Developing a chip-scale optical clock

Abstract. We report our in-house R&D efforts of designing and developing key integrated photonic devices and technologies for a chip-scale optical oscillator and/or clock. This would provide precision sources to RF-photonic systems. It could also be the basic building block for a photonic technology to provide positioning, navigation, and timing as well as 5G networks. Recently, optical frequency comb (OFC)-based timing systems have been demonstrated for ultra-precision time transfer. Our goal is to develop a semiconductor-based, integrated photonic chip to reduce the size, weight, and power consumption, and cost of these systems. Our approach is to use a self-referenced interferometric locking circuit to provide short-term stabilization to a micro-resonator-based OFC. For long-term stabilization, we use an epsilon-near-zero (ENZ) metamaterial to design an environment-insensitive cavity/resonator, thereby enabling a chip-scale optical long-holdover clock.


Introduction
A wide range of scientific and commercial systems require time synchronization with increasingly high precision and accuracy, including very long baseline interferometry (VLBI), particle colliders, global navigation satellite systems, and 5G communications networks. Such time synchronization is typically achieved through an external communication network, such as the internet, dedicated satellite links, or free-space optical time transfer. 1,2 Holdover clocks can reduce the required update rate of time synchronization signals for such systems and also allow operation in the event that these synchronization signals are temporarily unavailable, for example during a network outage. Many of these same applications, such as VLBI at microwave frequencies, would benefit from a frequency reference with not just long-term frequency stability, but also with low phase noise-or equivalently short-term frequency stability.
Most holdover clocks are based on architectures for probing atomic transitions, which enable good long-term frequency stability. However, the physics package containing the atoms to be probed as well as environmental shielding contribute to an overall size and cost that limits the systems' size, weight, and power consumption, and cost (SWaP-C). The chip-scale atomic clock (CSAC), 3 which is commercially available from vendors such as Microchip and Teledyne, is a state-of-the-art example of a miniaturized atomic clock, and indeed the size, and cost of a CSAC is much lower than other atomic clocks. Nevertheless, the required vapor cells are not fully integrated, adding to the size, and the environmental shielding adds cost.
Furthermore, the vast majority of holdover clocks operate at relatively low frequencies (e.g., less than 10 GHz). In this case, low-phase-noise operation can be achieved straightforwardly by *Address all correspondence to Weimin Zhou, weimin.zhou.civ@mail.mil the addition of a quartz oscillator in the loop. However, for holdover clocks that operate at frequencies at or above 10 GHz, low-phase-noise operation is more challenging to achieve, as low-phase-noise references at frequencies above 10 GHz are not widely available.
In this paper, we propose a chip-scale optical holdover clock (CSOC) that would use a CMOS-compatible epsilon-near-zero (ENZ) optical resonator to provide a laser with enhanced long-term frequency stability, an optical frequency comb (OFC) to divide the stabilized optical frequency to a microwave, and a CMOS-compatible delay-line-based stabilization scheme to reduce the phase noise of the generated microwave. The all-CMOS-compatible architecture of the proposed clock may allow it to have SWaP-C that is even lower than that of a miniaturized atomic clock, such as the CSAC. Such a reduction in SWaP-C would allow the CSOC to be used in a wider array of circumstances. The remainder of the paper is structured as follows. In Sec. 2, we discuss the overall architecture of the proposed CSOC and the design philosophy. In Sec. 3, we discuss our experimental progress toward achieving a short-path-delay-line-based selfstabilization system that would allow the proposed clock to generate a microwave signal with low phase noise. The architecture was designed such that it could also be realized on a photonic integrated chip (PIC) in a CMOS-compatible process. In Sec. 4, we discuss our experimental progress toward achieving the environmentally insensitive ENZ resonator that will grant the CSOC its long-term frequency stability. This includes developing wavelength thick ENZ indium-tin-oxide (ITO) metamaterial and fabrication of air-core in ENZ ITO resonator devices.

