Optical computing architectures are moving beyond conventional planar chip surfaces through three-dimensional laser micro- and nanoprinting methods that pattern light-manipulating structures across volume and directly on functional interfaces. In a comprehensive review published in PhotoniX, researchers led by Tian-Wei Wang, Yi-Yi Zhu, Linhan Lin, Hong-Bo Sun, and colleagues surveyed how additive, maskless direct laser writing enables the manufacturing of diffractive neural networks, interface-integrated processors, and non-planar photonic interconnects. By utilizing two-photon polymerization, scanning focused laser pulses locally induce nonlinear absorption within a photoresist, confining material crosslinking to sub-diffraction focal voxels without requiring lithographic masks. This processing capability allows researchers to organize multiple diffractive modulation planes with tailored axial spacings, trace non-planar waveguides in free space, and deposit micro-optical elements directly onto functional device surfaces without separate macroscopic alignment stages.
The physical motivation behind transitioning from planar integrated photonics to spatial laser-written architectures stems from the intrinsic multidimensional characteristics of light. Information can be processed simultaneously across amplitude, phase, polarization, wavelength, and spatial mode profiles through physical phenomena such as diffraction, dispersion, interference, and light-matter interactions. While planar semiconductor platforms utilize Mach–Zehnder interferometer meshes and microring resonator banks for matrix multiplication and optical weighting, their optical paths and phase modulators remain confined to two dimensions. Layered free-space diffractive networks overcome planar constraints by stacking phase masks along the propagation axis, but conventional implementations rely on bulky external holders and mechanical spacers that impede miniaturization. Three-dimensional laser printing circumvents these physical constraints by directly positioning successive diffractive layers with microscale axial precision, enabling high-density integration and spatial mode transformations within monolithic microscale devices.
In diffractive optical computing implementations, input optical wavefronts undergo predetermined phase and amplitude transformations across laser-printed layers to generate target output intensity distributions. Fabricating these structures demands precise spatial control, as local height variations and surface imperfections alter optical path lengths and redistribute optical power into unwanted scattering channels. Demonstrating benchmark capabilities, researchers fabricated binary Dammann gratings using two-photon polymerization, achieving an average per-spot diffraction efficiency of 3.46 percent for an order of four array, closely matching the theoretical target of 3.54 percent alongside spot nonuniformities below 1.4 percent. To mitigate phase errors arising from substrate tilt and stitching boundaries in diffractive neural networks, index-matched two-photon polymerization was developed by capping an IP-Dip phase-modulation layer with an IP-S polymer cover layer, shrinking refractive-index contrast to expand height tolerances. This index-matching technique increased experimental classification accuracy on a three-class FashionMNIST benchmark from 91.67 percent to 94.67 percent, approaching the simulated limit of 95.03 percent.
Further structural innovations in diffractive processors have leveraged spatial geometry to expand computational functionality across visible wavelengths and multimodal tasks. Micrometer-scale visible-light diffractive neural networks fabricated with cubical diffraction neurons demonstrated enhanced phase and amplitude modulation efficiencies at 532 nanometers compared to conventional cylindrical designs. A single-layer network achieved an 89.3 percent blind-testing accuracy classifying handwritten digits zero and one, while a two-layer ten-class network demonstrated 83.3 percent accuracy over digits zero through nine. In parallel, researchers integrated interlayer rotation as a physical degree of freedom using twisted diffractive neural networks formed by stacking Pancharatnam-Berry neuro-metasurfaces. By mechanically adjusting the twist angle between adjacent layers, a single meta-disk reconstructed distinct lunar-phase holograms, with numerical simulations indicating storage and retrieval capacities exceeding one hundred holographic images at structural similarity index measure values near 0.8. Demonstrating high-dimensional quantum gate operations, four-layer printed multiplane light-conversion structures achieved a process fidelity of 90.1 percent for a three-dimensional H1 gate.
Interface-integrated optical computing establishes direct functional coupling by printing micro-optical elements straight onto active optoelectronic surfaces, including complementary metal–oxide–semiconductor sensors, vertical-cavity surface-emitting lasers, and fiber facets. This interface integration eliminates the mechanical complexity and insertion loss associated with separate discrete coupling optics. Nanoprinted optical linear perceptrons fabricated directly on CMOS chips performed near-infrared optical decryption, achieving 100 percent accuracy in symmetric decryptor configurations and up to 86.67 percent in multi-decryptor systems. Utilizing CMOS-compatible two-photon fabrication, diffractive networks featuring two-micrometer neurons printed on image sensors executed visible-light optical logic operations at 515 nanometers with 100 percent experimental accuracy across seven logic functions. Additionally, integrated diffractive deep neural networks mapped aberrated point spread functions directly to Zernike pupil phase distributions, recovering the first fourteen Zernike polynomials with average root-mean-square errors of 0.036 pi for individual modes and 0.043 pi for combined wavefront profiles.
Direct laser writing on optical emitters and waveguides has similarly enabled compact source-integrated beam transformation and fiber imaging. Inverse-designed diffractive layers printed directly onto vertical-cavity surface-emitting laser facets converted complex transverse emission profiles into targeted holographic projections with peak signal-to-noise ratios exceeding 15.1 decibels, operating at approximately 79 percent diffraction efficiency and 10.5 percent total optical loss. For remote imaging, miniaturized three-layer diffractive neural networks measuring 150 by 150 micrometers were fabricated directly onto the distal facets of multimode fibers. Trained against vectorial fiber transmission models, these fiber-facet processors executed all-optical single-shot image transportation, resolving 4.90-micrometer feature sizes across a 65 by 65 micrometer field of view without computational digital reconstruction. These interface demonstrations confirm that aligning printed elements with specific source modalities can co-locate signal generation, passive filtering, and detection within compact operational volumes.
