Subwavelength Phase Engineering

• Subwavelength phase engineering is a design strategy that uses structures smaller than the wavelength to control wave phase, amplitude, and dispersion with precision.
• It employs various mechanisms such as effective medium synthesis, resonance engineering, and geometric-phase control to achieve full 0–2π phase coverage and beam steering.
• This approach enables advanced applications in integrated photonics, acoustic devices, and dynamic imaging while addressing challenges in fabrication and error tolerance.

Subwavelength phase engineering denotes the deliberate control of wave phase—and, in many implementations, the associated amplitude, dispersion, polarization, or radiation-channel interference—by using structures or modulations whose critical dimensions are below the relevant wavelength or Bragg scale. In this regime, the medium can act as an effective homogeneous material, a locally resonant cavity, a symmetry-organized scattering system, or a near-field converter that transfers information between evanescent and propagating channels. The concept appears across dielectric superscatterers, subwavelength grating waveguides, metasurfaces, freeform flat optics, photorefractive microcavities, magnonic resonators, acoustic labyrinths, matter-wave lattices, and wave-vortex systems (Krasikov et al., 2020, Badri et al., 8 May 2026, González-Andrade et al., 2019, Li et al., 2014).

1. Concept and scope

The term subwavelength is medium-dependent. In buried silicon metaoptics, the in-medium wavelength is $\lambda_{\mathrm{in}}=\lambda_0/n_{\mathrm{Si}}$, and with $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$ one obtains $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$; pitches of $300$–$410\,\mathrm{nm}$ are therefore subwavelength inside silicon (Bütün et al., 28 Jul 2025). In integrated silicon photonics, a subwavelength grating core with pitch $\Lambda=275\,\mathrm{nm}$ suppresses diffraction and Bragg reflection so that the waveguide behaves as a homogeneous metamaterial with an effective index set by fill factor and geometry (Badri et al., 8 May 2026). In acoustic reflection control, the unit-cell cross-section was chosen as $a_x=a_y=\lambda/8$, making the interface an ultrathin phase mask relative to the acoustic wavelength (Li et al., 2014).

The concept is not restricted to a single physical mechanism. In some platforms, subwavelength phase engineering is essentially effective-medium synthesis, with phase accumulation set by a designed $n_{\mathrm{eff}}$. In others it is resonance engineering, in which small geometric or material variations produce large phase excursions near a Fabry–Perot, Gires–Tournois, multipolar, or guided-mode resonance. It can also be geometric-phase control, where local element orientation contributes a Pancharatnam–Berry phase, or Fourier-spectrum control, where selected Fourier harmonics of a periodic modulation are suppressed to eliminate unwanted diffraction orders beyond the conventional subwavelength regime (Chen et al., 2020, cao et al., 2020, Lee et al., 2021).

The requirement of full local $0$–$2\pi$ phase coverage is likewise not universal. A distinct metasurface formulation based on partial control of phase demonstrated that phase shifts covering less than the full $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$0–$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$1 range can still support extreme-angle deflection and immersion metalenses with $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$2, including $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$3 efficiency for dielectric metasurfaces (Hail et al., 2018). This suggests that “subwavelength phase engineering” is best understood as a family of design strategies for manipulating the complex wave response below the scale at which conventional geometric optics or coarse grating pictures remain sufficient.

2. Governing formalisms

A central description is phase accumulation through an effective medium or guided section,

$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$4

This appears explicitly in subwavelength grating filters, where changing the crystallization fraction of a phase-change material modifies $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$5 and therefore the Bragg response, and in passive phase shifters, where the relevant quantity is the differential phase $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$6 with $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$7 (Badri et al., 8 May 2026, González-Andrade et al., 2019). In the ultra-broadband SWG phase shifter, the flatness criterion is

$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$8

so the design objective becomes group-index equalization rather than merely setting a target phase at one wavelength (González-Andrade et al., 2019).

A second formalism is symmetry-organized scattering. For non-spherical resonators, incident and scattered multipolar amplitudes satisfy

$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$9

with the $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$0-matrix block-diagonal in a basis adapted to irreducible representations of the scatterer’s point group. The optical theorem,

$\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$1

connects total extinction to the complex forward-scattering amplitude. For lossless particles, $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$2, so constructive phase alignment of the relevant multipolar channels maximizes $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$3 and therefore the scattering cross-section (Krasikov et al., 2020).

A third formalism is Jones-matrix control of co- and cross-polarized channels. For a rotated anisotropic nanopillar,

$\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$4

with retardance $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$5. In the circular basis, the cross-polarized channel acquires a geometric phase $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$6, while the co-polarized channel retains propagation phase. This separation permits simultaneous wavefront and polarization control in a single metasurface (Chen et al., 2020).

