Photonic Integration for UV to IR Applications

Photonic integration opens the potential to reduce size, power and cost of applications normally relegated to tableand rack-sized systems. Today, a wide range of precision, highend and ultra-sensitive, communication and computation, measurement and scientific applications, including atomic clocks, quantum communications and processing, and high resolution spectroscopy, are ready to take the leap from the lab to the chip. However, many of these applications operate at wavelengths not accessible to the SOI-based silicon photonics integration platform due to absorption, power handling, unwanted nonlinearities and other factors. Next generation photonic integration will require ultra-wideband photonic circuit platforms that scale from the UV to the IR and that offer a rich set of linear and nonlinear circuit functions as well as low loss and high power handling capabilities. This article provides an assessment of the field in ultra-wideband photonic waveguides to bring power efficient, ultra-high performance systems to the chip-scale and enable compact transformative precision measurement, signal processing, computation and communications techniques. Keywords— Photonic integration, microwave photonics, comb generation, microwave photonics,


I. INTRODUCTION
Photonic integrated circuits (PICs) enable many applications to be realized at the chip-scale and provide a path to reduced cost, complexity, power, size and improved manufacturability. Examples include digital coherent communications [1], fiber radio frequency (RF) transmission [2,3], integrated microwave photonics (IMWP) [4] and positioning and navigation [5]. Emerging precision applications including atomic clocks [6], precision metrology and transformative applications like quantum communications and information processing [7,8] will also benefit from photonic integration. However, many of these applications operate at wavelengths and at a level of performance not achievable with today's large scale SOI-based silicon photonic platforms.
The drive to produce high volume PICs has leveraged scaling of the silicon CMOS foundry and the associated manufacturing ecosystem [9]. The demand for silicon compatible photonics has accelerated the development of SOIbased silicon photonics, a combination of silicon on insulator (SOI) [10] and III-V [11] photonics with CMOS electronics. Both SOI and III-V photonics deliver solutions at the communications wavelengths and IR. However, for applications that operate at wavelengths shorter than 1 um, the relatively small bandgap of silicon leads to high waveguide loss or non-transparency as well as limited power handling capabilities that occur due to nonlinear losses, primarily to two photon absorption or TPA [12]. High losses also limit the optical bandwidth of desired nonlinear optical effects, like fourwave mixing (FWM) and second harmonic generation (SHG) that provide the ability to span the visible to IR. Examples include octave spanning and self-referenced optical frequency comb (OFCs) [13,14].
A wide range of emerging applications will drive the need for new wide-bandwidth photonic integration, that offer low optical losses extending from the ultraviolet (UV) to the infrared (IR). The application space over various wavelength windows in the UV (~200 nm) to the IR (~2350 nm) regime is illustrated in Fig. 1. In addition to low linear losses, there are high performance emerging applications that require waveguides that can handle high optical power, minimizing unwanted nonlinear phase shifts and nonlinear losses. These new applications focus particularly on the UV to Near IR (NIR) range (~200 nm -900 nm). Yet at the same time, there are applications that require optical nonlinearities, and waveguide technology is required that supports desired nonlinear interactions as well as required dispersion. These nonlinear and dispersion properties must be supported at the application wavelengths of interest with some applications generating frequencies that span optical octaves. These waveguides also must support more complex functions on-chip, including low loss linear passive components (e.g. splitters, demultiplexers, filters) and compatible on-chip optical gain mechanisms that support the UV to IR. Compatible waveguide actuation mechanisms (e.g. electrooptic, stress, thermal) that operate across this wide bandwidth must support building blocks like modulators, phase shifters, switches and detectors.
with wafer-scale processing to provide cost effective scalable solutions.
SOI provides a 300 mm wafer-scale PIC process, based on complementary metal oxide on semiconductor (CMOS) foundry processes with an ecosystem that supports device design, fabrication and testing geared towards high volume applications operating at wavelength above 1 um. This platform does not meet the needs of next generation UV to NIR light photonics due to the bandgap mismatch as well as an economic mismatch with lower volume applications, and more specialized requirements of visible light PICs. New end-to-end PIC platforms with next generation materials and CMOS-like design and fabrication will most likely utilize smaller waferscale (e.g. 200 mm) PIC processes (see [15] for example).
