The challenge of on-chip isolation

Isolators, as the photonics counterpart of electrical diodes, play a critical role in photonic integrated circuits (PICs) by preventing harmful feedback between different parts of the circuit. For example, an isolator is often used to protect a laser source from destabilizing feedback or damage from back-reflected light. The need for on-chip isolation becomes imperative as the level of photonic integration continues to scale, since unwanted reflections between the many integrated devices in a PIC are common and can be highly disruptive in a complex optical network.

Fundamentally, optical isolation requires breaking the time-reversal symmetry of light propagation. Such optical nonreciprocity can be realized by using magneto-optical (MO) effects, or nonmagnetic approaches such as indirect interband photonic transitions^{Reference Yu and Fan1,Reference Lira, Yu, Fan and Lipson2} and nonlinear optics.^{Reference Shi, Yu and Fan3} Traditional optical isolators used in free-space and fiber-optic systems are almost exclusively based on Faraday rotation in MO crystals, in which a magnetic field is applied along the light propagation direction to induce circular birefringence and polarization rotation. These Faraday rotation isolators typically feature an isolation ratio (the transmittance contrast between forward and backward propagating light) of the order of 30 to 40 dB, an insertion loss (loss of the forward propagating light passing through the isolator) below 1 dB, and an operation bandwidth exceeding 50 nm in the near-infrared (NIR) telecommunication window around 1.5-µm wavelength. Despite the high performance of these free-space isolators, they are bulky discrete devices not suitable for planar on-chip integration. In addition, at present, the price of a discrete isolator ranges anywhere from USD$30 up to more than USD$1,000, a prohibitive cost for large-scale PICs.

On-chip magneto-optical isolation can be achieved by Faraday rotation,^{Reference Karki, Stenger, Pollick and Levy4–Reference Zhang, Dulal, Stadler and Hutchings6} nonreciprocal phase shift (NRPS),^{Reference Dötsch, Bahlmann, Zhuromskyy, Hammer, Wilkens, Gerhardt, Hertel and Popkov7} nonreciprocal loss or gain,^{Reference Zaets and Ando8} and nonreciprocal mode conversion.^{Reference Montoya, Parameswaran, Hensley, Allen and Ram9–Reference Hammer, Ozgur, Evans and Butler11} The Faraday rotator configuration used in free-space isolators is sensitive to waveguide birefringence and is not suitable for on-chip optical isolation,^{Reference Levy5,Reference Dionne, Allen, Haddad, Ross and Lax12} unless a quasi-phase match scheme is implemented (Figure 1b).^{Reference Zhang, Dulal, Stadler and Hutchings6,Reference Hutchings, Holmes, Zhang, Dulal, Block, Sung, Seaton and Stadler13} Schemes based on nonreciprocal optical loss in an InGaAsP waveguide coated with iron^{Reference Shimizu and Nakano14} and active devices, including an amplifier to compensate for the loss,^{Reference Van Parys, Van Thourhout, Baets, Dagens, Decobert, Le Gouezigou, Make, Vanheertum and Lagae15} have been demonstrated. Nonreciprocal mode conversion,^{Reference Shintaku16} cutoff,^{Reference Tang, Drezdzon and Yoshie10,Reference Hemme, Dötsch and Hertel17} resonant delocalization,^{Reference El-Ganainy, Kumar and Levy18} and coupling^{Reference Montoya, Parameswaran, Hensley, Allen and Ram9,Reference Hammer, Ozgur, Evans and Butler11} have also been theoretically explored, although experimental realization of on-chip optical isolators and circulators based on these mechanisms has not yet been reported.

Figure 1.