Design Concept for CSOC
The principal objective of our design effort is to develop an optical holdover clock that generates a microwave signal with good frequency stability and low phase noise while minimizing its SWaP-C. The current state-of-the-art optical clock consists of three major components: [4][5][6][7][8] (1) an ultra-low-expansion (ULE) cavity stabilized laser that generates an optical frequency with ultra-narrow linewidth (low phase noise), (2) an atomic reference that disciplines the ULE-cavity-stabilized laser in order to provide it with long-term frequency stability while preserving its low phase noise, and (3) an octave-spanning OFC that coherently divides the low-phase-noise and long-term-stable optical frequency down to a microwave frequency. Our approach to achieve a CSOC is to develop an on-chip replacements for each of these three laboratory components while maintaining the same purpose for each of these components. The goal is to eliminate the high SWaP-C ULE locked laser and the atomic reference. With our new innovative design concept, our proposed CSOC is made by a micro-resonator-based OFC that is simultaneously stabilized with an on-chip delay-line interferometer and an environmentally insensitive ENZ resonator.

Concept of Micro-Resonator OFC for Frequency Division
An immediate choice for achieving a reduction in size is to replace the traditional mode-lockedlaser-based OFC with a micro-resonator-based OFC. [9][10][11][12][13] Indeed, other groups are developing octave-spanning micro-resonator-based OFCs 14,15 including as part of at least one effort to develop an entire on-chip optical atomic clock. 14 Hence, in the remainder of this paper, we will not focus on the development of the micro-resonator-based OFC.

Concept for Achieving Short-Term Stability
The ULE cavities that provide optical clocks with exceedingly low phase noise are typically 10 s of cms long 16 and typically require high-SWaP-C external housing to provide environmental isolation. Here, we propose to replace the ULE-cavity-stabilized laser with a photonic-integrated short-path delay-line interferometer that stabilizes the OFC without the need for an intermediate laser. [17][18][19] Although optical-fiber-based delay-line interferometers are a well-established tool to stabilize single-frequency lasers, 17 they are not commonly used on PICs. By bringing this technique to the PIC platform, we face a major challenge to overcome the waveguide propagation loss, which is orders of magnitude higher than the propagation loss of optical fiber.
Buried waveguides on the common 220-nm silicon-on-insulator platform typically achieve propagation losses of approximately 2 dB/cm at wavelengths near 1550 nm. 20 By moving to more complicated waveguide geometries, silicon waveguides have been reported with losses below 3 dB/m. [21][22][23] Additionally, some approaches toward ultra-low-loss Si 3 N 4 waveguides have been reported demonstrating less than 1 dB/m loss. 24,25 Nevertheless, regardless of the technology, the propagation loss of an integrated photonic delay line is significantly higher than the ∼0.17 dB∕km propagation loss of commercial optical fiber.
Our expectation of increased propagation loss of the integrated photonic delay line had a significant impact on the design of our interferometer. First, we limited the interferometer delay to less than 10 m, which significantly decreased the sensitivity of the interferometer. Second, even with a relatively short delay, we expect a high power imbalance between the delayed and non-delayed components of the light at the output of the interferometer, which may result in an increased contribution from laser relative intensity noise. Consequently, our development strategy was to first construct a fiber-optic prototype to demonstrate that a short-path interferometer has sufficient sensitivity to enable the generation of a low-phase-noise microwave. Future work will extend this prototype to an integrated photonic chip that stabilizes a micro-resonator-based OFC. Toward that end, we also designed and fabricated an integrated-photonic tunable coupler in order to mitigate the expected power imbalance. The results of these experiments are presented in Sec. 3.