Gain elements provide silicon with lasing, amplification, and other gain based functions, requiring addition of other material systems including III/V [16], that can be heterogeneously integrated with silicon [17]. For example, in the telecommunications waveband, 4 inch III-V wafers (e.g. indium phosphide) followed by hybrid SOI/III-V integration [18,19] or selective bonding of III-V to SOI or direct growth of III-V on SOI [20] are common approaches. SOI gain material systems operating at 1310 nm/1550 nm will need to be replaced with UV and UV-visible compatible semiconductor gain solutions (e.g. III-Nitride) [21] as well as NIR gain materials (e.g. GaAs) for next generation UV-NIR heterogeneous integration and/or nonlinear light generation techniques such as second harmonic generation [22].

II. NEXT GENERATION PHOTONIC INTEGRATION
Next generation wideband PIC technology will need to leverage wide bandgap semiconductors. Materials that have been reported, and that can support wafer-scale and heterogenous integration include silicon nitride (Si 3 N 4 ) [15], tantalum pentoxide (Ta 2 O 3 ) [23], aluminum nitride (AlN) [24,25], and aluminum oxide (Al 2 O 3 ) [26,27], shown in Fig. 2. Heterogeneous integration using thin film lithium niobite (LiNbO 3 ) waveguides is also possible [28], and can provide important second order nonlinearities and electrooptic modulation properties. Other possibilities include ultra-low loss etched silica waveguides and resonators [29] and the Hydex waveguide platform [30] that incorporates a dope silica core embedded in a silica cladding. Silica, LiNbO 3 , and Hydex waveguide technologies are not the subject of this perspective.
Optical loss contributions in general are due to material absorption and waveguide scattering as well as material impurity and inhomogeneities. Thin film loss measurements help access the material absorption loss component while actual waveguides need to be fabricated to access the waveguide scattering contributions. Other types of waveguide losses, for example material surface effects, may contribute to losses in the waveguide core materials.  [15], Ta 2 O 5 [23], Al 2 O 3 [26], AlN [25].
A summary of published material film and waveguide losses in the 200 nm to 650 nm range, adapted from [31], is shown in Fig. 3 for Si 3 N 4 waveguides [27], Ta 2 O 3 waveguides [32], bulk AlN single crystal [33], and atomic layer deposition (ALD) Al 2 O 3 waveguides [26]). While this data is representative of specific material and waveguide cases, a more detailed assessment of loss requires consideration of the waveguide geometry, cladding materials, mode confinement factor, device footprint, bend radius, and details of the actual fabrication process. The four materials shown have trade-offs in loss, potential waveguide geometry and optical confinement, nonlinear optical and electrooptical and other attributes like background autofluorescence [34] for spectroscopy applications, all of which are application dependent and can lead to choice of one material over others for reasons in addition to low loss alone. An overview of progress in low loss wide bandgap UV to IR waveguide technology is given in [27]. In this paper focus is placed on stochiometric silicon nitride deposited by low pressure chemical vapor deposition (LPCVD) due to the success of ultra-low optical losses [15], in part due to the high temperature anneals used to remove absorbing hydrogen bonds (e.g. N-H). Other silicon nitride waveguide compositions and processes have been widely used with great success including silicon-rich nitride (SiRN) [35] to reduce stress induced cracking for thick waveguides and PECVD nitride [36], however, these silicon nitride waveguides are not covered in this perspective article. Nitride (AlN) [33] (adapted from [31]).
The Si 3 N 4 material system offers transparency over an extremely wide wavelength range (~400 nm -2350 nm). The lowest reported loss designs employ a silica-based planar waveguide with a high-aspect ratio stochiometric Si 3 N 4 core, low H absorption through annealing [37,38] and a wafer bonded upper cladding to achieve a record low loss (0.045 ± 0.04) dB/m [39]. The subsequent development of a low-loss PECVD deposited upper cladding instead of wafer bonded cladding was an important step to wafer-scale manufacturable Si 3 N 4 PICs with losses in the range ~ 0.3 dB/m [38] and ring resonator loaded Q factors of 30 million and higher [38,40]. The Si 3 N 4 waveguide system is relatively mature and can be used to fabricate complex PICs typically on 4" or 6" wafers [15]. Today's ultra-low loss waveguides rely on low optical confinement in the thin Si 3 N 4 core suitable for large area devices with bend radius on the order 10 mm. For compact devices (sub-mm diameter) with a high optical mode confinement and nonlinear optical applications, processing techniques that overcome stress induced cracking for waveguide thickness greater than 400 nm and that minimize sidewall scattering losses are required [41,42]. At wavelengths close to 400 nm, Si 3 N 4 bandgap material losses dominate, and waveguide losses move towards the 20 dB/cm or higher. This loss may be acceptable for certain applications and designs. An overview of the history of and progress in Si 3 N 4 photonics is given in [15].
