Inverse-Designed Microstructured Phase Masks

• Inverse-designed microstructured phase masks are engineered optical elements that use computational optimization to tailor wavefront transformations with sub-wavelength precision.
• They integrate diverse fabrication techniques—such as grayscale lithography, direct laser writing, and 3D nanoprinting—to achieve reconfigurability and high efficiency in 3D light manipulation.
• Advanced gradient-based methods and adjoint simulations ensure robust scalability and performance metrics, including high focusing efficiency and extended depth-of-focus.

Inverse-designed microstructured phase masks are engineered optical elements whose geometric or material profiles are optimized through computational inverse-design methodologies to achieve specified wavefront transformations, spatial intensity distributions, or device functionalities. Unlike canonical analytic phase masks—such as cubic phase plates or log-aspheres—these devices exploit digital microstructural control (often at the sub-wavelength scale) and algorithmic optimization to realize precise target responses, including reconfigurability, arbitrary shaping, and single-shot 3D light patterning. Architectures span diffractive, metasurface, and phase-change material platforms, and fabrication approaches encompass grayscale lithography, direct-laser writing, UV casting, and nanoprinting. Recent demonstrations include reconfigurable photonic multiplexers, ultra-low-f/# lens arrays, depth-of-focus engineering, and volumetric holographic 3D printing (Wu et al., 2024, Hayward et al., 2024, Lin et al., 10 Jan 2026, Sturges et al., 2024, Bayati et al., 2021, Rondi et al., 2010, López-Pastor et al., 2019).

1. Mathematical Formulation and Optimization Workflows

Inverse-designed microstructured phase masks are formulated as constrained optimization problems over a set of spatially resolved (continuous or discrete) design variables such as surface-relief heights, pillar dimensions, or local dielectric permittivities. The optimization seeks to minimize a cost functional $J$ (often a weighted sum of squared errors between target and simulated field quantities), with regularization and fabrication constraints imposed via penalty terms or direct variable bounds.

A representative formulation for phase-relief masks involves optimizing the height map $h(x,y)$ (possibly quantized) to reproduce a target 3D energy distribution in a photopolymerizable resin. The cost functional is:

$$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$

where $I(Z_i;x,y;h)$ is the simulated intensity via angular-spectrum propagation, $T(Z_i;x,y)$ is the target volumetric dose, and $R[h]$ ensures surface smoothness and manufacturability (Lin et al., 10 Jan 2026).

For metasurface phase masks, the design variables may encode pillar diameters $d_j$ or fin widths $w_j$, usually parameterized over a unit-cell lattice. The objective often maximizes the minimum on-axis intensity over a prescribed focal depth for extended depth of focus applications:

$$\max_{p,\,t}\;\;t\qquad\textrm{subject to}\;\;I(z_i;p)\ge t,\;\;p_\mathrm{min}\le p\le p_\mathrm{max}$$

(Bayati et al., 2021, Sturges et al., 2024).

Adjoint-based gradient computation is typical, enabling efficient updates via backpropagation of field sensitivities for both electromagnetic and Fourier-optics models. Optimization algorithms span steepest-descent, L-BFGS, MMA (method-of-moving-asymptotes), gradient-free Bayesian routines, and direct binary search, selected according to problem differentiability and dimensionality (Hayward et al., 2024, Sturges et al., 2024).

2. Microstructural Platforms: Geometries and Materials

Inverse-designed phase masks utilize diverse microstructural platforms depending on application bandwidth, reconfigurability requirements, and fabrication capabilities:

