Dual-Stack Metamaterial Design

• Dual-stack metamaterial design is a strategy that integrates two functional metamaterial layers to broaden bandwidth and merge multiple resonances.
• It leverages mechanisms such as cascaded anisotropy, eigenmode hybridization, and symmetry breaking to tailor electromagnetic or thermal responses.
• Practical implementations in W-band retarders, MRI coils, and thermal pixels demonstrate scalable applications across microwave, infrared, and optical regimes.

Dual-stack metamaterial design denotes a family of architectures in which two stacked, cascaded, or co-designed metamaterial subsystems are used to broaden bandwidth, realize dual-frequency operation, merge multiple resonances, or decouple functions such as heating and emission. In published work, the term spans rotated anisotropic retarders in the W-band, orthogonal strip-array resonators for preclinical dual-nuclei MRI, conductor-backed dual-band Huygens metasurfaces for Ku-band steering, step-like epsilon-near-zero multilayers, material–structure integrated absorbers, and graphene microheater–metasurface thermal pixels (Mohamed et al., 2014, Hurshkainen et al., 2017, Budhu et al., 2021, Sun et al., 2013, Peng et al., 2021, Zhong et al., 4 Jun 2025).

1. Taxonomy of dual-stack architectures

The expression is not attached to one canonical geometry. In some works it denotes literal vertical stacking of two or more patterned layers separated by spacers; in others it denotes two functionally distinct resonant subsystems co-designed within one device. A related usage appears in silicon photonics, where both the core and the lateral cladding are patterned metamaterials rather than a single strip surrounded by homogeneous media (Dinh et al., 2021).

Domain	Stack definition	Reported function
W-band polarization optics	Rotated anisotropic DBT plates with air gaps	Achromatized half-wave retardance in W-band (Mohamed et al., 2014)
Preclinical MRI	Long-wire resonator plus short-wire resonator, driven by one non-resonant loop	Simultaneous tuning and matching at two Larmor frequencies (Hurshkainen et al., 2017)
Microwave absorption	Patterned lossy top layer over spacer and metal-backed bottom stack	Ultra-broadband absorption from 5.3 to 18 GHz (Peng et al., 2021)
Ku-band beam steering	Electric and magnetic meta-atoms on opposing sides of a dielectric	Dual-band transmission and reflection steering (Cho et al., 2022)
Thermal infrared pixels	Monolayer graphene microheater beneath an Au metasurface emitter	Narrowband ultrafast thermal emission (Zhong et al., 4 Jun 2025)
Silicon nanophotonics	Patterned core and patterned lateral cladding	Increased calculated overlap with air while preserving measured quality factor near 30,000 (Dinh et al., 2021)

This diversity is substantive rather than terminological. In the W-band retarder, the stack is a cascade of rotated birefringent plates; in the MRI coil, it is a pair of orthogonal multimode strip resonators inside a shield; in the thermal pixel, it is a transparent heater beneath a spectrally selective emitter. What remains common is deliberate partition of electromagnetic or electrothermal tasks across two interacting substructures.

2. Electromagnetic principles

A first recurrent mechanism is retardance flattening by cascaded anisotropy. In the W-band half-wave-plate work, each dog bone triplet plate is anisotropic, with positive refractive index along one axis and negative refractive index along the orthogonal axis. The single-plate retardance is

$$\delta(\omega)=\frac{2\pi}{\lambda}\,\Delta n(\omega)\,t,$$

and for two cascaded plates with relative angle $\phi$ the equivalent retardance satisfies

$$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$

The same work uses a three-plate Pancharatnam arrangement with $\theta_1=+30^\circ$, $\theta_2=-29^\circ$, $\theta_3=+30^\circ$ to flatten $\delta(\omega)$ near $\pi$ (Mohamed et al., 2014).

A second mechanism is eigenmode hybridization in coupled resonant arrays. The dual-nuclei MRI coil operates through resonant excitation of hybridized eigenmodes in two mutually orthogonal periodic strip structures. Each six-strip array supports six eigenmodes; mode order controls penetration depth and homogeneity, and odd modes couple efficiently to the centrally placed feed loop because they have a non-zero center field (Hurshkainen et al., 2017). Related hybridization logic appears in the absorber literature, where quarter-wavelength interference cancellation, spoof surface plasmon polariton mode, dielectric resonance mode, and grating mode are merged across a patterned lossy layer and a spacer-ground subsystem (Peng et al., 2021).