When operation below 400 nm wavelength is desired, the wider bandgap tantalum pentoxide (Ta 2 O 3 ), also referred to as Tantala, is a strong candidate. Tantala waveguides have shown record low waveguide losses (~ 3dB/m) at 1550 nm [23], with measured losses below 5 dB/cm just below 500 nm. Transparency of this material based on bandgap is theoretically close to 250 nm. For moderate core thickness (e.g. 90 nm) Tantala loss at 1550 nm are lower than the equivalent a 90 nm thick core Si 3 N 4 waveguide. Tantala can also be deposited to make thick cores (e.g. 800 nm -1 um) without the stress cracking issues of stochiometric Si 3 N 4 , making it an attractive material for ultra-compact and high optical confinements structures for applications like optical frequency comb generation [43]. Ring resonator structures with loaded Q values in the several million have been reported [43]. Consideration needs to be given to the annealing process used to drive out hydrogen in oxide cladding to achieve ultra-low losses in silicon nitride waveguides. Tantala anneal temperatures are limited to approximately 600 C due to material crystallization, compared to ~ 1200 C that can be used for Si 3 N 4 . This makes it important to limit if not eliminate the presence of hydrogen during fabrication of oxide cladded Tantala waveguides. Additionally, Tantala has an optical nonlinear coefficient greater than Si 3 N 4 [44,45], making it an attractive material for broadband nonlinear frequency applications including optical frequency combs [43] and supercontinuum generation [46]. Tantala has other advantages for applications involving generation and detection of low levels of light like nanophotonic waveguide enhanced Raman spectroscopy (NWERS) where the level of background Raman autofluorescence is low compared the desired Raman signal [34].
Aluminum nitride (AlN), while much higher loss than silicon nitride and Tantala, offers a bandgap that supports transparency down to wavelengths of 350 nm and shorter with the potential for UV operation. Published loss of single crystal bulk AlN down to 350 nm wavelength [33] is shown in Fig. 3. and AlN waveguide ring resonators on sapphire substrate with intrinsic (different than loaded) Q of > 170,000 at 638 nm and > 20,000 at 369.5 nm have been reported [24]. The tradeoff in higher loss for shorter wavelength operation may be desirable for certain applications. AlN also offers the important capability of electrooptic modulation and signal processing out to GHz frequencies [25] as well as second order nonlinearity . The AlN platform is one of the several discussed here that offers the exciting potential for quantum optics applications but rely on a different set of substrate and cladding materials. The AlN platform is one of the three main candidates that offers possibilities for integrated quantum optics that require operation near the UV in addition short wavelength nonlinear optics spectroscopy.
Finally, aluminum oxide (Al 2 O 3 ) also referred to as Alumina, shows strong promise for low loss integration from the UV to visible (208 nm -633 nm). Low loss waveguides fabricated using an Atom layer deposition (ALD) process [31] and sputter deposition [34] have achieved the best results with waveguide losses < 3 dB/cm at 371 nm wavelength and ring resonator intrinsic qualify factor > 470,000 at 405 nm wavelength [24]. An advantage of Alumina is the low but sufficient index of refraction relative to an oxide cladding at UV wavelengths so that dilute optical modes that result in low sidewall scattering losses can be supported.
New applications will drive the demand for photonic integration platforms that support operation from the UV to IR will drive next generation photonic integration beyond today's silicon photonic integration. These platforms will be based on wide bandgap materials and waveguide fabrication processes that provide low propagation losses. The most promising This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset. A summary of materials, losses and integration platform compatibility at sample wavelengths and applications is given in Table 1

III. LOW LINEWIDTH LASERS
Ultra-low linewidth lasers [49], the mainstay of high-end scientific experiments and applications, will benefit greatly from next generation photonics. Translating the performance of these spectrally-pure lasers to wafer-scale integrated devices will bring lower cost, size, weight and power along with increased environmental robustness to applications that are today confined to the laboratory. Realizing lasers with this level of performance and that produce ultra-low linewidth across the visible to IR will require wide bandwidth photonics. These applications include coherent communications [1], nextgeneration data center interconnects [50], atomic and quantum sensing [51,52], atomic clocks [6], precision metrology and time-frequency transfer [53] and quantum communications and computation [7,8]. Other applications include microwave frequency synthesizers [54] and optical laser gyros [55].
