Biomimetic apposition compound eye fabricated using microfluidic-assisted 3D printing

Survival of the fittest has continuously driven the evolution and improvement of compound eyes. Even early examples of arthropods dating back to the Cambrian era had evolved faceted compound eyes which enabled them to perceive their environment based on visual phototaxis. A compound eye consists of a group of ommatidia oriented in different directions to provide panoramic vision with advantages including depth perception, low-aberration detection, and high-sensitivity motion tracking. Each ommatidium includes a corneal facet lens, a crystalline cone, a rhabdom for light transmission, and pigment cells for optical isolation to minimize crosstalk. These features have inspired numerous artificial microlens-array-based optical systems for applications such as endoscopic examinations, robot navigation, and surveillance. Most artificial compound eyes rely on conventional microfabrication techniques. Planar microlens arrays can be fabricated relatively easily and then transferred onto hemispherical surfaces, but this transfer can degrade lens uniformity and performance. Advanced 3D microfabrication (e.g., 3D laser writing, laser lithography, chemical etching, two-photon polymerization) can produce true 3D eyes; however, a key mismatch remains between the curved compound-eye image and commercial planar imaging sensors, necessitating complex optics or image processing. Deformable optoelectronics with curved photodetector arrays can address this, but the required curvature can limit the eye size. To overcome these shortcomings, the authors developed a biomimetic apposition compound eye (BAC-eye) using a fabrication method that combines 3D printing with microfluidic-assisted moulding to pattern 522 microlenses omnidirectionally on a hemispherical surface. Each microlens is optically connected to the flat base through refractive-index-matched waveguides that mimic rhabdoms, enabling direct coupling to planar image sensors for full-colour, wide-field imaging. The BAC-eye closely recreates natural compound eye architecture (5 mm diameter) with a large viewing angle (170°). Proof-of-concept demonstrations include full-colour panoramic imaging and position tracking of a point source. With its unique 3D-to-2D mapping capability, the 3D BAC-eye opens applications in photonics, sensing, and imaging.

Prior work on artificial compound eyes largely used planar microlens arrays fabricated by thermal reflow, laser-induced forward transfer, laser ablation, jet printing, or microfluidic manipulation, then transferred onto hemispherical substrates. While accessible, this transfer compromises microlens uniformity and performance. Advanced 3D microfabrication (3D laser writing, laser lithography, chemical etching, two-photon polymerization) achieves complex 3D geometries, yet resulting images remain poorly matched to planar sensors, imposing additional optics or processing. Deformable optoelectronics with curved detector arrays can match curvature but are constrained by mechanical limits on bending radius, potentially restricting device size. These limitations motivated a design that preserves 3D optical collection while mapping outputs to a flat sensor via internal waveguides to avoid crosstalk and curvature mismatch.