Phase gradients then connect local phase design to beam steering or focusing. In optical freeform metasurfaces,

$\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$7

while in the acoustic reflective formulation the corresponding relation is

$\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$8

These equations formalize the idea that engineered subwavelength phase discontinuities act as momentum-transfer interfaces (Zhan et al., 2016, Li et al., 2014).

3. Resonant phase control in compact scatterers and cavities

A recurrent route to strong phase control is to exploit a local resonance whose phase response is steep while losses remain low. In symmetry-engineered dielectric superscattering, a finite-height ceramic cylinder with $\lambda_{\mathrm{in}}\approx 445\,\mathrm{nm}$9 symmetry was tuned so that an electric-dipolar-like $300$0 mode and a magnetic-dipolar-like $300$1 mode overlapped spectrally and interfered constructively in the forward direction. The optimized homogeneous resonator exceeded the dipolar single-channel scattering limit for a sphere by up to a factor of four, with measured and simulated maxima above three over several aspect ratios. The same analysis generalized the non-spherical partial-channel bound to

$300$2

twice the spherical single-channel bound, because multipoles mix within symmetry-defined blocks (Krasikov et al., 2020).

An asymmetric Fabry–Perot cavity provides a different resonant implementation. A nanopost on a distributed Bragg reflector behaves as a Gires–Tournois all-pass etalon: in the ideal lossless limit the reflected amplitude remains unity while the phase varies rapidly near resonance. In the reported silicon nanopost platform, less than $300$3 refractive-index modulation, from $300$4 to $300$5, produced nearly a full $300$6 phase swing with amplitude close to unity, enabling lenses, axicons, and vortex generators on a subwavelength lattice (Colburn et al., 2016).

The same cavity logic extends into the bulk of crystalline silicon. Laser-written buried meta-atoms with width $300$7 and pitch $300$8–$300$9 were modeled as low-finesse Fabry–Perot slabs embedded in silicon. By tuning the effective index through pitch while keeping the modified height fixed at $410\,\mathrm{nm}$0, the design achieved full $410\,\mathrm{nm}$1 phase coverage, simulated individual meta-atom transmission up to $410\,\mathrm{nm}$2, and a buried metalens focusing efficiency of about $410\,\mathrm{nm}$3 (Bütün et al., 28 Jul 2025).

Resonant phase amplification is not confined to optics. A magnonic Gires–Tournois interferometer formed by a permalloy film and a ferromagnetic stripe used spin-wave Fabry–Perot resonances of a slow bilayer mode to control the phase of reflected spin waves at subwavelength distances. At $410\,\mathrm{nm}$4 the reflection phase exhibited approximately $410\,\mathrm{nm}$5 jumps every $410\,\mathrm{nm}$6 as the stripe width changed, yielding full $410\,\mathrm{nm}$7 coverage and demonstrating a magnonic counterpart of a metasurface (Sobucki et al., 2020). Collectively, these examples indicate that resonance is often used not to create phase control from nothing, but to amplify an otherwise weak geometric or material perturbation into a usable subwavelength phase response.

4. Guided-wave, metasurface, and grating implementations

In integrated photonics, subwavelength phase engineering often appears as decoupled control of phase velocity, dispersion, and periodic perturbation strength. An ultra-broadband passive $410\,\mathrm{nm}$8 phase shifter based on subwavelength metamaterial waveguides used anisotropy and Floquet–Bloch dispersion engineering to obtain a phase shift error below $410\,\mathrm{nm}$9 over $\Lambda=275\,\mathrm{nm}$0–$\Lambda=275\,\mathrm{nm}$1, with a measured phase slope of about $\Lambda=275\,\mathrm{nm}$2 and insertion loss below $\Lambda=275\,\mathrm{nm}$3 (González-Andrade et al., 2019). A different SWG platform used a silicon core with evanescently coupled GST loading segments in the cladding; for $\Lambda=275\,\mathrm{nm}$4, $\Lambda=275\,\mathrm{nm}$5, and $\Lambda=275\,\mathrm{nm}$6, the central wavelength was about $\Lambda=275\,\mathrm{nm}$7 in the amorphous state, and switching to crystalline GST produced an extinction ratio of $\Lambda=275\,\mathrm{nm}$8. Partial crystallization yielded a blueshift of more than $\Lambda=275\,\mathrm{nm}$9, demonstrating nonvolatile reconfigurability through controlled phase accumulation and loss modulation (Badri et al., 8 May 2026).

