Integration of metasurface with photodetectors and image sensors: towards ultra-compact, multi-functional and on-chip optoelectronic imaging systems

Abstract

The miniaturization of imaging systems is essential for modern optoelectronics, while traditional bulk optics fundamentally limit integration density. Metasurfaces offer a transformative planar alternative by providing nanoscale control over incident light. However, realizing their full potential requires transitioning from discrete optical components to fully integrated sensing architectures. This review surveys the progressive integration of metasurface with photodetectors and image sensors. We examine the developmental trajectory across three distinct levels: free-space optical path placement, hybrid adhesive mounting and direct monolithic fabrication. Particular emphasis is placed on monolithic architectures and their role in enabling multi-dimensional information extraction. Finally, we discuss emerging challenges in material compatibility and scalable manufacturing to provide clear directions for future ultracompact optoelectronic device development.

Graphical abstract

Keywords

Metasurface / Image sensor / Optoelectronic integration / Multidimensional imaging / Optoelectronic metadevices

1 Introduction

Optoelectronic imaging systems are rapidly evolving toward extreme miniaturization and expanded functionality. Fields such as autonomous navigation, portable medical diagnostics and drone inspection drive this trend. The demand for ultra-compact, high-performance imaging modules continues to surge. However, traditional refractive optics impose fundamental physical barriers to this miniaturization trend. Conventional systems rely on cascaded lens groups and microlens arrays to correct optical aberrations and ensure high imaging quality. These macroscopic optical components inevitably occupy a large spatial footprint and introduce excessive weight. Consequently, traditional bulk optics severely restrict the compactness and integration density of the imaging system.

Metasurfaces offer a transformative planar alternative, providing nanoscale control over the phase, amplitude and polarization of incident light [1,2]. These subwavelength arrays eliminate the reliance on spatial phase accumulation [3]. Since the initial proposal of generalized refraction laws, metasurface technology has rapidly advanced from theory to applications. Dielectric metalenses now achieve diffraction-limited focusing in the visible spectrum with high efficiency [4]. Advanced group delay dispersion engineering enables continuous broadband achromatic focusing [5–7]. Beyond simple focusing, metasurface demonstrates exceptional capabilities in multi-parameter light-field acquisition, including single-shot full-Stokes polarization imaging and multidimensional spectral perception [8–10]. Despite these key advantages, replacing traditional optics with metasurface remains a gradual process. Evolution proceeds from partial substitutions via hybrid meta-refractive systems [11,12] to nearly full replacements in compact planar modules [13]. Subsequent architectures advance to on-chip mounting integration. For instance, researchers bond metasurfaces directly to sensors for complex computational imaging [14,15]. Ultimately, the technology reaches direct monolithic integration. This advanced paradigm fuses metasurfaces directly with active materials like colloidal quantum dots [16,17]. This overarching trajectory defines the direction of the field. This review systematically examines this three-stage evolution.

Initially, researchers deployed metasurfaces within free-space optical paths, replacing bulk refractive elements to reduce module volume. Key milestones include large-scale hybrid metalenses that correct high-order aberrations and coma [18,19]. For instance, Liu et al. [19] used hybrid meta-optics for secondary spectrum correction, enabling compact, high-resolution thermal imaging. This approach successfully reduces optoelectronic system footprints. However, it focuses exclusively on optics. Fully exploiting metasurfaces requires direct physical mounting or monolithic fabrication. These advanced architectures demand codesign to minimize the optical-electronic physical gap.

To further collapse footprints, the field advanced to physical mounting integration. Bonding metasurface arrays to image sensors via optical clear adhesives drastically shortens optical paths, transforming standard sensors into computational nodes. Progress includes on-chip multiplexed diffractive neural networks (MDNN) achieving >93% image classification accuracy via all-optical inference [14]. Hybrid mounting also enables chip-integrated full-Stokes polarimetric sensors using super-pixel layouts [15]. Additional breakthroughs include real-time brain spectral monitoring via reconfigurable supercells [20] and compact wide-angle cameras using metalens arrays [21]. Despite these successes, hybrid mounting requires mechanical alignment. Resulting tolerances degrade performance as pixel pitches shrink. Furthermore, modulated light converts into plane waves while propagating through the bonding substrate, severely weakening near-field enhancement between the metasurface and photosensitive layer.

Monolithic fabrication represents the ultimate integration frontier. Defining metasurfaces directly onto active materials such as silicon, 2D materials, colloidal quantum dots (CQDs) and perovskites places resonant nanostructures in the immediate near-field of the photoelectric depletion region. Consequently, metasurfaces actively participate in exciton generation. In silicon platforms, disordered metasurfaces with upconversion nanoparticles extend infrared responsivity to 0.22 A/W at 1550 nm [22]. For 2D materials, metasurface-integrated graphene and MoS₂ detectors realize high-extinction polarization sensitivity and zero-bias operation [23,24]. Notably, monolithic CQD integration resolves the contradiction between optical absorption depth and carrier diffusion length. Resonant cavity enhancement ensures near-perfect absorption (>90%) and extraordinary responsivity (up to 8000 A/W) in PbS/PbSe CQD photodetectors [16,17]. This architecture establishes a new design paradigm. Codesigned metasurfaces and photosensitive materials in subwavelength proximity unlocks novel light-matter interactions. These mechanisms break intrinsic absorption limits and directly convert multi-parameter optical data into electrical signals at different device ports, realizing true single-shot, highly efficient multi-dimensional perception [25].

This review systematically surveys the progressive integration of metasurfaces with photodetectors and image sensors. The structure is organized as follows: Section 2.1 reviews the initial deployment in free-space systems. Section 2.2 discusses hybrid mounting for computational imaging and spectral routing. Section 2.3 deeply analyzes monolithic integration paradigms across silicon, 2D materials, and quantum dot platforms. Finally, Section 3 summarizes current manufacturing challenges and provides an outlook on the future of autonomous, intelligent optical frontends.

2 Main

2.1 Integration of metasurface with image sensors in optical systems

This section explores the optical-system-level integration of metasurfaces. In this paradigm, metasurfaces function as discrete components within free-space optical paths. The internal logic of this section follows a clear architectural evolution, progressing from fundamental aberration correction to complete system consolidation. Section 2.1.1 addresses the foundational challenge of broadband achromaticity. Building upon this, Section 2.1.2 introduces hybrid metasurface-refractive systems, which strategically distribute optical burdens to achieve large scalable apertures. Section 2.1.3 then shifts the focus from basic focusing to spatial multiplexing, utilizing free-space propagation to physically decode multidimensional information. Finally, Section 2.1.4 presents the ultimate free-space architecture: standalone metalens systems that entirely discard traditional refractive elements. Together, these subsections illustrate the progressive miniaturization of macroscopic optics before transitioning to physical bonding.

2.1.1 Optical-system-level integration of achromatic metalenses

Realizing broadband achromatic metalenses requires fulfilling a strict frequency-dependent spatial phase profile. To eliminate chromatic aberration, the metasurface must precisely satisfy the target group delay (GD) at every radial coordinate r:

(1)\mathrm{G}\mathrm{D}\left(r\right)=\frac{\mathrm{\partial }\varphi }{\mathrm{\partial }\omega }=-\frac{1}{c}\left(\sqrt{r^2+f^2}-f\right),

where \omega is the angular frequency, c is the speed of light, and f is the focal length. Fulfilling this exact GD demand fundamentally governs the evolution of all achromatic metasurface architectures.

Early metalenses suffered from severe single-layer group delay limitations. Researchers first achieved high-performance focusing at discrete visible wavelengths (Fig. 1a) [4]. High-aspect-ratio TiO₂ nanopillars delivered diffraction-limited spots with a high numerical aperture (NA=0.8). However, continuous broadband operation remained theoretically challenging.

To overcome this, subsequent designs introduced simultaneous control of spatial phase and dispersion. Researchers demonstrated a broadband achromatic metalens using coupled TiO₂ nanopillars (Fig. 1b) [5]. This breakthrough relies on expanding the target phase profile via a Taylor series around a design frequency {\omega }_{\text{d}}:

(2)\varphi \left(r,\omega \right)=\varphi \left(r,{\omega }_{\mathrm{d}}\right)+{\frac{\mathrm{\partial }\varphi }{\mathrm{\partial }\omega }|}_{{\omega }_{\mathrm{d}}}\left(\omega -{\omega }_{\mathrm{d}}\right)+\frac{1}{2}{\frac{{\mathrm{\partial }}^2\varphi }{\mathrm{\partial }{\omega }^2}|}_{{\omega }_{\mathrm{d}}}{\left(\omega -{\omega }_{\mathrm{d}}\right)}^2.

By independently tuning the basic phase, group delay, and group delay dispersion (GDD), this design covered 470−670 nm. Concurrently, researchers utilized GaN-based integrated-resonant units to fully eliminate chromatic aberration across 400−660 nm (Fig. 1c) [6].

Expanding beyond single elements, array formats extend achromatic functionality. Researchers developed a broadband achromatic metalens array using Si₃N₄ nanopillars (Fig. 1d) [7]. This array achieved zero effective material dispersion over 430−780 nm. It successfully reconstructed white-light 3D scenes through macroscopic integral imaging.

