🔬 optics

Multifunctional Imaging with an Inverse-Designed Nonlocal Metasurface

This paper presents a topology-optimized, inverse-designed nonlocal metasurface that enables switchable all-optical imaging between brightfield and phase-contrast modalities for transparent biological samples simply by changing the polarization state, thereby eliminating the need for mechanical repositioning or computational post-processing.

Original authors: Lincoln Clark, Yang Xu, Cat-Uyen Phan, David J. Byrne, Shikun Ma, Lukas Wesemann, Elizabeth Hinde, Kylie L. Gorringe, Ann Roberts

Published 2026-06-23

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lincoln Clark, Yang Xu, Cat-Uyen Phan, David J. Byrne, Shikun Ma, Lukas Wesemann, Elizabeth Hinde, Kylie L. Gorringe, Ann Roberts

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are looking at a clear glass marble. To the naked eye, it's invisible against a clear background. To see it, you usually need to paint it, shine a special light on it, or use a giant, complex machine with many lenses and mirrors to trick your eyes into seeing its shape.

This paper introduces a tiny, flat chip (called a metasurface) that acts like a "smart switch" for a microscope. Instead of needing a big machine or moving parts, this chip can instantly change how it sees the world just by turning a knob on a light switch (specifically, by rotating the polarization of the light).

Here is how it works, broken down with simple analogies:

1. The Problem: The Invisible Ghost

Transparent things like living cells or tissue don't absorb light; they just let it pass through. Because they don't block light, they look invisible in a standard "brightfield" microscope (like looking at a ghost in a foggy room). Scientists usually have to stain them with dye (which kills the cells) or use complex, expensive equipment to make them visible.

2. The Solution: The "Magic Glasses"

The researchers built a microscopic chip made of silicon that sits right in front of the sample. This chip is designed using a super-smart computer algorithm (called inverse design or topology optimization). Think of this algorithm as a master chef who tries millions of different recipes in a simulation to find the perfect one, rather than just guessing.

This chip has a special superpower: It can change its "personality" based on the light hitting it.

Mode A (The "Identity" Mode): When you shine light with a horizontal orientation (let's call it "Horizontal Light"), the chip acts like a clear window. It lets the image pass through exactly as it is. If you look through it, you see a standard, flat, boring image where the transparent cells are still invisible.
Mode B (The "Edge-Detecting" Mode): When you rotate the light to be vertical ("Vertical Light"), the chip suddenly changes its behavior. It acts like a highlighter pen that only traces the edges and slopes of the object.

3. How It Creates the "3D" Look

The paper explains that this chip performs a mathematical operation called a derivative. In plain English, it looks for changes.

Imagine a smooth hill. If you walk across it, the ground doesn't change much.
But if you hit a cliff or a steep slope, the ground changes rapidly.

This chip ignores the flat, smooth parts of the cell (the "hill") and only highlights the steep parts (the "cliffs"). Because it highlights the slopes, the flat, invisible cells suddenly look like they have 3D shadows and depth, similar to how a topographic map shows mountains. This is called Phase Contrast Imaging.

4. The Secret Trick: The "One-Way Street"

How does a flat chip know which way is "up" or "down" to create these shadows? Usually, you need to tilt the chip or use complex lenses to do this.

The researchers used a clever trick involving diffraction (the way light bends around corners).

Imagine the chip is a busy highway.
When the "Vertical Light" hits it, the chip kicks some of the light into a "side road" (higher diffraction orders) that gets trapped inside the glass substrate and disappears.
Because some light is kicked away in one direction but not the other, the remaining light that reaches your camera has an imbalance. This imbalance creates the "slope" effect.
When the "Horizontal Light" hits it, the chip doesn't kick any light away; everything passes straight through.

5. What They Actually Did

The team didn't just simulate this; they built it and tested it.

The Test: They made fake transparent shapes (like a "spoke" wheel and a "bullseye" target) and real biological samples (HeLa cells and human ovarian cancer tissue).
The Result:
- Without the chip (or with Horizontal light), the cells were hard to see.
- With Vertical light, the cells popped out with a 3D-like appearance, showing their shapes and textures clearly.
- They were even able to mathematically calculate the exact "steepness" of the cell's surface just by taking two pictures (one with each light setting) and doing a quick calculation.

Summary

This paper presents a tiny, flat, silicon chip that replaces bulky, expensive microscope parts. By simply rotating the polarization of the light source, it switches between seeing a normal, flat image and a high-contrast, 3D-like image of transparent objects. It does this by using a computer-designed pattern that tricks light into highlighting the edges of invisible objects, making it possible to see clear cells without staining them or moving any parts of the microscope.

Technical Summary: Multifunctional Imaging with an Inverse-Designed Nonlocal Metasurface

Problem Statement
Transparent biological samples, such as unstained cells and tissues, exhibit poor contrast in conventional brightfield microscopy because they primarily modulate the phase rather than the amplitude of transmitted light. While established techniques like Zernike phase contrast and differential interference contrast (DIC) can visualize these phase variations, they typically require bulky optical setups, complex alignment, and the insertion of additional components (e.g., phase rings or Wollaston prisms) to switch between imaging modalities. Furthermore, computational phase retrieval methods often necessitate capturing multiple images or mechanical scanning, introducing latency and complexity. There is a need for ultra-compact, real-time imaging systems capable of performing multiple image processing modalities—specifically switching between brightfield and phase-contrast imaging—directly with light, without moving parts or post-processing.

Methodology
The authors propose a nonlocal, free-form dielectric metasurface designed using topology optimization to perform switchable image processing in the object plane.

Design Approach: The metasurface is designed using a topology optimization algorithm that iteratively determines the spatial distribution of permittivity (amorphous silicon) within a unit cell. Unlike methods that optimize parameters of predefined geometries, this approach optimizes the material density directly. To ensure robustness against fabrication defects (specifically under- and over-etching), the algorithm incorporates a weighted cost function that evaluates the optical response of the regular design alongside dilated and eroded variants.
Operating Principle: The device relies on nonlocal spatial filtering, where the transmission depends on the angle of incidence (spatial frequency). The metasurface is engineered to exhibit an asymmetric transfer function about normal incidence for one polarization state while maintaining a symmetric, constant response for the orthogonal state.
- Brightfield Mode: Under x-polarized illumination, the metasurface acts as an identity operator, providing a flat optical transfer function (OTF) for conventional brightfield imaging.
- Phase Imaging Mode: Under y-polarized illumination, the metasurface implements a first-order derivative operator with respect to the x-direction ( $H(k_x) \propto k_x$ ). This converts phase gradients into intensity modulations, generating pseudo-3D images similar to DIC.
Mechanism of Asymmetry: To achieve the required asymmetry at normal incidence without tilting the sample or the metasurface, the design utilizes diffraction. The metasurface period is chosen such that higher-order diffracted modes ( $\pm 1$ orders) are trapped by total internal reflection within the quartz substrate. This diffraction removes energy from the zeroth order in an angle-dependent manner, creating the necessary linear OTF slope for the y-polarization.
Fabrication: The device consists of a 300 nm thick patterned amorphous silicon layer on a quartz substrate. It was fabricated using electron-beam lithography and a pseudo-Bosch etching process. A 20 nm $Al_2O_3$ hard mask was retained on the surface.

Key Results

Optical Characterization: The fabricated metasurface was characterized at 886 nm. Measurements confirmed that the x-polarized OTF is flat (constant transmission), while the y-polarized OTF exhibits a linear dependence on spatial frequency ( $k_x$ ) over a numerical aperture (NA) of 0.05. The measured transmission coefficients and slopes aligned well with design targets, despite minor discrepancies attributed to fabrication non-uniformity and refractive index variations.
Phase Test Targets: The system successfully imaged fabricated phase test targets (a spoke pattern and a "bullseye" structure). By rotating the input polarization, the system switched between a standard brightfield image and a phase-contrast image revealing phase gradients. Quantitative recovery of the phase gradient was demonstrated by solving a quadratic equation derived from the intensity measurements of both polarization states, yielding root mean square errors (RMSE) of approximately 0.05–0.06 rad/µm.
Biological Imaging: The metasurface was applied to unstained HeLa cells and human ovarian cancer tissue samples. In brightfield mode, the cells were nearly invisible. Upon switching to y-polarized illumination, the cell morphology and tissue textures (including intra-cystic spaces) became clearly visible with a pseudo-3D appearance. The results were comparable to DIC microscopy, though with a lower NA (0.05 vs. 0.4). To mitigate diffraction artifacts caused by the finite size of the metasurface and the coverslip separation, the authors employed an averaging technique where the metasurface was slightly offset between 10 image captures.

Significance and Claims
The paper claims to demonstrate the first inverse-designed, nonlocal metasurface capable of switching between brightfield and phase-contrast imaging modalities simply by rotating the illumination polarization, without requiring mechanical movement or additional optical components.

Compactness and Integration: The work extends inverse-design methods to reconfigurable phase-contrast microscopy, offering an ultra-compact form factor suitable for integration into standard microscope objectives or computational camera systems.
All-Optical Computation: The device performs analogue optical computation (differentiation) at the speed of light, eliminating the power consumption and latency associated with digital image processing.
Robustness: The use of topology optimization with fabrication-aware constraints resulted in a design that is robust to manufacturing defects, a critical step toward scalable production.
Limitations and Scope: The authors modestly note that the current implementation operates at a single wavelength (886 nm) and requires spatially coherent light. The phase gradient recovery is limited to one dimension (x-axis), and the current NA is relatively low (0.05), which restricts the resolution compared to high-NA DIC systems. The paper suggests that future work could focus on broadband operation, dual-axis gradient recovery, and extending the approach to image-plane processing.

In conclusion, this work establishes a pathway for developing low-cost, energy-free imaging systems for biological diagnostics and materials characterization by leveraging nonlocal metasurfaces for real-time, switchable optical image processing.