Metamaterial Silicon Waveguides

• Metamaterial silicon waveguides are optical structures defined by subwavelength silicon-dielectric patterns that engineer anisotropic effective indices for precise modal confinement and dispersion control.
• They achieve significant performance gains, such as over 10 dB crosstalk reduction and reduced bending loss, enabling dense photonic integration and nonlinear optical applications.
• Advanced CMOS-compatible fabrication using high-resolution lithography ensures precise control of critical dimensions, supporting applications in sensing, frequency comb generation, and hybrid photonic systems.

Metamaterial silicon waveguides are optical waveguide structures in which subwavelength patterning of silicon and dielectric materials imposes new, designer effective optical properties, enabling unprecedented control over modal confinement, dispersion, crosstalk suppression, and device functionality within the silicon photonics platform. These waveguides exploit the effective-medium regime, where periodic structuring on a scale much smaller than the wavelength yields a tailored anisotropic dielectric response, inaccessible in homogeneous or conventionally patterned dielectric waveguides. The integration of metamaterial engineering with CMOS-compatible silicon platforms is central to dense photonic integration, nonlinear optics, sensing, and interfacing with external and hybrid photonic systems.

1. Fundamental Principles and Effective-Medium Description

Metamaterial silicon waveguides derive their functional properties from subwavelength structuring, such as alternating layers or rods of Si and SiO₂ (or air), forming an effective uniaxial medium characterized by a dielectric tensor. In the deep-subwavelength limit (lattice period $\Lambda \ll \lambda_0$), the region is described by an anisotropic permittivity tensor: $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$ with fill factor $f = w/\Lambda$ (for rod or layer width $w$), and

$$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$

Typical values at $1.55\,\mu$m: $\varepsilon_\perp \approx 7.4$, $\varepsilon_{||} \approx 3.5$ for $f\approx0.4$–$0.5$ (Khavasi et al., 2016). This anisotropy fundamentally modifies boundary conditions for electromagnetic field decay, directly engineering the evanescent "skin depth" and supporting much tighter or tailored modal confinement than isotropic claddings.

In multi-refractive-index metamaterials (MRIMs), spatial multiplexing of dissimilar single-mode silicon waveguides with different widths results in a composite that's not describable by a single effective index, but rather supports beams with distinct phase velocities $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$0, each determined by modal analysis of the constituent waveguide strips (Yu et al., 2018).

2. Device Architectures and Fabrication

The dominant device architectures include: (a) strip waveguides with metamaterial claddings or gap regions, (b) suspended silicon cores with subwavelength lateral claddings, (c) metamaterial slabs in both core and cladding ("dual-metamaterial" geometries), and (d) patterned or multi-channel arrays realizing MRIMs. Critical design parameters include lattice period (typically $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$1 nm), fill factor $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$2 (engineered via feature width and gap), and the overall arrangement of metamaterial and bulk regions.

Representative fabrication steps are:

• Electron-beam lithography (100–400 keV) with HSQ (hydrogen silsesquioxane) or similar resists achieves $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$310–20 nm critical dimension control.
• Fully CMOS-compatible pattern transfer via ICP-RIE or Cl₂/O₂ reactive-ion etching; a single etch can define both core and metamaterial features (Khavasi et al., 2016, Dinh et al., 2021, Dinh et al., 2021).
• In suspended THz or mid-IR geometries, vapor-phase HF etching removes the buried SiO₂, exploiting engineered SWG claddings for both optics (index contrast) and mechanical support (Dinh et al., 2021, Dinh et al., 2022).
• Transition tapers and adiabatic mode-converters (e.g., from conventional strip into SWG sections) minimize insertion loss (Farnesi et al., 2021).

Feature sizes down to 50 nm and period control within ±5 nm are routinely implemented, ensuring minimal modal mismatch and consistent effective-index engineering.