Concept of Environment-Insensitive Device for Long-Term Stability
Any oscillator is susceptible to long-term frequency drift. One of the most pervasive sources of this drift arises from the oscillator's sensitivity to environmental fluctuations, including acoustic, vibrational, and temperature perturbations. The atomic reference in the laboratory-scale clock serves to mitigate this drift, because the atomic transition is relatively insensitive to the environment. Cold atoms are a particularly high-performing reference since they are extremely well isolated from the environment. Nevertheless, atomic sources inherently increase the SWaP-C of a clock.
Rather than pursuing an on-chip atomic reference, we propose an approach to obtain longterm frequency stability. We are developing an environmentally insensitive hollow-core microresonator that is clad with an ENZ metamaterial. 26 Although we do not expect the frequency accuracy or frequency stability at very long timescales (e.g., >10 3 s) of the ENZ reference to be competitive with cold atom references, we expect its frequency stability to compete with miniaturized atomic clocks such as the CSAC. Moreover, when we consider the possibility to incorporate the fabrication of the ENZ resonator into a CMOS process, we expect that the overall SWaP-C will be greatly reduced, resulting in a cheap, compact, and easily manufacturable frequency reference.
Our concept is to use the ENZ metamaterial as waveguide cladding to form an air or vacuum core resonator. 26 Figure 1 shows a schematic illustration of the air-ring, ENZ-metamaterialcladding resonator device. The idea is to make an intrinsic environment-insensitive cavity/ resonator device. For a typical ring resonator, the resonant optical frequency is f op ¼ m × f 0 , where m is the number of modes and f 0 is the fundamental frequency. The fundamental wavelength λ 0 ¼ C∕f 0 , which is the round-trip optical path. Also, λ 0 ¼ n × d, where n is the effective refractive index of the ring path and d is the physical dimension of the resonator. Both temperature and vibration may change the physical dimension of the resonator. In general, the environmentally induced frequency change is given by Eq. (1): E Q -T A R G E T ; t e m p : i n t r a l i n k -; e 0 0 1 ; 1 1 6 ; 6 6 3 In a typical micro-resonator containing SiO 2 , Si 3 N 4 , Si, etc., the change in refractive index, Δn, can be significant due to temperature changes and vibration-induced strains. In case of a material with thermal expansion α th , Eq. (1) can be written as Eq. (2): ; t e m p : i n t r a l i n k -; e 0 0 2 ; 1 1 6 ; 5 8 1 where α th is the thermal coefficient. The change in resonator physical dimension, Δd, can be significant due to thermal expansion (≅ α th ΔT) and deformation by vibration.
In contrast, in a vacuum/air resonator embedded in ENZ metamaterial (e.g., in ENZ cladding layer), as provided by the embodiments herein, Δn is negligible. The Δd can be less significant due to the geometry of the material where vacuum is enclosed by the ENZ cladding layer. In addition, with the ENZ condition, optical phase changes very slowly in the ENZ media, thus environmentally induced small geometry variations have less effect on optical phase variation, and therefore generates less phase noise.

Self-Referencing-Stabilization Locking of an OFC
In this section, we review our experimental efforts toward realizing an integrated-photonic delayline interferometer that can stabilize an OFC. We constructed a prototype 8-m fiber-optic interferometer that stabilized a commercially available OFC in order to demonstrate that a short-path interferometer had sufficient sensitivity to significantly reduce the free-running phase noise of the OFC, thereby enabling low-phase-noise microwave generation. We found that the phase noise of the tone was reduced by over 35 dB when the OFC was stabilized with the short-path interferometer, reaching −130 dBc∕Hz at a 1-kHz offset from the 10-GHz carrier frequency. We also designed and tested tunable couplers that were fabricated by a commercial foundry (AIM Photonics). These tunable couplers will be used in future work to ensure that the delayed and non-delayed light components in an integrated-photonic interferometer have similar power.