Design and fabrication of the BAC-eye: - Architecture: The BAC-eye replicates apposition compound eye anatomy. Each ommatidium comprises a surface microlens (corneal facet analogue) atop a cylindrical post (crystalline cone analogue) connected to a refractive-index-matched silicone-elastomer waveguide (rhabdom analogue) embedded within a dyed photosensitive polymer substrate that serves as an absorbing matrix (pigment cells analogue). - Mould preparation: A hemispherical mould with an open pit was 3D-printed (projection micro-stereolithography; nanoArch P140, 10 μm resolution). The hemispherical surface contained 522 cylindrical microholes (diameter 180 μm), omnidirectionally distributed up to ±85° polar angle. - Microfluidic-assisted lens moulding: The pit was filled with acrylate resin, degassed at −0.1 MPa for 10 min, and spun about its central axis (typical: 1500 rpm, 4 min). Centrifugal forces partially expelled resin, leaving a volume dependent on spin speed and hole orientation. Upon stopping, surface tension shaped the resin meniscus into a uniform concave profile in each microhole; the equilibrium contact angle θe was measured as 13.2°, yielding a predicted microlens radius R = d/(2 cos θe) ≈ 92 μm. The resin was UV-cured for 15 min to fix the concave moulds. Experimental microlens curvature after replication was 91.9 ± 0.8 μm, independent of hole orientation and spin speed. Post heights varied systematically with spin speed and hole position as predicted by simulation. - Substrate and waveguide formation: A complementary hemispherical substrate with 522 tapered hollow pipelines connecting the dome surface to a flat base was 3D-printed from a photosensitive polymer (n ≈ 1.46) doped with Sudan Black 3 dye (1500 μg/mL). The substrate was aligned and inserted into the mould using six supports/slots. The assembly was immersed in liquid-state two-part RTV silicone (n ≈ 1.50) and evacuated at −0.1 MPa for 20 min to fully fill the microholes and pipelines; after 4 h room-temperature curing, the elastomer formed microlenses and continuous waveguides. Demoulding yielded the BAC-eye: each microlens (radius ~90–92 μm) sat on a post, with a tapered silicone waveguide narrowing from d1 ≈ 157 μm (at the ommatidium) to d2 ≈ 100 μm (at the base). Outputs were arranged hexagonally at the flat base, enabling direct registration to planar CMOS sensors without additional optics. Optical optimization and characterization: - Crosstalk suppression: The dyed substrate absorbed stray light. At 1500 μg/mL dye concentration, a 9.8 μm-thick film achieved optical density ~3 across the visible, minimizing inter-waveguide crosstalk. RTV silicone was transparent across 400–1100 nm. - Crosstalk experiments: Three 100 μm-diameter, 90°-bend silicone waveguides (600 μm bend radius) embedded in dyed polymer were tested with 450 nm light coupled into the central channel. Only the illuminated channel emitted at the output; extinction ratios relative to adjacent channels were 16.1 dB and 15.2 dB. - Ray-tracing simulations: Incidence on microlenses at various orientations (e.g., α = 0°, 36°, 79.2°) was traced through straight and curved waveguides. Lower-index substrate and higher-index waveguide ensured total internal reflection; despite non-axisymmetric internal reflections in curved guides, confinement and efficient transmission to the base were maintained. - Loss and angular sensitivity: Overall optical loss (coupling in/out plus propagation) was measured at 5.37 dB, attributed to waveguide bending and tapering. The angular sensitivity function for ommatidia at α = 0°, 36°, and 79.2° showed maxima near expected incident angles; acceptance angle (FWHM) was ~44°, consistent with large multimode waveguide diameters. Panoramic imaging: - A BAC-eye was placed directly on a colour CMOS sensor. Each ommatidium projected light over ~80×80 pixels; per-ommatidium outputs were homogenized by averaging and computationally mapped back to a hemispherical panorama for 3D visualization. Demonstrations included imaging a square mask and dynamic patterns (a red cross fixed at α = 60°, β = 0° and a blue triangle moving from α = 85° to 20° at β = 180°), producing uniform, aberration-free panoramic reconstructions. Coherent laser illumination produced speckle within sub-images due to interference. 3D point-source tracking: - A diverging green point source delivered via optical fiber was placed at various angles and distances. The BAC-eye produced Gaussian-like light spots on the CMOS; spot center encoded angular position and spot width encoded distance. A calibration between spot FWHM and source distance yielded a high-quality fit (R² = 0.9999). Unknown test positions were recovered with root-mean-square deviation < 0.16. Accuracy can be improved by increasing ommatidia count and camera bit depth.