• Phase-change PCM platforms: Sb$_2$Se$h(x,y)$0 thin films are pixelated into grids (100 nm × 100 nm × 30 nm) allowing pixel-wise reversible switching between amorphous ($h(x,y)$1) and crystalline ($h(x,y)$2) states, $h(x,y)$3; implemented via direct-laser writing for device adaptability (Wu et al., 2024).
• Multilevel diffractive relief: MLA structures are designed as concentric rings, each with quantized heights ($h(x,y)$4 levels, $h(x,y)$5 μm, ring width $h(x,y)$6 μm); these structures serve as ultra-low-f/#, high-NA microlenses (Hayward et al., 2024).
• Meta-optics: SiN pillars and nanofins (diameters and widths ~100–300 nm; heights ~600 nm) form locally periodic metasurfaces for broadband EDOF imaging; symmetry constraints (C$h(x,y)$7) are imposed for polarization independence (Bayati et al., 2021).
• 3D nanoprinted free-form phase plates: Phase mask surfaces are parameterized by radial height $h(x,y)$8, fabricated by two-photon laser nanoprinting with sub-wavelength accuracy (mean error <25 nm over a 1–2 mm aperture) (Sturges et al., 2024).

3. Fabrication Protocols and Resolution Limits

Fabrication methods are tailored to microstructural constraints and material platforms:

• Grayscale optical lithography: For large-area phase masks and high-resolution diffractive structures, exposure-dose calibration yields up to 256 height levels per pixel, with relief tolerances <10 nm (Lin et al., 10 Jan 2026, Hayward et al., 2024).
• Direct laser writing: Utilized for reconfigurable PCM devices, enabling feature sizes as small as 200–300 nm and pixel-wise phase control (Wu et al., 2024).
• UV-casting replication: Facilitates mass production of MLAs on flexible polymer films, transferring master-etched grayscale patterns (Hayward et al., 2024).
• Two-photon 3D nanoprinting: Provides sub-micron lateral resolution and axial control (~300–500 nm), enabling “stitched” free-form masks with post-fabrication pre-compensation for systematic deviations (Sturges et al., 2024).

Feature size limits (typically ~100–700 nm lateral, 10–100 nm vertical precision) and quantization granularity constrain designable phase profiles; regularization during optimization manages these constraints. For PCM devices, minimum written feature approaches ~200 nm (Wu et al., 2024).

4. Device Performance Metrics and Experimental Validation

Performance metrics capture focusing efficiency, extinction ratio, depth of focus, chromatic bandwidth, imaging fidelity, and throughput:

• Reconfigurable PCM photonics: MDM and WDM multiplexers report extinction ratios (ER) >15 dB simulated, >10 dB measured, insertion loss (IL) <2 dB, channel bandwidth >40–100 nm, repeatability within ±2 dB (Wu et al., 2024).
• Diffractive micro-optics: MLAs exhibit submicron focal spots (FWHM <1 μm), simulated focusing efficiency $h(x,y)$9, measured $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$00.60–0.65 across RGB; thickness reduction >3$$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$1 vs refractive analogues (Hayward et al., 2024).
• EDOF meta-optics: Measured depth of focus extension from 0.1 mm (traditional) to $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$20.4 mm (EDOF phase mask), PSF-invariance bandwidth $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$3290 nm, SSIM (structural similarity) indices markedly increased for broadband imaging (Bayati et al., 2021).
• Single-shot 3D printing: Volumetric throughput $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$41 mm$$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$5/s (%%%%24$h(x,y)$225%%%% voxels/s); lateral resolution $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$824 μm, axial $$J[h] = \sum_{i=1}^n w_i \int \left[ I(Z_i;x,y;h) - \alpha T(Z_i;x,y) \right]^2\,dx\,dy + \lambda R[h]$$922 μm, space–bandwidth product scaling to %%%%28$h(x,y)$229%%%% voxels for large-field masks (Lin et al., 10 Jan 2026).
• Depth-of-field extension: Nanoprinted phase plates extend DOF by factors $I(Z_i;x,y;h)$24 (simulated and measured), resolution broadening only $I(Z_i;x,y;h)$313% at best focus, multiple object planes imaged simultaneously (Sturges et al., 2024).