Planar metasurfaces realize analogous control in free space. A silicon nitride metasurface quarter-wave plate in the blue combined propagation phase and Pancharatnam–Berry phase to control co- and cross-polarized channels simultaneously; meta-holograms reached $a_x=a_y=\lambda/8$0 diffraction efficiency, while a metalens with $a_x=a_y=\lambda/8$1 gave an experimental focusing efficiency of $a_x=a_y=\lambda/8$2 and an $a_x=a_y=\lambda/8$3-polarized focus with $a_x=a_y=\lambda/8$4 under right-circular polarization (Chen et al., 2020). A freeform visible metasurface platform based on cylindrical Si$a_x=a_y=\lambda/8$5N$a_x=a_y=\lambda/8$6 nanoposts implemented a cubic phase plate with enhanced depth of field over $a_x=a_y=\lambda/8$7 and an Alvarez lens with a tunable focal-length range of over $a_x=a_y=\lambda/8$8 for $a_x=a_y=\lambda/8$9 of total displacement (Zhan et al., 2016). In high-contrast infrared optics, the annular groove phase mask used a concentric subwavelength grating in synthetic diamond with $n_{\mathrm{eff}}$0 to realize a vector vortex coronagraph; after RCWA-guided re-etch optimization, L-band starlight rejection reached $n_{\mathrm{eff}}$1 (Catalan et al., 2016). Partial-phase metasurfaces established that high-angle deflection and immersion focusing need not rely on full local $n_{\mathrm{eff}}$2–$n_{\mathrm{eff}}$3 phase coverage, reaching $n_{\mathrm{eff}}$4 and $n_{\mathrm{eff}}$5 dielectric efficiency at visible wavelengths (Hail et al., 2018).

Periodic phase engineering also includes explicit control of diffraction pathways. In transmission-type slit metasurfaces based on local Fabry–Perot resonances, each slit was filled with the same dielectric permittivity but different heights, and adjacent resonant slits acquired a constant phase difference

$n_{\mathrm{eff}}$6

allowing full $n_{\mathrm{eff}}$7–$n_{\mathrm{eff}}$8 coverage across a supercell with near-unity transmission; a representative $n_{\mathrm{eff}}$9 design achieved $0$0 in the anomalous channel (cao et al., 2020). Fourier-component engineering generalized this idea beyond the conventional subwavelength limit by showing that resonant diffraction is governed by the superposition of scattering processes mediated by higher Fourier harmonics. By setting selected $0$1 to zero, unwanted diffraction orders can be suppressed even when $0$2, enabling bound states in the continuum and highly efficient zero-order spectral responses beyond the subwavelength regime (Lee et al., 2021).

Near-field phase coding at the aperture level extends the same principle to emitters. A silicon nanophotonic antenna on a $0$3 SOI platform used transversally interleaved subwavelength gratings to introduce a specific near-field phase factor in the Fraunhofer transform. The resulting antenna combined $0$4 radiation efficiency, a footprint of $0$5, and far-field beamwidths of $0$6 and $0$7 in the longitudinal and transversal directions, respectively (Khajavi et al., 2022). Across these platforms, subwavelength phase engineering is less a single architecture than a design grammar for distributing phase, resonance, and coupling among features too small to be interpreted as ordinary optical elements.

5. Imaging, reconstruction, and dynamically written phase structures

Subwavelength phase engineering can also be used to recover or transform information that would otherwise be lost to diffraction. A solid-immersion diffractive optical processor placed the object on a high-index slab, allowing spatial frequencies up to $0$8 to remain propagating inside the slab. A learned diffractive encoder converted those high-$0$9 components into lower spatial-frequency modes that propagate in air, and decoder layers reconstructed a magnified image. At $2\pi$0, the system experimentally resolved phase-line gratings with linewidths of about $2\pi$1 and $2\pi$2; in simulations, increasing the slab index and reducing encoder feature size to about $2\pi$3 enabled resolution near $2\pi$4 at $2\pi$5 (Hu et al., 2024).

Time reversal offers a complementary route. In optical phase conjugation of multiply scattered light, a subwavelength NSOM source first had its near field scattered by random ZrO$2\pi$6 nanoparticles, converting evanescent information into propagating far-field channels. About $2\pi$7 spatial-frequency modes were optimized for coupling into a single-mode reflector, increasing the detector signal by about $2\pi$8. When the conjugated beam was launched back, it retraced the multiple-scattering paths and regenerated a focus with $2\pi$9 at $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$00, compared with a diffraction-limited control focus of about $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$01 in a homogeneous medium (Park et al., 2016).