Ultimately, a fundamental physical bound limits the maximum GD a single nanostructure layer can provide. To break this delay-bandwidth limit, researchers utilized 3D-printed multilayer structures (Fig. 1e) [26]. Most recently, researchers pushed achromatic focusing to cover the full visible spectrum using dispersion-matched layered meta-atoms (Fig. 1f) [27]. This architecture provides two-level group delay compensation. It explicitly sums the fine GD of nanoantennas (\mathrm{G}{\mathrm{D}}_{\text{nano}}) and the coarse GD of dispersion-matched layers \mathrm{G}{\mathrm{D}}_{\text{layer}}:

(3){\mathrm{G}\mathrm{D}}_{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}}\left(r\right)={\mathrm{G}\mathrm{D}}_{\mathrm{n}\mathrm{a}\mathrm{n}\mathrm{o}}\left(r\right)+{\mathrm{G}\mathrm{D}}_{\mathrm{l}\mathrm{a}\mathrm{y}\mathrm{e}\mathrm{r}}\left(r\right).

This cascaded dispersion matching overcomes the insufficient focusing power of single layers. It supports highly scalable large apertures while maintaining an exceptional peak focusing efficiency of 88%.

A comprehensive comparison of the design parameters and performance metrics for these representative devices is summarized in Table 1. The free-space integration of achromatic metasurfaces represents a fundamental paradigm shift. It transitions flat optics from simple phase profiling to multidimensional dispersion engineering. This developmental trajectory reveals a persistent effort to break inherent physical trade-offs. Standalone single-layer nanostructures inherently possess restricted degrees of freedom. Strategic transitions toward multilayer cascades and spatially multiplexed arrays effectively decouple phase control from dispersion constraints. This architectural evolution actively introduces advanced computational modalities directly into the primary light path. Mastering these complex dispersion-matched strategies lays essential theoretical groundwork. It ensures the feasibility of subsequently bonding these broadband optical frontends directly onto semiconductor sensor platforms.

2.1.2 Metasurface-refractive optics and hybrid refractive-metalens system

Hybrid refractive-metalens systems (HRMS) strategically integrate metasurfaces with conventional refractive lenses. This architecture establishes a highly pragmatic synergy. It substantially alleviates the efficiency-performance trade-offs inherent in standalone metalenses.

Standalone metasurfaces suffer from severe inherent chromatic dispersion. To resolve this, researchers introduced the metacorrector concept (Fig. 2a) [11]. By placing TiO₂ nanofins behind a spherical plano-convex lens, they corrected broadband aberrations. A major advance followed with the hybrid achromatic metalens (HAML) (Fig. 2b) [12]. This design merges a refractive phase plate with a diffractive metalens. The core physical mechanism relies on precise group delay (GD) compensation. The metasurface must provide an anomalous group delay to cancel the refractive lens dispersion:

(4){\mathrm{G}\mathrm{D}}_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}\left(r\right)=\frac{\mathrm{\partial }{\varphi }_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}}{\mathrm{\partial }\omega }=-{\mathrm{G}\mathrm{D}}_{\mathrm{r}\mathrm{e}\mathrm{f}}\left(r\right),

where \omega is the angular frequency, and \mathrm{G}{\mathrm{D}}_{\text{ref}} is the refractive group delay. Coupling the metasurface with a bulk lens dramatically reduces the requisite phase delay bounds. The bulk lens provides the primary optical power, while the metasurface solely manages residual dispersion.

Scaling pure metalenses to macroscopic apertures critically degrades absolute light efficiency. To advance large-area implementation, researchers applied metasurface correctors to commercial centimeter-scale refractive lenses (Fig. 2c) [18]. Furthermore, global optimization of a single refractive lens with a metasurface corrects secondary spectra (Fig. 2d) [28]. At the physical interface, the metasurface corrects localized ray trajectories via the generalized Snell’s law:

(5)n_{\mathrm{t}}\sin {\theta }_{\mathrm{t}}-n_{\mathrm{i}}\sin {\theta }_{\mathrm{i}}=\frac{\lambda }{2\text{π}}\frac{\mathrm{d}{\varphi }_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}}{\mathrm{d}r}.

Here, n_{\text{i}} and n_{\text{t}} represent refractive indices. {\theta }_{\text{i}} and {\theta }_{\text{t}} denote the incident and refracted angles. The term {\text{d}\varphi }_{\text{meta}}/\text{d}r represents the spatial phase gradient. Placing the metasurface within a free-space optical system leverages the high transmission efficiency of bulk glass. The metasurface acts merely as a localized momentum modulator, ensuring high total system throughput.

Achieving an ultra-wide field-of-view (FOV) typically demands complex, bulky cascaded lens assemblies. Researchers simplified this by combining a metasurface corrector with a refractive lens. This combination enables high-performance long-wave infrared thermal imaging (Fig. 2e) [19]. Similarly, researchers integrated a metalens with a plastic refractive system. This achieved a 100° FOV in the visible spectrum (Fig. 2f) [29]. The metasurface specifically neutralizes Seidel aberrations (like coma and spherical aberration) accumulated through the bulk lenses. It utilizes an even-order polynomial phase profile:

(6){\varphi }_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}\left(r\right)=\frac{2\text{π}}{\lambda }\sum _{n=1}^N{a}_{2n}{r}^{2n},

where \lambda is the wavelength, and r is the radial coordinate. The variables a_{2n} represent the optimized polynomial coefficients. Deploying the metasurface alongside curved refractive interfaces physically uncouples macroscopic focusing from microscopic aberration correction. The refractive surfaces minimize extreme ray bending. This physically prevents the metasurface from exceeding spatial Nyquist sampling limits at large incident angles.

The hybrid refractive-metalens architecture represents a crucial transitional framework in modern optics. It effectively redistributes the optical burden between macroscopic and nanoscopic domains. The macroscopic refractive bulk provides the primary optical power and high absolute transmission. Concurrently, the ultrathin metasurface assumes the complex task of high-order aberration correction. It also performs arbitrary dispersion tailoring. This collaborative strategy immediately scales flat optics to centimeter-level apertures and ultra-wide FOVs. It completely circumvents the fundamental efficiency bottlenecks of purely nanostructured focusing interfaces. However, this approach inherently requires precise spatial alignment between curved bulk lenses and planar metasurfaces. Furthermore, it fails to achieve the ultimate form-factor miniaturization. Fully planar or monolithic optoelectronic integration ultimately promises this extreme miniaturization.

2.1.3 Multifunctional metasurface for spatial multiplexing in the image plane

In image plane spatial multiplexing, researchers place multifunctional metasurfaces within the optical path. These structures generate spatially separated patterns based on incident light properties.

Researchers introduced division-of-focal-plane (DoFP) devices using dielectric metasurfaces (Fig. 3a) [8]. These structures split incident light according to orthogonal polarization bases. To achieve this, the metasurface encodes a multiplexed phase profile \varphi \left(x,y\right):

(7)\mathrm{e}\mathrm{x}\mathrm{p}\left[\mathrm{i}\varphi \left(x,y\right)\right]=c_1\mathrm{e}\mathrm{x}\mathrm{p}\left[\mathrm{i}{\varphi }_1\left(x,y\right)\right]+c_2\mathrm{e}\mathrm{x}\mathrm{p}\left[\mathrm{i}{\varphi }_2\left(x,y\right)\right],

where {\varphi }_1 and {\varphi }_2 dictate independent focal positions, and c_1, c_2 are complex weighting coefficients. Deploying this device in the optical path allows free-space propagation to physically separate these multiplexed wavefronts into distinct intensity spots. Applying this principle, researchers developed generalized Hartmann−Shack arrays (Fig. 3b) [30]. Silicon metalens sub-arrays produce six polarization-dependent focal spots. This enables simultaneous extraction of amplitude, phase, and polarization profiles.

To map polarizations to full images, researchers proposed matrix Fourier optics (Fig. 3c) [9]. A metagrating applies a spatially varying Jones matrix J\left(x,y\right). Its far-field angular spectrum J\left(k_x,k_y\right) follows the spatial Fourier transform:

(8)\tilde{J}\left(k_x,k_y\right)=\iint J\left(x,y\right)\mathrm{e}\mathrm{x}\mathrm{p}\left[-i\left(k_{x}x+k_{y}y\right)\right]\mathrm{d}x\mathrm{d}y,

where k_x and k_y represent the transverse spatial frequencies. The metagrating diffracts distinct polarization states into different far-field angles. A subsequent standard optical lens is strictly required to focus these angular components. This optical accumulation effectively translates the far-field intensities into spatially separated images on different regions of the CMOS sensor.

Alternatively, disordered metasurfaces achieve spatial multiplexing through random scattering (Fig. 3d) [31]. These structures utilize weak dichroism and random phase distributions. They project the incident Stokes vector S into a 2D speckle intensity pattern I:

(9)I\left(x,y\right)=H\left(x,y\right)\cdot S,

where H(x,y) represents the calibrated spatially varying measurement matrix. Placing the metasurface at a specific distance in the imaging path generates these unique speckle fingerprints. Reconstruction algorithms then decode the full-Stokes information from these broadband patterns without traditional beam splitters.