Proof of Concept Demonstration Using Fiber Delay-Line Interferometer
The theory of operation of the delay-line-interferometer-stabilized OFC is as follows. The OFC illuminates a fast photodiode to generate a microwave with a frequency, f m , that is an integer multiple, m, of the OFC's repetition rate, f r , such that f m ¼ mf r . Meanwhile, the optical power at the output of the interferometer, dPðtÞ, is proportional to fluctuations of the OFC's carrier frequency, df c ðtÞ, according to dPðtÞ ∼ τ g df c ðtÞ, where τ g is the group delay imposed by the interferometer delay. Notably, the carrier frequency of the OFC is nearly an integer multiple, N, of the repetition rate; it is given by f c ¼ Nf r þ f CEO , where f CEO is the carrier-envelope offset frequency of the OFC. If f CEO is stabilized so that its noise contribution is negligible, then the output power of the interferometer is proportional to the noise of the repetition rate, df r ðtÞ, according to dPðtÞ ∼ Nτ g df r ðtÞ. This signal is filtered and fed back to the OFC in order to stabilize the repetition rate. The key point is that the interferometer's sensitivity to the repetition rate noise is proportional both to the length of the interferometer delay and to the integer N, which corresponds approximately to the ratio of the carrier frequency to the OFC's repetition rate and is typically on the order of 10 5 or 10 6 .
One consequence of the higher propagation loss on an integrated chip is that the path imbalance in the interferometer will be limited to the scale of meters or less. This limitation is also related to the limited area available on a PIC. By contrast, the delay lines used to stabilize single-frequency lasers are typically kilometers long since the additional length increases the interferometer sensitivity. 17 Hence, the integrated-photonic interferometer will necessarily have orders of magnitude lower sensitivity than the typical fiber-optic interferometer. Nevertheless, we expect that the fact that the interferometer is sensitive to frequency noise at the carrier frequency-and therefore proportional to the large integer N-will afford us some leeway in the sensitivity. To prove our concept for a short-path interferometer, we constructed a fiberoptic prototype interferometer with a 40-ns delay line, corresponding to 8 m of optical fiber. We stabilized a commercially available OFC with this interferometer and generated a 10-GHz microwave signal with phase noise of −130 dBc∕Hz at 1-kHz offset frequency. 18 The architecture of the system used to generate the 10-GHz signal is shown in Fig. 2. The OFC was a commercially available fiber-based comb from Menlo Systems with a center wavelength of approximately 1550 nm and a f rep of 250 MHz. A commercially available f-2f interferometer (Menlo Systems) measured the f CEO of the comb. The output of the OFC was split into two parts.
The first portion of the OFC output was used to generate the 10-GHz signal. The light from the OFC entered a fast PIN photodetector (Optilab PD-20), where it generated a microwave frequency comb, with frequency components at the f rep of 250 MHz and its integer harmonics. We used a microwave bandpass filter (BPF) with a center frequency of 10 GHz to isolate the 40th harmonic and amplified it with a low-phase-noise RF amplifier. Since the frequency of the microwave output was directly related to the OFC repetition rate, its phase noise was also directly proportional to the phase noise of the OFC repetition rate.
The second portion of the OFC output was used to stabilize the OFC's repetition rate, thereby reducing the phase noise of the 10-GHz microwave. The light from the OFC passed through an acousto-optic frequency shifter (AOFS1), which imparted a frequency shift on the OFC according to the frequency of the signal at its voltage input. This signal was the f CEO of the OFC and was provided by the output of the f-2f interferometer. The polarity of the frequency shift imparted by AOFS1 was chosen such that it canceled the f CEO . The f CEO -free comb light entered a 0.8-nm optical BPF and then a fiber-optic Michelson interferometer with a path imbalance of 40 ns. We used single-mode fibers and non-polarization-maintaining components. The Faraday rotating mirrors mitigated the impact of polarization fluctuations. The output of the Michelson interferometer illuminated a set of balanced photodiodes, which converted the fluctuations of the f rep into a photocurrent. We used balanced photodiodes to reduce the current associated with the laser intensity noise. The photocurrent from the balanced photodiodes passed through a loop filter and fed back to an electro-optic modulator within the OFC cavity in order to stabilize the f rep .
The phase noise of the 10-GHz microwave generated by the OFC is shown in Fig. 3. When the OFC was free-running (i.e., when the self-stabilization circuit is turned off), the microwave had a phase noise of approximately −93 dBc∕Hz at a 1-kHz offset frequency. When the OFC was stabilized, the microwave had a phase noise of approximately −130 dBc∕Hz at a 1-kHz offset frequency. Hence, the self-stabilization circuit reduced the phase noise of the microwave by over 35 dB in this offset frequency range. In the stabilized case, the noise at offset frequencies below 1 kHz is consistent with environmental fluctuations in the laboratory, such as temperature variations, vibrations, and acoustics. Hence, we believe that with improved packaging, the noise in this offset-frequency range will decrease.
These results demonstrate that a 40-ns fiber-optic interferometer has sufficient sensitivity to stabilize an OFC such that we could generate a microwave signal with ultra-low phase noise. We expect the interferometer on the integrated platform to introduce additional loss, so further work may be required to achieve similar performance with a PIC. On the other hand, due to its smaller size, an integrated circuit can be more easily isolated from environmental fluctuations than the fiber-optic prototype, which suggests that the noise at some offset frequencies may decrease. In any case, these results are a significant and encouraging step toward the realization of an on-chip stabilization circuit.