- Fabricated a hemispherical biomimetic apposition compound eye (BAC-eye) with 522 omnidirectionally oriented microlenses on a 5 mm-diameter dome, achieving a 170° field of view (edge ommatidia oriented at ±85°). - Introduced microfluidic-assisted moulding integrated with 3D printing to form uniform-curvature microlenses (radius of curvature 91.9 ± 0.8 μm; theory ~92 μm), independent of position and spin speed. - Implemented intracorporal refractive-index-matched silicone waveguides tapering from ~157 μm to ~100 μm, mapping 3D surface sampling to a flat, hexagonally arranged 2D array compatible with planar CMOS sensors. - Suppressed optical crosstalk using a dyed polymer substrate (Sudan Black 3, 1500 μg/mL; optical density ~3 at 9.8 μm thickness across visible). Measured extinction ratios between illuminated and adjacent channels: 16.1 dB and 15.2 dB. - Total optical loss (coupling + propagation) measured at 5.37 dB; loss attributed to waveguide bending and tapering. - Angular sensitivity showed wide acceptance (~44° FWHM) and peak responses consistent with ommatidia orientation; light remained well confined in waveguides. - Achieved full-colour, wide-angle panoramic imaging with uniform reconstructed features; sub-images spanned ~80×80 pixels before homogenization. - Demonstrated 3D point-source tracking: Gaussian fitting of spots on the CMOS enabled distance estimation with a strong calibration fit (R² = 0.9999) and reconstruction of unknown positions with RMS deviation < 0.16.

The BAC-eye closely mimics the architecture and function of natural apposition compound eyes while solving a key integration barrier: matching curved eye images to flat, commercial sensors. By guiding each ommatidium’s collected light through refractive-index-matched, crosstalk-suppressed waveguides to a flat base, the system provides seamless 3D-to-2D mapping without auxiliary matching optics or curved detector arrays. This approach simplifies system optics and electronics and is readily scalable: each ommatidium effectively contributes one pixel to the panoramic image, so resolution can be increased by increasing ommatidia count and filling factor without redesigning the sensor. Optical characterization confirms efficient coupling and confinement with minimal crosstalk and acceptable loss given the multimode, curved geometry. The wide acceptance angle of each ommatidium, advantageous for panoramic coverage and motion sensitivity, supports robust 3D perception, enabling simple and precise 3D point-source localization. Demonstrations of full-colour panoramic imaging and 3D tracking highlight the technology’s potential for applications such as endoscopy, robotics, surveillance, and machine vision. The platform’s compatibility with planar sensors and adaptability via 3D printing and microfluidic-assisted moulding offer a flexible route to tailored compound-eye designs across sizes and fields of view. Moreover, the device can serve as a biomimetic model for studying insect vision and perception, and could operate in an emitting mode if coupled to a 2D display for projection or volumetric display applications.

This work introduces a microfluidic-assisted 3D-printing method to fabricate a biomimetic apposition compound eye that integrates 522 uniform-curvature microlenses and refractive-index-matched waveguides within a dyed polymer substrate. The BAC-eye directly interfaces with planar CMOS sensors to deliver aberration-free, full-colour panoramic imaging over ~170° and enables precise 3D point-source tracking using simple Gaussian fitting. Optical crosstalk is effectively eliminated by substrate absorption, and overall loss is moderate given the tapered, curved waveguides. Main contributions include: (1) a scalable fabrication strategy for complex 3D micro-optics inside curved geometries; (2) intrinsic 3D-to-2D optical mapping compatible with flat sensors; and (3) validated panoramic imaging and 3D tracking performance. Future work can focus on increasing ommatidia count and filling factor to enhance resolution and sensitivity, optimizing waveguide geometry to reduce loss, leveraging multimode speckle reconstruction for ultra-high-resolution panoramas, and exploring emitting-mode applications (e.g., planetarium projection, volumetric 3D displays). The approach is promising for compact endoscopic imaging, microrobotics, and machine vision.

- Optical loss (~5.37 dB) arises from waveguide bending and tapering; reducing curvature or optimizing tapers could improve throughput. - Wide acceptance (~44° FWHM) benefits coverage but limits per-ommatidium angular resolution; overall image resolution is constrained by the number and packing density of ommatidia. - Speckle artifacts can appear under coherent laser illumination, affecting image uniformity (less relevant under incoherent light). - Fabrication relies on 3D-printing resolution; initial mould curvature cannot be directly printed at current resolutions and requires microfluidic-assisted shaping. - Post heights vary with position/spin speed (though lens curvature is uniform), which could introduce minor variability in optical path lengths and coupling.