Miniature Optoelectronic Compound Eye Camera

The study addresses the long-standing challenge of integrating bioinspired compound-eye (CE) optics with planar CCD/CMOS sensors at insect-like scales. Natural CEs provide wide field-of-view (FOV), distortion-free imaging, and motion sensitivity through thousands of hemispherically arranged ommatidia. Artificial CE efforts have been limited by a fundamental mismatch between the curved focal surface produced by multi-ommatidial optics and the flat surface of imaging sensors, leading to defocus when miniaturized. The research question is whether specially designed ommatidia can extend depth-of-field and focus range sufficiently to enable direct, compact integration with planar CMOS without auxiliary relay optics or curved detectors. The purpose is to design, fabricate, and validate a miniature optoelectronic CE camera using logarithmic-profile ommatidia to overcome defocus while retaining large FOV and motion sensitivity, enabling real-time imaging and 3D tracking in a device comparable in size to insect eyes.

Prior artificial CE cameras have used planar microlens arrays (MLAs) integrated with CCD/CMOS, achieving reasonable resolution but limited FOV due to planar geometry. 3D curved CEs fabricated via microlens templating, self-writing, reconfigurable micromachining, and femtosecond laser additive/subtractive manufacturing demonstrated large FOV but typically required microscope-based image acquisition and lacked integrated photodetectors, limiting portability and real-time use. Optoelectronic CE systems have been realized by assembling microlenses with arrays of photodetectors on curved surfaces (e.g., Forero et al.'s cut-and-assembled system with enhanced unidirectional FOV; Rogers et al.'s flexible lens array on hemispherical deformable silicon, achieving 140°–180° FOV). These approaches suffer from assembly complexity, relatively low resolution, and size larger than insect eyes. Aberration correction traditionally relies on aspheres or multi-lens/multi-aperture systems, which are difficult to implement at microscale with high fidelity using conventional fabrication (thermal reflow, inkjet printing, wet etching, embossing). Two-photon polymerization (TPP) enables arbitrary 3D freeform micro-optics and has been used for spherical-ommatidia μ-CEs, but defocus remains with planar detectors. The logarithmic lens concept can produce an extended focal line with uniform intensity, potentially expanding depth-of-field and addressing the mismatch without complex assemblies.

Design: Ommatidia designed as logarithmic-profile microlenses to extend focus range and depth-of-field, mitigating the mismatch between curved focal surfaces and planar CMOS. Comparative designs included spherical ommatidia of identical footprint for benchmarking. COMSOL Multiphysics simulations analyzed focused light fields for spherical vs logarithmic ommatidia and focal surface curvature.

Fabrication: Polymer μ-CEs were fabricated via femtosecond laser two-photon polymerization (FL-TPP) in negative-tone hybrid photoresist SZ2080 (with 1% 4,4-Bis(diethylamino)benzophenone photosensitizer). A gray-scale printer-based FL-TPP system (Malvern Nano Series, Jicheng ultrafast equipment) used a near-infrared laser (central wavelength 780 nm, pulse length 150 ps, repetition 80 MHz) focused with a 60× objective (NA 0.6) to write the 3D structures. Post-fabrication characterization used laser scanning confocal microscopy (OLS4100, Olympus) for 3D profiles and transmission optical microscopy (CX41, Olympus). Transmission spectra were measured with a micro-area spectrometer (ULS204x64×EVO, Avantes).

Geometries and parameters: Single-lens benchmarks: spherical lens radius 25 μm (D=50 μm), focal length ~355 μm; logarithmic lens radius 25 μm with focusing range ~100–800 μm. μ-CE domes: typical dome diameter ~400 μm, height ~90 μm. Optimized ommatidia for integrated camera: D=110 μm, focal line endpoints d1=500 μm and d2=1000 μm. Arrays fabricated with 19 to 160 ommatidia. Uniformity statistics showed standard deviations for diameter of 0.62 μm (spherical) and 0.57 μm (logarithmic) and for height of 0.16 μm (spherical) and 0.15 μm (logarithmic).

Focusing and imaging tests: He–Ne laser (λ=632.8 nm) focusing tests were performed in water. A 10×/0.25 NA objective-based microscope system recorded focal spot distributions while varying imaging distance to probe focusing across ommatidia. Point spread functions (PSF) and intensity profiles along X and Y were measured at incidence angles 0°, 30°, 45°, and Gaussian fits used to quantify FWHM. Field-of-view (FOV) was derived from dome geometry and confirmed experimentally via angle-resolved focusing micrographs.

Optoelectronic integration: μ-CEs were directly bonded/integrated atop commercial planar CMOS detectors (reported as OmniVision OY9734/OY9374 and OmniVision OV3742, sensor size ~2.5×1.7 mm). A 19-ommatidia CE with real-space size ~400 μm covered >80,000 pixels, providing adequate resolution. The total device weight, including CMOS, was ~230 mg.

3D position estimation and motion tracking: The object–image relationship for the μ-CE camera was calibrated. For macroscopic tests, a triangular target (20 mm side) was placed at known positions (220 mm/0°, 233 mm/19.3°, 282 mm/38.8°). Images from multiple ommatidia were used to reconstruct distance and azimuth. For motion tracking, time-lapse imaging of a beetle was recorded; image definition metrics across ommatidia over time enabled simultaneous distance and azimuth estimation and trajectory reconstruction.

