Programmable Metasurfaces: Principles & Applications

• Programmable metasurfaces are engineered 2D arrays of meta-atoms whose electromagnetic, acoustic, or optical responses are tunable via external stimuli.
• They enable dynamic wavefront transformations such as beam steering, dual-polarization multiplexing, and on-chip neural computing through integrated digital and material-based control mechanisms.
• Key performance metrics include phase tuning range, reconfiguration speed, and error tolerance, with applications spanning 5G/6G communications, adaptive sensing, and quantum information processing.

Programmable metasurfaces are engineered two-dimensional arrays of subwavelength "meta-atoms" whose individual electromagnetic, acoustic, or optical response is tunable in real time by external, often electronic, stimuli. By harnessing integration of actuation, embedded control, and/or software-defined networks, programmable metasurfaces synthesize arbitrary wavefront transformations far exceeding the constraints of static or globally tunable planar structures. Such platforms underpin advances in reconfigurable communications, adaptive sensing, analog signal processing, and emerging quantum and photonic information technologies, with realized functionalities spanning dynamic beam steering, dual-polarization multiplexing, on-chip neural computing, and quantum algorithm execution.

1. Fundamental Principles of Programmable Metasurfaces

A programmable metasurface comprises a planar array of engineered unit cells—meta-atoms—each designed to support externally controllable values of effective amplitude, phase, and/or polarization. The EM response at each site is typically modeled as a complex reflection or transmission coefficient: $\Gamma_{mn}(t) = A_{mn}(t)\, e^{j\varphi_{mn}(t)}$ where $(m,n)$ denotes the cell position, $A_{mn}$ the amplitude, and $\varphi_{mn}$ the phase, both of which are programmable via electronic (varactor/PIN/MEMS), thermal (phase change), or optical (carrier injection) actuation (Liu et al., 2018, Abou-Hamdan et al., 16 May 2025).

Wave propagation and metasurface scattering are governed by Maxwell's equations with imposed boundary conditions determined by the spatial map of $\Gamma_{mn}$ (or its transmission analog $t_{mn}$), with key design relations encapsulated by generalized Snell's law: $$\sin\theta_r - \sin\theta_i = \frac{\lambda}{2\pi}\frac{d\varphi(x)}{dx}$$ enabling dynamic beam steering, focusing, and anomalous refraction. In software-defined metasurfaces, digital controllers and embedded networks provide addressable, real-time reconfiguration of each meta-atom, abstracting physical control as programmable APIs (Tasolamprou et al., 2018, Petrou et al., 2019).

Multi-modal generalizations extend the concept beyond electromagnetics to acoustics and quantum optics: programmable acoustic metasurfaces modulate sound waves via spatiotemporal coding of Helmholtz-cavity elements (Rajabalipanah et al., 2020); programmable quantum metasurfaces explicitly encode quantum unitary operations into the complex scattering matrix of nanostructured lens arrays (Tanuwijaya et al., 2023).

2. Architectures, Materials, and Control Mechanisms

The realization of programmable metasurfaces spans diverse embodiments depending on bandwidth, frequency, and required functionality:

• Electronic/diode-based (RF–microwave): Arrays of metallic patches with per-cell varactors/PINs permit continuous phase tuning, binary coding, or multi-bit quantization. Control is delivered via direct wiring or on-chip ASICs implementing asynchronous handshakes for scaling (Petrou et al., 2019). Electromagnetic interference, routing congestion, and power consumption are mitigated through delay-insensitive communication and mesh topologies (Tasolamprou et al., 2018, Petrou et al., 2019).
• Phase-change and volatile materials (THz/optical): Programmable THz metasurfaces exploit voltage-driven phase transitions in VO$_2$ microwires for sub-nanosecond, dual-polarization binary coding (Shabanpour, 2020). Nanostructured GST elements achieve 6-bit-level phase and amplitude modulation on integrated photonic waveguides, enabling in-memory matrix-vector computation for optical neural networks (Wu et al., 2020, Abou-Hamdan et al., 16 May 2025).
• Anisotropic and polarization-multiplexed designs: Complex 2D unit-cell geometries with orthogonally positioned active elements allow independent manipulation of x- and y-polarized channels, enhancing information density and dual-mission communication in dynamic environments (Shabanpour, 2020).
• Space-time coding: Meta-atom temporal modulation, often executed by high-speed electromechanical or PIN-diode switching, underpins space–time coding metasurfaces (STCMs) for programmable calculus operations or multi-harmonic acoustic/EM wave control (Shi et al., 4 Jan 2026, Rajabalipanah et al., 2020).
• Embedded networks and resource sharing: Architectures such as the HyperSurface (HSF) platform integrate wireless intercell communication transceivers, with design trade-offs between layer-sharing (co-designing EM and communication functions) and dedicated layers for higher SNR and bandwidth (Tasolamprou et al., 2018).