The stimulated Brillouin scattering (SBS) laser [56] is an important laser class capable of producing highly coherent, spectrally-pure emission that can operate from the visible [57] to the infrared [58]. With sub-Hz fundamental linewidth and the capability for Hz level integral linewidth emission through cavity stabilization [59][60][61][62], this class of laser is an ideal candidate for integration. Brillouin is an elegant nonlinearity, based on interaction between sound and light and can support wide bandwidth lasing from visible to infrared.
There has been significant progress towards integration of the SBS laser, historically limited to fiber implementations, into a photonic integrated platform, as summarized in Fig. 4. The underlying principle falls into two general categories, designs that involve waveguides that are engineered to phase match the light (photons) with the acoustic (phonon) waves, and designs that do not require this phase matching. The later involves non-confined short-lived acoustic waves that are generated in a high Q resonator. Early micro-optical SBS lasers incorporated free-space coupled CaF2 whispering gallery mode resonators capable of sub-Hz fundamental linewidth emission [63]. Moving an SBS laser to on-chip required fabrication of integrated waveguide structures that demonstrated sufficient Brillouin gain and low loss structures to fabricate a high Q resonator. Chalcogenide (As 2 S 3 ), a material with extremely high Brillouin and Kerr nonlinearities, was notably the first material used to demonstrate SBS gain in a waveguide [67] and was subsequently embedded in an external optical fiber resonator to demonstrate SBS lasing [64]. Brillouin gain was later demonstrated in a suspended silicon waveguide [68] and an SBS laser demonstrated by embedding the suspended waveguides in an integrated silicon waveguide bus-coupled racetrack resonator [65]. Chemically etched silica resonators coupled to a tapered-fiber coupled chemically etched ultra-high Q silica microresonator [69] was implemented as an on-chip air clad etched silica microresonator coupled to an integrated Si 3 N 4 waveguide bus [29]. These prior works required dispersion engineering between optical and acoustic modes or nonlinear absorption in the UV and visible (for silicon) making wide bandwidth operation, even with materials that have low loss, difficult as well as other limitations related to wafer-scale This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5131683
compatibility of the fabrication process. By utilizing wide bandwidth Si 3 N4 waveguides designed to support dilute optical modes and high power handling capability, an all-waveguide wafer scale SBS laser was recently demonstrated, showing onchip record sub-Hz fundamental linewidth emission [66] is possible. By realizing spectrally pure SBS lasers in a wide bandgap integration platform that also supports integration with a wide variety of components [15], paves the way to integrated ultra-low linewidth SBS lasers for a wide range of UV to IR applications. A history of SBS photonics is given in a recent review article [70].

IV. OPTICAL FREQUENCY COMBS
A key building block is the optical frequency comb (OFC), a single source of multiple well-defined, stable, evenly spaced frequencies [71]. Wide bandgap semiconductors have been critical towards realizing chip-scale OFCs with the capability to producing hundreds of frequencies that span tens of Terahertz [72] [73] [74] [75].
Further integration of OFCs with other components will enable a wide range of precision applications traditionally relegated to bench-top systems. Example applications include low noise microwave frequency generation [76], astro-combs [77], atomic clocks [62], spectroscopy and laser sensing, waveform synthesis [78] and optical metrology. The large multioctave comb emission and the ability to lithographically define combs with different line spacing and other nonlinearities like second harmonic generation, makes OFCs central to precision microwave readout using self-referencing techniques [79] to create a precision 'optical ruler' [71]. Wide bandgap semiconductor OFCs enable comb lines emitted at the lowfrequency IR side of the emission spectrum to be referenced to comb lines at twice the frequency on the high-frequency visible side of the emission spectrum. The OFC spectral line spacing and their common offset can be locked to a radio frequency standard to create these optical rulers. Other advantages of integrated OFCs include realizing combs with spacing >100 GHz, difficult to achieve with non-integrated comb techniques. Phase stabilization for microwave synthesis using interlocking frequency combs [76] is another key application of this technology. In the fiber communications world, integrated comb generators can replace large arrays of power hungry WDM lasers for high capacity fiber communications systems [80] and the rack-scale to the chip-scale leveraging progress in next generation wide bandgap photonics [73,81].
