The many applications of high optical quality (Q)-factor microresonators [1] have stimulated remarkable progress in boosting its value across a range of material systems and device platforms (see results and table in Ref. [2]). The Q factor frequently enters quadratically into device performance. For example, in Raman [3], parametric [4,5], and Brillouin [6–8] nonlinear optical oscillators, the threshold varies inverse quadratically with Q. Moreover, this same Q dependence enters the fundamental linewidth of laser oscillators [9] such as microcavity Brillouin lasers [10]. Ultimately, these considerations are key performance drivers in soliton microcomb systems [11], microresonator gyroscopes [12–15], and microwave sources [14,16,17]. For chip-based systems, silica microresonators using thermally grown oxide offer the highest Q factors [8]. Here, by optimizing fabrication methods for silica whispering-gallery resonators, the Q factor of this system has been boosted to over 1.1 billion. The results are confirmed through linewidth, ring-down, and parametric oscillation measurements.

 

Fig. 1. Microresonator images and mode profiles. (a) Scanning electron microscope (SEM) image of microresonator (${\sim}22\;{\rm GHz}$ FSR; 3 mm diameter). Dotted white line indicates cross section in panel b. (b) False-color SEM cross section with narrow trench structure and 120 μm undercut. (c) Left (right) panel, fundamental TE (TM) mode electrical field distribution from numerical simulation. Arrows indicate electric field vectors.

Download Full Size | PDF

In the investigation, wedge whispering-gallery resonator devices are fabricated using a combination of thermal oxidation ($8\;\unicode{x00B5}{\rm m}$ thickness on 4 in. (100 mm) high-purity float-zone silicon wafers), optical lithography, buffered HF silica wet etch, and ${{\rm XeF}_2}$ silicon dry etch, as detailed in reference [8]. The wedge angle was $27 {-} {{30}^ \circ}$. To boost overall Q performance, several process and mask layout improvements were implemented. The resist pattern and subsequent buffered HF silica wet-etch-defined trenches are narrowed, as illustrated in Figs. 1(a) and 1(b). This lowers dry-etch time for silicon removal and thereby reduces unintended microscopic silica dry etching. The temperature of each fabrication step is also strictly controlled within a ${\pm}{0.5^ \circ}{\rm C}$ range. This includes both the dry-etch temperature and the ${{\rm XeF}_2}$ source temperature, both of which are servo controlled to ensure reproducible results. The importance of this latter temperature control is that it permits a more precise determination (and maximization) of silicon undercut to reduce optical loss. The ultimate limit of undercut is determined by silica buckling [18]. Finally, the silica microresonators are annealed 2–3 times for 20 h in an ultra-pure nitrogen ambient at ${1000^ \circ}{\rm C}$ to remove water and to release bulk stress in the suspended silica structure.

 

Fig. 2. Q measurements. (a) Measured intrinsic ${{\rm Q}_{0}}$ versus resonator diameter. FSR, resonator diameters, and ${{\rm Q}_{0}}$ values are indicated in millions (M). Red dotted line corresponds to a finesse of 60,000. (b) Left panel: resonance linewidth measurement of a 10 GHz FSR device at 1585 nm. Upper trace is resonance transmission (blue dots) with Lorentzian lineshape fitting (red curve). The linewidth is 220 kHz, corresponding to an intrinsic ${{\rm Q}_{0}} = 1130$ M and loaded ${{\rm Q}_{\rm L}} = 860$ M. Lower trace is a frequency calibration (black dots) from a Mach–Zehnder interferometer (FSR is 5.979 MHz) with sinusoidal fitting (cyan curve). Right panel: ring-down measurement (blue) of the device measured in the left panel. An exponential decay fitting is shown in red. Photon lifetime is 704 ns, corresponding to loaded ${{\rm Q}_{\rm L}} \approx 840$ M. (c) Intrinsic ${{\rm Q}_{0}}$ and loaded ${{\rm Q}_{\rm L}}$ data measured from 1530 nm to 1630 nm. (d) Distribution of ${{\rm Q}_{0}}$ values measured from different mode families for TE (left panel) and TM (right panel) mode polarization measured in a single FSR.

Download Full Size | PDF

Intrinsic Q factors (${{\rm Q}_{0}}$) are measured by characterization of resonance linewidths (accounting for waveguide loading effects) and through cavity ring-down measurements. Coupling to the resonators using a tapered fiber coupler [19] and a polarization controller was used to excite modes that are primarily TE or TM like [see Fig. 1(c)]. As an aside on the optical coupling, mechanical vibrations are present on the tapered fiber coupler. To suppress these vibrations, it is possible to place the tapered fiber on the top of the wedge edge in contact with the silica surface. As a result, the coupling condition becomes stable without inducing much scattering loss to the resonator. At the same time, the amount of coupling can be readily controlled by simply moving the contact location of the taper on the resonator. This has the effect of adjusting the distance between the taper waveguide and the location of the optical mode [see Fig. 1(c)] to thereby adjust the resonator loading by the waveguide.

