Optical microresonators are essential to a broad range of technologies and scientific disciplines. However, many of their applications rely on discrete devices to attain challenging combinations of ultra-low-loss performance (ultrahigh Q) and resonator design requirements. This prevents access to scalable fabrication methods for photonic integration and lithographic feature control. Indeed, finding a microfabrication bridge that connects ultrahigh-Q device functions with photonic circuits is a priority of the microcavity field. Here, an integrated resonator having a record Q factor over 200 million is presented. Its ultra-low-loss and flexible cavity design brings performance to integrated systems that has been the exclusive domain of discrete silica and crystalline microcavity devices. Two distinctly different devices are demonstrated: soliton sources with electronic repetition rates and high-coherence/low-threshold Brillouin lasers. This multi-device capability and performance from a single integrated cavity platform represents a critical advance for future photonic circuits and systems.
Optical microresonators or microcavities1 provide diverse functions that include frequency microcombs2, soliton mode-locked microcombs3,4,5,6,7, Brillouin lasers8,9,10,11,12, bio- and nanoparticle sensors13,14, cavity optomechanical oscillators15, parametric oscillators16,17, Raman lasers18, reference cavities/sources19,20,21,22 and quantum optical devices23. Key performance metrics improve with increasing Q factor across all applications1. For example, power consumption as well as phase and intensity noise in signal sources can be dramatically reduced, because these quantities scale inverse quadratically with Q factor. Also, higher Q improves the ability to resolve a resonance for sensing or for frequency stabilization. Such favourable performance scalings have accounted for a sustained period of progress in boosting Q factor by reducing optical loss in resonators across a range of materials24,25,26,27,28. Likewise, the need for complex microcavity systems that leverage high Q factors has driven interest in low-loss monolithically integrated resonators25,26,29,30,31,32,33,34,35. For example, Q values in waveguide-integrated devices as high as 80 million (ref. 32) and 67 million (strongly confined resonators, ref. 35) have been attained.
Nonetheless, the highest Q factor resonators remain discrete devices that are crystalline36 or silica-based1,10,37,38. These discrete resonators are, moreover, unique in the microcavity world in terms of overall performance and breadth of capability. This includes generation of electronic-repetition-rate soliton streams required in optical clocks39 and optical synthesizers40, rotation measurement at near-earth-rate sensitivity in micro-optical-gyros41,42, synthesis of high-performance microwave signals43,44,45,46, and operation as high-stability optical frequency references19,20,21 and reference sources22. Functions such as these belong to a new class of compact photonic systems that rely on ultrahigh-Q fabrication methods that have so far defied photonic integration.
Here, a monolithic microcavity having a record high Q factor is demonstrated. The materials, process steps and, in particular, the use of a plasma-enhanced chemical vapour deposition (PECVD) silicon nitride waveguide enable full integration of these ultrahigh-Q resonators with other photonic devices. They also enable wafer-scale hybrid integration to III–V active devices for optical pumping and detection47,48. Critically, and as required for new system-on-a-chip applications, the resonator supports design controls required to realize many device functions previously possible using only discrete (non-waveguide-integrated) devices. To demonstrate its capability, electronic-rate solitons (15 GHz) are generated at a low pumping power level. Also, as a distinctly different capability, high-coherence and low-threshold stimulated Brillouin laser oscillation is demonstrated. Beyond the necessity of ultrahigh-Q (UHQ) factor, these demonstrations illustrate resonator dispersion control to support dissipative Kerr soliton generation and precise diameter control to phase match the Brillouin laser process. Finally, the higher-index silicon nitride waveguide to lower-index silica resonator optical coupling is beneficial for coupling ideality49,50.
Integrated UHQ resonator
In this section, the process steps required to microfabricate the resonator are reviewed. Also, the resonator optical spectrum and Q factor are characterized. Then, the design requirements and properties of the silicon nitride waveguide are presented.