Complementing free-space diffractive networks, three-dimensional laser printing provides advanced routing capabilities through free-standing polymer waveguide interconnects that circumvent planar routing congestion. By utilizing fractal branching architectures where successive bifurcation layers multiply channel capacity, printed three-dimensional couplers route optical signals across multi-tiered spatial configurations. Fabricated 1-to-81 fractal waveguide couplers and interconnect arrays established 225 input and 529 output routing paths within a footprint of 0.46 by 0.46 square millimeters. Beyond simple power division and signal distribution, printed three-by-three waveguide networks implemented fixed spatial filtering matching Haar convolution kernels. Advanced couplers achieved coupling losses down to 0.06 decibels for optimized 1-to-2 splitters and maintained insertion losses below 2 decibels across a broad spectral window spanning 520 to 980 nanometers, illustrating how three-dimensional interconnect topologies can embed linear matrix transformations directly into transmission geometries.
Realizing these optical processing architectures depends critically on material properties, as refractive index determines required modulation depths and waveguide confinement, while material absorption, scattering, and thermal instability degrade transmitted optical power. Standard commercial photoresists like IP-Dip and IP-S provide transparent structural matrices, yet extending functionality requires integrating active materials. Stimuli-responsive liquid–crystal elastomers and woodpile photonic crystals generate tunable birefringence and structural color, while chemically cleavable resins serve as sacrificial templates for multimaterial assembly. Polymer-assisted functionalization incorporates cadmium sulfide precursors, multiwalled carbon nanotubes, or colloidal quantum dots into laser-written matrices to introduce photoluminescence and electrical conductivity. Beyond polymers, direct laser writing routes now process inorganic compositions, including two-photon decomposition of metallic alloys, calcination of zinc-oxide microstructures, and low-temperature conversion of siloxanes into fused silica glass, creating pathways for high thermal stability and active optoelectronic interfacing.
Printing resolution and throughput present complementary fabrication challenges that dictate the physical scale and channel density of optical computing modules. Because two-photon polymerization voxels are inherently elongated along the optical axis, active confinement techniques have been developed to sharpen spatial definitions. Methods such as Resolution Augmentation through Photo-Induced Deactivation, stimulated emission depletion lithography, and light-matter co-confinement compress voxel dimensions below 50 nanometers by combining central excitation with peripheral photoinhibition or chemical quenching. To address processing speeds when scaling to large-area diffractive arrays and extensive waveguide interconnects, high-throughput approaches parallelize exposure. Digital micromirror devices and metalens arrays generate programmable multifocal arrays exceeding 120,000 coordinated focal spots, achieving fabrication rates up to 1.2 times 10 to the eighth voxels per second, while spatiotemporal projection and volumetric tomographic synthesis assemble three-dimensional components in seconds.
Looking forward, the review authors highlight four primary development vectors necessary to transition laser-printed optical processors from passive laboratory proof-of-concepts into practical computing hardware. First, reconfigurable modulation requires embedding phase-change materials, liquid crystals, or electro-optic media into printed voxels to enable dynamic weight updating and post-fabrication calibration without inducing thermal crosstalk or optical scattering. Second, cascaded deep neural networks demand all-optical or electro-optic nonlinear activation layers, which can leverage printed nanocrystal and semiconductor cavities to lower activation thresholds and minimize optical power requirements between linear diffractive stages. Third, fully freeform three-dimensional optical layouts must replace planar layer stacks by optimizing optical component positions, orientations, and non-planar propagation paths throughout the entire 3D volume. Finally, heterogeneous optical packaging must unite laser sources, diffractive processing cores, non-planar photonic wire bonds, and sensor arrays into unified multichip modules that maintain structural stability, low coupling loss, and thermal resilience under continuous operation.
Subject of Research: Three-dimensional laser micro- and nanoprinting for optical computing hardware
Article Title: Three-dimensional laser micro/nano printing for optical computing
Article References: Wang, T.-W., Zhu, Y.-Y., Liu, W., Xu, K.-M., Sun, H.-B., & Lin, L. (2026). Three-dimensional laser micro/nano printing for optical computing. PhotoniX, 7(1), Article 67. https://doi.org/10.1186/s43074-026-00289-5
Image Credits: AI Generated
DOI: 10.1186/s43074-026-00289-5
Keywords: optical computing, laser micro printing, two photon polymerization, diffractive neural networks, photonic interconnects, integrated photonics, micro optics, nanolithography, waveguide arrays, CMOS integration, spatial mode transformation, photoresist materials, multifocus printing, optical neural networks, volumetric printing, optical logic
Cite Scienmag News
Cassandra Pierce. (November 17, 2026). Three-Dimensional Laser Printing Advances Diffractive Optical Computing and Photonic Interconnect Networks. Scienmag. https://scienmag.com/three-dimensional-laser-printing-advances-diffractive-optical-computing-and-photonic-interconnect-networks/
Cassandra Pierce. "Three-Dimensional Laser Printing Advances Diffractive Optical Computing and Photonic Interconnect Networks." Scienmag, 17 November 2026, https://scienmag.com/three-dimensional-laser-printing-advances-diffractive-optical-computing-and-photonic-interconnect-networks/. Accessed 17 November 2026.
Cassandra Pierce. "Three-Dimensional Laser Printing Advances Diffractive Optical Computing and Photonic Interconnect Networks." Scienmag. November 17, 2026. https://scienmag.com/three-dimensional-laser-printing-advances-diffractive-optical-computing-and-photonic-interconnect-networks/