A third mechanism is co-located electric and magnetic sheet responses. In the Ku-band Huygens surface, an electric meta-atom and a magnetic meta-atom are placed on opposing sides of a dielectric. With electric sheet admittance $Y_{se}$ and magnetic sheet impedance $Z_{sm}$, the normal-incidence coefficients are

$\phi$0

The Huygens condition $\phi$1 yields a reflectionless sheet with arbitrary transmission phase, which is then sampled spatially to implement generalized Snell steering (Cho et al., 2022).

A fourth mechanism is symmetry breaking and mode mixing. In the chiral U-shaped resonator assembly, two metallic layers rotated by $\phi$2 produce collinear induced electric and magnetic dipoles at the resonances, maximizing the chirality parameter and enabling circular-polarization-dependent negative refractive index (0912.3736). In the dual-layer graphene nanoribbon metamaterial, one layer supports an $\phi$3-polarized localized surface plasmon and the other a $\phi$4-polarized plasmon. Independent control of $\phi$5 and $\phi$6 tunes $\phi$7 and $\phi$8 separately, so amplitude, phase, ellipticity, and handedness can be selected through their relative magnitudes and phase difference (Matthaiakakis et al., 9 Jul 2025).

A fifth mechanism is parallel propagation through dissimilar subwavelength channels. In the multi-refractive-index metamaterial, one period contains two dissimilar waveguides with distinct effective indices $\phi$9 and $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$0, so the same incident TM field excites multiple phase velocities simultaneously. The structure therefore cannot be described by a single refractive index and impedance even at fixed frequency and polarization (Yu et al., 2018).

3. Representative implementations

Several implementations illustrate how dual-stack design is specialized to disparate frequency ranges and figures of merit.

Platform	Stack description	Key reported performance
W-band Pancharatnam HWP	Three rotated DBT plates, air gaps $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$1 mm	Usable band 87.4–92.2 GHz modeled and 88.9–92.7 GHz measured; modeled fractional bandwidth 5.3%, measured 3.1% (Mohamed et al., 2014)
Dual-nuclei MRI coil	Long-wire array, short-wire array, and feed loop in a 90 mm shield	Good agreement of measured and simulated $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$2 dB at 282.6 and 300.1 MHz; phantom SNR $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$3 for $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$4F and $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$5 for $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$6H (Hurshkainen et al., 2017)
Dielectric microwave absorber	Honeycomb PLA layer, lossy anti-honeycomb CB/PLA layer, PLA spacer, copper ground	Absorption from 5.3 to 18 GHz with RBW $$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$7 (Peng et al., 2021)
Ku-band “Wall-E” surface	Dual Huygens resonator with one electric and one magnetic meta-atom per cell	Up to 94% transmission efficiency, 85% reflection efficiency, and at most 6 dB power loss over a 150-degree field of view (Cho et al., 2022)
Thermal metamaterial pixel	Au bottom gate/Al$$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$8O$$\cos(\delta_{\mathrm{eq}}/2)=\cos(\delta_1/2)\cos(\delta_2/2)-\sin(\delta_1/2)\sin(\delta_2/2)\cos(2\phi).$$9/MLG/Au metasurface on Si/SiO$\theta_1=+30^\circ$0	10–90% rise/fall times 1.87 $\theta_1=+30^\circ$1s and 1.33 $\theta_1=+30^\circ$2s; $\theta_1=+30^\circ$3 kHz; 3×3 array rendered 26 alphabetical letters (Zhong et al., 4 Jun 2025)
Graphene nanoribbon dual layer	Orthogonal monolayer graphene ribbon arrays separated by 30 nm dielectric	Operation from $\theta_1=+30^\circ$4 to $\theta_1=+30^\circ$5m; empirical $\theta_1=+30^\circ$6 eV gives $\theta_1=+30^\circ$7m (Matthaiakakis et al., 9 Jul 2025)

The W-band retarder is notable because the single DBT plate is strongly birefringent but intrinsically narrowband: the usable band for a single plate is 91.3–93.6 GHz, whereas the built three-plate Pancharatnam stack broadens the modeled band to 87.4–92.2 GHz at the price of reduced transmitted intensity. The same study explicitly states that a dual-stack configuration is not reported as a built device, but that the analytic dual-retarder combination can broaden $\theta_1=+30^\circ$8 around $\theta_1=+30^\circ$9 relative to a single plate (Mohamed et al., 2014).