In Fig. 1a, a scanning electron microscope (SEM) image shows two of the silica ridge resonators connected by a common silicon nitride waveguide (false colour red in the image). Zoom-in and cross-sectional views of portions of the resonator are shown in Fig. 1b and Fig. 1c, respectively. Detailed fabrication steps are presented in Fig. 1d. The fabrication process begins by growing a thermal silica layer on a high-purity float-zone silicon wafer (additional details are provided in the figure caption). A phosphoric acid etch is used to define the silicon nitride waveguide. It is important that the silicon nitride is fully removed from the silica in this step as otherwise it will adversely impact the optical Q factor. In the penultimate step, a ring aperature is created as opposed to fully removing the interior silica. This reduces etch loading effects in the final dry-etch step. Not shown in the process flow is the deposition of a thin layer (20 nm) of silica by atomic layer deposition to the etched waveguide. This serves to protect the silicon nitride waveguide during the final dry etch step. The resonator has a maximum thickness of approximately 8.5 μm and the silica layer under the waveguide has a thickness of 2.5 μm. The silicon nitride waveguide has a thickness of approximately 250 nm. It is initially 3 to 3.5 μm in width at the edge of the wafer and as discussed later is tapered to about 900 nm near the resonator so as to phase match to the resonator optical mode. Two designs of the waveguide were used to vary the coupling strength: straight and pulley32. Measurements provided in the Supplementary Information show the effect of pulley length on coupling strength. The spectral uniformity of the coupling strength is also verified from 1,500 nm to 1,560 nm.
Spectral measurements of the integrated microcavity were performed by end-fire coupling to a tunable external cavity laser and monitoring the transmission through the silicon nitride waveguide as the laser was scanned. The devices measured had a free spectral range (FSR) of approximately 15 GHz (ridge ring diameter of approximately 4.3 mm) and Fig. 2a presents a spectral scan containing over three FSRs measured near 1,550 nm. Transverse mode families featured hybrid polarizations and were identified by measuring their dispersion curves as described below and then comparing with numerical modelling. The waveguide used in this scan (and in Fig. 2b) was a pulley version with length 200 μm that supported the transverse electric (TE) fundamental mode but that coupled to both TE and transverse magnetic (TM) resonator modes. Strictly speaking, these resonator modes are hybrids on account of the wedge cross-section.
High-resolution scans of the fundamental TE and TM resonator modes are presented in Fig. 2b. A linewidth fitting algorithm gives an intrinsic Q factor of 120 million for the TE mode and 205 million for the TM mode. To further confirm the high optical Q factor of the TM mode, cavity ring-down was performed37. Figure 2c is a superposition of ten ring-down traces of the highest Q factor measured from the fundamental TM mode family. The ring-down gives an intrinsic Q factor of 230 million. Ring-down of the fundamental TM mode spectrum in Fig. 2b gives an intrinsic Q factor of 216 million in close agreement with the linewidth inferred Q. As shown in the Supplementary Information, comparable results were confirmed across the tuning range of the test laser (1,520–1,560 nm). Silica has a spectrally wide low-optical-loss window and it is expected that these results could be reproduced over an even broader band of wavelengths.
Silicon nitride waveguide design
In Fig. 2d, the phase matching of the silicon nitride waveguide to the silica ridge resonator is studied by plotting the waveguide effective index versus the waveguide width (250 nm waveguide thickness). Mode profiles within the waveguide–resonator coupling section are given in Fig. 2e. Phase matching to achieve critical coupling of the TE fundamental mode could be obtained with approximately 60% yield (measured on 21 devices). An important feature of this system is that phase matching to the fundamental TE resonator mode occurs when the waveguide is single mode, which is optimal for high coupling ideality49. As has been noted elsewhere50, high-Q resonators tend to be overmoded and therefore phase matching to a waveguide generally involves careful design to avoid non-ideality arising from the required larger waveguide dimensions. In our work, the use of a waveguide material (silicon nitride) with index larger than that of the resonator material (silica) allows phase matching to occur when the waveguide is single mode.
In this section, the integrated UHQ resonator is applied to demonstrate two kinds of device. The first is a solition microcomb that features both a detectable repetition rate and low operating power. The second is a high-coherence Brillouin laser with a sub-milliwatt threshold.
