On-chip high-Q optical microresonators constitute a pivotal type of photonic device and have been widely used for quantum optics, nonlinear optics, as well as sensing [1–3]. So far, chip-based optical microresonators have been demonstrated in many material platforms and structures [4–20]. Among them, chip-based silica [5,6] and thin-core Si3N4 optical microresonators [9–11] exhibit the highest optical Q-factors, because most of the optical modes in these structures are confined in the silica materials so that the absorption loss can be significantly reduced [21]. To date, such ultra-high-Q optical microresontors have become essential for applications in ultra-narrow linewidth lasers [5,22–25], Kerr soliton microcombs [26,27], and gyroscopes [28,29], whose performances heavily rely on the Q-factor. For example, the fundamental laser linewidth of the Brillouin lasing or optical parametric oscillation threshold is found to be inversely proportional to Q2.

Recently, with the development of the fabrication techniques of the integrated photonics, several microresonator platforms with relatively higher material absorption loss, such as aluminum gallium arsenide (AlGaAs) [18,21] and tantalum pentoxide (Ta2O5) [20,21] at 1550 nm, as well as Si3N4 [30] at visible wavelength, have achieved the optical Q-factors that approach their absorption limits. However, the optical Q-factors of the on-chip microresonators with ultra-low material absorption [5,6,9–11,21] are still limited by the scattering loss introduced during the fabrication process. On the other hand, compared with the standard optical fibers, the optical losses of the on-chip optical resonators or waveguides are still much higher. How to reduce them to the level of optical fibers is an open issue to the community of integrated photonics.

In this work, we report the chip-based silica microtoroid resonators with their record-high optical Q-factors of 3.3 billion and 3.2 billion at the wavelengths of 1560 nm and 1064 nm, respectively. Compared with the highest Q-factor reported in the past [5,6,31], we have achieved an improvement by three times and five times at the wavelengths of 1560 nm and 1064 nm, respectively. By employing cavity-enhanced photothermal spectroscopy measurement [21], we see that these optical Q-factors are near the material-limited Q-factors, i.e., 80.5% at 1560 nm and 60.8% at 1064 nm. If compared with the standard optical fiber [32–34], their associated optical losses are only 38.4 times at 1560 nm and 7.7 times at 1064 nm. To prove the good performance of such microresonators of ultra-high Q-factors but with the relatively large size of millimeter, we implement optical parametric oscillation at a record-low threshold (31.9 μW). Moreover, we realize a single-soliton microcomb with a record-low operation power of 220.2 μW for the known soliton microcombs thus far.

The fabrication process of the microtoroid resonators is based on the optimization of the method in earlier works [4,35]. Figure 1(a) shows the fabrication process of the optical microtoroid resonators. A stepper is used to pattern the photoresist on a thermally grown silicon oxide film with a thickness of 12 μm. The silica microdisk is formed through over-etching in buffered hydrofluoric solution. After a cleaning step using organic solvents to remove the photoresist, the silica microdisk is undercut using xenon difluoride dry etching to ensure a circular silicon pedestal. It is noteworthy that multiple cleaning processes are required to avoid potential contamination of the silica microdisk. A CO2 laser with a Gaussian beam is applied to heat and reflow the silica microdisk. Since the diameter of the silica microdisk is much larger than the beam waist (∼1mm) of the CO2 laser, the position of the microdisk is continuously adjusted by using a three-dimensional linear translation stage to circulate the laser around the periphery of the sample. This circulating step is iterated as the laser power increases. The typical optical power of the CO2 laser at the conclusion of the reflow process is around 40–45 W. Following the reflow process, a 40-h annealing in an oxygen ambience at 1000°C is performed to drive down the water and release the stress induced by the reflow process. It should be noted that, compared to the annealing in a nitrogen ambience, the oxygen ambience is more appropriate to improve the stress stability for silica-based devices on the chip [36]. In our experiment, the maximum Q-factors of the microtoroid resonator at the wavelength of 1560 nm, after the annealing in an oxygen or in a nitrogen ambience, are measured to be 3.3 billion and 1.9 billion, respectively. Figure 1(c) shows a scanning electron microscopy image of the fabricated microtoroid resonator with a diameter of 2.8 mm. In addition, an atomic force microscopy (AFM) measurement is performed to evaluate the surface roughness. As shown in Fig. 1(b), the measured root-mean-squared (RMS) roughness on the surfaces of resonators is found to be ∼0.3nm.

