Grayscale Electron-Beam Lithography

• Grayscale electron-beam lithography is a nanofabrication technique that uses spatially modulated electron doses to create continuous or discretized 3D resist profiles with sub-10 nm vertical and lateral precision.
• The process integrates advanced CAD-based grayscale encoding and precise dose control, optimizing exposure through methods like proximity effect correction and iterative Fourier-transform algorithms.
• Applications span high-resolution photonics, metasurface fabrication, and device calibration, enabling practical implementations such as Mie void metasurfaces and astronomical-grade x-ray gratings.

Grayscale electron-beam lithography (GEBL) is a direct-write nanofabrication methodology enabling continuous or discretized three-dimensional (3D) resist profiles with sub-10 nm vertical and lateral precision. Originating in electron‐scattering physics and resist chemistry, GEBL uses spatially modulated electron doses to produce surface relief and volumetric patterns for advanced applications in photonics, diffractive optics, nanomechanics, and instrument calibration. The process decouples the traditional binary patterning model and provides robust control over resist residual height, sidewall geometry, and, via pattern transfer, enables the direct realization of freeform nanostructures in a wide range of materials (Li, 2015, Goldberg et al., 29 Mar 2026, McCoy et al., 2021, Wang et al., 2024, Shiloh et al., 2014).

1. Physical Principles of GEBL

GEBL operates by varying the local electron dose $d(x, y)$ delivered to the resist surface, which induces a depth- and laterally-resolved energy deposition profile determined by the electron point-spread function (PSF). The PSF, $\text{psf}(r, z)$, combines narrow forward-scattering (nm scale) and broad backscattering (hundreds of nm), typically modeled by a double Gaussian:

$$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$

with backscattering coefficient $\eta \approx 0.2–0.3$. The local energy dose $e(x,y,z)$ is given by the convolution:

$$e(x, y, z) = \iint d(x', y')\,\text{psf}(x-x', y-y', z)\,dx' dy'$$

(Li, 2015).

Following exposure, the resist development process proceeds at a rate $r(x, z)$ that is a highly nonlinear function of local energy deposition. For PMMA, a representative model is a third-order polynomial:

$r(x, z) = F[e(x, z)] = a_3e^3 + a_2e^2 + a_1e$

with coefficients empirically determined for each resist/process (Li, 2015). The net result is a spatially varying residual resist thickness $h(x, y)$ as a function of delivered dose, calibrated via profilometry or AFM, typically exhibiting sigmoidal or exponential dose–thickness dependence (McCoy et al., 2021, Goldberg et al., 29 Mar 2026, Wang et al., 2024).

2. GEBL Process Flow and Calibration

A canonical GEBL sequence entails:

• Substrate Preparation: Selection and baking of substrate (e.g., Si wafer, conductive glass) for resist adhesion/stability.
• Resist Coating: Spin-coating of positive resists (PMMA) or material-precursor films (e.g., Sb₂S₃ precursor) to final thickness $h_0$ (Goldberg et al., 29 Mar 2026, Wang et al., 2024).
• Dose Design and Pattern Definition: Generation of spatial dose maps $\text{psf}(r, z)$0 from CAD or grayscale encodings (see Section 3), with optional proximity effect correction (PEC) to compensate for electron scattering and substrate backscatter (Borghi et al., 2024).
• E-beam Exposure: E-beam tool delivers modulated local dose, typically across a range spanning the linear and saturation regions of the resist contrast curve. Practical lateral resolutions of $\text{psf}(r, z)$1 nm and vertical precision of $\text{psf}(r, z)$2 nm are attainable for PMMA, with the lower bound set by tool resolution, proximity effects, and resist chemistry (Goldberg et al., 29 Mar 2026).
• Development: Developed in suitable solvent (e.g., MIBK:IPA) to reveal relief structure. Calibration structures (dose arrays) are employed to empirically establish $\text{psf}(r, z)$3 or $\text{psf}(r, z)$4 relations, where $\text{psf}(r, z)$5 is local electron dose (McCoy et al., 2021, Goldberg et al., 29 Mar 2026).
• Pattern Transfer: For devices requiring material etching (e.g., Si), the developed resist is used as an etch mask; features such as void depths or pillar heights are transferred according to the known selectivity ratio $\text{psf}(r, z)$6 (Goldberg et al., 29 Mar 2026).
• Optional Postprocess Steps: Thermally activated reflow (e.g., TASTE) to selectively smooth and facet resist profiles, or direct formation of dielectric/semiconductor structures (e.g., Sb₂S₃ via beam-induced conversion) (McCoy et al., 2021, Wang et al., 2024).

The table below summarizes critical calibration data:

Material/Method	Dose Range	$\text{psf}(r, z)$7 Type	Depth Resolution	Reference
PMMA/Si	0.05–0.8 μC/cm²	Sigmoidal, linear	≤10 nm	(Goldberg et al., 29 Mar 2026)
PMMA (TASTE)	50–180 μC/cm²	Exponential	N/A	(McCoy et al., 2021)
Sb₂S₃ direct	2k–30k μC/cm²	Linear	<10 nm	(Wang et al., 2024)

3. Grayscale Encoding: Design and Software Integration

GEBL has incorporated advanced methodologies for encoding arbitrary 3D reliefs and phase profiles. Standard workflows now leverage CAD-based parametric surfaces, iterative Fourier-transform algorithms (e.g., Gerchberg-Saxton) for holographic phase-masks, and direct mapping of 3D models (from Blender or equivalent) into grayscale templates (Shiloh et al., 2014, Borghi et al., 2024). The process can be specified as:

• Model-to-Grayscale Mapping: Analytical or mesh geometric height $\text{psf}(r, z)$8 is normalized, then mapped to an 8-bit (0–255) grayscale image:

$\text{psf}(r, z)$9

with further correction for contrast nonlinearity via empirically fitted transfer curves (Borghi et al., 2024).