Detectable-rate soliton microcombs
Microcombs and in particular soliton mode-locked microcombs3 represent a major new application area of microcavities. These tiny systems are being studied as a way to transfer large-scale frequency comb technology to an integrated photonic chip. They provide highly reproducible optical spectra, have achieved two-third and full octave-span coverage5,51 and have highly stable repetition rates4. However, mode-locked pulse repetition rates that are both detectable and readily processed by electronics are required in all frequency comb systems in order to self-reference the comb52. To achieve self-referenced octave-span operation in microcombs at practical power levels, frequency comb formation is divided into a THz-rate comb (micrometre-scale resonator diameter) and an electronic-rate comb (centimetre-scale resonator diameter)40. However, while the smaller-diameter, THz-repetition-rate, soliton combs have been demonstrated with integrated waveguides51,53, the large-diameter electronic-rate soliton microcomb has so far only been possible using discrete silica and crystalline resonators3,4. Part of the challenge here is achieving a sufficiently high Q factor to overcome the increased pumping volume of the larger electronic-rate soliton comb. Also compounding this problem are resonator design requirements imposed by the soliton physics. These include minimization of avoided mode crossings and anomalous mode family dispersion (bright solitons). Attaining the combined features of ultrahigh-Q factor to overcome a large optical pumping volume, while incorporating these resonator design features within an integrated microcavity has not been possible. Here, the integrated ridge resonator is used to demonstrate a new capability for an integrated design, 15 GHz soliton generation. The pumping power level is also low.
By using modelling and measurement techniques described elsewhere3,4, the ridge resonator was designed to minimize avoided mode crossings as required for soliton generation. Along these lines, the fundamental TE mode family dispersion was characterized by measuring the frequency of all modes between 1,530 nm and 1,580 nm using an external cavity laser calibrated by a Mach–Zehnder interferometer4. These data were then plotted by removing both an offset frequency (ω0) and a linear dispersion term proportional to the FSR (D1) of the mode family at ω0 (D1/2π = 15.2 GHz was measured). Modes are indexed using the label μ with μ = 0 assigned to the mode at frequency ω0, which is also the mode that is pumped to generate the solitons. The resulting data are plotted in Fig. 3a. The red curve is a parabola showing that the mode family is dominated by second-order dispersion over the range of modes measured. A fitting gives D2/2π = 6.4 kHz, where ω μ = ω0 + D1μ + D2μ2/2. Significantly, there is no observable mode-crossing-induced distortion of the mode family, therefore making the mode family well suited for soliton formation3,4.
Solitons were triggered and stabilized using the power kick5 and capture-lock technique54. An optical spectrum is shown for a single soliton state in Fig. 3b. Visible in the spectrum is the optical pump wave. Despite the large (4.3 mm diameter) resonator required to attain the 15 GHz repetition rate, the ultrahigh Q ensured a low parametric oscillation threshold16 power near 5 mW (waveguide coupled) and pump power as low as 25 mW for stable soliton operation. To confirm the soliton repetition rate, the soliton pulse stream was detected and analysed on an electrical spectrum analyser. The electrical spectrum is provided as an inset to Fig. 3b and shows a measured repetition rate near 15.2 GHz. The pulse-like nature of the soliton stream was measured using the frequency resolved optical gate (FROG) method and is presented in Fig. 3c. As a representative photonic circuit based on the device demonstrated here, Fig. 3d conceptualizes a dual-comb spectrometer modelled after a recent demonstration using discrete ultrahigh-Q devices55.
Low-threshold Brillouin lasers
The Brillouin process has attracted considerable interest in microdevices56. Brillouin laser action has been demonstrated in discrete resonators based on silica8,10,57 and CaF2 (ref. 9). Laser action has also been realized in integrated resonators using silicon12 and chalcogenide waveguides11. However, reference sources22, microwave synthesizers44,45 and Brillouin gyroscopes41 require the highest possible optical Q factors for generation of narrow-linewidth signals and have so far relied on discrete devices. Moreover, ultrahigh Q allows devices to operate at low power despite increased size. For example, discrete centimetre-scale Brillouin laser gyroscopes have recently achieved rotation rate sensitivities near the Earth’s rotation rate41. In this approach, sub-Hertz Brillouin laser linewidths result from the inverse quadratic scaling of fundamental phase noise on Q factor57. These narrow linewidths endow the device with high rotation-rate sensitivity. However, the gyroscope performance is also boosted by increased resonator size (increased Sagnac effect) that normally would require higher pumping power. Here, the ultrahigh-Q resonance provides an offsetting effect, enabling gyro operation at low (milliwatt-scale) power levels.