Experimentally, we first measure the optical Q-factors of the microtoroid resonators by characterizing the linewidths of the cavity modes at the wavelength of 1560 nm band. During the linewidth measurement, a tunable narrow linewidth laser is coupled to the microresonator via a tapered fiber [37] to sweep across the resonance of the cavity mode. Meanwhile, a fiber Mach–Zehnder interferometer (MZI) calibrated by an electro-optic modulator is employed for the characterization of resonance linewidth. The left panel of Fig. 2(a) shows the measured transmission spectra of the singlet and doublet cavity modes at the wavelength of ∼1560nm. Then, through the transmission measurement, the corresponding intrinsic Q-factors are thus inferred to be 2.45 billion and 3.37 billion, respectively. A noteworthy phenomenon is that a backscattering-induced mode splitting [38] can be easily observed in our millimeter-sized microtoroid resonators with ultra-high Q-factors. In contrast, this mode splitting is not observable in the wet-etched microdisk resonators of similar size [5,31] due to the relatively lower optical Q-factor. Specifically, optical mode with the splitting rate smaller than 100 kHz is observed in the linewidth measurements. We also count the distribution of splitting rate and further investigate the origin of mode splitting. By comparing the measured and calculated results, we can conclude that the mode splitting is induced by the bulk inhomogeneities rather than by the surface roughness. More details can be found in Appendix A.

The above-mentioned results of Q-factors characterization can be confirmed further by the ringing measurements [39–41]. In the right panel of Fig. 2(a), we show the ringing spectrum that is obtained by sweeping the laser frequency across the resonance at a relatively fast speed. According to the theoretical fitting of the measured ringing spectra [41], the intrinsic optical Q-factors are found to be 2.44 billion for the singlet mode and 3.33 billion for the doublet mode. These values are consistent with the results obtained from the linewidth measurements. To investigate the dependence of the scattering loss on the wavelength, we also perform the measurements at the wavelength of ∼1064nm, and the corresponding results are presented in Fig. 2(b). The inferred intrinsic Q-factors for the singlet and doublet cavity modes in the linewidth measurement are 2.55 billion and 3.25 billion, respectively. Similarly, the ringing measurements yield the intrinsic Q-factors of 2.59 billion and 3.23 billion for the singlet and doublet cavity modes, respectively, which are in good agreement with the linewidth measurements.

To the best of our knowledge, these results present the record values for the on-chip optical microresonators at both the 1560 nm and 1064 nm wavelength bands, which are three times and five times higher than the previous works [5,6,31] at the wavelengths of 1560 nm and 1064 nm, respectively. The enhancement of the optical Q-factor is attributed to the reduced surface roughness of our microtoroid resonators. In fact, their equivalent propagation losses reach 7.7 dB/km at 1560 nm and 11.5 dB/km at 1064 nm, being only 38.4 times and 7.7 times larger than those of the standard optical fibers [32–34], respectively, at these bands of wavelength.