(a) Schematic of a nonreciprocal optical resonator structure and scanning electron microscopy (SEM) cross-section image (inset) through the garnet-clad section of the Si waveguide.^{Reference Bi, Hu, Jiang, Kim, Dionne, Kimerling and Ross38} H denotes the direction in which the magnetic field was applied. (b) Schematic showing a Faraday rotation waveguide isolator and SEM cross section (inset) of the garnet-clad waveguide. Adapted with permission from Reference Zhang, Dulal, Stadler and HutchingsReference 6. © 2017 Nature Publishing Group. Note: HSQ, hydrogen silsesquioxane; RPC, reciprocal polarization converter; L _G and L _N, length of waveguides with and without magnetic material cladding, respectively. (c) Cross-sectional image of a fabricated one-dimensional magnetophotonic crystal consisting of magneto-optical garnet (CeYIG, cerium-doped yttrium iron garnet) between two SiO₂ and Ta₂O₅ Bragg mirrors on a substituted gadolinium gallium garnet (SGGG) substrate.^{Reference Yoshimoto, Goto, Isogai, Nakamura, Takagi, Ross and Inoue52} (d) Schematic and SEM cross section (inset) of a Mach–Zehnder transverse electric (TE)-mode isolator with CeYIG mesas adjacent to the side walls of the Si waveguide. Note: a-Si:H, hydrogenated amorphous silicon; TM, transverse magnetic. Adapted with permission from Reference Ishida, Miura, Shoji, Yokoi, Mizumoto, Nishiyama and AraiReference 56. © 2017 The Optical Society.

Unlike Faraday rotation, NRPS is implemented with magnetization perpendicular to the light propagation direction and is insensitive to birefringence.^{Reference Dötsch, Bahlmann, Zhuromskyy, Hammer, Wilkens, Gerhardt, Hertel and Popkov7} In a typical NRPS device configuration, an applied magnetic field perpendicular to light propagation direction biases the MO material and lifts the degeneracy between forward and backward propagating waveguide modes by inducing a difference in the effective refractive indices of the two modes. The directional dependence of waveguide mode effective index can be leveraged to realize asymmetric light transmission in various device platforms, including Mach–Zehnder interferometers (MZIs),^{Reference Yokoi, Mizumoto and Shoji19–Reference Shoji and Mizumoto21} microring resonators,^{Reference Kono, Kakihara, Saitoh and Koshiba22,Reference Zhu and Jiang23} multimode interferometers,^{Reference Yang, Roh, Ok, Woo, Byun, Lee, Mizumoto and Lee24,Reference Shui, Nie, Zhang, Peng, Xie, Deng and Bi25} and photonic crystals.^{Reference Kono and Koshiba26} Examples of devices based on a resonator, photonic crystal, and an interferometer are given in Figure 1a, c, and d. These device configurations suffer from limitations such as the large footprint of interferometers and the low bandwidth and high sensitivity to environment of resonators.

Development of on-chip isolators has been hampered by the incompatibility of traditional MO materials with monolithic integration on semiconductor substrate platforms. Unlike doped-glass-based fiber-optic isolators, the very limited real estate on a chip requires the use of materials with high Verdet constants (i.e., large Faraday rotation) in on-chip isolators. These materials, typically single-crystalline bismuth or cerium-doped yttrium iron garnet (BiYIG, Bi_xY_3–xFe₅O₁₂ and CeYIG, Ce_xY_3–xFe₅O₁₂), have attractive figures of merit ([FOM], the ratio of Faraday rotation to optical loss) exceeding 100 degrees/dB in the NIR (Figure 2f), but cannot be monolithically grown on common semiconductor substrates (Si or InP) due to large lattice and thermal expansion mismatch.^{Reference Boudiar, Payet-Gervy, Blanc-Mignon, Rousseau, Le Berre and Joisten27}

Figure 2.