Development of Integrated-Photonic Tunable Couplers
Another consequence of the high propagation losses in a photonic integrated waveguide is that the loss imparted by a centimeters-or meters-long delay line imparts a large power imbalance between the interferometer paths. Such an imbalance may lead to exaggerated noise contributions from laser intensity noise. So, we designed, fabricated, and tested a 1-by-2 splitter with a tunable splitting ratio to balance the optical power of the short delay with a meter-long delay at the output combining point of the asymmetric interferometer.
The tunable 1-by-2 splitter was fabricated using a commercial integrated-photonics foundry (AIM-Photonics). We also designed and had fabricated a delay-line waveguide, an asymmetric Mach-Zehnder interferometer, and balanced photodetectors using the same process. Optical images of these components are depicted in Fig. 4.
The splitting ratio of the 1-by-2 power splitter is adjusted using a doped silicon heater that runs parallel to the evanescent coupling region. The silicon heater is partially etched to prevent mode coupling from the evanescent coupling region into the heater itself, which would otherwise cause excess loss. The thermal tuning performance is depicted in Fig. 5. This adjustable power split allows the unequal loss in the two interferometer paths to be offset so that the output power is roughly equal. The relative splitting ratio can be adjusted by up to −35 dB.

Developing ENZ-ITO Metamaterial Based Environment-Insensitive Devices
We outline our effort of developing the air-core, ENZ-metamaterial cladding resonator as environment-insensitive cavity device. First, we need to develop a suitable ENZ metamaterial for such a device. Although there are many reported ENZ metamaterials such as multi-layer metal-dielectric super lattice 27 and nano-artificial metastructures, 28 those metamaterials involve high-cost fabrication and may not be compatible with Si-photonics and CMOS processing. We chose ITO that is a practical thin-film material widely used in semiconductor optoelectronic devices. ITO has shown particular promise because of its potential for low-cost fabrication and  compatibility with Si-based integrated photonics. However, in all the previously reported ENZ ITO film studies, the film thicknesses were significantly less than the wavelength region where ENZ properties occur. In our air-core, ENZ-metamaterial cladding resonator design, ENZ-ITO film needs to be around one wavelength thick. Since we would like to work at the telecom wavelength (1.55 μm) in order to benefit from the majority of telecom photonic devices, we need to develop thick ENZ ITO metafilms.