The basic OFC structure is in principle relatively simple, a bus-coupled ring resonator with the correct optical nonlinear and dispersive properties. Dissipative Kerr soliton (DKS) microresonator comb generators [82] [80] [73] [81] are an important type of OFC that have found application to frequency synthesis [78], terabit optical communications [80] and dual-comb spectroscopy [83]. Kerr OFCs balance nonlinear four-wave mixing parametric gain, cavity loss, and dispersion in a microresonator to generate continuously circulating stable soliton pulses that yield a low noise frequency comb output. There have been a wealth of successful designs realized in the Si 3 N 4 waveguide platform. The absence of nonlinear losses in the wide bandgap semiconductors, like two-photon absorption in silicon, enable these comb generators to bridge the visible to IR. Given the flexibility in Si 3 N 4 to design a wide range of low loss waveguide resonators with different FSRs, the comb generator line spacing can be optimized for a given application.
Producing frequency combs that extend to the visible regime have proved more difficult. Typically combinations of parametric Kerr nonlinear processes ( 3 ) in the IR are combined with second harmonic generation ( 2 ) in Si 3 N 4 [84] and AlN waveguide resonators [85,86]. An example of visible light comb production is shown in Fig. 5 for an AlN waveguide resonator with a first nonlinear process, four wave mixing (FWM) comb production, followed by a second nonlinear process, second harmonic generation. Looking forward, there is much progress to be made in visible light combs in other lower loss wide bandgap semiconductors e.g. Si 3 N 4 , Alumina (Al 2 O 3 ) and Tantalum Pentoxide (Ta 2 O 5 ), that generate directly in the visible, and bridge the UV to IR. Progress in visible to near infrared (NIR) supercontinuum light generation in Si 3 N 4 has shown promise for ultra-broadband frequency generation [13]. As an example of progress towards combs in visible capable wide band materials, the wide bandgap Tantala platform has been used to demonstrate a Kerr soliton waveguide OFC with a resonator loaded Q factor > 3 million (see Fig. 6

) [43].
This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset. Tantala provides low loss [23], an index of refraction comparable to silicon nitride and a nonlinear Kerr coefficient several times larger than Si 3 N 4 [23]. The low stress of deposited Tantala films provides the ability to fabricate thick waveguide structure, ~ 1 um, needed to engineer and balance nonlinearities with group velocity dispersion (GVD) without the cracking issues of Si 3 N 4 . The wide bandgap of Tantala offers the potential to move operation of nonlinear and linear ring resonators to the 300 nm wavelength operating regime. Initial results with Tantala Kerr Soliton OFCs have shown operation from 1.4 um to 1.8 um.

PLEASE CITE THIS ARTICLE AS
An intriguing application of wide bandgap nonlinear frequency combs is quantum information processing. Optical quantum frequency combs [87][88][89][90], or 'quantum combs', are the single photon counterpart to classical OFCs. Quantum optical computations benefit from having many entangled photon pair modes. A quantum OFC can produce many frequency modes, leveraging the wavelength parallelism of comb teeth produced by single photon highly correlated nonlinear processes like spontaneous four-wave mixing (SFWM) [91]. An example of a processing chip for multi-photon entangled state quantum processing using an OFC is shown in Fig. 7. This approach shows promise for integrating a high number of entangled modes on a chip, using techniques like FWM signal and idler interferometers and single photon detectors for quantum processing [87]. Quantum information processing can benefit from the frequency tooth spacing > 100 GHz achievable with integrated OFCs, making processing of photons (e.g. filtering) manageable, whereas fiber quantum frequency combs produce narrow frequency spacing making filtering and other tasks difficult. An issue with increased spacing using integrated OFCs is the reduction in number of frequency quantum modes limited to a certain wavelength band (e.g. C-band). In order to provide more quantum modes, SFWM Kerr waveguide OFCs that extend from the visible to IR with thousands of modes such as that reported in [84] or further progress with DKS OFCs that span the visible to IR will be needed.

V. MULTI-COUPLED RESONATORS AND PHOTONIC MOLECULES
Combining multiple optical resonators into linear chains or two dimensional arrays opens up possibilities in optical signal processing, spectral engineering and linear and nonlinear optics. Research in Photonics Molecules seeks to use photons in optically coupled resonant microstructures to realize functions and behaviors analogous to electronic atomic and molecular systems [92,93]. In this sense, optical resonators can be thought of as photonic atoms and collections of photonic atoms can be designed to act like photonic molecules (PMs).
Photonic molecules have the potential to impact a wide variety of UV to IR applications. Example applications include lasers for communications and sensing, spectral engineering [92] [93], optical signal processing [92,94], nonlinear optics [95], sensing and manipulation of small biological samples, nano-particle detection [96], label-free single molecule detection [97], ultra-low power optical switching [98,99], and many body physics simulations [100,101]. Integrated resonators fabricated with wide bandwidth photonics that can achieve high loaded Q and can be fabricated into arrays will impact the field of photonic molecules and its applications.