As a second device demonstration, the integrated ridge resonator is applied to generate high-coherence Brillouin laser action. To phase match the Brillouin process when pumped near 1,550 nm devices with diameters of approximately 6 mm were fabricated. Figure 4 shows the optical spectrum of the lasing Stokes wave. The weaker pump signal peak in the spectrum results from the need to collect the lasing Stokes wave in the propagation direction opposite to the pumping direction. Its strength is determined by residual reflection and backscattering in the measurement. The upper left inset in Fig. 4 shows the Stokes power versus pumping and gives a threshold power of approximately 800 μW in the waveguide. The upper right inset is the microwave beat signal between the pump wave and the Stokes wave. It has a high coherence as evidenced by the resolution bandwidth.
Beyond the specific device demonstrations provided here, performance considerations relating to noise and fluctuations frequently require combination of high Q and large mode volume that present challenges for scalable fabrication processes. For example, thermorefractive and photothermal fluctuations19,58 destabilize cavity frequency and vary inversely with resonator volume. Likewise, the fundamental timing jitter noise of a microcomb soliton stream is predicted to scale as n2/V, where n2 is the Kerr nonlinear coefficient and V is the mode volume59. To the extent that these sources of fluctuation can be managed or reduced in active devices by increased cavity volume (or reduced n2 as in the case of soliton timing jitter), higher Q factors can counteract the accompanying effect on pumping power. Also, in dual-comb spectrometers, reduced FSR (large diameter) devices are necessary to prevent under-sampling of chemical spectra. These larger microcomb devices also benefit from the highest possible Q factor to enable operation at practical power levels.
As a practical consideration, it is important to protect devices from the environment (that is, moisture and contamination). In the present context, a hermetic seal for the resonator can be applied at the wafer level or to individual devices60 (Fig. 5). Silicon-to-glass anodic bonding at temperatures of 300–450 °C (ref. 60) is possible, and as shown in the Supplementary Information we have confirmed that long-term annealing (>50 h) at temperatures as high as 1,000 °C does not impact device performance.
In summary, we have demonstrated an integrated resonator with a record optical Q factor. Low pump-power soliton generation at 15 GHz as well as high-coherence Brillouin laser action were demonstrated to illustrate functionality previously possible in only discrete devices, but nonetheless required for self-referenced microcomb photonic systems and for integrated systems requiring high-coherence signal sources. The waveguide material, PECVD silicon nitride, is among the most widely used waveguide materials in the photonics industry and provides a nearly universal interface to other photonic devices fabricated on a common silicon wafer.
PECVD SiN x fabrication and material characterization
The temperature of PECVD SiN x deposition was 350 °C, and a thermal annealing (1,000 °C for 3 h) was also carried out to reduce the presence of N–H bonds. We measured the elemental composition of the PECVD SiN x film via X-ray photoelectron spectroscopy (XPS), and it was found to have Si:N = 0.852:1.0 compositional ratio. In addition, the refractive index of the film is 2.089 and 2.044 at 632 nm and 1,550 nm, respectively.
The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank O. Painter and B. Baker for assistance with the PECVD silicon nitride process, H. Atwater and W.-H. Cheng for assistance with silica atomic layer deposition, M. Hunt for assistance with electron-beam microscopy, Y.-H. Lai for technical assistance, and A. Matsko and J. Bowers for helpful discussions. We also gratefully acknowledge the Defense Advanced Research Projects Agency under the DODOS (award no. HR0011-15-C-0055, sub award KK1540) and PRIGM:AIMS (grant no. N66001-16-1-4046) programs and the Kavli Nanoscience Institute.