Thick ENZ-ITO Meta-Film Development
We fabricated a variety ∼2-μm-thick ENZ ITO films via pulsed DC sputtering onto a SiO 2 layer placed on top of a Si substrate and subsequently thermally annealed them. We performed a systematic study of these metafilms and discovered a non-uniform variation of the ENZ property. 29 The experimental results of these ENZ properties are shown in Fig. 6. Since no equipment exists that can directly measure an unknown variable (vertically) permittivity, ε, of a multi-layer material system, we developed a comprehensive method to obtain the permittivity as a function of vertical location in the film. The method involves taking variable angle spectroscopic ellipsometry and fitting the measured data with an appropriate physical model, which also sliced the thick ITO film region into 30 thin layers.
The different zero-crossing frequencies are shown in Fig. 6(a), for 10 min annealing with temperatures ranging from 250°C to 400°C. Figure 6(b) shows that for all annealing temperatures, the ITO layer permittivity was not constant with respect to the depth. At lower annealing temperatures, the samples' permittivity depth profile showed a relatively small amount of non-uniformity, which is qualitatively consistent with the behavior reported for thin ITO films as shown in Fig. 6(c). However, at higher annealing temperatures, a sharp transition region appeared midway through the depth of the ITO film. Above this transition region-i.e., near the top of the ITO film-the real part of the permittivity was positive, whereas below the transition region-i.e., near the bottom of the ITO film-the real part of the permittivity was negative.
To understand this variation behavior, we conducted several microanalysis experiments on three of our ITO samples: as-deposited, annealed at 250°C, and annealed at 350°C. 29 Specifically, we examined the crystallinity and texture using x-ray diffraction (XRD), the crystallite morphology using cross-sectional transmission electron microscopy (TEM), and the chemical composition using energy dispersive x-ray spectroscopy (EDS). The XRD revealed that the as-deposited sample was predominantly amorphous, whereas the annealed samples were polycrystalline with no dominant crystal orientation. The TEM picture as shown in Fig. 7(a) revealed that the crystal grains in the high-temperature-annealed sample changed size and shape partway through the film's thickness, whereas the crystal grains in the low-temperature-annealed samples showed no such change. The EDS result as shown in Fig. 7(b) further showed that in the high-temperature-annealed sample, the oxygen concentration increased sharply partway through the film, whereas the low-temperature-annealed sample showed no such change. These results indicate that the crystal grain morphology and oxygen vacancy density may contribute significantly to the permittivity of the ITO film.
We are developing a new fabrication technique to improve the uniformity of the ENZ ITO thick films. The goal is to make both ε 1 and ε 2 closer to zero through the thickness of the film because higher ε 2 will increase the absorption loss for our future device. For this, we deposit the ITO film with high-power impulse magnetron sputtering (HiPIMS), a relatively recent innovation in sputter deposition tools that allows for more control over film growth. Below is the real part and imaginary permittivity behavior of 80-min annealed sample at 650°C. The results are shown in Fig. 8. The permittivity depth profile is still not uniform. The real part of the permittivity crosses 0 at 1557 nm, and the imaginary part of the permittivity remains at 0.39. However, ε 2 is closer to zero through the layer thickness compared with our previous ENZ ITO samples.
We are also studying different annealing techniques such as annealing under oxygen gas to control formation of oxygen vacancies in ITO. The detailed results will be discussed in future publications.

Design and Modeling of Air-Core, ENZ-ITO Cladding Ring Resonator
Light-guiding properties of the air-core, ENZ-ITO-cladding ring resonator were numerically modeled using a finite-element-method (FEM) eigenmode solver. The ENZ-ITO resonator consists of a bus waveguide and an optical ring cavity. The bus waveguide and ring cavity guide light in air through a hollow channel embedded in the ENZ-ITO film. The light-guiding mechanism in the hollow waveguide is based on total external reflection (TER) at the air-ITO interface.
In our TER-based waveguide analysis, the fabricated ITO was modeled as a superlattice with 30 sublayers due to the non-uniform distribution of its dielectric constant. The thickness of the individual layer was set to be 60 nm. According to the ellipsometry analysis, the fabricated ITO sample exhibits ENZ properties near the bottom of the film. The region where the absolute value of the real part, ε 1 , of the dielectric constant of the ITO is less than 1 was defined as a low-epsilon region. In the air-core waveguide design, this low-epsilon region was used as a cladding layer directly interfacing the air core. Figures 9(a) and 9(b) present the dielectric constant distribution of the air-core ENZ-ITO waveguides. Both real and imaginary parts of the dielectric constant vary gradually along the deposition direction. In this design, a ridge-shape air-core, enclosed by the ITO film, provides the lateral mode confinement. Figure 10 shows the calculated electric field distribution of the fundamental transverse magnetic mode using a commercial FEM solver, COMSOL Multiphysics TM . The electric field in Fig. 9 (a) The real ε 1 and (b) the imaginary ε 2 parts of the dielectric constant distribution across the air-core ENZ waveguide.  the air-core waveguide is described by Eðx; y; zÞ ¼ Eðx; yÞ · e −jβz , where β is the propagation constant of the guide mode. As shown in the figure, the electric field is tightly confined in the ridge waveguide. Due to the large refractive index contrast between the core (air) and the cladding (ITO), the evanescent tail of the guided mode rapidly decays in the low epsilon region in the ITO film. The simulation parameters used in the mode analysis are t 1 ¼ t 2 ¼ 2 μm, and w 1 ¼ w 2 ¼ 4 μm. The calculated effective refractive index of the fundamental mode is β∕k 0 is 0.96641 − j 8.3207 × 10 −4 , where k 0 ð¼ 2π∕λ 0 Þ is the wave vector and λ 0 is the free-space wavelength. The estimated propagation loss is 29.3 dB/mm.