Control complexity and reconfiguration fidelity are balanced by protocols including row–column cross-modulation (BCM/MPCM), reducing control-line and memory requirements while preserving multi-bit phase quantization up to K bits via K-fold partitioning (Zhang et al., 2024).

3. Functionality, Encoded Operations, and Performance Metrics

Programmable metasurfaces support a spectrum of reconfigurable functionalities:

• Dynamic beam steering and focusing: Beam redirection is achieved by encoding spatially varying phase gradients, with main-lobe gain, side-lobe level (SLL), and beamwidth determined by aperture size, unit-cell quantization (N states), and spatial phase resolution (Taghvaee et al., 2020). 4-state (2-bit) coding on arrays ≥5λ × 5λ with unit-cell size λ/3 secures HPBW ≈ 15°, SLL < –12 dB, and ±5° pointing accuracy up to θ ≈ 60° (Taghvaee et al., 2020).
• Dual-polarization and OAM multiplexing: Anisotropic 1-bit or higher-order coding allows independent or simultaneous control of orthogonal polarization channels, e.g., two VO$_2$ microwires enable four digital states (00, 01, 10, 11) for THz dual-polarized focusing and OAM-mode beam generation with intrinsic reconfiguration <500 fs and wide operational bandwidth (Shabanpour, 2020).
• Analog and quantum information processing: Chalcogenide phase-change metasurfaces implement analog matrix-vector multiplication (64-level precision, 6 bits) in optical CNNs with direct encoding of both positive/negative weights via modal contrast tuning (Wu et al., 2020). Programmable quantum metasurfaces encode Grover’s search and quantum Fourier transform algorithms onto geometric-phase metalens arrays, with unitary operations programmed into real-space amplitude/phase profiles and selected by spatial light modulators (SLMs) (Tanuwijaya et al., 2023).
• Space–time processing and analog calculus: By applying optimized temporal coding sequences to multi-bit meta-atoms, STCMs directly implement spatial differentiation/integration of incident EM fields at selected harmonics. FPGA control and PIN-diode settling enable function switching in <1 μs, supporting multi-operation on a single input by frequency multiplexing (Shi et al., 4 Jan 2026).

Key performance metrics include phase tuning range (e.g., >300° for THz (Xu et al., 2024)), reflection/transmission amplitude (insertion loss <2 dB in optimized designs), SLL, HPBW, operational bandwidth (up to 30%), reconfiguration speed (tens of ns to sub-ps), digital coding resolution, and net system-level impact, e.g., >95% simulated interference contrast for classical trials and single-photon RMSE <0.22 for QFT operations (Tanuwijaya et al., 2023).

4. Programmable Metasurfaces in Communications, Sensing, and Computing

Wireless communications: Programmable metasurfaces synthesize dynamically controlled scattering environments for MIMO, full-space coverage (STARS), and reconfigurable intelligent surfaces (RIS) in 5G/6G systems (Gan et al., 7 Dec 2025). They enable hardware-free transmitters and receivers by delegating modulation and mixing to the physically reconfigurable aperture (Tang et al., 2019); e.g., QPSK and 16QAM can be modulated directly by real-time phase encoding of unit cells, with practical demonstration at 2 Mbps and <20 ms latency (Tang et al., 2018). Hardware complexity is reduced from N parallel PAs/mixers to a single PA and O(N) DACs, at a cost of modest SNR penalty due to pulse shaping and nonideal switching.

Sensing and imaging: Spatiotemporal coding of metasurface phase profiles supports high-resolution ranging (sub-cm at mmWave (Gan et al., 7 Dec 2025)), angle-of-arrival estimation by near-field modeling, and adaptive mutual coupling for radar/sonar applications (Gan et al., 7 Dec 2025, Rajabalipanah et al., 2020). Multi-beam, multi-band, and cooperative sensing scenarios leverage PMs to maximize spatial diversity and SNR under practical deployment constraints.

Analog computing and photonic neural networks: Stacked intelligent metasurfaces (SIMs) implement high-rank unitary or non-unitary operators in the EM domain, enabling direct over-the-air computation, analog convolutions, and DNN inference with sub-ns latency and petasample/s throughput (Abou-Hamdan et al., 16 May 2025, Gan et al., 7 Dec 2025). Nonlinear activations are physically induced via structural nonlinearity—circuital or material dependence of the Green’s function on programmable bias states. In-memory computing with GST-based metasurfaces achieves robust positive/negative weight encoding in CNN convolution and fully-connected layers, approaching digital accuracy at MHz programming rates (Wu et al., 2020).