3. Modal Engineering, Crosstalk Suppression, and Bending Loss

The introduction of anisotropic, all-dielectric metamaterial claddings or gap regions leads to orders-of-magnitude suppression of evanescent coupling and bend-induced radiative loss compared to conventional device stacks:

• Coupling length ($$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$4) in directional couplers is increased up to $\sim$69%%%%214%%%% by replacing a silica gap with a metamaterial gap, with $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$7 rising from $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$80.35 mm to $$\varepsilon = \mathrm{diag}(\varepsilon_\perp, \varepsilon_{||}, \varepsilon_\perp)$$93.3–3.9 mm at $f = w/\Lambda$0m, reflecting a $f = w/\Lambda$1 dB crosstalk reduction for $f = w/\Lambda$2m waveguide pitches (Khavasi et al., 2016).
• In lateral-cladding designs ("e-skid" waveguides) with five periods ($f = w/\Lambda$3 nm), crosstalk suppression exceeds $f = w/\Lambda$4 relative to conventional strip guides, and bend loss is reduced by $f = w/\Lambda$5 (e.g. from $f = w/\Lambda$6 dB/turn down to $f = w/\Lambda$70.3 dB/turn at $f = w/\Lambda$8m), with propagation loss maintained at $f = w/\Lambda$9 (Jahani et al., 2017).
• In MRIM geometries, spatial isolation and independent modal velocities enable simultaneous multi-angle beam deflection, dual-focus lenses, and multi-resonance etalons, with effective index and beam fraction determined by channel geometries and numerically by modal solve (Yu et al., 2018).

These advances underpin the scaling of photonic integrated circuits to sub-$w$0m dimensions without crosstalk and with negligible insertion penalty over $w$1 bandwidths (Khavasi et al., 2016, Jahani et al., 2017, Cabo et al., 2023).

4. Dispersion and Nonlinear Optical Control

Metamaterial silicon waveguides offer new degrees of freedom for dispersion engineering and nonlinear optical processes:

• Subwavelength-grating (SWG) metamaterial claddings enable independent and broad tuning of zero-dispersion wavelengths, soliton phase-matching points, and phase velocity, crucial for frequency comb and supercontinuum generation (Dinh et al., 2022).
• By adjusting core width $w$2 and cladding gap $w$3, researchers demonstrate independent control of two dispersive-wave emission wavelengths, enabling ultra-broad ($w$4–$w$5m) on-chip supercontinuum—currently the widest in silicon photonics (Dinh et al., 2022).
• Dual-metamaterial waveguides—patterned in both core and cladding—enable independent control of vertical and lateral index contrast, raising the external mode overlap fraction with upper cladding from $w$6 (strip) to $w$7 (dual-metamaterial), enhancing sensitivity for sensing and hybrid integration, and allowing high-Q operation in ring resonators ($w$8) (Dinh et al., 2021).

Mid-IR and THz operation is enabled by fully suspended cores surrounded by metamaterial claddings, eliminating buried oxide absorption while maintaining strong confinement and losses below $w$9 at $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$0m (Dinh et al., 2021).

5. Functional Elements: Power Splitters, Universal Couplers, and MRIM Devices

Metamaterial silicon waveguides underpin several advanced functional photonic elements:

• Symmetric Y-junctions, using SWG stems and arms, achieve $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$1 excess loss (TE$$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$2) over $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$3 bandwidth at sub-100 nm minimum feature size, with robust fabrication tolerance ($$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$4 bias tolerance, $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$5 dB penalty) and low first-order mode excess loss over extended bandwidths (Cabo et al., 2023).
• SWG tapers facilitate universal coupling between planar silicon waveguides and bulk whispering-gallery-mode (WGM) resonators covering resonator indices from $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$6–$$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$7 and sizes $$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$8–$$\varepsilon_\perp = f\,\varepsilon_\text{Si} + (1-f)\,\varepsilon_\text{SiO₂} \ , \quad \dfrac{1}{\varepsilon_{||}} = \frac{f}{\varepsilon_\text{Si}} + \frac{1-f}{\varepsilon_\text{SiO₂}}$$9 mm, with coupling efficiencies of $1.55\,\mu$0–$1.55\,\mu$1 and loaded $1.55\,\mu$2 up to $1.55\,\mu$3, opening heterogeneous integration with LiNbO$1.55\,\mu$4, CaF$1.55\,\mu$5, and SiO$1.55\,\mu$6 devices (Farnesi et al., 2021).
• MRIM structures constructed from parallel arrays of silicon strips of distinct widths and spacings yield devices (prisms, lenses, etalons) with multiple effective refractive indices, supporting multi-angle deflection and enhanced resonance densities. Fabrication on SOI by direct etching, with minimal crosstalk due to deep subwavelength gap sizing, enables CMOS-compatible multi-functional refractive photonics (Yu et al., 2018).