Beyond polarization, spatial multiplexing efficiently separates spectral information. Researchers demonstrated transverse chromatic dispersion multiplexing using metalens arrays (Fig. 3e) [10]. The structural dispersion induces a wavelength-dependent lateral focal shift \text{Δ}x\left(\lambda \right):

(10)\mathrm{\Delta }x\left(\lambda \right)\approx f\cdot \frac{\lambda -{\lambda }_0}{\mathrm{\Lambda }},

where f is the focal length, {\lambda }_0 is the central wavelength, and \mathrm{\Lambda } is the effective grating period. Free-space propagation provides the necessary distance for these angular and spectral differences to accumulate. They eventually form macroscopic lateral displacements on the image plane, enabling snapshot 4D light-field imaging.

Finally, metasurfaces can multiplex phase information into intensity via optical shearing (Fig. 3f) [32]. A single-layer metalens generates polarization-dependent shear point spread functions. The interference of laterally sheared fields (E_1 and E_2) produces an intensity I:

(11)I\left(x,y\right)={\left|E\left(x+\mathrm{\Delta }x/2,y\right)+E\left(x-\mathrm{\Delta }x/2,y\right){\mathrm{e}}^{\mathrm{i}\delta }\right|}^2,

where \text{Δ}x is the spatial shear distance and \delta is the introduced phase delay. Deploying the metalens before the image plane allows these slightly shifted replicas to physically overlap. This interference directly converts invisible 2D phase gradients into measurable intensity variations.

Deploying metasurfaces at the image or focal plane signifies a conceptual leap. It transitions the field from fundamental wavefront shaping to physical information decoding. Rather than merely focusing light, the metasurface functions as a powerful spatial multiplexer. Crucially, macroscopic free-space propagation acts as the necessary computational medium. It translates elusive optical properties into spatially distinct focal spots or diffraction patterns. This paradigm entirely eliminates bulky mechanical rotating parts and cascaded beam splitters. It directly links high-dimensional optical degrees of freedom to standard 2D sensors. Ultimately, this free-space spatial multiplexing serves as a direct conceptual predecessor. It verifies the decoding capability of metasurfaces, setting the theoretical stage for direct physical mounting integration onto sensor pixels.

2.1.4 Imaging based on metalenses without refractive elements

Constructing optical systems entirely from metalenses marks the ultimate realization of standalone flat optics. By orchestrating complex phase profiles, standalone metasurfaces completely assume the dual burden of primary image formation and aberration correction.

To separate chiral states, researchers demonstrated a multispectral chiral metalens (MCHL) (Fig. 4a) [33]. This design interleaves TiO₂ nanofin arrays using geometric phase control. The phase profile {\varphi }_{\pm }\left(x,y\right) independently targets left- and right-circularly polarized (LCP/RCP) light:

(12){\varphi }_{\pm }\left(x,y\right)=-\frac{2\text{π}}{{\lambda }_{\mathrm{d}}}\left(\sqrt{{\left(x\pm x_0\right)}^2+y^2+f^2}-f\right)\pm 2\theta \left(x,y\right),

where x_0 defines the lateral focal shift, {\lambda }_{\text{d}} is the design wavelength, and \theta \left(x,y\right) represents the local nanostructure rotation angle. Eliminating macroscopic polarizers allows single-layer nanostructures to intrinsically route chiral states to distinct spatial locations. This enables simultaneous multispectral imaging and circular dichroism mapping.

Addressing severe off-axis aberrations, researchers developed wide-angle metasurface doublet cameras (Fig. 4b) [13]. Cascading two amorphous silicon metasurfaces successfully corrects monochromatic aberrations. The focusing metasurface employs a highly optimized even-order polynomial phase profile \varphi \left(r\right):

(13)\varphi \left(r\right)=\frac{2\text{π}}{\lambda }\sum _{n=1}^N{a}_{2n}{\left(\frac{r}{R}\right)}^{2n},

where r is the radial coordinate, R is the normalization radius, and a_{2n} are the optimized coefficients. Cascading two planar metasurfaces completely replaces bulky refractive lens groups for wide-angle monochromatic aberration correction. This doublet achieves a FOV exceeding 60°.

To capture three-dimensional spatial data, researchers achieved monocular passive 4D imaging (Fig. 4c) [34]. A single-layer metalens generates polarization-decoupled conjugate single-helix point spread functions. The metasurface encodes a helical phase mask \varphi \left(r,\phi \right):

(14)\varphi \left(r,\phi \right)=-\frac{2\text{π}}{\lambda }ef\left(\sqrt{r^2+f^2}-f\right)+l\phi +k_{x}x,

where l is the topological charge governing the spiral phase, \phi is the azimuthal angle, and k_x introduces a lateral spatial carrier. Engineering complex helical wavefronts directly into the focusing element extracts depth information without macroscopic moving stages. This extracts intensity, depth, and polarization in a single shot.

Furthermore, minimizing the macroscopic imaging cavity volume remains a fundamental challenge. Researchers designed folded-lens configurations to diagonally fold the optical path (Fig. 4d) [35]. Three horizontally stacked metasurfaces on a glass substrate provide focusing and total internal reflection (TIR). To maintain TIR, the metasurface imparts an anomalous phase gradient \mathrm{\nabla }\varphi:

(15)\mathrm{\nabla }\varphi =\frac{2\text{π}}{\lambda }\left(n_{\mathrm{s}\mathrm{u}\mathrm{b}}\sin {\theta }_{\mathrm{T}\mathrm{I}\mathrm{R}}-n_{\mathrm{i}}\sin {\theta }_{\mathrm{i}\mathrm{n}}\right),

where n_{\text{sub}} is the substrate index, {\theta }_{\text{in}} is the incident angle, and {\theta }_{\text{TIR}} is the folding angle exceeding the critical angle. Using reflective metasurfaces to diagonally fold the optical path dramatically compresses the macroscopic imaging cavity volume. This yields a quasi-diffraction-limited camera with a 0.7 mm thickness.

Finally, researchers developed meta-grating-lens (MGL) single-layer polarization cameras (Fig. 4e) [36]. This device integrates a vector metagrating with a focusing metalens. The complex spatial Jones matrix J\left(x,y\right) superimposes four independent polarization channels:

(16)J\left(x,y\right)=\sum _{i=1}^4{P}_i\mathrm{e}\mathrm{x}\mathrm{p}\left[\mathrm{i}\left({\varphi }_{\mathrm{f}}\left(x,y\right)+k_i\cdot r\right)\right],

where P_{\mathrm{i}} denotes the polarization projection matrix, {\varphi }_{\text{f}} is the base focusing phase, and k_{\mathrm{i}} defines distinct spatial carrier wavevectors. Consolidating vectorial gratings and focusing profiles into one layer achieves full-Stokes imaging without separate macroscopic optics. This enables real-time multidimensional imaging within a fingernail-sized module.

Optical systems relying solely on metalenses represent a radical functional consolidation. This paradigm entirely discards bulky refractive elements. The ultrathin metalens independently executes primary focusing, aberration correction, and multidimensional information extraction. Innovations like helical point spread functions and folded optical paths aggressively miniaturize the volumetric footprint. They extract spectral, polarimetric, and depth data simultaneously. However, operating without refractive elements forces the metasurface to bear the entire optical burden. This often introduces severe trade-offs regarding broadband efficiency and numerical aperture. Nevertheless, this standalone architecture proves that nanostructured layers can autonomously resolve complex imaging tasks. It provides the definitive theoretical foundation for advanced optoelectronic integration. This naturally leads to physically bonding these ultrathin optical frontends directly onto semiconductor image sensors.

2.2 Mounting integration of metasurface with image sensors

This section investigates the mounting integration of metasurfaces. This hybrid approach physically bonds the nanostructured layer directly to commercial image sensors. It effectively collapses macroscopic optical paths into micrometer-scale adhesive gaps. The subsections herein are logically organized by the specific high-dimensional optical properties being decoded. Section 2.2.1 focuses on polarization management, detailing how bonded metasurfaces replace bulky cascaded polarimeters. Section 2.2.2 examines spectral routing, where non-absorptive metasurfaces supersede traditional Bayer filters to enable high-efficiency hyperspectral imaging. Finally, Section 2.2.3 explores the decoding of complex structured light, specifically focusing on orbital angular momentum (OAM) modes. This progression demonstrates how mounting integration systematically compresses specialized macroscopic optical benches into compact, functionalized sensor modules.

2.2.1 Mounting integration of metasurface array with image sensors

Hybrid mounting integration transitions traditional optoelectronics to a computational sensing paradigm. Bonded metasurfaces function as physical encoders. They precisely modulate the incident wavefront before photoelectric conversion.

Traditional microscopy fundamentally struggles with the trade-off between field-of-view (FOV) and numerical aperture (NA). To solve this, researchers integrated silicon metalenses directly onto CMOS sensors using polarization-multiplexed dual-phase designs (Fig. 5a) [37]. This architecture maintains high NA while significantly expanding the FOV.

