Optical microresonators [1], owing to their long photon storage time in small spatial volumes, enable strong enhancement of optical interactions, especially in the generation of soliton microcombs [2,3]. Specifically, the need to enhance desired nonlinear processes has motivated the development of different material platforms in improving the Q factors of such microresonators. In recent years, there has been progress made in many chip-based material platforms, such as silica [4–7], silicon nitride [8–12], lithium niobate [13,14], and others [15–22]. Among them, the chip-based silica microresonators in the forms of microtoroid, microsphere, as well as microdisk exhibit the highest Q factors [4–7]. Compared to the microtoroid and microsphere resonators, the silica microdisk resonators are more widely used for Brillouin lasers [5] and Kerr soliton microcombs [23] owing to the controllabilities of the free spectral range (FSR) and dispersion in such structures with the diameter, wedge angle, as well as thickness [23]. Also, due to the low material dispersion and ultrahigh Q factor, Kerr soliton microcombs with a repetition rate as low as 1.86 GHz have been achieved using a 35-mm-diameter microdisk resonator [24]. However, due to the fabrication limitations, such ultrahigh-Q silica microdisk resonators have only been obtained by using a chemical wet-etching technique [5,7]. In that work, extended hydrofluoric acid (HF) wet etching was performed to remove the “foot” region of the silica layer induced by the wet etching, which makes the diameter of the fabricated microresonator much smaller than the initial design [5].

In a previous work, a dry-etched silica microdisk resonator [25] with an optical Q factor of 2.4×107 was fabricated and used for visible Kerr microcomb generation by controlling the dispersion with the wedge angle [26]. Very recently, full integration of silica-based optical resonators with a waveguide using one material platform has been presented [27]. However, silica optical microresonators based on dry etching have never achieved the same ultrahigh-Q (Q>100 million) values as demonstrated by those that are chemically etched.

In this work, by optimizing the fabrication process, we have obtained ultrahigh-Q silica microdisk resonators on a silicon chip by employing an inductively coupled plasma (ICP) dry-etching technique. The intrinsic optical Q factor as high as 194 million is demonstrated by fabricating a 1-mm-diameter silica microdisk resonator with a thickness of 4 μm. To the best of our knowledge, this is the first report of chip-based ultrahigh-Q silica microresonators fabricated by dry etching.

Our standard fabrication process is similar to that shown in our previous work [25,26,28]. Compared with the HF wet etching [5], the dry-etching process usually produces larger sidewall roughness and increases the difficulty in removing photoresist, which significantly affects the Q factors of the microresonators. In this work, by improving the fabrication process such as reducing the sidewall roughness and adding a silicon layer to make the residual photoresist on top surface of the silica microdisk be easily removed, we have boosted the dry-etched silica microresonators into the ultrahigh-Q region.

So far, the optical Q factor of the chip-based silica microdisk resonator is limited by the scattering losses induced by the surface roughness of the microresonator. For the dry-etched microdisk resonators, the sidewall roughness predominantly results from two pattern-transfer steps: lithography-induced roughness and dry-etch-induced roughness. During photolithography, we optimized the hard-bake temperature and duration time of the patterned photoresist to obtain a smooth sidewall of the mask. For the dry-etching process, we employed an etching system based on neutral loop discharge (NLD) plasmas (ULVAC, NLD 570) [29]. These NLD plasmas tend to have high plasma density with a relatively low electron temperature [30], which lowers the substrate temperature and thus reduces the difficulty in subsequently removing the photoresist. This is of great importance, particularly for the long-time plasma etching in this work. During the ICP etching, CF4 is used as the main etch gas with a small amount of SF6. In our experiment, due to the limited selectivity between the photoresist and the silica layer, the etching recipe such as the flow rate of the SF6 (40.0 sccm) and CF4 (3.0 sccm) (sccm, standard cubic centimeters per minute) as well as the ICP power (400 W) is specifically adjusted to make a balance between the etching rate and the selectivity. Figures 1(a) and 1(b) show the optical and scanning electron microcopy images of the dry-etched silica microdisk with a wedge angle of around 35°.

To characterize the silica microdisk resonator, a tapered optical fiber is used to excite the optical modes of the microresonator via the evanescent wave [31]. We obtain the transmission spectrum by scanning an external semiconductor laser at the 1550 nm wavelength across the resonance frequency. By adjusting the distance between the taper fiber and the resonator, the resonator can be undercoupled to avoid the thermo-optic effect [32], which makes the measured value close to the intrinsic optical Q factor. Further adjusting the distance between the taper fiber and the resonator, the critical coupling and overcoupling regimes can also be accessed. During the measurement, the optical modes are calibrated using a fiber Mach–Zehnder interferometer.

To further confirm the measured Q factors of the microresonators, accurate measurement using the ringing method [35,36] was also performed to minimize the influence of thermal effects and laser frequency noise. The laser excites the optical mode, which results in the interference between the delayed light from the resonator and the direct transmitted pump laser. Figure 4(b) shows the ringing curve of the optical mode in Fig. 4(a). The theoretical fit shows an intrinsic Q factor of 196.2 million, which matches quite well with the Lorentzian fit in Fig. 4(a). Particular attention should be paid to the fact that the Q factor is even higher than that for a chemically etched analogue with the same diameter and thickness [5].

To evaluate the surface roughness of the fabricated microdisk resonators, we have performed atomic force microscope (AFM) measurement. As shown in Figs. 1(c) and 1(d), the root-mean-squared (RMS) roughness is 0.72 nm for the top surface and 0.67 nm for the sidewall. The relatively larger roughness of the top surface may be attributed to the residual photoresist or silicon. The roughness of the sidewall is similar to the measured value of the wet-etched silica microdisk resonator [5]. Also, the autocorrelation lengths for the top surface and the sidewall are around 20 nm and 30 nm, respectively, which are much smaller than that obtained from wet etching [5]. In the future, the optical Q factor can be further increased by optimizing the fabrication process such as the removal of the photoresist on the top surface of the microresonator.

In conclusion, ultrahigh-Q (1.94×108) silica microdisk resonators on a silicon chip have been successfully demonstrated by dry etching. Such dry-etched devices will be useful for Brillouin lasers with precise FSR control. The fabrication process will also be applicable to multiwedge resonators [37] for broadband dispersion engineering by only controlling the dry-etching process with photolithography and ultralow-loss delay line on chip [38]. Our method also paves the way for the on-chip integration of ultrahigh-Q microresonators with the bus waveguides within silica material.

Acknowledgment. The authors thank Jiayu Zhang, Hongyu Yang, and Kuo Yang for assistance with AFM measurement.