6. Theoretical and Numerical Frameworks

Modeling of metamaterial silicon waveguides is grounded in a combination of:

• Analytic effective-medium theory (EMT) for deep-subwavelength unit-cell lattices, providing expressions for effective permittivity and refractive index as functions of fill factor and lattice geometry (Khavasi et al., 2016, Sifat et al., 2016, Dinh et al., 2021).
• Eigenmode analysis (e.g., Lumerical MODE, COMSOL) to solve for supermode propagation constants ($1.55\,\mu$7), effective indices, modal profiles, and coupling coefficients (Khavasi et al., 2016).
• Full-3D FDTD for full dispersion, loss, mode overlap, and extraction of quantities such as external confinement fraction $1.55\,\mu$8, group index $1.55\,\mu$9, and supercontinuum phase-matching (Dinh et al., 2021, Dinh et al., 2022).
• Coupled-mode theory for directional coupling and crosstalk analysis, and analysis of beat length, $\varepsilon_\perp \approx 7.4$0, where $\varepsilon_\perp \approx 7.4$1 (Khavasi et al., 2016).
• Calculation of normalized mode area $\varepsilon_\perp \approx 7.4$2 for plasmonic and deeply confined regimes, as well as energy overlap in engineered-splitting devices (Sifat et al., 2016, Cabo et al., 2023).

In SWG-assisted universal couplers and MRIMs, detailed band-structure and Bloch-mode analysis is employed, supplemented with overlap integrals for coupling optimization (Farnesi et al., 2021, Yu et al., 2018).

7. Integration, Limitations, and Outlook

Metamaterial silicon waveguides are inherently CMOS-compatible, relying on processes and materials already established for commercial silicon photonics. Limitations are dictated by:

• Minimum feature size ($\varepsilon_\perp \approx 7.4$3 nm in advanced e-beam; $\varepsilon_\perp \approx 7.4$4 nm in standard DUV stepper); this constrains unit-cell period and fill-factor accuracy (Cabo et al., 2023, Khavasi et al., 2016).
• Fabrication variability; device performance (e.g., coupling length, insertion loss, mode overlap) is robust to $\varepsilon_\perp \approx 7.4$5 nm dimensional errors, with measured device response changing by $\varepsilon_\perp \approx 7.4$6 in external confinement or $\varepsilon_\perp \approx 7.4$7 dB in splitter loss (Dinh et al., 2021, Cabo et al., 2023).
• For deeply subwavelength and multi-refractive-index structures, lithographic variation and sidewall roughness determine low-loss operation ($\varepsilon_\perp \approx 7.4$82–4 dB/cm typical) (Jahani et al., 2017, Dinh et al., 2021).
• Bandwidth is fundamentally broadband, tied to non-resonant effective-medium physics (typ. $\varepsilon_\perp \approx 7.4$9100 nm), exceeding performance of photonic crystal or inverse-designed structures at comparable loss (Jahani et al., 2017).

Future directions include further scaling of subwavelength periodicity ($\varepsilon_{||} \approx 3.5$0 nm), advanced multi-index structures for programmable optics, extension into hybrid photonic-electronic regimes, and on-chip access to the full transparency window of silicon (mid-IR to telecom) (Dinh et al., 2022, Dinh et al., 2021, Jahani et al., 2017). The use of metamaterial engineering for fully tailorable dispersion, crosstalk, and modal profiles will remain central to high-density photonic integration, nonlinear/mid-IR photonics, and the creation of universal interconnects and interfaces across disparate platforms.