Addressing the high power consumption of electronic computing, researchers developed on-chip multiplexed diffractive neural networks (MDNN) (Fig. 5b) [14]. According to the Huygens−Fresnel principle, each meta-atom acts as a secondary spherical wave source. Its local complex amplitude response follows the Jones matrix J_{\text{meta}}:

(17)\begin{array}{rl}& J_{\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{a}}\left(r^{\prime }\right)\\ & =R\left(\theta \left(r^{\prime }\right)\right)\left[\begin{array}{cc}{a}_x\left(r^{\prime }\right){\mathrm{e}}^{\mathrm{j}{\varphi }_x\left(r^{\prime }\right)}& 0\\ 0& a_y\left(r^{\prime }\right){\mathrm{e}}^{\mathrm{j}{\varphi }_y\left(r^{\prime }\right)}\end{array}\right]R\left(-\theta \left(r^{\prime }\right)\right),\end{array}

where R\left(\theta \right) is the rotation matrix, \theta \left(r^{\prime }\right) is the spatial orientation angle of the asymmetric nanostructure, while a_{x,y} and {\varphi }_{x,y} represent the amplitudes and phases along the orthogonal axes. Following this modulation, the light field undergoes Rayleigh−Sommerfeld diffraction along the z-direction to form the output layer. Crucially, the optical clear adhesive (OCA) defines the exact micron-scale propagation distance z required for this diffractive optical computing. This design executes parallel image classification with over 93% accuracy.

Furthermore, standard image sensors discard polarization data. To reconstruct this, researchers bonded metasurface polarization filter arrays onto CMOS chips (Fig. 5c) [15]. The measured intensity I_j at the j-th metasurface super-pixel relies on its specific structural Mueller matrix elements:

(18)I_j=\frac{1}{2}\sum _{i=0}^3{M}_{0i}^{\left(j\right)}{S}_i,

where M_{0i}^{\left(j\right)} represents the first-row elements of the Mueller matrix for the j-th pixel, and S_{\mathrm{i}} denotes the incident Stokes parameters. Direct UV-gluing perfectly aligns these structural filters with the underlying CMOS pixels. This precise mechanical fixation minimizes optical crosstalk for accurate matrix inversion.

Similarly, researchers accomplished single-shot spectroscopic ellipsometry using silicon metasurface arrays (Fig. 5d) [38]. The array’s structural dispersion encodes the broadband polarization state into spatial intensity distributions. The sensor captures the intensity vector I\left(\lambda \right):

(19)I\left(\lambda \right)=W\left(\lambda \right)S\left(\lambda \right),

where W\left(\lambda \right) is the structurally defined broadband measurement matrix, and S\left(\lambda \right) is the wavelength-dependent Stokes vector. Adhesive mounting ensures the sensor precisely captures this spatially encoded intensity before macroscopic diffraction blurring occurs. Iterative algorithms invert this matrix to reconstruct thin-film parameters.

Expanding dimensionality further, researchers monolithically integrated achromatic metalens arrays onto commercial CMOS sensors (Fig. 5e) [39]. These arrays correct chromatic dispersion across the visible spectrum, achieving diffraction-limited full-color light-field imaging.

Finally, extreme wide-angle perception typically demands bulky fisheye lenses. Integrating metalens arrays directly onto CMOS sensors solves this constraint (Fig. 5f) [21]. To compensate for severe off-axis aberrations, each sub-metalens incorporates an incident angle-dependent phase shift \varphi \left(x,y\right):

(20)\varphi \left(x,y\right)=-\frac{2\text{π}}{\lambda }\left(\sqrt{{\left(x-f\tan \alpha \right)}^2+y^2+f^2}-f\right),

where \lambda is the operational wavelength, \alpha is the incident angle, f is the focal length, and f\tan \alpha denotes the off-axis focal shift. Bonding the array to the sensor physically fixes the focal length f exactly to the thickness of the microscopic adhesive layer.

Mounting integration successfully unifies metasurface arrays and commercial CMOS sensors via OCA or UV-glue. This hybrid architecture effectively collapses macroscopic free-space optics into microscopic adhesive layers. It eradicates bulky discrete filters and enables single-shot computational sensing. However, this approach presents inherent physical limitations. It fundamentally relies on the precise mechanical alignment of separate substrates. Any submicron mismatch severely degrades performance and introduces optical crosstalk. Furthermore, the modulated light must propagate through the bonding substrate before reaching the photodiodes. During this transit, it converts into plane waves. This propagation weakens the near-field electromagnetic enhancement effect between the metasurface and the photosensitive layer.

2.2.2 Spectral separation and hyperspectral imaging in mounting integration

This subsection focuses on mounting metasurfaces onto image sensors for spectral management. It fundamentally transforms standard intensity detection into a precise spectral-spatial mapping process. The measured intensity I(x,y) at any sensor pixel follows a strict spectral integration:

(21)I\left(x,y\right)=\underset{{\lambda }_{\mathrm{m}\mathrm{i}\mathrm{n}}}{\overset{{\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}}{\int }}T\left(x,y,\lambda \right)S\left(\lambda \right)d\lambda ,

where T\left(x,y,\lambda \right) represents the structurally engineered spectral transmission or routing function, and S\left(\lambda \right) is the incident spectrum. Directly mounting the metasurface onto the sensor array physically defines the minimum focal distance. This ensures the routed photons accurately hit the corresponding microscopic photodiodes without macroscopic diffraction spreading. Fulfilling this mapping condition governs the evolution from simple color routing to complex hyperspectral reconstruction.

Traditional commercial color filters absorb over two-thirds of incident light. To overcome this photon waste, researchers integrated silicon nitride metasurfaces directly with monochromatic CMOS sensors (Fig. 6a) [40]. Inverse engineering designs deflect RGB light into specific RGGB quadrants within 2 µm unit cells. This pixel-level routing achieves an average energy utilization of 84%, significantly outperforming absorptive Bayer filters.

However, scaling pixel pitches down to the deep submicron regime severely worsens optical crosstalk. To break this geometric limit, researchers developed freeform metasurface color routers (Fig. 6b) [41]. They abandoned regular geometric nanopillars. Instead, they utilized fully differentiable topology optimization. This method maximizes the target photodiode area A_c for each color channel c:

(22)F={\sum }_{c\in \{R,G,B\}}{\int }_{{\lambda }_c}{\iint }_{A_c}\frac{1}{2}\mathrm{R}\mathrm{e}\left(E\times H^{\ast }\cdot \hat{z}\right)\mathrm{d}S\mathrm{d}\lambda ,

where E and H denote the local electromagnetic fields. Adhesive mounting rigidly locks this highly optimized freeform layer exactly above the 0.6 µm pixel array. This extreme physical proximity is absolutely mandatory. It allows the sensor to capture the tightly confined near-field routed photons before they diverge. This topology-optimized design yields a remarkable 3.5× flux increase over conventional filters.

Beyond simple RGB routing, researchers realized real-time ultraspectral imaging chips (Fig. 6c) [20]. They mounted an array of 155216 reconfigurable metasurface supercells directly onto a CMOS sensor. This system continuously covers the 450−750 nm range with an ultrahigh 0.8 nm resolution. It enables dynamic brain spectral monitoring and real-time hyperspectral video acquisition.

Fusing spectral and polarization dimensions typically demands massive optical benches. Researchers overcame this by demonstrating real-time hyperspectro-polarimetric imaging (Fig. 6d) [42]. An encoding metasurface integrated with a conventional sensor mathematically compresses both dimensions into a single spatial snapshot. Lightweight neural networks then rapidly reconstruct the high-dimensional data. This computational architecture achieves simultaneous hyperspectral and full-Stokes imaging at video rates (30 fps).

Table 2 provides a detailed quantitative comparison of the key performance metrics and structural architectures across the discussed integration approaches. The direct integration of metasurfaces for spectral management signifies a profound paradigm shift. It systematically replaces lossy organic dyes with nonabsorptive dielectric nanostructures. This fundamental transition resolves the severe photon waste inherent in standard commercial image sensors. Metasurfaces evolve the spectral acquisition mechanism from simple intensity attenuation to precise spatial photon routing. This intelligent redirection drastically amplifies the absolute quantum efficiency of the underlying semiconductor pixels. It completely shatters the discrete trichromatic limitation of conventional color imaging. Furthermore, encoding continuous broadband spectra directly into complex spatial patterns transforms the sensor array into a high-dimensional optical encoder. Coupled with neural network reconstruction, this architecture shifts the spectral resolution burden from physical path lengths to robust digital processing. Ultimately, this seamless fusion establishes a highly scalable framework for real-time ultraspectral perception.

2.2.3 Mounting integration of metasurface with image sensors for orbital angular momentum (OAM) mode sorting and detection

This subsection focuses on the direct integration of metasurfaces onto CMOS camera chips. This architecture enables the chip-scale sorting and detection of orbital angular momentum (OAM) modes. It overcomes the bulkiness and high crosstalk of traditional free-space optical systems. Traditional mode demultiplexing relies on cascaded optical assemblies to perform complex coordinate unwrapping. To compress this volume, researchers demonstrated a compact OAM sorter by integrating a double-layer TiO₂ metasurface doublet directly onto a CMOS chip (Fig. 7a) [43]. The core metasurface functions as a phase unwrapper, performing exact log-polar coordinate mapping via a continuous phase profile \varphi \left(x,y\right):

(23)\varphi \left(x,y\right)=\frac{2\text{π}a}{\lambda f}\left(y\arctan \left(\frac{y}{x}\right)-x\ln \left(\frac{\sqrt{x^2+y^2}}{b}\right)+x\right),

where a and b are conformal mapping scaling constants, \lambda is the wavelength, and f is the focal length. This conformal transformation strictly requires an exact Fourier propagation distance. Bonding the metasurface directly to the CMOS chip via an adhesive layer physically locks in this exact focal gap, preventing the unwrapped field from diffracting unpredictably. This device spatially separates modes from m = −3 to +3 with an average crosstalk of −6.43 dB.