As illustrated in Fig. 8, a single bus coupled resonator (Fig.  8a) behaves like an atom in resonance linewidth. Stability will not be that of an atom unless locked to an atomic resonance. Passive or active locking can lead to very stable optical This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5131683
resonances [102]. The bus coupled resonator is also known as an optical all-pass filter function. Two coupled resonators (Fig.  8b) behave like a diatomic molecule with six energy supermodes [103]. Coupling between optical cavities using quantum dots and micro-disc resonators has been shown to yield molecular behaviors like frequency splitting and bonding and anti-bonding orbitals [98]. Attractive and repulsive forces between optical resonators similar to Coulomb forces in electronic molecules can be demonstrated and utilized as well as ultrafast optical switching [104] effects and functions. An assembly of mutually-coupled photonic atoms (e.g. microresonators) can realize a photonic molecule with modes and behavior very much like its electronic counterpart with atomic energy or resonant states and molecular super-modes [105].
A linear array of bus coupled resonators (Fig. 8c) leads to the formation of optical crystal bands [106], also analogous to a thin film filter. Properties of the linear array can be designed by varying the coupling strength and sign [107]. Forming a 2D array of resonators with a line defect leads to the well-known photonic crystal (see Fig. 8d) structure with discrete allowable energy states and forbidden bandgaps [108], enabling dispersion engineering and slow light [109,110] as well as other devices and functions that take advantage of these properties (e.g. ultra-compact optical filters [111] [112]). By varying the size of the coupled resonators in a 2D array (see Fig. 8e) the photonic equivalent of defect states [113] as well as defect states in a linear resonator chain [114] can be realized.

VI. OTHER WIDE BANDGAP OPTICAL COMPONENTS
Waveguide actuation and dynamic active components like modulators are critical for generating sideband control signals for feedback loops, e.g. Pound Drever Hall (PDH) stabilization [61], and for RF analog modulation and digital data modulation.
Progress in wide optical bandwidth photonic modulators has been limited. Electrooptic and strain modulation are the primary effects employed but tend to be either weak or bandwidth limited with good progress in AlN and silicon nitride. The Pockels effect was employed for electrooptic modulation in AlN to demonstrate modulation up to 4.5 Gb/s [25]. Electrooptic modulation in an AlN waveguide was employed to demonstrate a push-pull MZI modulator with 21 volt V at 1550 nm and a 12.9 volt V at 1064 nm. A modulation rate of 140 MHz with 12 dB extinction ratio (ER) was shown for the MZI, useful for signal processing and feedback loops and a dual-ring modulator was shown to operate at 4 GHz with 3 dB ER. The stress optic effect has been employed for MHz modulation speeds using ferroelectric lead zirconate titanate (PZT) films on Si 3 N 4 waveguides with MHz response RF modulation response [115] and low energy modulation for 780 nm light at cryogenic temperatures [116]. Higher speed electro-optic modulation in the O-band and C-band was achieved with PZT films deposited on Si 3 N 4 bus-coupled ring modulators demonstrating modulation > 25 GHz with a half-wave voltage-length product of ~ 1 V cm [117]. Slower tuning and switching is achievable with thermally controlled actuators in many of these wide bandgap photonics, for example Si 3 N 4 [15].
The performance of demonstrated thermal, Pockels and PZT actuators is satisfactory for relatively low bandwidth RF communications links and switching applications e.g. quantum (e.g. quantum heralding) and optical phase array switchable delay lines [118]. Other active and passive components have been demonstrated in wide bandwidth photonics [15,27,31], however, many open issues remain as well as performance progress especially for modulators and switches.

VII. APPLICATIONS
Applications that benefit from low loss across the visible to IR will also leverage the linear and nonlinear properties as well as ability to integrate high performance passive and active components and support traditional RF and digital system requirements as well as next generation quantum, sensing and metrology applications.
RF optical communications, radar systems and RF optical signal processing require complex, low loss, high performance optical filters to shape the spectral or temporal contents as well as optimize optical signal to noise ratio (OSNR) and perform other signal processing functions in the radio-frequency (RF) (~300 kHz -300 GHz) and microwave (~300 MHz -300 GHz) domains. Low loss, ability to support high optical powers and high linearity are critical features for replacing power hungry electronics and maintaining transmission integrity in RF optical links. Optical filters that can handle high power before exhibiting nonlinear loss and filter distortion and that cover the multiple wavebands from visible to infrared will be indispensable for systems that combine RF and sensor systems.