(a) X-ray diffraction (ω − 2θ) scans of pulsed laser deposition cerium-doped yttrium iron garnet (CeYIG) and YIG multilayer garnet films after rapid thermal annealing at 800°C for 5 min on silicon, showing characteristic garnet peaks.^{Reference Sun, Du, Goto, Onbasli, Kim, Aimon, Hu and Ross46} (b) Scans of sputtered terbium iron garnet (TIG) on Si and quartz substrates, showing characteristic garnet peaks. Adapted with permission from Reference Dulal, Block, Gage, Haldren, Sung, Hutchings and StadlerReference 51. © 2016 American Chemical Society. (c) In-plane hysteresis loops and (d) out-of-plane Faraday rotation hysteresis loops at a wavelength (λ) of 1550 nm, with magnetic field and light propagation perpendicular to the film, for sputtered CeYIG on Si substrates annealed at 800°C for various times.^{Reference Goto, Onbasli, Kim, Singh, Inoue, Kimerling, Ross and Ross43} Inset in (c) is a zoomed-in view at a low magnetic field range, and shows the dependence of saturation Faraday rotation on annealing time. (e) Bright-field transmission electron microscope cross-sectional image of YIG/CeYIG/silica/Si, showing large-grained YIG top seed layer and fine-grained CeYIG.^{Reference Sun, Du, Goto, Onbasli, Kim, Aimon, Hu and Ross46} (f) Figure of merit (FOM) of single-crystal CeYIG films with different crystal orientations on a gadolinium gallium garnet (GGG) substrate, with magnetic field and light propagation perpendicular to the film.^{Reference Onbasli, Beran, Zahradník, Kučera, Antoš, Mistrík, Dionne, Veis and Ross44}

Two routes have been pursued to resolve the integration challenge—the hybrid approach, which relies on wafer bonding of garnet crystals,^{Reference Shoji, Mizumoto, Yokoi, Hsieh and Osgood28–Reference Pintus, Huang, Zhang, Shoji, Mizumoto and Bowers33} and the monolithic approach, which directly deposits polycrystalline ferromagnetic oxides or doped semiconductors or polymers^{Reference Zaman, Guo and Ram34–Reference Amata, Royer, Choueikani, Jamon, Parsy, Broquin, Neveu and Rousseau37} on substrates. The former approach can leverage the large Verdet constant and low material absorption in single-crystalline garnets to achieve good isolation performance; however, the wafer-bonding process required by this approach is expensive and limits fabrication throughput and yield.

Deposited polycrystalline ferromagnetic oxides, on the other hand, offer a promising path toward monolithic integration of optical isolators directly on semiconductor substrates with improved fabrication throughput and yield. One limitation with this approach is the inferior optical performance of these polycrystalline thin films, which so far exhibit lower Faraday rotation and higher optical absorption compared to their single-crystalline counterparts. The isolator performance is ultimately bounded by the material characteristics, and more specifically, the material MO FOM in units of degrees of rotation per dB loss. Therefore, improving the FOM of MO films is a critical research objective. In addition to enhancing material quality, further reducing the film growth and heat-treatment temperatures to ensure compatibility of the material deposition process with standard complementary metal oxide semiconductor (CMOS) back-end fabrication constitutes another important research thrust.

In this article, we review the progress to date in the growth and characterization of deposited MO garnet thin-film materials for NIR photonics applications. We specifically focus our attention on materials prepared using pulsed laser deposition (PLD) and magnetron sputtering, the two most commonly adopted deposition techniques for MO oxides. Integration of these materials into a planar device platform will be discussed, and the prospects toward efficient on-chip optical isolation will be evaluated in the context of material FOM and processing compatibility.

Monolithic MO thin films: Growth and characterization

The growth of high FOM garnets such as BiYIG and CeYIG requires processes that promote growth of the garnet phase over competing phases such as CeO₂ or Bi₂O₃, particularly on nongarnet substrates where there is no epitaxial stabilization of a garnet phase. Iron garnet phases have been grown directly on substrates such as silica, and use of seed layers of YIG combined with rapid thermal annealing (RTA) has been shown to promote the growth of MO garnets. Figure 2 shows examples of structural characterization, magnetic hysteresis, Faraday rotation, and FOM of MO garnet films on various substrates.