Simultaneous identification of spin and orbital angular momentum introduces high complexity. Researchers resolved this by mounting a spin-controlled periodic gradient metasurface (PGS) onto commercial sensors (Fig. 7b) [44]. The metasurface modulates incident light through the Pancharatnam−Berry (PB) phase. Its slow-axis orientation angle \theta \left(x,y\right) follows a precise spatial distribution:

(24)\theta \left(x,y\right)=\frac{\text{π}x}{a-by}+\frac{\text{π}\left(a-c\right)x}{ac},

where a, b, and c act as specific transformation parameters. The first term provides the specific PGS phase for mode conversion. The second term introduces a linear gradient phase. This gradient spatially separates orthogonal spin angular momentum (SAM) components by a defined diffraction angle. Directly mounting this metasurface onto the image sensor rigidly defines the short propagation path for these diverging beams. This mechanical stability ensures the backend deep learning network receives highly repeatable spatial intensity arrays without macroscopic vibration errors. This system successfully identifies high-order modes (\left|l\right|\le 10) in real time.

The direct mounting of metasurfaces onto image sensors represents a critical milestone in structured light perception. By encoding spatial transformations into planar metasurfaces, researchers compress macroscopic optical benches into chip-scale devices. When physically bonded to a standard semiconductor sensor, this architecture yields an integrated mode receiver capable of real-time topological charge identification. However, structured light beams possess intricate phase singularities. They are inherently sensitive to spatial misalignment. The adhesive mounting process inevitably introduces mechanical alignment variations. These variations exacerbate modal crosstalk and restrict the detection fidelity of higher-order optical states. Resolving this sensitivity to lateral registration errors fundamentally necessitates the transition toward monolithic fabrication. In this ultimate paradigm, researchers must lithographically define the nanostructured metasurface directly upon the active pixel array.

2.3 Monolithic integration of metasurface with photodetectors

This section examines the monolithic integration of metasurfaces. By fabricating nanostructures directly upon active photoelectric semiconductors, this advanced paradigm leverages intense near-field coupling to fundamentally alter charge generation. The internal logic systematically navigates through distinct optoelectronic material platforms. Section 2.3.1 begins with silicon, demonstrating how monolithic localized resonances successfully break intrinsic bandgap limitations. Section 2.3.2 transitions to atomically thin 2D materials, utilizing engineered optical anisotropy to overcome inherently weak absorption. Section 2.3.3 explores solution-processed quantum dots and perovskites, achieving near-perfect absorption through precise critical coupling. Finally, Section 2.3.4 culminates this material evolution. It illustrates how these monolithic architectures empower true single-chip multidimensional parameter detection.

2.3.1 Metasurface-integrated silicon-based photodetectors

This subsection focuses on integrating metasurfaces directly with silicon or hybrid photodetectors. This monolithic architecture aims to boost fundamental responsivity, extend spectral detection bands, and introduce multidimensional discrimination capabilities.

Exploiting near-field enhancement, researchers miniaturized color-sensitive photodetectors using hybrid nanoantennas integrated with silicon structures (Fig. 8a) [45]. This design enables wavelength-selective absorption and RGB color routing at nanoscale pixel sizes. It achieves high color discrimination in highly compact sensors. Similarly, researchers realized compact angle-resolved spectrometers by monolithically integrating metasurfaces with photodetectors (Fig. 8b) [46]. Direct on-chip lithography eliminates macroscopic optical paths. This architecture utilizes phase-engineered dispersion to provide joint angle-spectral resolution.

Fundamentally, intrinsic silicon possesses a 1.1 eV bandgap. This physical limit renders it transparent and unresponsive to infrared (IR) telecommunication wavelengths (e.g., 1550 nm). To break this limitation, researchers combined structurally disordered silicon metasurfaces with upconversion nanoparticles (UCNPs) (Fig. 8c) [22]. To maximize the highly nonlinear multiphoton upconversion process, the design intentionally introduces spatial disorder into a periodic nanopillar array. The modified positions \left(x^i,y^i\right) of the nanopillars follow:

(25)\begin{array}{c}{x}^i=x_0^i+\left(\sigma \ast P_0\right)U_1\left(X,Y\right),\\ y^i=y_0^i+\left(\sigma \ast P_0\right)U_2\left(X,Y\right),\end{array}

where \left(x_0^i,y_0^i\right) denotes the original periodic coordinates, P_0 is the lattice periodicity, and U_1, U_2 are independent uniform distributions within \left[-1,1\right]. The parameter \sigma \in \left[0,1\right] strictly quantifies the degree of disorder. Introducing this structural disorder purposefully breaks translational symmetry and distorts the photonic band structure. Monolithic fabrication embeds the UCNPs exactly within the resulting Mie or Bragg resonant hotspots. This extreme spatial overlap exponentially amplifies the nonlinear upconversion efficiency. It achieves a room-temperature responsivity of 0.22 A/W and an external quantum efficiency of 17.6% at 1550 nm, successfully extending silicon’s response deep into the IR band.

Beyond pure silicon, researchers fabricated ultrafast single-chip optical receivers by wafer-bonding functional silicon nanopost metasurfaces to InGaAs membrane photodetector arrays (Fig. 8d) [47]. This hybrid monolithic platform executes intensity, phase, and polarization modulation simultaneously. Direct wafer bonding ensures the modulated near-field reaches the active InGaAs layer without free-space diffraction losses. It enables ultrahigh-speed 320 Gbit/s PAM4 and 240 Gbit/s 64QAM detection with a responsivity of 0.27 A/W and extremely low crosstalk.

The specific operational bandwidths, responsivity metrics, and underlying physical mechanisms of these silicon-integrated photodetectors are systematically compared in Table 3. Monolithic integration on silicon platforms uniquely leverages mature semiconductor manufacturing to upgrade classical photodetectors. Traditional silicon sensors are inherently blind to telecommunication wavelengths due to their rigid 1.1 eV bandgap. Fabricating disordered or resonant nanostructures directly upon the silicon substrate bypasses this fundamental material constraint. These integrated metasurfaces concentrate optical energy to trigger localized nonlinear multiphoton absorption. This direct manipulation successfully extends standard silicon responsivity deep into the infrared spectrum. Additionally, integrating phase-engineered structures facilitates ultrahigh-speed multidimensional signal demultiplexing directly at the detection interface. Ultimately, this architecture transforms cost-effective silicon arrays into broadband, high-speed optical receivers without requiring exotic bulk semiconductors.

2.3.2 Monolithic integration of metasurface with graphene and 2D materials

This subsection focuses on integrating metasurfaces directly with graphene and other 2D materials. This architecture achieves high polarization discrimination and enables zero-bias operation. It leverages nanoscale anisotropy for polarization selectivity while minimizing power consumption. Pristine 2D materials like graphene exhibit extraordinary carrier mobility but suffer from exceedingly weak optical absorption. To overcome this, researchers realized zero-bias mid-infrared graphene photodetectors using metasurface-assisted bulk photoresponse (Fig. 9a) [23]. The metasurface structurally breaks the uniform absorption profile. This generates a steep local electron temperature gradient \mathrm{\nabla }{T}_{\text{e}}\left(x\right). The resulting zero-bias photo-thermoelectric (PTE) current I_{\text{PTE}} follows:

(26)I_{\mathrm{P}\mathrm{T}\mathrm{E}}=\frac{1}{R}\int S\left(x\right)\mathrm{\nabla }{T}_{\mathrm{e}}\left(x\right)\mathrm{d}x,

where R is the device resistance, and S\left(x\right) is the spatially varying Seebeck coefficient. Monolithically fabricating asymmetric metasurfaces directly onto graphene strictly dictates this highly localized thermal gradient. This direct contact maximizes asymmetric near-field absorption within the atomic lattice before thermal diffusion occurs. This mechanism provides calibration-free, highly sensitive polarization detection without dark current noise.

Expanding functionality to multi-polarization, researchers integrated chiral plasmonic metasurfaces with graphene-silicon heterojunctions (Fig. 9b) [48]. This structure enables near-infrared full-Stokes polarization detection. The metasurface dictates chiral selection, converting specific polarization states into localized near-field enhancements to drive the underlying heterojunction.

Transitioning to other van der Waals materials, researchers integrated metallic metasurface gratings with 2D material photodiodes (e.g., black phosphorus and MoS₂) (Fig. 9c) [24]. This configuration demonstrates highly polarization-sensitive mid-infrared detection at zero bias. The metasurface gratings provide targeted absorption enhancement. They achieve high responsivity and polarization extinction ratios exceeding 20 dB. Finally, researchers achieved broadband multidimensional detection using metasurface-assisted graphene structures (Fig. 9d) [49]. This multi-port design realizes the simultaneous detection and discrimination of polarization and wavelength. It covers a massive 1–8 µm spectral bandwidth with a wavelength prediction accuracy of 0.5 µm. This efficiently extracts multiple parameters within a highly compact footprint.