Dynamic or optically written subwavelength phase patterns push the concept further. In a thin-film lithium niobate microdisk, two counter-propagating pumps formed a standing wave whose photorefractive space-charge field wrote a subwavelength grating with period

$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$02

For a $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$03 pump with $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$04, this gave $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$05. The resulting grating coupled CW and CCW whispering-gallery modes, producing mode splitting up to $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$06 and reflection near $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$07, and it enabled first-order quasi-phase-matching for backward second-harmonic generation with about $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$08 enhancement (Hou et al., 2024). In Bose–Einstein condensates, a short 2D optical-lattice pulse imprinted a phase $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$09; when the phase excursion within a lattice cell exceeded $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$10, principal-value phase wraps produced subwavelength phase structure. The thresholds were $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$11 for $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$12 and $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$13 for $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$14, with the resulting sub-cell structure inferred from higher-order momentum populations after time of flight (Wen et al., 2019).

A different dynamical manifestation is the type-II vortex around an oscillating subwavelength hole. In water-wave experiments, a dipole-like hole field interfered with a single incident plane wave to produce a phase winding of $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$15 around the excluded region, even though the physical field outside the hole need not vanish. The topological charge

$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$16

and the orbital angular momentum were both controlled by the relative phase $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$17 between the dipole and incident wave, with handedness switching under $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$18 (Ye et al., 29 Apr 2026). These cases show that subwavelength phase engineering can mean not only imposing a target phase map, but also converting, retrieving, or dynamically writing phase information that is initially inaccessible in the far field.

6. Limits, misconceptions, and research directions

A persistent misconception is that effective-medium theory makes subwavelength order and disorder optically indistinguishable. Deep-subwavelength multilayers organized by stealthy hyperuniformity show otherwise. For TE waves at $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$19 in a system with $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$20, $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$21, and $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$22, the critical angles are about $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$23, $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$24, and $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$25. In this regime, interface phase effects and Goos–Hänchen-like contributions dominate transport, causing substantial deviations from EMT predictions and enabling angle-selective localization control even when layer thicknesses are deep-subwavelength (Park et al., 2024).

A second misconception is that superscattering or beyond-diffraction performance necessarily implies a violation of fundamental limits. In multipolar dielectric resonators, the enlarged scattering response does not evade constraints; it reflects a different organization of channels in low-symmetry structures, where multipoles mix within irreducible-representation blocks and the appropriate upper bound differs from the spherical partial-wave bound (Krasikov et al., 2020). A related misconception is that full local $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$26–$n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$27 phase coverage is mandatory for high-performance flat optics. Partial-phase metasurfaces demonstrate otherwise, but their performance depends on careful control of the interacting area, scatterer compactness, and mutual coupling (Hail et al., 2018).

The practical limits recur across platforms. Miniaturization narrows bandwidth in compact resonators and antennas, even when overlapping resonances can partially mitigate the penalty (Krasikov et al., 2020). In PCM-loaded waveguides, higher $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$28 improves extinction but increases insertion loss and constrains practical reconfigurability; in the reported device, partial crystallization beyond $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$29 was limited by rising absorption (Badri et al., 8 May 2026). In SWG phase shifters, fabrication errors of $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$30 still kept the phase-shift error within $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$31, but this robustness depended on wide guides and $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$32 duty cycle (González-Andrade et al., 2019). By contrast, the diamond AGPM was highly sensitive to small errors: $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$33 or $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$34 could strongly degrade starlight rejection, motivating post-fabrication re-etch tuning (Catalan et al., 2016). Solid-immersion diffractive imaging remains limited by proximity, material loss, and encoder feature size (Hu et al., 2024), while buried silicon metaoptics require control of sidewall roughness to about $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$35 and show Monte Carlo phase-error levels of about $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$36 under concurrent geometric and refractive-index variations (Bütün et al., 28 Jul 2025).

Current directions follow directly from these constraints. Buried volumetric metaoptics aim to preserve a pristine wafer surface while adding internal optical functionality (Bütün et al., 28 Jul 2025). PCM-loaded SWG devices point toward nonvolatile, multi-level reconfigurable filters and phase shifters (Badri et al., 8 May 2026). Solid-immersion diffractive processors motivate higher-index, lower-loss materials and smaller feature sizes for operation further below $n_{\mathrm{Si}}(1.55\,\mu\mathrm{m})\approx 3.48$37 (Hu et al., 2024). Deep-subwavelength stealthy-hyperuniform structures suggest that interface-phase design can bridge disordered photonics and metamaterials (Park et al., 2024). Taken together, the literature indicates that subwavelength phase engineering has evolved from a metasurface-specific phrase into a general framework for controlling how waves accumulate, exchange, convert, and refocus phase when the relevant structure is smaller than the wavelength but not simpler than the wave physics.