Beating bandwidth limits for large aperture broadband nano-optics

Johannes E. Fröch, Praneeth Chakravarthula, Jipeng Sun, Ethan Tseng, Shane Colburn, Alan Zhan, Forrest Miller, Anna Wirth-Singh, Quentin A. A. Tanguy, Zheyi Han, K. F. Böhringer, Felix Heide, Arka MajumdararXiv 2024 · DOI · arXiv

Abstract

Flat optics have been proposed as an attractive approach for the implementation of new imaging and sensing modalities to replace and augment refractive optics. However, chromatic aberrations impose fundamental limitations on diffractive flat optics. As such, true broadband high-quality imaging has thus far been out of reach for low f-number, large aperture, flat optics. In this work, we overcome these intrinsic fundamental limitations, achieving broadband imaging in the visible wavelength range with a flat meta-optic, co-designed with computational reconstruction. We derive the necessary conditions for a broadband, 1 cm aperture, f/2 flat optic, with a diagonal field of view of 30° and an average system MTF contrast of 30% or larger for a spatial frequency of 100 lp/mm in the visible band (> 50 % for 70 lp/mm and below). Finally, we use a coaxial, dual-aperture system to train the broadband imaging meta-optic with a learned reconstruction method operating on pair-wise captured imaging data. Fundamentally, our work challenges the entrenched belief of the inability of capturing high-quality, full-color images using a single large aperture meta-optic.

Introduction

The "lens bump" on modern smartphones is a physical manifestation of the limitations of traditional refractive optics, which require significant bulk to correct for chromatic aberrations. Meta-optics—flat surfaces covered in sub-wavelength nanostructures—offer a path toward ultra-thin cameras. However, scaling these optics to large apertures (centimeter-scale) has historically been hindered by severe chromatic aberrations, where different colors focus at different points, resulting in unusable, blurry images. This work presents a 1 cm aperture, f/2 flat meta-optic that achieves high-quality, full-color video imaging, proving that large-aperture diffractive optics are not fundamentally limited by color distortion.

Methods: End-to-End Co-Design

The core innovation lies in an "end-to-end" co-design approach, where the physical lens and the digital reconstruction algorithm are optimized simultaneously.

The end-to-end design pipeline co-optimizing the meta-optic phase profile and computational reconstruction backend.

The design process involves two main steps:

  1. Extended Depth of Field (EDOF) Optimization: A differentiable pipeline designs a radially symmetric phase profile to maximize focal intensity across the visible spectrum (450–650 nm). This "stretches" the focus, ensuring all colors hit the sensor with a consistent blur pattern.
  2. Joint Optimization: The lens is optimized alongside a probabilistic diffusion-based neural network. This ensures the lens produces a Point Spread Function (PSF) that the algorithm is specifically trained to sharpen.

To manage the complexity of $10^9$ scatterers on a 1 cm lens, the team utilized radial symmetry, reducing design parameters to $10^4$. The lens was fabricated using Silicon Nitride (SiN)-on-quartz, resulting in a diffractive layer only 1 µm thick—four orders of magnitude thinner than traditional lenses. Training was facilitated by a custom coaxial dual-aperture system that captured simultaneous "ground truth" (refractive lens) and "blurred" (meta-optic) images for the same scene.

Results and Physical Implementation

The resulting meta-optic demonstrates that large-scale flat optics can compete with traditional refractive lenses. The system achieved an average Modulation Transfer Function (MTF) contrast of >30% at 100 lp/mm across the visible band.

Physical implementation of the 1 cm aperture meta-optic, including SEM images of the nanostructures and the integrated camera system.

Key findings include:

  • Contrast Enhancement: By designing for EDOF behavior, the reconstruction task is simplified from complex deconvolution to contrast enhancement, as information for all colors is preserved in the raw capture.
  • Off-Axis Performance: Surprisingly, the meta-optic outperformed traditional refractive lenses at large field angles. While refractive lenses suffer from coma and astigmatism at edges, the meta-optic's PSF remained more confined, leading to clearer details at the periphery.
  • Real-Time Imaging: The system supports full-color broadband imaging at 11–19 frames per second, capable of capturing fast-moving objects in ambient light.

Conclusion

This work challenges the long-held belief that large-aperture meta-optics cannot produce high-quality color images. While some fine details remain slightly more blurred than high-end refractive counterparts, the 1 cm aperture proves commercial viability for consumer electronics like laptops and smartphones. Future work will focus on non-symmetric designs to further improve off-axis performance and optimizing diffusion models for mobile hardware.

Summary updated

Abstract

Flat optics have been proposed as an attractive approach for the implementation of new imaging and sensing modalities to replace and augment refractive optics. However, chromatic aberrations impose fundamental limitations on diffractive flat optics. As such, true broadband high-quality imaging has thus far been out of reach for low f-number, large aperture, flat optics. In this work, we overcome these intrinsic fundamental limitations, achieving broadband imaging in the visible wavelength range with a flat meta-optic, co-designed with computational reconstruction. We derive the necessary conditions for a broadband, 1 cm aperture, f/2 flat optic, with a diagonal field of view of 30° and an average system MTF contrast of 30% or larger for a spatial frequency of 100 lp/mm in the visible band (> 50 % for 70 lp/mm and below). Finally, we use a coaxial, dual-aperture system to train the broadband imaging meta-optic with a learned reconstruction method operating on pair-wise captured imaging data. Fundamentally, our work challenges the entrenched belief of the inability of capturing high-quality, full-color images using a single large aperture meta-optic.
BibTeX
@misc{froch2024beating,
  author = {Johannes E. Fröch and Praneeth Chakravarthula and Jipeng Sun and Ethan Tseng and Shane Colburn and Alan Zhan and Forrest Miller and Anna Wirth-Singh and Quentin A. A. Tanguy and Zheyi Han and K. F. Böhringer and Felix Heide and Arka Majumdar},
  title = {{Beating bandwidth limits for large aperture broadband nano-optics}},
  year = {2024},
  eprint = {2402.06824},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2402.06824},
}
Figure