A 3 rd order low loss Si 3 N 4 bus-coupled CROW filter [38] demonstrated an ultra-high 80 dB extinction ratio (ER) with low insertion loss (< 1.3dB) and a flat passband. Filters with this This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5131683
level of ER and filter shape performance can be used for nonlinear pump-signal separation in Brillouin, four-wave mixing (FWM), second order nonlinear generation and in precision spectroscopy. Filter channelization for RF photonic and WDM optical communications, where individual channel bandwidths can be tuned in addition to center frequency tuning is possible with Si 3 N 4 due to low loss elements. A CROW-based bandpass filter fabricated can select channel in a frequencydivision subcarrier satellite communication system [119]. Programmable RF filter networks leverage Si 3 N 4 low loss to realize networks of tunable filter arrays, enabling multiple dynamically configurable complex filter functions on the same chip to implement a wide class of filter types (e.g. FIR and IIR) [120]. Programmable multistage lattice filters for dispersion compensation [121] and other analog signal conditioning applications, are possible due to the ability to integrate multiple low loss elements, like switches and delay lines, on the same chip. These programmable elements can deliver low loss, low linearity needed for many RF optical systems.
Optical delay lines are used in a wide variety of applications including RF and digital filtering, optical beam forming, optical signal processing, information coding, digital data storage and synchronizers and pulse shaping. Non-resonant delays are used for RF and analog functions like transversal and FIR filters, IIR filters and other discrete time signal processors [122]. Broadband Si 3 N 4 delays provide discrete, medium to large delays, up to 250 ns on a single chip (tens of meters in length). Resonant delay lines, like optical ring resonators (ORRs), are more compact and are continuously tunable over ranges of picoseconds to order of a nanoseconds [123]. Non-resonant Si 3 N 4 waveguides provide about 12.5ns per meter delay and lengths up to 25-m with 250ns delay have been fabricated. Spiral geometries with waveguide crossovers on a large area chip (2 cm x 2 cm) [124] have been demonstrated for applications like optical gyroscopes [125]. Larger true time delays are possible for phased array antenna applications by combining long delay lines with Si 3 N 4 optical switches [126]. Next generation signal processing elements and delay lines will see further size reduction and integration using advanced techniques like SBS in wide bandgap photonics [70].
In the digital application space, some of the more difficult problems today are found in the fiber data center interconnect (DCI) environment where exploding data center capacity pushes traditional fiber interconnects to their power and space limitations. The power and engineering limitations that DCIs are facing are due to the evolution of switch ASIC chips, the engines of the Data Center Interconnect (DCI). Switch ASICs will grow from 12.8 Tbps today to 100 Tbps, placing a strain on energy resources [127] and traditional engineering approaches. Continued scaling of today's fiber transmission technologies in the DCI requires novel solutions and technologies. Solutions will involve bringing high capacity coherent WDM into the DCI. However, coherent WDM links rely on power consuming technologies like digital signal processors (DSPs) and large-bandwidth phase locked loops [128]. Solutions incorporating high-capacity WDM links will have to be free of DSPs and other high power analog or digital electronics.
A new WDM coherent link architecture utilizes capacity and power scaling advantages of highly-integrated, ultra-stable, narrow-linewidth laser and transceiver technology. The ARPAe funded FREquency Stabilized Coherent Optical (FRESCO) project supports coherent Tbps transmission per wavelength and scaling to support future 100 Tbps switch ASICs with order several pJ per bit efficiency. FRESCO brings narrow linewidth and laser stabilization technology developed for frequency standards [61] and atomic clocks [6] to the coherent fiber DCI. The FRESCO architecture is based on a shared ultra-stable, spectrally-pure laser using wide bandgap photonics, to realize an ultra-stable, spectrally pure, shared optical comb (Tx and LO) that is modulated with highly-integrated silicon photonic coherent transceiver. A high level illustration of the FRESCO integrated transceiver is shown in Fig. 9. A FRESCO shared optical source consists of a silicon photonic tunable laser that serves as a pump for a silicon nitride SBS laser [66]. The SBS laser is stabilized to an ultra-high Q factor frequency stabilized resonator that reduces the SBS emission integral linewidth and stabilizes the optical carrier frequency and phase. The stabilized ultra-low linewidth SBS laser output pumps a nonlinear optical frequency comb (OFC) source [78] that serves as a shared WDM Tx and LO source for a multi-channel silicon photonic coherent modulator and receiver [10]. Other application for wide bandgap next generation photonics will leverage the ultra-low phase noise and highly level of integration including microwave synthesis [54], optical gyroscopes [130,131] and photonic quantum circuits [8] for quantum information processing and communications [132].