On-chip microfluidic integration and calibration: The μ-CE camera was integrated with a microfluidic chip for on-chip imaging. Due to short focal length, clear imaging was achieved for objects at distances >1.4 mm. A backpropagation (BP) neural-network-based calibration used targets with known parameters to map image arrays to 3D positions. Post-calibration, microscale shapes (square, triangle) at different positions were reconstructed in 3D, and the 3D trajectory of a living Paramecium over 5 s was reconstructed from video. Imaging FOV radius examples: 80 mm (macroscale triangle test) and 150 μm (microfluidic Paramecium).

• Logarithmic ommatidia significantly extend focus range and depth-of-field compared to spherical ommatidia, enabling compatibility with planar CMOS without additional relay optics or curved sensors.
• Single-lens benchmarks: spherical lens (D=50 μm) focal length ~355 μm; logarithmic lens (D=50 μm) focusing range ~100–800 μm. Measured focus ranges: spherical ~360 μm vs logarithmic ~500 μm.
• μ-CE geometry yields theoretical FOV ~96.9°, with practical device FOV ~90°. The spherical μ-CE control exhibited FOV ~20°.
• Angular sensitivity function (ASF) for logarithmic CE had measured FWHM ~12.1°, versus spherical CE ASF FWHM ~19.3°, indicating higher angular sensitivity for the logarithmic CE across a wide FOV.
• PSF under varying incidence angles (0°, 30°, 45°) remained consistent (FWHM along X and Y: ~4.1 ± 0.2), indicating low aberration and high fabrication fidelity.
• Integrated μ-CE camera (≈400 μm CE on planar CMOS, covering >80,000 pixels; device weight ≈230 mg) captured wide-FOV images and enabled spatial position identification and motion tracking.
• Position reconstruction of a 20 mm triangular target at known positions yielded reconstructed values close to ground truth: 220 mm/0° → 226 mm/0°; 233 mm/19.3° → 233 mm/19.4°; 282 mm/38.8° → 262 mm/38.8°.
• Real-time beetle trajectory reconstructed from μ-CE video by analyzing ommatidial image definition over time.
• On-chip integration with a microfluidic device enabled 3D trajectory reconstruction of a living Paramecium over 5 s; clear imaging for objects at distances >1.4 mm; micro-scale FOV radius ~150 μm.

The findings demonstrate that using logarithmic-profile ommatidia effectively addresses the defocus caused by the mismatch between a CE’s curved focal surface and a planar CMOS detector. By extending the focus range and depth-of-field, the focal line from each ommatidium can be captured on a flat sensor without resorting to complex relay optics, multilayer lens stacks, or curved photodetectors, simplifying fabrication and integration at insect-scale feature sizes. Quantitative comparisons show superior FOV (~90° vs ~20° for spherical control), improved angular sensitivity (ASF FWHM ~12.1°), and consistent PSF across incident angles, validating the optical design and fabrication fidelity. The integrated μ-CE camera further demonstrates practical capabilities in spatial localization and trajectory tracking of macroscopic (beetle) and microscopic (Paramecium) moving targets, highlighting its utility for real-time, wide-FOV, multi-view imaging in compact platforms. There is a trade-off of some energy throughput (slightly darker images) due to the logarithmic lens design, but the benefits in DOF and integration outweigh this for many applications. The approach provides a scalable path to merge complex micro-optics with standard planar sensors, promising for robot vision, endoscopy, micro-navigation, and on-chip microscopy.

The work introduces an insect-scale optoelectronic μ-CE camera by fabricating 19–160 ommatidia with logarithmic profiles via FL-TPP and directly integrating them with planar CMOS sensors. Compared to spherical ommatidia, the logarithmic design increases depth-of-field and focus range, enabling defocus-free imaging on flat detectors while maintaining a large FOV (~90°) and high angular sensitivity. The miniaturized device (~400 μm CE; ~230 mg including CMOS) achieves wide-FOV imaging, accurate spatial position identification, and real-time trajectory monitoring of moving targets. Integration with microfluidics further enables on-chip, time-resolved 3D tracking of microorganisms. Future research could focus on increasing ommatidia count and pixel utilization, optimizing lens profiles for higher throughput and brightness, improving calibration and reconstruction accuracy (including learning-based methods), integrating custom low-noise sensors, and exploring application-specific packaging for endoscopy, microrobotics, and wearable or implantable micro-vision systems.

• Energy throughput loss with logarithmic lenses leads to slightly darker images compared to spherical lenses.
• Reported FOV optimization suggests diminishing returns at very large FOV due to marginal image quality and information utilization at the edges.
• Reconstruction accuracy, while good, shows some deviation (e.g., 282 mm true distance reconstructed as 262 mm), indicating calibration and modeling limitations.
• Imaging artifacts can arise from mixed bright-field/dark-field conditions and blind spots, affecting uniformity across ommatidia.
• Dependence on precise fabrication; although TPP provides high fidelity, scalability and throughput for mass production may be constrained.
• Sensor model and integration details may vary; device performance can depend on sensor characteristics (pixel size, sensitivity, noise).