To provide a clearer overview of the technological progression, Table 4 outlines the core metrics, working wavelengths, and fabrication compatibility of the aforementioned monolithic devices. The monolithic integration of metasurfaces with atomically thin 2D materials establishes an advanced paradigm for low-noise sensing. Pristine materials like graphene possess exceptional carrier mobility. However, their innate optical responsivity is fundamentally limited by minimal atomic-layer absorption. Fabricating symmetry-breaking metasurfaces directly upon these van der Waals interfaces overcomes this physical barrier. It provides intense near-field electromagnetic confinement. More crucially, the engineered optical anisotropy actively governs the spatial distribution of generated electron-hole pairs. This asymmetric localized excitation perfectly facilitates zero-bias operation. It completely eradicates dark current noise while delivering exceptional polarization extinction ratios. Consequently, the integrated device seamlessly extracts complex polarization vectors and broad spectral data simultaneously. This direct material fusion elegantly transforms intrinsically isotropic planar nanomaterials into highly sensitive, self-powered multidimensional optical decoders.

2.3.3 Monolithic integration of metasurface with quantum dot and perovskite

This subsection focuses on the monolithic integration of metasurfaces with quantum dots (QDs) and perovskites. These solution-processed materials often suffer from weak optical absorption in ultrathin films. To achieve near-perfect absorption, the integrated metasurface must fulfill the critical coupling condition. According to coupled-mode theory, the total absorption A\left(\omega \right) of the integrated system follows:

(27)A\left(\omega \right)=\frac{4{\gamma }_{\mathrm{r}\mathrm{a}\mathrm{d}}{\gamma }_{\mathrm{a}\mathrm{b}\mathrm{s}}}{{\left(\omega -{\omega }_0\right)}^2+{\left({\gamma }_{\mathrm{r}\mathrm{a}\mathrm{d}}+{\gamma }_{\mathrm{a}\mathrm{b}\mathrm{s}}\right)}^2},

where {\omega }_0 is the resonant frequency. The term {\gamma }_{\text{rad}} is the radiative decay rate of the nanostructure. The term {\gamma }_{\text{abs}} represents the intrinsic absorptive decay rate of the active photoelectric material. Monolithic integration precisely tailors the nanostructure geometries directly upon the active film. This physical proximity allows engineers to match the radiative leakage ({\gamma }_{\text{rad}}) exactly with the material’s absorption rate ({\gamma }_{\text{abs}}). This critical coupling maximizes near-field confinement directly within the active layer.

Applying this principle, researchers utilized plasmonic enhancement in HgTe colloidal QD devices for mid-infrared detection (Fig. 10a) [50]. Integrated plasmonic cavities boost responsivity by aligning localized resonances with the 3–5 μm atmospheric window. This enables highly sensitive single-pixel scanning thermal imaging.

Further advancing monolithic optics-electronics integration (MOEI), researchers combined silicon metasurface cavities with PbS/PbSe CQDs (Fig. 10b) [16]. This low-temperature micro-nano integration process achieves near-perfect absorption (>90%) in the short-wave infrared (SWIR). By satisfying the impedance matching condition, the structure yields an exceptional photoconductive responsivity up to 8000 A/W.

Extending to the ultraviolet and visible spectra, researchers integrated hybrid organic-inorganic perovskites (HOIP) directly into metasurfaces (Fig. 10c) [51]. This architecture leverages strong Mie resonances to increase broadband photocurrent by over ten times. It maintains rapid response times (<5.1 μs), supporting dynamic optical communications and image transmission. Finally, researchers demonstrated high-responsivity mid-infrared photodetectors using metamaterial-enhanced PbSe/PbS heterojunctions (Fig. 10d) [17]. Integrating sintered CQDs directly with metallic perfect absorbers generates intense localized dipole resonances. This architecture achieves a 20-fold overall responsivity enhancement. Consequently, it enables highly robust mid-infrared detection at room temperature.

Table 5 presents a systematic comparison of the resonant mechanisms, target spectra, and key performance enhancements for the discussed metasurface-integrated quantum dot and perovskite photodetectors. Monolithic integration with solution-processed semiconductors resolves the contradiction between optical absorption depth and carrier diffusion length. While ultrathin films ensure efficient charge extraction, they inherently suffer from insufficient light-matter interaction. Embedding resonant metasurfaces directly onto these active layers engineers profound near-field electromagnetic confinement. This architecture traps incident photons via localized resonances, amplifying absorption to near unity. Consequently, this synergistic fusion delivers exceptional photoelectric responsivity. It crucially empowers robust SWIR and mid-infrared perception at room temperature, successfully eliminating the need for bulky cryogenic cooling.

2.3.4 Monolithic integration of metasurface with multidimensional and multi-parameter detectors

This subsection focuses on the direct monolithic integration of metasurfaces with photodetectors. This architecture enables simultaneous multidimensional parameter detection, including full-Stokes polarization, wavelength, and angle. It supports single-chip, high-precision multidimensional optoelectronic responses.

Traditional polarimetry fundamentally relies on macroscopic Stokes−Mueller matrices. To bypass this, researchers realized chip-scale full-Stokes polarimeters utilizing optoelectronic polarization eigenvectors (OPEVs) (Fig. 11a) [52]. At the fundamental level, the photocurrent density J_{\text{ph},i} originates from the phenomenological expansion of electric fields:

(28)J_{\mathrm{p}\mathrm{h},i}={\beta }_{ijk}\widetilde{E_j}\widetilde{E_k^{\ast }}+{\gamma }_{ijkl}\widetilde{E_j}\widetilde{E_k^{\ast }}{E}_l,

where \tilde{E} represents the oscillating optical field. E_{\mathrm{l}} denotes the static electric field, such as a built-in depletion field. The tensors {\beta }_{ijk} and {\gamma }_{ijkl} dictate the material's intrinsic photoelectric response. Monolithic integration precisely engineers the local oscillating field \tilde{E} exactly within the semiconductor’s active region. This tight near-field coupling directly programs the intrinsic optoelectronic tensors. This single-shot measurement achieves high Stokes accuracy within a CMOS-compatible chip. Furthermore, researchers designed chiral dielectric metasurfaces for mid-infrared detection, where the physical image matches a Fabry−Pérot (FP) cavity composed of a high-low-high refractive index ternary system (Fig. 11b) [53]. To optimize the chiral response, the device is modeled as a series of radiant power sources using the multipole moment expansion method. The total scattering power P is expressed as the summation of contributions from electric and magnetic multipoles:

(29)P=\frac{2{\omega }^4}{3c^3}{\left|p\right|}^2+\frac{2{\omega }^4}{3c^3}{\left|M\right|}^2+\frac{{\omega }^6}{5c^5}{Q}_{\alpha \beta }{Q}_{\alpha \beta }+\frac{{\omega }^6}{20c^5}{M}_{\alpha \beta }{M}_{\alpha \beta },

where p and M are the electric and magnetic dipole moments, while Q_{\alpha \beta } and M_{\alpha \beta } represent the electric and magnetic quadrupole moments, respectively. Monolithic integration effectively couples these high-order radiant sources directly to the active HgCdTe or InGaAs layers. By suppressing non-radiative losses in the ternary FP cavity, this architecture achieves efficient circular-to-linear polarization conversion. This six-pixel technique supports high-extinction-ratio full-Stokes measurement at room temperature.

Monolithic integration for multidimensional detection fundamentally transcends simple intensity measurement. Rather than relying on cascaded free-space optical matrices, this architecture fuses wavefront modulation directly with the carrier excitation process. By programming specific optoelectronic polarization eigenvectors (OPEVs) into the active region, the metasurface dictates the localized near-field coupling. This direct physical interaction seamlessly maps high-dimensional optical parameters, such as full-Stokes states, directly into measurable photocurrents. Ultimately, this paradigm establishes a highly scalable, CMOS-compatible framework. It completely transforms standard pixels into powerful, single-shot multidimensional optical decoders.

3 Conclusion and outlook

Metasurface technology has fundamentally transformed modern optical design. This review has traced its evolutionary trajectory from isolated wavefront modulation components to fully integrated optoelectronic systems. This progression spans optical-system-level integration, mounting integration, and monolithic integration. While each level offers functional advantages, they present unique physical and engineering challenges.

Optical-system-level integration scales metasurfaces to macroscopic apertures. However, it faces limits regarding volumetric constraints. Decoupling complex dispersion across massive bandwidths without inducing aberrations remains difficult. Future efforts require global inverse design algorithms. These must co-optimize macroscopic refractive lenses and microscopic phase distributions simultaneously. This approach minimizes the physical footprint while maximizing absolute transmission efficiency.