An intriguing emerging chip-scale application is atom cooling [62,133,134]. Atomic clocks are constructed using tightly coordinated sets of free-space laser beams and bulky optics to cool, pump and probe an atomic species. Certain quantum information processing architectures are based on cooled atoms. Wider deployment of atomic clocks will require integration of the lasers and optical interfaces to reduce cost, size and power consumption and decrease sensitivity to environmental conditions. To drive down the cost and complexity as well as improve the robustness of these systems, PICs will be required to interface pump and probe lasers, This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5131683
delivered by optical fibers, to free-space beams with required geometry and beam quality. Such PICs pose unique integration challenges including low loss visible light fiber-coupled waveguides, visible light passive components including splitters and combiners, and low loss waveguide-to-free-space ultra-large-area gratings uniformly fabricated over mm-scale areas that emit non-diverging beams at pre-designed angles.
Recently, the design, fabrication and measurement of a Si 3 N 4 photonic integrated circuit (PIC) that interfaces a fibercoupled 780 nm laser to serve as laser cooling beams for a 3D magneto optic trap (3D-MOT) [135] was reported. The 3D laser cooling beam interface PIC is shown schematically in Fig. 10. A 780 nm laser is coupled via a single mode fiber to a single mode waveguide at the Si 3 N 4 PIC input. The input guided light is directed to a 1 x 3 multimode interference (MMI) waveguide splitter and routed to three slab waveguide beam expanders that in turn uniformly illuminate three large-area free-space surface grating emitters. The gratings are located on a 13.5 mm diameter circle and emit free-space collimated 780 nm beams, emitted at 54.7⁰ from the PIC surface-normal, in order to produce 90-degree intersection between all three beams at a distance of 9 mm from the chip surface (in the center of the atom vapor cell). The large area gratings are designed to deliver freespace collimated beams to cool of a large volume of Rb-87 atoms as well as provide a good beam intensity uniformity for all six cooling beams.

VIII. SUMMARY AND FUTURE PROSPECTS
Next generation photonics will employ wide bandgap semiconductor PIC technology. Among the most intriguing platforms that support wafer-scale and heterogenous integration are silicon nitride (Si 3 N 4 ), tantalum pentoxide (Ta 2 O 3 ), aluminum nitride (AlN), and aluminum oxide (Al 2 O 3 ). These wide bandgap technologies bring to the designer's toolbox an integration platform capable of addressing key performance features not available in SOI and other platforms including ultra-low linear and nonlinear losses, transparency from the UV and visible through the infrared, and high power handling capabilities. A wide array of passive, active, linear and nonlinear components, can be combined on-chip, bringing high performance systems to the chip scale and open the potential for new signal processing, computation and sensing systems-onchip. Low linewidth lasers, multi-octave spanning self-referenced frequency combs, ultra-high Q resonators and programmable optical filter networks, show promise for photonic circuits that can addresses high bandwidth problems that are becoming too power-intensive for analog and digital electronic processing. Emerging applications will leverage the ultra-low phase noise in including photonic quantum circuits for quantum information processing and communications and precision timing and quantum metrology using atom cooling.
Problems that remain to be tackled include development of a true integration PIC platform that can support UV-NIR applications and wavelength specific designs without platform modifications as well as full foundry compatibility. Heterogeneous gain across the full UV-NIR range is a major issue that needs to be addressed with promise for great advances. Gain for lasing and periodic signal boosting (e.g. SOAs) that span this wavelength range and enable direct signal boosting or amplification of non-linear generated frequencies (e.g. SHG) will be critical. Other milestones to be reached include to further lower losses by another order of magnitude and reach waveguide ring resonator quality factors (Q) > 100 million heading to 1 Billion, down to the UV. Another critical area includes platform compatible actuation, particularly modulation. A wide range of modulation frequency, both narrow band and broad band, are needed. At the low end (e.g. 1 MHz -500 MHz), techniques like acousto-optic modulation (AOM) will be needed to sweep frequencies with high conversion efficiency. Signal and data modulation at high RF frequencies (> 25 GHz) and high bit rates (>25 Gbps) will also be needed. With increased sensing and atom interaction applications, techniques to bring atomic and molecular materials onto these chips will be combined with ultra-low loss materials.