Garnet films have been used in demonstrations of integrated nonreciprocal photonic devices such as isolators and modulators.^{Reference Bi, Hu, Jiang, Kim, Dionne, Kimerling and Ross38–Reference Sun, Du, Goto, Onbasli, Kim, Aimon, Hu and Ross46} Bi et al.^{Reference Bi, Hu, Jiang, Kim, Dionne, Kimerling and Ross38} demonstrated the first monolithically integrated optical isolator on a silicon-on-insulator (SOI) platform (Figure 1a) in which the MO garnet was deposited by PLD on a single-mode silicon “racetrack” resonator through a window in the SiO₂ cladding layer. A 20-nm-thick YIG seed layer was deposited and crystallized by RTA, and then 80-nm CeYIG (composition Ce₁Y₂Fe₅O₁₂) was grown. The YIG layer stabilizes the CeYIG, and the use of RTA to crystallize the YIG seed layer significantly reduces the overall thermal budget for the garnet film stack. This is beneficial to reduce cracking in the garnet films and to prevent degradation of other parts of the photonic circuit.^{Reference Sung, Qi and Stadler47,Reference Sung, Qi and Stadler48} Films grown using this method exhibited high MO activity with Faraday rotation up to ∼3000 degrees/cm at 1550-nm wavelength (Figure 2d), compared to single-crystal values of 3300 degrees/cm^{Reference Hu, Sun, Du, Onbasli and Ross45} to 5500 degrees/cm,^{Reference Onbasli, Beran, Zahradník, Kučera, Antoš, Mistrík, Dionne, Veis and Ross44} and bulk-like saturation magnetization (Figure 2c).

Seed layers have also been applied in forming sputtered polycrystalline CeYIG films.^{Reference Stadler and Mizumoto32,Reference Hansen and Krumme49} The seed-layer process can be applied to various substrates, including silica and quartz, although oxidation of a silicon nitride-coated substrate was observed during PLD garnet growth.^{Reference Onbasli, Goto, Sun, Huynh and Ross50} Sun et al.^{Reference Sun, Du, Goto, Onbasli, Kim, Aimon, Hu and Ross46} introduced a simplified PLD process, by placing the YIG seed layer on top of the MO garnet, then crystallizing both at the same time by RTA. The film structure resulting from this process is shown in Figure 2e. This process gives direct contact between the MO garnet and the waveguide, which maximizes the coupling of light from the waveguide to the MO cladding. This avoids the reduction in net Faraday rotation caused by the opposite sign of FR in YIG versus BiYIG and CeYIG.

Rare-earth garnets have also been developed for MO applications. Dulal et al.^{Reference Dulal, Block, Gage, Haldren, Sung, Hutchings and Stadler51} sputter-deposited Bi-doped terbium iron garnet (BiTIG), which has the same sign of Faraday rotation as CeYIG and BiYIG, and with opposite sign of Faraday rotation as TIG. The TIG and BiTIG can be grown directly on Si and quartz without a seed layer (Figure 2b), and the magnitude of Faraday rotation was up to 500 degrees/cm at 1550 nm.

Garnet films have also been incorporated into multilayer photonic structures to form magnetophotonic crystals (MPCs). For example, Yoshimoto et al.^{Reference Yoshimoto, Goto, Isogai, Nakamura, Takagi, Ross and Inoue52} demonstrated a MPC structure consisting of a sputtered 423-nm-thick CeYIG polycrystalline film integrated between two Bragg mirrors of alternating layers of Ta₂O₅/SiO₂ (Figure 1c). The resulting Faraday rotation of the MPC was 30× higher (–2.92 degrees at 1570-nm wavelength) compared to a CeYIG film without the Bragg mirrors.