5. Scalability, Control, and Reliability Engineering

Scaling programmable metasurfaces to apertures comprising $>10^6$–$10^9$ meta-atoms necessitates innovations in control architecture, memory efficiency, power distribution, and error tolerance:

• Hierarchical and partitioned addressing: Techniques such as multiple-partition cross-modulation (MPCM) reduce wiring and per-beam memory by representing coding sequences as products of row/column vectors, achieving O(M + N) complexity (versus O(M × N) for true per-element addressing), with ≤1.5° beam pointing error for 100-cell arrays (Zhang et al., 2024).
• Asynchronous and distributed mesh communication: Strictly planar, delay-insensitive ASIC meshes support scalable, clockless control with per-hop handshake times of 40–50 ns and full-array updates in microseconds, while minimizing power (<1 μW quiescent, <100 μW active) and mitigating signal-integrity issues (Petrou et al., 2019).
• Error modeling and robustness: Monte Carlo and analytical frameworks reveal that independently random (“uncorrelated stuck-at”) faults degrades beam-steering performance gracefully, with main-lobe directivity maintained within 3 dB up to ~25% error rate; in contrast, clustered or deterministic faults are highly deleterious, emphasizing architectural design to spatially decorrelate failures and enable power-gating without global loss (Taghvaee et al., 2019, Taghvaee et al., 2020).
• Rapid inverse design and physical-model embedding: Physics-informed neural networks, embedding coupled-mode theory as differentiable layers, can inversely design metasurfaces supporting >300° phase tuning without amplitude loss in seconds, with >60% reduction in target error and dramatic acceleration over conventional "black-box" DNNs (Xu et al., 2024).

6. Application-Specific Implementations and Experimental Demonstrations

Quantum and photonic information: Programmable metasurfaces directly encode quantum unitaries in the geometry of metalens arrays, supporting sequential algorithmic selection via SLM and readout via single-photon cameras. Demonstrations of Grover’s search and quantum Fourier transform unitaries achieve RMSE <0.22 for the QFT, with single-step, lensless execution, indicating a pathway to on-chip, single-photon quantum processors (Tanuwijaya et al., 2023).

Wave-based analog calculus: Space–time-coding metasurfaces (STCMs) implement electromagnetic analog computing primitives such as spatial differentiation and integration at selected spatiotemporal harmonics, with FPGA-driven code switching <1 μs and experimental confirmation of real-time edge enhancement (Shi et al., 4 Jan 2026).

IoT and channel engineering: LLAMA metasurface arrays provide up to +15 dB link gain in polarization-mismatched 2.4 GHz IoT links, extending range by 5.6× and delivering capacity improvements of 100–180 kbps/Hz, accomplished with low-cost, FR4-based multilayer structures and real-time optimization over two bias voltages (Chen et al., 2020).

THz and beyond: Multi-bit LC-loaded metasurfaces and PCM-based photonic mode converters support dynamic beam steering up to 68°, dual-polarization multiplexing, and in-memory analog computation in the THz band, with experimentally verified phase tuning >270°, near-unity amplitude, and full waveform control over broad bandwidth (Shabanpour, 2020, Wu et al., 2020, Xu et al., 2024).

7. Challenges, Trade-offs, and Future Directions

Challenges remain in scaling reconfigurability to nanophotonic and THz domains, achieving fine-grained, multi-bit coding, reducing actuation energy, and integrating robust, low-latency control. Hardware constraints motivate hybrid analog-digital architectures, active–passive trade-offs, and adaptive machine-learning–based optimization.

Key open problems include:

• Full co-design of meta-atom, circuitry, and network for GHz+ update rates and >10$^9$ pixel densities (Abou-Hamdan et al., 16 May 2025).
• Unified modeling frameworks capturing near-field, multi-layer, and mutual-coupling effects beyond geometric optics for accurate system-level synthesis (Gan et al., 7 Dec 2025).
• Pareto-optimization of communication, sensing, and computation metrics in multi-functional, AI–co-designed metasurface platforms.
• Robustness to fabrication tolerances, runtime variability, and adversarial control threats, with fault-aware beamforming and adaptive online reconfiguration (Taghvaee et al., 2019, Taghvaee et al., 2020).
• Integration of physics-informed inverse design pipelines and self-adaptive coding sequence discovery via differentiable simulation (Xu et al., 2024, Shi et al., 4 Jan 2026).

Programmable metasurfaces thus stand at the nexus of custom hardware, multidisciplinary physics, and embedded intelligence, poised to reshape functional electromagnetic media for 6G/B6G, photonic AI, quantum information, and beyond.