The unexpectedly high similarity of the measured intrinsic Q-factors at 1560 nm and 1064 nm as well as the ultra-smooth surface of the microtoroid resonators indicates that optical Q-factors of the microresonators are not determined by the scattering loss. To further explore the loss mechanisms of the microresonators, we employ the cavity-enhanced photothermal spectroscopy (CEPS) [21,42,43] to find the material absorption-limited Q-factor defined as Qabs=ω0κa,(1)where κa is the energy loss rate for the material. This method quantifies the correlation between the intracavity energy density (ρ) and resonance shift (δω0) induced by the photothermal effect (α) and Kerr effect (g) through the linear relation δω0∼(α+g)ρ. Since the Kerr nonlinear effect is much smaller than the photothermal effect (g≪α) for the suspended silica resonators, it is reasonable to neglect the resonance shift caused by the Kerr effect [21] (α+g≈α). Then, the energy loss rate (κa) can be calculated with the following equation: κa=α(δT/Pabs)(−δω0/δT)Veff.(2)Here, δT is the change of the temperature in the mircoresonator, Pabs denotes the absorbed optical power, and Veff represents the mode volume of the optical mode. These involved parameters are determined as follows. The value of Veff can be calculated by a finite-element method (FEM) modeling of the optical mode in the resonator. δT/Pabs is determined by incorporating the simulated optical mode field as a heat source in the finite-element simulation and the resonance tuning coefficient δω0/δT is measured by using a thermoelectric cooler (TEC) with a proportional-integral-differential (PID) temperature controller to precisely adjust the temperature of the microresonator. More details of the parameters used for the simulations can be found in a previous work [21].

Figure 3(a) shows the typical transmission spectra with thermal broadening [44] at the wavelength of 1560 nm. By fitting the thermal triangle line shape [43], the normalized resonance shift δω0/ω0 versus the intracavity energy density ρ is determined and presented in Fig. 3(c), so that the photothermal coefficient α can be inferred from the previously mentioned linear dependence of these two variables. We need to point out that, during the measurement, only singlet cavity modes are employed to simplify the calculation of the intracavity energy and make our measurement more accurate. Figure 3(b) shows the results of the measured δf0/δT at both 1560 nm and 1064 nm bands, which indicates the corresponding thermo-optic coefficients of 1.097×10−5K−1 and 1.141×10−5K−1, respectively. These results align well with the previously reported first order thermo-optic coefficient [45].

With all parameters we have determined, the obtained absorption-limited Q-factors (Qabs) at both 1560 nm and 1064 nm are presented in Fig. 3(d). The measurements conducted on several cavity modes with varying intrinsic Q0 demonstrate a consistent Qabs ranging from 3.5 billion to 4.2 billion at 1560 nm, and from 5.1 billion to 5.5 billion at 1064 nm.

These measured absorption-limited Q-factors at the communication wavelength agree with the previously measured result of the state-of-the-art material on chip [21], while it is the first experimental measurement for the absorption-limited Q-factor at the wavelength of 1064 nm. By comparing these results to the measured intrinsic Q-factors, we can conclude that the achieved intrinsic Q-factors of the microtoroid resonators are mainly limited by the material absorption loss.

To determine the contribution of different loss mechanisms, we have measured the intrinsic Q-factors of five samples in multiple wavelength bands, as shown in Fig. 4(a). The corresponding values of the intrinsic loss γ0 (dB/km) are also plotted in Fig. 4(b). To analyze these measurement results, we also model the surface scattering loss of the microtoroid resonators based on the three-dimensional volume current method [5,31,46–48], considering the measured surface roughness of 0.3 nm and the correlation length of 30 nm. Due to the small roughness and the relatively large mode volume of the microtoroid resonator, the calculated scattering losses are much lower than the measured material absorption losses at both 1560 nm and 1064 nm, which further confirms that the intrinsic Q-factors of the microtoroid resonators are mainly limited by the absorption loss.

Interestingly, there is a decreasing trend of the intrinsic optical loss with the wavelength increasing from 1480 nm to 1570 nm [Fig. 4(b)], while no significant variation is observed with the wavelength of 1020 nm to 1070 nm. Since the scattering loss is negligible at these wavelengths, it suggests that there is an additional loss mechanism beyond the scattering and material absorption losses of the microtoroid resonator. We attribute this extra loss to the surface water absorption [49], which can be simulated by FEM modeling. More details are provided in Appendix B. At the wavelength of the 1064 nm band, the simulated intrinsic losses are slightly lower than the measured values, and it could be due to the contamination induced scattering loss during the fabrication or measurement of the samples. Also, the scattering induced optical loss is more sensitive at 1064 nm than that of 1560 nm. The comparison between the modeled and measured results is summarized in Table 1. It is seen that the absorption loss contributes to 80.5% of the intrinsic loss at 1560 nm and 60.8% at 1064 nm, as the dominant factor for the intrinsic Q-factors.Table 1.