The MRI coil represents a different use of stacking. Its long-wire resonator of length $\theta_2=-29^\circ$0 mm and its capacitively loaded short-wire resonator of length $\theta_2=-29^\circ$1 mm occupy parallel planes separated by 9 mm, and the feed loop sits 7 mm from the long-wire plane and 2 mm from the short-wire plane. The selected eigenmodes are short-wire Mode 1 for $\theta_2=-29^\circ$2F at 282.6 MHz and long-wire Mode 3 for $\theta_2=-29^\circ$3H at 300.1 MHz, permitting independent control of depth profiles at the two nuclei without retuning (Hurshkainen et al., 2017).

The active infrared literature divides the stack even more sharply by function. In the thermal pixel array, monolayer graphene is simultaneously a broadband transparent microheater and an electrical switch, while the Au nano-antenna array above it acts as a narrowband emitter within a metal–insulator–metal cavity. The reported pixel uses a 20 $\theta_2=-29^\circ$4m square active region, Au square patches of side length 500, 600, or 700 nm, and a representative emissivity maximum near $\theta_2=-29^\circ$5m (Zhong et al., 4 Jun 2025). In the all-optical graphene nanoribbon system, by contrast, both stacked layers are optical resonators; their orthogonal plasmon axes are independently gated and selectively heated by pump polarization, enabling ultrafast mode mixing rather than heater–emitter decoupling (Matthaiakakis et al., 9 Jul 2025).

4. Modeling, retrieval, and synthesis methodologies

Across the literature, dual-stack design is rarely performed with a single monolithic solver from the outset. A recurring workflow is unit-cell or single-layer characterization, reduced-order composition of the stack, and then full-wave or system-level validation.

For layered electromagnetic media, effective-parameter retrieval is a common first stage. The W-band retarder retrieves refractive index and impedance from slab S-parameters as

$\theta_2=-29^\circ$6

with anisotropic retrieval performed independently for the principal polarizations. The same study combines HFSS unit-cell simulations with a transmission-line model seeded by the single-plate S-parameters, then inserts Jones rotation matrices and explicit spacer sections to optimize rotated cascades (Mohamed et al., 2014).

For broadband epsilon-near-zero stacks, inverse spectral synthesis is used instead of direct shape sweeps. The Bergman representation writes

$\theta_2=-29^\circ$7

with $\theta_2=-29^\circ$8; equating this to the step-like multilayer expression yields an inverse map from spectral poles and residues to layer filling ratios and thicknesses. The dual-stack extension then treats two ENZ meta-atoms in cascade and evaluates the combined response with a transfer-matrix method rather than a simple static average (Sun et al., 2013).

For dual-band metasurfaces, integral-equation synthesis has been used to retain inter-sheet and intra-sheet mutual coupling explicitly. The conductor-backed stacked metasurface of (Budhu et al., 2021) is homogenized into two spatially varying sheet impedances $\theta_2=-29^\circ$9 and $\theta_3=+30^\circ$0, and an EFIE is discretized by the method of moments. The dual-band nonlinear problem is then solved by an iterative “ping-pong” scheme between $\theta_3=+30^\circ$1 and $\theta_3=+30^\circ$2, followed by passivation to purely reactive sheets through optimization over far-field mismatch.

For broadband absorbers, material–structure co-optimization is central. The dielectric microwave absorber uses a genetic algorithm coupled to CST MWS, with 58-bit chromosomes consisting of 3 bits for material selection and 53 bits for geometry. The fitness functional is

$\theta_3=+30^\circ$3

subject to polarization insensitivity, thickness $\theta_3=+30^\circ$4 mm, and manufacturability constraints (Peng et al., 2021).

For multifunctional and layered metamaterials more generally, reduced analytical operators are used to keep the optimization tractable. The discrete-dipole-approximation framework writes

$\theta_3=+30^\circ$5

and optimizes positions against multiple scenario-indexed objectives (Capers et al., 2022). The mechanical metamaterial framework of (Wang et al., 2020) instead uses a VAE with latent dimension 16, a regressor for $\theta_3=+30^\circ$6, and a graph-based MRF assembly stage to enforce inter-tile compatibility. At the most reduced end, the quantum-graph treatment of layered metamaterials replaces each resonant cell by a vertex scattering matrix $\theta_3=+30^\circ$7 and uses the Bloch secular condition

$\theta_3=+30^\circ$8

together with a transfer-matrix formulation for $\theta_3=+30^\circ$9 layers (Lawrie et al., 2023).