Device integration toward on-chip optical isolation

Correlating material FOM with device performance

For both Faraday rotators and NRPS-based isolators, performance is ultimately dictated by the FOM of the MO material. In Faraday rotators, the polarization plane of light has to be rotated by 45 degrees per pass, hence the insertion loss (IL) is given by:

(1)$$IL = 45/FOM \left( {{\rm{in\ dB}}} \right)$$.

For NRPS-based isolators, the waveguide design affects the NRPS (measured by phase delay per unit waveguide length in degrees/cm) and device performance as well.^{Reference Sun, Du, Goto, Onbasli, Kim, Aimon, Hu and Ross46} Figure 3c plots the ratio of waveguide NRPS and the MO material’s Faraday rotation (both with the unit of degrees/cm) in an SOI waveguide capped with an MO thin film whose mode profile and cross-sectional configuration are illustrated in Figure 3a–b, respectively. Note that for all waveguide dimensions, the waveguide NRPS is less than the material’s Faraday rotation.

Figure 3.

(a) Quasi-transverse magnetic (TM)-mode profile and (b) cross-sectional structure of a magneto-optical (MO) film capped silicon-on-insulator (SOI) waveguide; (c) variation in the ratio of a waveguide nonreciprocal phase shift and the Faraday rotation of the MO material across the SOI waveguide cross-section illustrated in (b); the dashed line delineates the waveguide single-mode (left) and multi-mode (right) regimes. (d) Simulated insertion losses of isolators versus the MO material figure of merit (FOM). Three different device configurations: microring, Faraday rotator, and Mach–Zehnder interferometers (MZIs) are compared. Here, the microring and MZI isolators consist of SOI waveguides shown in (b) with a core width of 450 nm and a height of 220 nm. The dashed line corresponds to the best experimentally assessed FOM in monolithically integrated cerium-doped yttrium iron garnet (CeYIG) films.

Figure 3d presents the computed ILs of three types of isolator devices (Faraday rotators, MZIs, and microrings, where the latter two are both based on NRPS) versus the material FOM. Here the devices considered are ideal—for example, the only source of optical loss results from MO material attenuation, and birefringence of the rotator is assumed to vanish. A semistandard SOI waveguide with dimensions of 450 nm (width) × 220 nm (height) is used for the MZI and microring devices. In all cases, the isolation ratio can, in theory, reach infinity (e.g., for a critically coupled microring or for a power-balanced MZI), and therefore, IL is the proper device performance measure. Figure 3d indicates that Faraday rotators offer the best-in-class performance for a given MO material FOM, provided that their sensitivity to birefringence is properly addressed. This is a direct consequence of the less-than-unity ratio between waveguide NRPS and material Faraday rotation shown in Figure 3c.

Figure 3d allows us to project the performances of monolithic isolators based on available deposited MO thin films. Reliable quantification of FOM can be challenging as accurate loss evaluation in low-attenuation (extinction coefficient k < 10^–3) MO thin films is only possible using waveguide measurements. Taking the best experimentally assessed loss data of 40 dB/cm^{Reference Bi, Hu, Dionne, Kimerling and Ross54} and Faraday rotation of 2650 degrees/cm^{Reference Goto, Onbasli, Kim, Singh, Inoue, Kimerling, Ross and Ross43} in PLD-deposited CeYIG films, we obtain a material FOM of 66 degrees/dB. The metric corresponds to ILs of 0.11 dB (microring), 0.54 dB (Faraday rotator), and 1.1 dB (MZI). The result highlights the viability of achieving high-performance optical isolation based on monolithically deposited MO thin films.

Summary

Integrated isolators present a challenge due to the need for waveguide-compatible designs with high bandwidth, good fabrication tolerance, polarization diversity, and a small footprint, coupled with the requirement for incorporating magneto-optical materials with a high FOM. Progress, both in device designs based on nonreciprocal phase shift or Faraday rotation as well as the growth of substituted iron garnet films on photonic substrates, have enabled demonstrations of monolithically integrated isolators and suggest the feasibility of manufacturing high-performance integrated devices.