Mounting integration effectively compresses optical benches into chip-scale devices. Yet, adhesive bonding inherently introduces mechanical alignment variations and micrometer-scale gaps. This physical separation exacerbates modal crosstalk and prevents sensors from capturing strongly enhanced near-field energy. To address this, researchers must optimize advanced experimental equipment. Advancing toward high-precision wafer-scale alignment and bonding processes is crucial. Ultimately, resolving this extreme sensitivity to lateral registration errors necessitates a complete transition toward direct monolithic integration.

Monolithic integration represents the ultimate optoelectronic convergence. This architecture involves fabricating nanostructures directly upon active photoelectric regions. Alternatively, it involves patterning metasurfaces directly onto commercial CMOS image sensors without additional substrates. Despite its immense potential, this paradigm faces significant practical challenges. These include complex optoelectronic co-design, severe material compatibility issues, and the immaturity of expensive large-scale manufacturing. Standard high-temperature CMOS fabrication often degrades chemically sensitive optoelectronic layers. Therefore, developing highly adaptable low-temperature micro-nano integration processes is essential. This is particularly critical for emerging optoelectronic materials like colloidal quantum dots and organic compounds. Furthermore, advancing large-area wafer-scale nanoimprint lithography will significantly reduce massive manufacturing costs. This development simultaneously ensures high detection efficiency and advanced imaging functionality.

Translating these integrated metasurfaces from laboratory prototypes to commercial manufacturing remains the ultimate objective. Overcoming the distinct challenges across these three integration levels will establish highly robust sensing platforms. These ultracompact computational sensors will directly empower several advanced optical applications. First, fusing high-dimensional metasurface encoders with deep learning enables single-shot computational imaging. This paradigm supports real-time biomedical diagnostics and complex material inspection without macroscopic optical benches. Second, integrated broadband achromatic metalenses provide the foundation for next-generation spatial computing. They facilitate ultra-compact, aberration-free near-eye displays for augmented and virtual reality systems. Finally, monolithically integrating metasurfaces with emerging active materials yields highly efficient room-temperature infrared detectors. This structural innovation accelerates the direct deployment of low-cost, uncooled short-wave and mid-infrared imaging arrays.

References

[1] Yang , Y. , Seong , J. , Choi , M. , Park , J. , Kim , G. , Kim , H. , Jeong , J. , Jung , C. , Kim , J. , Jeon , G. , Lee , K. , Yoon , D.H. , Rho , J. : Integrated metasurfaces for re-envisioning a near-future disruptive optical platform. Light Sci. Appl 12(1), 152(2023)

[2] Ha , S.T. , Li , Q. , Yang , J.K.W. , Demir , H.V. , Brongersma , M.L. , Kuznetsov , A.I. : Optoelectronic metadevices. Science 386(6725), eadm7442(2024)

[3] Yu , N. , Genevet , P. , Kats , M.A. , Aieta , F. , Tetienne , J.P. , Capasso , F. , Gaburro , Z. : Light propagation with phase discontinuities: generalized laws of reflection and refraction. Science 334(6054), 333–337(2011)

[4] Khorasaninejad , M. , Chen , W.T. , Devlin , R.C. , Oh , J. , Zhu , A.Y. , Capasso , F. : Metalenses at visible wavelengths: Diffraction-limited focusing and subwavelength resolution imaging. Science 352(6290), 1190–1194(2016)

[5] Chen , W.T. , Zhu , A.Y. , Sanjeev , V. , Khorasaninejad , M. , Shi , Z. , Lee , E. , Capasso , F. : A broadband achromatic metalens for focusing and imaging in the visible. Nat. Nanotechnol 13(3), 220–226(2018)

[6] Wang , S. , Wu , P.C. , Su , V.C. , Lai , Y.C. , Chen , M.K. , Kuo , H.Y. , Chen , B.H. , Chen , Y.H. , Huang , T.T. , Wang , J.H. , Lin , R.M. , Kuan , C.H. , Li , T. , Wang , Z. , Zhu , S. , Tsai , D.P. : A broadband achromatic metalens in the visible. Nat. Nanotechnol 13(3), 227–232(2018)

[7] Fan , Z.B. , Qiu , H.Y. , Zhang , H.L. , Pang , X.N. , Zhou , L.D. , Liu , L. , Ren , H. , Wang , Q.H. , Dong , J.W. : A broadband achromatic metalens array for integral imaging in the visible. Light Sci. Appl 8(1), 67(2019)

[8] Arbabi , E. , Kamali , S.M. , Arbabi , A. , Faraon , A. : Full-Stokes imaging polarimetry using dielectric metasurfaces. ACS Photonics 5(8), 3132–3140(2018)

[9] Rubin , N.A. , D’Aversa , G. , Chevalier , P. , Shi , Z. , Chen , W. , Capasso , F. : Matrix Fourier optics enables a compact full-Stokes polarization camera. Science 365(6448), eaax1839(2019)

[10] Hua , X. , Wang , Y. , Wang , S. , Zou , X. , Zhou , Y. , Li , L. , Yan , F. , Cao , X. , Xiao , S. , Tsai , D.P. , Han , J. , Wang , Z. , Zhu , S. : Ultra-compact snapshot spectral light-field imaging. Nat. Commun 13(1), 2732(2022)

[11] Chen , W.T. , Zhu , A.Y. , Sisler , J. , Huang , Y.W. , Yousef , K.M.A. , Lee , E. , Qiu , C.W. , Capasso , F. : Broadband achromatic metasurface-refractive optics. Nano Lett 18(12), 7801–7808(2018)

[12] Balli , F. , Sultan , M. , Lami , S.K. , Hastings , J.T. : A hybrid achromatic metalens. Nat. Commun 11(1), 3892(2020)

[13] Arbabi , A. , Arbabi , E. , Kamali , S.M. , Horie , Y. , Han , S. , Faraon , A. : Miniature optical planar camera based on a wide-angle metasurface doublet corrected for monochromatic aberrations. Nat. Commun 7(1), 13682(2016)

[14] Luo , X. , Hu , Y. , Ou , X. , Li , X. , Lai , J. , Liu , N. , Cheng , X. , Pan , A. , Duan , H. : Metasurface-enabled on-chip multiplexed diffractive neural networks in the visible. Light Sci. Appl 11(1), 158(2022)

[15] Zuo , J. , Bai , J. , Choi , S. , Basiri , A. , Chen , X. , Wang , C. , Yao , Y. : Chip-integrated metasurface full-Stokes polarimetric imaging sensor. Light Sci. Appl 12(1), 218(2023)

[16] Đorđević , N. , Schwanninger , R. , Yarema , M. , Koepfli , S. , Yarema , O. , Salamin , Y. , Lassaline , N. , Cheng , B. , Yazdani , N. , Dorodnyy , A. , Fedoryshyn , Y.M. , Wood , V. , Leuthold , J. : Metasurface colloidal quantum dot photodetectors. ACS Photonics 9(2), 482–492(2022)

[17] Schwanninger , R. , Koepfli , S.M. , Yarema , O. , Dorodnyy , A. , Yarema , M. , Moser , A. , Nashashibi , S. , Fedoryshyn , Y. , Wood , V. , Leuthold , J. : Highly responsive mid-infrared metamaterial enhanced heterostructure photodetector formed out of sintered PbSe/PbS colloidal quantum dots. ACS Appl. Mater. Interfaces 15(8), 10847–10857(2023)

[18] Sawant , R. , Andrén , D. , Martins , R.J. , Khadir , S. , Verre , R. , Käll , M. , Genevet , P. : Aberration-corrected large-scale hybrid metalenses. Optica 8(11), 1405–1411(2021)

[19] Liu , M. , Zhao , W. , Wang , Y. , Huo , P. , Zhang , H. , Lu , Y. , Xu , T. : Achromatic and coma-corrected hybrid meta-optics for high-performance thermal imaging. Nano Lett 24(25), 7609–7615(2024)

[20] Xiong , J. , Cai , X. , Cui , K. , Huang , Y. , Yang , J. , Zhu , H. , Li , W. , Hong , B. , Rao , S. , Zheng , Z. , Xu , S. , He , Y. , Liu , F. , Feng , X. , Zhang , W. : Dynamic brain spectrum acquired by a real-time ultraspectral imaging chip with reconfigurable metasurfaces. Optica 9(5), 461–468(2022)

[21] Chen , J. , Ye , X. , Gao , S. , Chen , Y. , Zhao , Y. , Huang , C. , Qiu , K. , Zhu , S. , Li , T. : Planar wide-angle-imaging camera enabled by metalens array. Optica 9(4), 431–437(2022)

[22] Chen , W. , Zhang , S. , Wang , C. , Wu , Y. , Shi , X. , Shen , J. , Liu , Y. , Zhang , X. , Tjiptoharsono , F. , Lee , H.Y.L. , Zhu , D. , Wang , Q. , Yang , J.K.W. , Zhu , J. , Dong , Z. : Enabling highly efficient infrared silicon photodetectors via disordered metasurfaces with upconversion nanoparticles. Sci. Adv 11(43), eadx7783(2025)

[23] Wei , J. , Li , Y. , Wang , L. , Liao , W. , Dong , B. , Xu , C. , Zhu , C. , Ang , K.W. , Qiu , C.W. , Lee , C. : Zero-bias mid-infrared graphene photodetectors with bulk photoresponse and calibration-free polarization detection. Nat. Commun 11(1), 6404(2020)