5. Generalization to Arbitrary Target Functions and Architectures

Inverse-designed microstructured phase masks are adaptable to a wide array of applications and architectural modalities:

• Arbitrary wavefront engineering: Any desired spatial phase or amplitude distribution (e.g., vortex beams, top-hat profiles, multichannel couplers) can be realized by modifying the FOM in the optimization, subject to microstructure constraints (Hayward et al., 2024, Bayati et al., 2021, López-Pastor et al., 2019).
• Programmable and reconfigurable elements: PCM-platform devices offer pixel-wise “writing”/“erasure” for on-demand functional reconfiguration, enabling holograms or beam-splitters reprogrammable over >10$I(Z_i;x,y;h)$4 cycles (Wu et al., 2024).
• Volumetric and tomographic printing: Inverse-designed phase masks enable single-shot volumetric fabrication in photoresists, coupling mask engineering with materials design (optical absorption, resin kinetics) to generate intricate 3D structures (Lin et al., 10 Jan 2026).
• Quantum and classical mode transformations: Succession of phase masks and Fourier transforms realizes arbitrary $I(Z_i;x,y;h)$5-mode unitary transformations, vital for universal interferometers in quantum photonics, scalable without the mesh complexity of beam-splitters (López-Pastor et al., 2019).
• Free-space and fiber platforms: Phase patterns may be mapped to large-area films or fiber facets; adaptation to OAM and multicore supermodes, though challenging due to alignment tolerances and thermal dissipation (Wu et al., 2024).

6. Computational and Physical Constraints, Robustness, and Scalability

Inverse-designed microstructured phase masks must rigorously account for sampling resolution, quantization effects, process variations, error sensitivity, and scalability:

• Sampling and quantization: Pixel size and height quantization are limited by fabrication (e.g., minimum feature width $I(Z_i;x,y;h)$6700 nm for grayscale lithography, lateral voxels $I(Z_i;x,y;h)$7300 nm for nanoprinting), and must be encoded in the optimization (Hayward et al., 2024, Sturges et al., 2024, Lin et al., 10 Jan 2026).
• Fabrication robustness: Post-fabrication measurements and pre-compensation routines correct systematic deviations, yielding sub-wavelength accuracy over mm-scale masks (Sturges et al., 2024).
• Error and fidelity analysis: For SLM-based or mask arrays, phase errors $I(Z_i;x,y;h)$8 on individual pixels degrade fidelity as $I(Z_i;x,y;h)$9; to achieve $T(Z_i;x,y)$0 for $T(Z_i;x,y)$1, $T(Z_i;x,y)$2 should be $T(Z_i;x,y)$3 rad (López-Pastor et al., 2019).
• Scalability: Space–bandwidth product scales quadratically with mask side length and inversely with pixel pitch ($T(Z_i;x,y)$4). Large-field lithography, parallel writing, and hierarchical optimization frameworks enable extension to %%%%43$h(x,y)$244%%%% addressable elements (Lin et al., 10 Jan 2026, Wu et al., 2024).
• Physical limitations: Resolution limits set by diffraction, chromatic aberration, and material dispersion; enforcement of physical regularization and symmetry during design mitigates systematic errors.

7. Key Research Directions and Outlook

Emerging trends in inverse-designed microstructured phase masks prioritize reconfigurability, volumetric throughput, and broadband functionality. The integration of phase-change materials with adjoint optimization algorithms enables programmable photonic systems with fine spatial resolution. Volumetric holographic printing, extended DOF multi-plane imaging, and ultra-low-f/# multiwavelength optics illustrate the expansive architecture and application space. The scalability of information capacity, through increased mask area and reduced pixel pitch, is a pivotal driver for next-generation high-throughput manufacturing and multiplexed optical processing (Lin et al., 10 Jan 2026, Wu et al., 2024, Hayward et al., 2024, Bayati et al., 2021, Sturges et al., 2024, López-Pastor et al., 2019).

A plausible implication is that the confluence of differentiable electromagnetic or Fourier-optics modeling, digital microstructuring, and adaptive fabrication will continue to extend the feasibility and precision of phase mask-based devices, offering new paradigms for programmable, multi-functional optics across photonic, quantum, and manufacturing domains.