Developing Fabrication Technique for the Air-Core, ENZ-ITO Cladding Ring Resonator
We have designed several fabrication methods for the air-core, ENZ-ITO-cladding ring resonator. The goal is to use an air cavity as the waveguide core in a ring resonator configuration that is then covered with 1-to 2-μm-thick ENZ ITO material as cladding for the waveguide as shown in Fig. 1. The fabrication process of such device structure involves multiple processing steps. The first processing step is depositing ITO on a SiO 2 ∕Si substrate. The second step is depositing a sacrificial material layer such as Si or Ge, and lithographically forming a disk or ring for the resonator. The third step is depositing additional ITO to cover the disk/ring and then annealing it to achieve the desired ENZ property. The final step is etching small windows on top of the disk to expose the sacrificial layer and then remove it using XeF 2 vapor. Due to the nature of ENZ materials, there is some absorption that may increase the propagation loss and reduce the Q-factor of the resonator device. Therefore, a major challenge is to reduce the propagation loss for our air-core/ENZ ITO cladding ring-resonator. For that we take two approaches in the fabrication: first, from our experimental data as shown in Figs. 6 and 8, the ENZ ITO layers have lower ε 1 and ε 2 near the substrate and high ε 2 at the top surface, which increases absorption near the surface. Therefore, we developed a fabrication method for a mushroom-shaped sacrificial Si/Ge structure, on which we conformally deposit ITO to cover the "mushroom," Figure 11 shows the SEM picture of the mushroom-shaped Si disk and pictures after the ITO is deposited on the sacrificial Si disk. After the annealing process, the ENZ ITO near the interface with the sacrificial Si/Ge mushroom has lowest ε 2 . Therefore, once the sacrificial Si/Ge is removed by the vapor etching process, the air-core ring-resonator will be surrounded by low loss ENZ ITO material. Second, using the lithographic process, we created the air-ring resonator with a notched ridge waveguide cross-section shown as Figs. 9 and 10 to reduce interaction between the optical mode and ITO cladding and reduce the amount of light leaking into the ITO.
We are in the process of fabrication and testing such air-ring, ENZ-ITO device structures; further fabrication details and testing results will be reported in future publications. Fig. 11 (a) An SEM picture of the mushroom-shaped Si disk; (b) and (c) ITO covered Si disk.

Summary/Conclusion
We are developing key building block technologies for a long-holdover CSOC. These include a self-locking concept to provide short-term stabilization to a micro-resonator-based OFC. For long-term stabilization, we are developing an environment-insensitive cavity using an ENZ metamaterial. This chip-scale device can be used as oscillator source or local clock in RF and communication systems as well as in 5G networks, and it can potentially be integrated with an optical transmitter and receiver for free-space OTT and ranging to provide positioning, navigation, and timing with low SWaP-C.