). First, we test the quality factor (Q) of our cavity by sending a scanning laser into the cavity and measuring the linewidth of the transmitted Lorentz peak, as illustrated in Fig. 3(a). The doublet resonance dip is due to the coupling between the clockwise and counter-clockwise traveling modes induced by surface roughness ^[44]. The microdisk resonator has the intrinsic and loaded Q factors of 1.08×107 and 0.98×107, respectively.

Next, we measure the threshold power of the optical parametric oscillation (OPO). We find the OPO process occurs at the threshold power of 1.62 mW (launched power in fiber). The formula of the OPO threshold power can be written as ^[14,45] where V0 is the effective mode volume, n0 is the refractive index of silica, ω/2π denotes the resonance frequency, n2 represents the nonlinear refractive index of silica, c is the speed of light in vacuum, η is the coupling rate, and Q denotes the loaded Q-factor of the cavity. The measured threshold power matches quite well with the theoretical value. The deviation of linear behavior of the last data point in threshold measurement is due to the energy transfer of the first sideband to the second sideband.

Finally, we pump the disk above its OPO threshold power and gradually increase the pump power, while observing the evolution of the optical spectrum of the generated visible comb without using any extra frequency conversion technique. Figure 4 shows the optical spectra produced by pumping the cavity with launched optical powers of 21.50 mW, 28.50 mW, and 36.50 mW, respectively. The pump wavelength is fixed at 780 nm, where the disk possesses an anomalous dispersion. During the whole experiment the microdisk resonator is over-coupled to the tapered fiber in order to maintain a mechanical stability. As shown in Fig. 4(a), only the primary combs dominate in the spectrum when the pumping power is at a relatively low level ^[46]. By controlling parameters like the frequency detuning of the pump laser or the coupling rate between the fiber taper and the optical microcavity, the position of the primary comb can be altered ^[46]. By further increasing the pump power after the emergence of the primary combs, sub-combs will be generated ^[46], filling the gaps between primary comb lines, as is illustrated in Figs. 4(b) and 4(c). In Fig. 4(c), the visible comb contains as many as 112 optical frequency lines covering the optical wavelengths from 700 nm to 897 nm. We should stress that 700 nm is the shortest wavelength achieved in any experimentally demonstrated visible Kerr combs that do not use a frequency conversion technique, and the wavelength range of our visible comb is also broader than any one reported in previous studies ^[29–38]. Importantly, the comb line at the shortest wavelength is very close to the S01−P03 transition at 698 nm of Sr87 atoms ^[2], and it has potential to be further exploited in precise and accurate, as well as compact, optical clocks.

We have successfully generated a visible Kerr microcomb using our specially designed large-wedge-angle silica microdisk. The geometry of the microdisk enables the appearance of anomalous dispersion to occur naturally in the 780 nm wavelength range without the use of mode hybridization. We can directly produce efficient visible Kerr combs by pumping the cavity at 780 nm without requiring any frequency conversion technique. The larger tolerances in the disk’s thickness and wedge angle greatly reduce the difficulties in the device fabrication, making it more versatile for practical applications. Additionally, our visible microcomb covers the wavelength range from 700 nm to 897 nm, which, to the best of our knowledge, is the broadest optical wavelength range of any experimentally demonstrated visible Kerr combs. Such chip-based visible microcombs will find applications in optical coherence tomography and strontium optical lattice clocks. In the future, we will attempt to pump our microcavity at a shorter wavelength so that the comb lines can be extended to the wavelengths below 698 nm to reach the Sr87 clock transition. Also, the disk wedge angle and other geometrical parameters can be further explored to shift the zero dispersion point to a shorter wavelength and expand the wavelengths of the microcomb in the visible and ultra-violet ranges. By optimizing the fabrication process of the microdisk cavity with a larger diameter, we expect to achieve a visible Kerr-comb to a shorter wavelength with a detectable repetition rate and access it in a stable soliton state ^[47] using the dry-etched silica microcavity.

Acknowledgment. The authors thank Dr. Hong Chang for helpful discussions. We also thank Han Wang and Dr. Xuewei Wu for technical assistance.