[24] Lien , M.R. , Wang , N. , Guadagnini , S. , Wu , J. , Soibel , A. , Gunapala , S.D. , Wang , H. , Povinelli , M.L. : Black phosphorus molybdenum disulfide midwave infrared photodiodes with broadband absorption-increasing metasurfaces. Nano Lett 23(21), 9980–9987(2023)

[25] Zeng , Y. , Zhong , H. , Long , Z. , Cao , H. , Jin , X. : From performance to structure: a comprehensive survey of advanced metasurface design for next-generation imaging. npj. Nanophotonics 2(1), 39(2025)

[26] Pan , C.F. , Wang , H. , Wang , H. , S , P.N. , Ruan , Q. , Wredh , S. , Ke , Y. , Chan , J.Y.E. , Zhang , W. , Qiu , C.W. , Yang , J.K.W. : 3D-printed multilayer structures for high-numerical aperture achromatic metalenses. Sci. Adv 9(51), eadj9262(2023)

[27] Chang , S. , Zhang , L. , Duan , Y. , Rahman , M.T. , Islam , A. , Ni , X. : Achromatic metalenses for full visible spectrum with extended group delay control via dispersion-matched layers. Nat. Commun 15(1), 9627(2024)

[28] Chu , Y. , Xiao , X. , Ye , X. , Chen , C. , Zhu , S. , Li , T. : Design of achromatic hybrid metalens with secondary spectrum correction. Opt. Express 31(13), 21399–21406(2023)

[29] Wu , P. , Li , X. , Xing , Y. , Wang , J. , Zheng , W. , Wang , Z. , Ma , Y. : Broadband achromatic hybrid metalens module with 100° field of view for visible imaging. Sensors (Basel) 25(10), 3202(2025)

[30] Yang , Z. , Wang , Z. , Wang , Y. , Feng , X. , Zhao , M. , Wan , Z. , Zhu , L. , Liu , J. , Huang , Y. , Xia , J. , Wegener , M. : Generalized Hartmann-Shack array of dielectric metalens sub-arrays for polarimetric beam profiling. Nat. Commun 9(1), 4607(2018)

[31] Fan , Q. , Xu , W. , Hu , X. , Zhu , W. , Yue , T. , Yan , F. , Lin , P. , Chen , L. , Song , J. , Lezec , H.J. , Agrawal , A. , Lu , Y. , Xu , T. : Disordered metasurface enabled single-shot full-Stokes polarization imaging leveraging weak dichroism. Nat. Commun 14(1), 7180(2023)

[32] Li , L. , Wang , S. , Zhao , F. , Zhang , Y. , Wen , S. , Chai , H. , Gao , Y. , Wang , W. , Cao , L. , Yang , Y. : Single-shot deterministic complex amplitude imaging with a single-layer metalens. Sci. Adv 10(1), eadl0501(2024)

[33] Khorasaninejad , M. , Chen , W.T. , Zhu , A.Y. , Oh , J. , Devlin , R.C. , Rousso , D. , Capasso , F. : Multispectral chiral imaging with a metalens. Nano Lett 16(7), 4595–4600(2016)

[34] Shen , Z. , Zhao , F. , Jin , C. , Wang , S. , Cao , L. , Yang , Y. : Monocular metasurface camera for passive single-shot 4D imaging. Nat. Commun 14(1), 1035(2023)

[35] Kim , Y. , Choi , T. , Lee , G.Y. , Kim , C. , Bang , J. , Jang , J. , Jeong , Y. , Lee , B. : Metasurface folded lens system for ultrathin cameras. Sci. Adv 10(44), eadr2319(2024)

[36] Li , F.J. , Feng , Z. , Wan , Z. , Shi , T. , Qiu , S. , Zeng , J.H. , Song , S. , Wang , S. , Deng , Z.L. , Li , X. : Meta-grating-lens-based monolithic polarization camera. Sci. Adv 11(43), eadx9886(2025)

[37] Xu , B. , Li , H. , Gao , S. , Hua , X. , Yang , C. , Chen , C. , Yan , F. , Zhu , S. , Li , T. : Metalens-integrated compact imaging devices for wide-field microscopy. Adv. Photonics 2(6), 066004(2020)

[38] Wen , S. , Xue , X. , Wang , S. , Ni , Y. , Sun , L. , Yang , Y. : Metasurface array for single-shot spectroscopic ellipsometry. Light Sci. Appl 13(1), 88(2024)

[39] Zhang , Y. , Jiang , X. , Qu , G. , Han , J. , Li , C. , Bo , B. , Ruan , Q. , Liu , Z. , Song , Q. , Xiao , S. : On-chip integration of achromatic metalens arrays. Nat. Commun 16(1), 7485(2025)

[40] Zou , X. , Zhang , Y. , Lin , R. , Gong , G. , Wang , S. , Zhu , S. , Wang , Z. : Pixel-level Bayer-type colour router based on metasurfaces. Nat. Commun 13(1), 3288(2022)

[41] Kim , C. , Hong , J. , Jang , J. , Lee , G.Y. , Kim , Y. , Jeong , Y. , Lee , B. : Freeform metasurface color router for deep submicron pixel image sensors. Sci. Adv 10(22), eadn9000(2024)

[42] Zhang , L. , Zhou , C. , Liu , B. , Ding , Y. , Ahn , H.J. , Chang , S. , Duan , Y. , Rahman , M.T. , Xia , T. , Chen , X. , Liu , Z. , Ni , X. : Real-time machine learning-enhanced hyperspectro-polarimetric imaging via an encoding metasurface. Sci. Adv 10(36), eadp5192(2024)

[43] Cheng , J. , Sha , X. , Zhang , H. , Chen , Q. , Qu , G. , Song , Q. , Yu , S. , Xiao , S. : Ultracompact orbital angular momentum sorter on a CMOS chip. Nano Lett 22(10), 3993–3999(2022)

[44] Li , L. , Gao , L. , Cheng , Y. , Zhou , S. , Wang , J. , Yu , H. , Gao , C. , Fu , S. : Metasurface-based intelligent identification of total angular momentum spectra for beams. ACS Photonics 12(3), 1418–1425(2025)

[45] Ho , J. , Dong , Z. , Leong , H.S. , Zhang , J. , Tjiptoharsono , F. , Daqiqeh Rezaei, S. , Goh , K.C.H. , Wu , M. , Li , S. , Chee , J. , Wong , C.P.Y. , Kuznetsov , A.I. , Yang , J.K.W. : Miniaturizing color-sensitive photodetectors via hybrid nanoantennas toward submicrometer dimensions. Sci. Adv 8(47), eadd3868(2022)

[46] Cai , G. , Li , Y. , Zhang , Y. , Jiang , X. , Chen , Y. , Qu , G. , Zhang , X. , Xiao , S. , Han , J. , Yu , S. , Kivshar , Y. , Song , Q. : Compact angle-resolved metasurface spectrometer. Nat. Mater 23(1), 71–78(2024)

[47] Soma , G. , Akazawa , T. , Kato , E. , Komatsu , K. , Takenaka , M. , Nakano , Y. , Tanemura , T. : Ultrafast one-chip optical receiver with functional metasurface. Nat. Commun 16(1), 10070(2025)

[48] Li , L. , Wang , J. , Kang , L. , Liu , W. , Yu , L. , Zheng , B. , Brongersma , M.L. , Werner , D.H. , Lan , S. , Shi , Y. , Xu , Y. , Wang , X. : Monolithic full-stokes near-infrared polarimetry with chiral plasmonic metasurface integrated graphene-silicon photodetector. ACS Nano 14(12), 16634–16642(2020)

[49] Jiang , H. , Chen , Y. , Guo , W. , Zhang , Y. , Zhou , R. , Gu , M. , Zhong , F. , Ni , Z. , Lu , J. , Qiu , C.W. , Gao , W. : Metasurface-enabled broadband multidimensional photodetectors. Nat. Commun 15(1), 8347(2024)

[50] Tang , X. , Ackerman , M.M. , Guyot-Sionnest , P. : Thermal imaging with plasmon resonance enhanced HgTe colloidal quantum dot photovoltaic devices. ACS Nano 12(7), 7362–7370(2018)

[51] He , J. , Li , C.Y. , Qi , D.X. , Cai , Q. , Liu , Y. , Fan , R.H. , Su , J. , Huo , P. , Xu , T. , Peng , R. , Wang , M. : Improving photoelectric conversion with broadband perovskite metasurface. Nano Lett 22(16), 6655–6663(2022)

[52] Deng , J. , Shi , M. , Liu , X. , Zhou , J. , Qin , X. , Wang , R. , Zhen , Y. , Dai , X. , Chen , Y. , Wei , J. , Ni , Z. , Gao , W. , Qiu , C.W. , Chen , X. : An on-chip full-Stokes polarimeter based on optoelectronic polarization eigenvectors. Nat. Electron 7(11), 1004–1014(2024)

[53] Cheng , B. , Song , G. : Full-Stokes polarization photodetector based on the hexagonal lattice chiral metasurface. Opt. Express 31(19), 30993–31004(2023)

RIGHTS & PERMISSIONS

The author(s)