Summary of Loss at 1064 nm and 1560 nm

For a demonstration of how well the fabricated ultra-high-Q microtoroid resonators can perform, we employ the aforementioned 2.8-mm-diameter microtoroid resonator with an intrinsic Q-factor of 3.3 billion for an optical parametric oscillation threshold measurement. By monitoring the power of the generated sidebands at the varying pump power levels, we infer a threshold of 31.9 μW as depicted in Fig. 5(a). This threshold is 30 times lower than the value obtained from a silica microdisk resonator [6,50] due to a significant enhancement in the Q-factor, making a new record for the millimeter-sized optical microresonators. Previously, the thresholds in similar orders were only obtained with the microresonators of much smaller sizes [18,51]. During the threshold measurement, the coupling condition is carefully optimized to achieve an under-coupled state [52,53], thereby minimizing the threshold.

On the other hand, the ultra-high Q-factor significantly diminishes the power required for soliton formation. By employing the auxiliary-laser-heating method [54,55], we successfully obtain a single soliton microcomb using the fabricated microtoroid resonator. As shown in the inset of Fig. 5(b), a soliton step is observed in the transmission spectrum of the pump light. Figure 5(b) displays the optical spectrum of a single soliton excited with a pump power of only 220.2 μW. To the best of our knowledge, this is the first demonstration of a soliton at such a low power level, even lower than those for the generations of dark pulses; see the comparison in Table 2. Among the available platforms listed in Table 2, the creation of solitons in our ultra-high-Q silica microtoroid demonstrates the advantages of low pump power and low repetition rate.Table 2.

Comparison of Coherent Microcomb Generation with Various Devices

In summary, we have realized significantly enhanced Q-factors for the on-chip optical resonators operating at both 1064-nm and 1560-nm wavelength bands. The record-breaking Q-factor is achieved with 3.3 billion at 1560 nm and 3.2 billion at 1064 nm. Owing to the extremely high Q-factors, mode splitting smaller than 100 kHz is observed in a millimeter-sized microtoroid resonator. By employing the cavity-enhanced photothermal spectroscopy, we determine that the absorption loss of the currently available material of silica contributes to 80.5% and 60.8% of the total loss at 1560 nm and 1064 nm, respectively, thus proving that the optical Q-factors of our fabricated microresonators are mainly limited by the material absorption. Further analysis has been performed by measuring the Q-factor across a wide range of wavelengths and modeling the scattering and surface water absorption losses. Notably, the surface water absorption is found to contribute to an increase of Q-factors with the wavelength increasing from 1480 nm to 1570 nm, but has no discernible impact at 1064 nm. To demonstrate the good performance of the fabricated millimeter-sized microtoroid resonators, we break the record for the threshold of parametric oscillation to 31.9 μW and observe a single-soliton microcomb formation with a record-low pump power of 220.2 μW. Our results significantly elevate the upper limit of Q-factors for on-chip optical resonators, providing valuable insights into the optical loss mechanisms in thermally grown silica. Furthermore, this chip-based silica microresonator with an ultra-high-Q factor and a relatively large mode volume offers a unified platform for realizing an ultra-narrow linewidth Brillouin laser and a high-performance Kerr soliton microcomb, as well as facilitates the exploration of the interplay between nonlinear optical mechanisms. By integrating a waveguide and the silica microtoroid resonator on the same chip [62], it will be possible to achieve a fully integrated ultra-high-Q microtoroid resonator for practical application. Since the material absorption limit has been nearly reached in the current research, we expect that a further improvement on the Q-factor of the chip-based optical microresonators will be made possible only by optimizing the quality of the silica material.

Acknowledgment. We thank Professor Limin Tong for helpful discussions.