5. Trade-offs, sensitivities, and common misconceptions

Bandwidth broadening is not automatic. The W-band half-wave-plate study states explicitly that the large birefringence produced by the negative-index resonance also introduces steep dispersion in $\delta(\omega)$0, so achromatization requires carefully chosen plate angles and spacings. Its measured three-plate broadening is accompanied by a drop in mean $\delta(\omega)$1 from about 0.8 for a single plate to about 0.6 for the three-plate device (Mohamed et al., 2014).

Stacked responses are not simply additive. The absorber work states that the design is “not merely additive; it is synergistic,” because the patterned lossy layer, spacer, and ground each activate different resonance families. The MRI coil likewise exhibits significant resonance shifts between isolated-eigenmode simulations and the assembled shield-and-phantom system: long-wire Mode 3 shifts from 309.9 MHz to 300.1 MHz, and short-wire Mode 1 shifts from 264.4 MHz to 282.6 MHz owing to mutual coupling, shield loading, and phantom dielectric loading (Peng et al., 2021, Hurshkainen et al., 2017).

Dual-stack does not necessarily mean dual-band, and dual-gate does not necessarily mean top-and-bottom gate control of one channel. The Pancharatnam retarder is a single-band achromatization strategy, not a dual-band device (Mohamed et al., 2014). The silicon nanophotonics literature uses “dual-metamaterial” for a patterned core and patterned cladding that independently tune vertical and horizontal confinement rather than for two vertically separated layers (Dinh et al., 2021). The graphene thermal pixel explicitly states that its “dual-gate” control refers to two independent gate electrodes per pixel, not a stacked top-and-bottom gate on a single channel (Zhong et al., 4 Jun 2025).

Tolerance budgets are architecture dependent but rarely negligible. In the W-band retarder, measured discrepancies are attributed mainly to grid alignment errors, especially in HWP 3, and the design guidance recommends conservative tolerances such as $\delta(\omega)$2 rotation error and $\delta(\omega)$3–$\delta(\omega)$4m translational misalignment (Mohamed et al., 2014). In the dual-metamaterial waveguide, applying $\delta(\omega)$5 nm to both longitudinal and lateral hole dimensions changes $\delta(\omega)$6 by less than 0.01 across 1.5–1.6 $\delta(\omega)$7m (Dinh et al., 2021).

Improved multifunctionality often trades against another figure of merit. In the thermal pixel, smaller pitch increases mutual heating through the substrate, while higher heater power raises temperature and radiance but can aggravate thermal crosstalk (Zhong et al., 4 Jun 2025). In the multi-refractive-index metamaterial, thinning the metal walls increases throughput, but the paper reports that sufficiently thin walls eliminate the multi-focus behavior and drive the array toward a single effective index (Yu et al., 2018).

6. Applications and extensions

Dual-stack design has been applied to polarization optics, MRI, infrared thermal photonics, absorbers, beam-steering surfaces, negative-index media, and silicon photonic confinement engineering. The W-band DBT methodology generalizes to other bands by scaling unit-cell dimensions and substrate thickness, while keeping the retrieval, Jones/TL modeling, and Pancharatnam achromatization framework unchanged (Mohamed et al., 2014). The MRI coil framework is likewise presented as scalable to other nuclei such as $\delta(\omega)$8Na and $\delta(\omega)$9P by recomputing $\pi$0, resizing $\pi$1, and re-optimizing the capacitively loaded short-wire resonator (Hurshkainen et al., 2017).

Several papers treat dual-stack as an intermediate point on a larger design ladder. The stacked-metasurface IE/MoM method is explicitly extended to multi-band operation by cycling the “ping-pong” iteration across $\pi$2 (Budhu et al., 2021). The absorber study identifies tri-stack and multi-stack variants, graded-index spacers, and hierarchical inclusions as natural bandwidth extensions, and suggests machine learning and advanced multi-objective optimizers for co-optimizing material recipes and geometry (Peng et al., 2021). The active thermal and graphene-plasmonic works point toward larger matrices, multispectral emission, CMOS-compatible routing, and multi-color pixels within a common heater or ribbon platform (Zhong et al., 4 Jun 2025).

At the modeling level, the same trend appears. The DDA framework is extended to multiple functions and, by implication, to more than two coupled layers through block Green operators and multi-scenario objectives (Capers et al., 2022). The quantum-graph formulation already writes an $\pi$3-layer transfer-matrix system and demonstrates positive refraction, negative refraction, and beam steering through layered compositions (Lawrie et al., 2023). This suggests that dual-stack design is best understood not as one specific geometry, but as a systems strategy for distributing spectral, spatial, polarization, and modal tasks across interacting metamaterial subsystems while retaining an explicit handle on coupling, dispersion, and fabrication tolerance.