The Scaling Bottleneck of Glass Optics
To focus light onto a sensor, a traditional refractive lens relies on its shape and physical thickness. Light travels slower in glass than in air; by varying the thickness across a lens, we introduce spatially varying phase shifts that bend an incoming wavefront into a sharp focal point. As discussed by Alan Zhan et al. in Low-Contrast Dielectric Metasurface Optics, this requirement for substantial physical thickness to accumulate phase shifts is the primary barrier to miniaturizing modern cameras.
Furthermore, a single curved surface introduces optical aberrations—such as chromatic aberration, where different colors focus at different points, or off-axis distortions. To correct these errors, optical engineers traditionally stack multiple lens elements together. This stacking dramatically increases the total track length and weight of the device, a challenge highlighted by Fan Yang in Wide field of view large aperture meta-doublet eyepiece. For large-aperture or long-range systems, the mass of these glass optics scales cubically with focal length and aperture size, making them impractical for weight-constrained applications like aerospace, drones, or portable electronics (Jing Wang et al. in A golay metalens for long-range, large aperture, thermal imaging via sparse aperture computational imaging).
Metasurfaces: Phase Control Without Bulk
To eliminate the bulk of glass while maintaining control over light, researchers have turned to meta-optics or metasurfaces. Instead of forcing light to travel through millimeters of curved glass, metasurfaces use a single flat layer of sub-wavelength nanostructures (such as silicon or silicon nitride pillars) to instantly alter the phase, amplitude, or polarization of the incoming wavefront.
As reviewed by Anna Brook in Meta-optical Imaging at Thermal Wavelengths, these flat optics leverage sub-wavelength structures to overcome traditional bottlenecks in field-of-view and bandwidth. By precisely engineering the geometry and arrangement of these pillars, a flat meta-optic can mimic the focusing power of a bulky refractive lens within a microscopic footprint. This allows engineers to bypass the traditional requirement for a minimum physical depth to maintain a specific focal length between the lens and the sensor (Alexander D. White et al. in A Silicon Photonics Computational Lensless Active-Flat-Optics Imaging System).
The Hardware-Software Co-Design Paradigm
If flat meta-optics are so thin, why haven't they completely replaced glass lenses? The challenge is that a single flat layer introduces severe optical aberrations when exposed to broadband (full-color) light or wide fields of view. Because nanostructures interact with light at specific wavelengths, a simple metalens will focus red, green, and blue light at completely different distances, resulting in blurry, unusable raw images.
The breakthrough that makes modern ultra-thin imaging work is hardware-software co-design. Instead of demanding that the physical meta-optic produce a perfect, human-interpretable image directly on the sensor, the hardware is designed to encode the scene into a multiplexed pattern, and a computational backend decodes it.
Incoming Light ──> [ Engineered Meta-Optic ] ──> [ Multiplexed Pattern on Sensor ] ──> [ Neural Network Backend ] ──> Final Image
For example, Johannes E. Fröch et al. demonstrate this in Beating spectral bandwidth limits for large aperture broadband nano-optics by co-designing a flat meta-optic with a diffusion-based neural reconstruction backend to achieve high-quality full-color imaging. Similarly, in the long-wave infrared spectrum, the authors of Broadband thermal imaging using meta-optics utilize a deep-learning-assisted inverse design framework to optimize complex subwavelength structures alongside a reconstruction network, achieving a six-fold improvement in broadband thermal image quality without adding physical lens elements.
Eliminating Lenses Entirely: Lensless Imaging
The logical conclusion of this computational paradigm is to remove the lens entirely. In lensless imaging, the focusing element is replaced by a thin, flat mask placed directly in front of the sensor array. As explained by Salman S. Khan et al. in FlatNet: Towards Photorealistic Scene Reconstruction from Lensless Measurements, the sensor captures multiplexed, non-interpretable light patterns instead of focused images, creating a significant computational challenge.
Without a lens, every point in the scene casts a wide, overlapping pattern (a point spread function) across the entire sensor. The raw measurement looks like uninterpretable noise to the human eye. However, because the mask's optical properties are mathematically known, the system can solve a computational inverse problem to reconstruct the original scene. Frameworks like FlatNet combine physics-inspired inversion layers with deep learning to transform these multiplexed patterns into photorealistic images.
By treating the optical mask as the "first layer" of a neural network, Yonggan Fu et al. show in SACoD: Sensor Algorithm Co-Design Towards Efficient CNN-powered Intelligent PhlatCam that the physical phase mask and the backend convolutional neural network can be jointly optimized. This co-design drastically reduces the computational overhead and physical thickness, enabling intelligent vision systems to fit into ultra-thin form factors like wearables or internet-of-things (IoT) devices.
Expanding Capabilities in Restricted Spaces
By shrinking the optical stack to a fraction of a millimeter, this computational flat-optics paradigm unlocks imaging capabilities in environments where traditional lenses cannot physically fit:
- Medical Endoscopy: Medical imaging in highly confined anatomical spaces requires extreme miniaturization. Ningzhi Xie et al. present an elegant solution in Inverse-designed large field-of-view polychromatic metalens for tri-color scanning fiber endoscopy, utilizing optimized silicon nitride meta-atoms to eliminate chromatic aberrations for tri-color scanning fiber endoscopy inside narrow blood vessels.
- Thermal and Night Vision: Traditional long-wave infrared systems rely on heavy, expensive materials like Germanium to correct off-axis aberrations. As demonstrated by Anna Wirth-Singh et al. in Large Field-of-View Thermal Imaging via All-Silicon Meta-Optics, a single-layer all-silicon meta-optic combined with computational post-processing can achieve wide-angle thermal sensing in a form factor orders of magnitude thinner and lighter than traditional multi-element refractive lenses. This is mirrored by Ilya Reshetouski et al. in Lensless Imaging with Focusing Sparse URA Masks in Long-Wave Infrared and Its Application for Human Detection, where expensive lenses are replaced with a thin etched silicon mask.
- Advanced 3D Sensing: Beyond standard 2D photography, flat optics can multiplex complex light fields. Fan Yang et al. demonstrate in Monocular depth sensing using metalens that a single ultra-thin metalens can use polarization to multiplex rotating light patterns, enabling high-precision 3D depth sensing and image reconstruction simultaneously without bulky dual-camera setups.
Key Foundational Work
Thin infrared imaging systems through multichannel sampling (2008) — contributed 3 concept elements to this direction.
Wide field of view catadioptrical head-mounted display (2003) — contributed 1 concept elements to this direction.