• Dose Rasterization: The grayscale map is converted to local dwell-time or beam-current levels via calibrated LUTs:

$$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$0

where $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$1 is the normalized geometric height and the mapping function is usually a low-order polynomial based on measured development/removal data.

• PEC and Field Stitching: After LUT application, proximity effect correction is applied to mitigate broadening from backscattering. Fields larger than the maximum EBL write field (e.g., $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$2m × $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$3m) are segmented with overlaps for seamless stitching (Borghi et al., 2024).
• Dynamic Range and Quantization: Practical systems can support 8–16 grayscale (dose) levels per spatial coordinate, with dithering schemes mitigating quantization artifacts where needed. Bit-depth, pixel pitch, and depth-step fidelity are determined by resist dynamics and tool capability (Li, 2015, Goldberg et al., 29 Mar 2026).

4. Optimization and Sidewall Shape Control

A core advantage of GEBL is deterministic sidewall engineering through spatial dose optimization. The target sidewall profile $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$4 is achieved by minimizing the deviation between simulated resist cross-section and the design, under total dose and process constraints. For PMMA/Si systems, this is implemented as a constrained optimization, solved via simulated annealing (SA):

• Objective Function: Minimize the least-squares error between simulated ($$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$5) and target ($$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$6) half-widths at discrete depths:

$$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$7

• Constraints: Fixed total dose, per-pixel dose bounds; mapped to discrete dose levels as required.
• Algorithmic Solution: SA proposes random dose adjustments under shrinking temperature, with local PSF-based exposure simulation, nonlinear development-rate mapping, and 2D path-based development simulation at each step. Typical spatial discretizations: $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$8 nm, $$\text{psf}(r, z) = (1–\eta)\frac{1}{\pi \alpha^2(z)}e^{-r^2/\alpha^2(z)} + \eta\frac{1}{\pi \beta^2(z)}e^{-r^2/\beta^2(z)}$$9 nm (Li, 2015).
• Empirical Rules: Verified experimentally, inclusion of edge-boosted doses ($\eta \approx 0.2–0.3$0–$\eta \approx 0.2–0.3$1 higher at pattern edges) achieves near-vertical or undercut profiles; uniform doses yield overcut sidewalls (Li, 2015).
• Workflow: Design, calibration, optimization, quantization, write/measure/test, iterate.

5. Applications in Nanophotonics, Metasurfaces, and Diffractive Optics

GEBL is now a principal enabling technology for high-efficiency diffractive and refractive micro-optics, 3D metasurfaces, and micro/nano-mechanical structures.

• Mie Void Metasurfaces: GEBL enables precise multi-depth definition in PMMA, transferred into Si, to form arrays of air voids with nanometer-tuned depth, resulting in resonant spectral tuning across the visible. Depth control in the $\eta \approx 0.2–0.3$2 nm regime enables full visible color gamut, as evidenced by a linear fit: $\eta \approx 0.2–0.3$3 nm for void depths $\eta \approx 0.2–0.3$4 from 200–500 nm (Goldberg et al., 29 Mar 2026).
• Direct Patterning of Sb₂S₃: Variable-dose GEBL in molecular precursor films followed by direct "develop" yields amorphous Sb₂S₃ optical structures with sub-10 nm step and $\eta \approx 0.2–0.3$5 nm lateral precision, indexed by empirical dose–height calibration, and refractive-index confirmed through genetic algorithm fitting to spectral data (Wang et al., 2024).
• 3D Diffractive Elements: Multilevel Fresnel zone plates and metalenses fabricated in a single GEBL step achieve >80% of the theoretical diffraction efficiency for $\eta \approx 0.2–0.3$6 phase levels at visible wavelengths, limited primarily by dose-control and material refractive index (Wang et al., 2024).
• Astronomy-Grade X-Ray Gratings with TASTE: GEBL-written staircases, when reflowed above local $\eta \approx 0.2–0.3$7 (TASTE), self-equilibrate into blazed gratings with precisely tunable blaze angle and facet smoothness at nanometer RMS levels; suitable for next-generation spectrometers with sub-nanometer groove stability (McCoy et al., 2021).

6. Full-Field Lithography and Holographic Phase Mask Approaches

Beyond serial scanning, GEBL can be implemented via full-field illumination of static holographic phase masks fabricated by FIB into low-stress Si₃N₄ membranes. By converting algorithmically derived phase profiles $\eta \approx 0.2–0.3$8 into mask thickness $\eta \approx 0.2–0.3$9:

$e(x,y,z)$0

where $e(x,y,z)$1 is the electron-matter interaction constant, one can produce arbitrary grayscale intensity distributions. These masks permit massively parallel, high-speed lithography over micrometer-scale fields with $e(x,y,z)$230 distinct gray levels, approaching the fidelity of advanced scanning approaches (Shiloh et al., 2014).

Compared to scanning GEBL, mask-based methods offer

• Orders-of-magnitude higher throughput (full-field writing in $e(x,y,z)$3 s).
• Fabrication and alignment complexity shifted to mask preparation.
• Intrinsic limits set by mask feature size, membrane stability, and projection demagnification; resolutions $e(x,y,z)$4–30 nm are achievable.

7. Limitations, Accuracy Metrics, and Practical Considerations

• