In the theoretical world of nanophotonics, researchers design devices with mathematical perfection. Waveguides have perfectly vertical walls, microresonators possess flawless boundaries, and metasurfaces feature arrays of identical, sharp geometric features.
In the physical world, however, the story changes completely. The manufacturing processes used to build these devices—such as electron-beam lithography and chemical etching—operate at their physical limits. As a result, fabricated structures suffer from nanometer-scale imperfections: sidewall roughness, non-vertical etching, layer thickness variations, and material impurities.
Because nanophotonic devices manipulate light at scales smaller than its wavelength, they are extraordinarily sensitive to these minor variations. As noted by Wei et al. and Azimi, a deviation of just a few nanometers can completely alter a device's optical response, dampening resonances or rendering the hardware entirely non-functional. To transition nanophotonics from laboratory curiosities to mass-producible technologies, researchers must design for fabrication tolerance—creating devices that maintain high performance despite unavoidable manufacturing errors.
Why Nanoscale Deviations Ruin Optical Performance
To understand why a few nanometers matter so much, we have to look at how nanophotonic devices control light. Many components rely on precise phase matching, constructive interference, or localized optical modes.
When the physical geometry of a structure changes, its effective refractive index shifts. For example, in complex photonic crystal slabs, imperfections like sidewall roughness and poor etching verticality distort the boundaries of the crystal lattice, degrading the device's quality factor (Hou et al.). In resonant structures, such as the metallic nano-structures studied by Tang et al., dimensional defects damp resonance peaks and lower overall filtering efficiency.
Furthermore, geometric variations break the symmetry of the system. This structural disorder triggers backscattering—where light reflects backward instead of moving forward—introducing severe signal loss and noise into integrated optical circuits (Sun et al.; Tang et al.).
To combat these issues, modern research focuses on four primary strategies: geometric engineering, topological protection, robust inverse design, and post-fabrication correction.
Strategy 1: Geometric Engineering and Mode Isolation
One way to build fabrication tolerance is to design geometries where the optical mode is inherently insulated from boundary defects. If light does not strongly interact with the rough edges of a device, those edges cannot easily scatter it.
Ideal Waveguide: [ Smooth Core ] --> Light flows perfectly
Fabricated Waveguide: [~ Rough Edge ~] --> Light scatters at boundaries
Air-Host Waveguide: Rough Wall | Air Gap (Light Trapped Here) | Rough Wall
(Light avoids the rough boundaries entirely)
An elegant example of this is the "air-host" plasmonic waveguide designed by Feng et al.. Instead of trapping light inside a solid dielectric layer where thickness variations alter performance, their design confines light within an air gap between a nanowire dimer. Because the mode is localized in the air rather than the material interfaces, the device exhibits high tolerance to substrate layering inconsistencies.
Similarly, Singh et al. achieved robustness in a multimode interference (MMI) wavelength combiner by introducing a non-uniform refractive index distribution. This design creates distinct optical modes that act collectively, making the device resilient against small variations in groove thickness and width during manufacturing.
Strategy 2: Topological Protection Against Disorder
Instead of trying to avoid boundaries, topological photonics alters the underlying physics of the device so that light is mathematically forbidden from scattering backward.
In a standard waveguide, a defect acts as a mirror, reflecting light and ruining transmission. However, by arranging ring resonators or photonic crystals into specific topological configurations, researchers can create "edge states." As demonstrated by Chen et al., using conjugated series coupled ring resonators creates topological interface states that are robust against manufacturing disorders and material impurities.
Because these topological states lack a physical mechanism to support backward propagation, light simply flows around defects or surface roughness without losing energy (Tang et al.). This approach allows high-performance filters and switches to operate reliably even when fabricated with standard, lower-precision manufacturing techniques.
Strategy 3: Robust Inverse Design
Traditional photonic design relies on human intuition to tweak basic shapes (like circles or rectangles). Inverse design, by contrast, uses optimization algorithms to search a massive mathematical space and discover non-intuitive, highly complex geometries that achieve a specific performance goal.
To make these algorithmic designs practical, the optimization process must explicitly account for manufacturing imperfections. If an algorithm designs a device with features that are only 2 nanometers wide, that device cannot be reliably manufactured.
Standard Inverse Design: Optimize for Perfect Geometry ---> Fails in Fab
Robust Inverse Design: Optimize for [Geometry ± 5nm] ---> Succeeds in Fab
To solve this, Butz et al. introduced a "dynamic binarization" technique for objective-first inverse design. By treating the transition from continuous material properties to discrete physical boundaries as an active variable during optimization, they closed the gap between idealized simulations and physical manufacturing limits. Designing with these constraints ensures that the resulting metasurfaces or mode converters remain high-performing even when their edges are slightly blurred or distorted by the lithography process (Makarenko et al.).
Strategy 4: Post-Fabrication Correction and Calibration
When hardware cannot be made perfectly tolerant through design alone, researchers turn to software and computational methods to mitigate the impact of physical errors.
One approach is to predict and correct for variations before the device is even deployed. Azimi developed SEMU-Net, a deep-learning framework that analyzes microscopic images of fabricated nanophotonic structures. By segmenting and identifying the exact process variations present on a chip, the system acts as a corrector, allowing engineers to map physical defects directly to their optical consequences.
Alternatively, the system-level architecture can be co-optimized to tolerate individual component failures. In large-scale integrated photonic arrays, such as the lensless active-flat-optics imaging system developed by Khial, fabrication inconsistencies can render individual sensor elements completely inoperative. However, by pairing the imperfect hardware array with a digital computational imaging decoder, the system can reconstruct high-quality images despite localized hardware faults, shifting the burden of precision from physical manufacturing to digital algorithms.
Key Foundational Work
Robust design of slow-light tapers in periodic waveguides (2009) — contributed 2 concept elements to this direction.
Spherical distributed dielectric resonators (1993) — contributed 1 concept elements to this direction.