Figure 2(a) plots the highest intrinsic Q factors obtained for several device diameters. Two device sizes have Q factors over 1.1 billion. A typical spectral measurement is illustrated in the left panel of Fig. 2(b), which shows a transmission spectrum for a device having a free-spectral-range (FSR) of approximately 10 GHz (diameter 6.5 mm). The full-width-at-half-maximum linewidth is measured (at 1585 nm) to be 220 kHz, corresponding to a loaded Q factor ${{\rm Q}_{\rm L}} = 860$ million (M). By measuring the transmission depth, a coupling Q factor of ${{\rm Q}_{\rm c}} = 3.6$ billion is determined, from which an intrinsic ${{\rm Q}_{0}} = 1.1$ billion is inferred (note: ${{1/{\rm Q}}_{\;\rm L}}{= {1/{\rm Q}}_{0}} + {{1/{\rm Q}}_{\;\rm c}}$). The rising dependence of intrinsic Q factor with increasing resonator diameter has been previously observed in the wedge resonator structure and results from round trip loss that is limited by surface scattering [8]. To further confirm the Q results, the ring-down of cavity power was also performed. Ring-down data from the same device measured in the left panel of Fig. 2(b) is shown in the right panel of Fig. 2(b). The ring-down decay gives a photon lifetime of about 704 ns, which corresponds to ${{\rm Q}_{\rm L}} = 840$ M in close agreement with the linewidth-inferred loaded Q. To the authors’ knowledge, these are the highest optical Q factors reported for on-chip devices. In particular, previous high-performance results for silica resonators are also plotted as blue and green points in Fig. 2(a). It is also worth noting that the ${{\rm Q}_{0}} \gt 1$ billion performance has been maintained for more than half a year to date by enclosing the resonators in a purged plexiglass box. The devices possess ultra-high Q over a broad spectral band, as shown in Fig. 2(c), where measured intrinsic and loaded Q factors in a 9 GHz FSR resonator (diameter 7.3 mm) are plotted versus wavelength.

In a typical resonator, there are dozens of modes from different mode families and polarizations that have ultra-high-Q factors. Histograms of measured ${{\rm Q}_{0}}$ values for TE (left panel) and TM (right panel) mode families in a 9 GHz FSR resonator (near 1550 nm) are plotted in Fig. 2(d). Q factors in the histograms are measured by linewidth fitting on spectra taken by fast frequency scanning of an external-cavity diode laser. The scan was intentionally performed from blue to red wavelengths so that resonator thermal hysteresis effects [20] would tend to degrade the apparent Q factor. As a result, the measured Q values are likely somewhat lower than the actual Q values. To further confirm the Q measurements, the parametric oscillation threshold was also measured using the experimental setup in Fig. 3(a). As shown in Fig. 3(b), a sub-milliwatt threshold (0.946 mW) was measured using a 9 GHz FSR device at 1550 nm. This is a record low threshold for this low FSR [21]. To perform this measurement, the pump line was filtered using a fiber Bragg grating, as shown in the setup. The pure frequency comb power was then measured and plotted. The inset to Fig. 3(b) shows both the filtered and unfiltered spectrum at 1.12 mW input power.

 

Fig. 3. Sub-milliwatt parametric oscillation threshold measured in the device having a 9 GHz FSR. (a) Experimental setup. AOM, acousto-optic modulator; PC, polarization controller; FBG, fiber Bragg grating filter; PD, photo detector; OSA, optical spectrum analyzer; OSC, oscilloscope; AFG, arbitrary function generator. (b) Plot of parametric oscillation power versus input power (1550 nm) showing oscillation threshold of 0.946 mW. Inset: parametric oscillation spectrum (solid blue) and filtered pump (dashed gray) at 1.12 mW input power.

Download Full Size | PDF

In summary, by optimizing fabrication methods and recipe parameters, we have demonstrated a record on-chip Q factor over 1.1 billion in silica microresonators. Future efforts will be directed towards implementing these improvements in fully waveguide-integrated ultra-high-Q resonators [22]. Moreover, besides pure silica structures [8,22], there is significant progress in ultra-high-Q structures where silica provides the cladding for low confinement silicon nitride waveguides [14,23]. In these structures, it is important to understand loss limits imposed by silica on the overall design. More generally, the results presented here establish an upper bound for optical loss in structures using thermally grown silica.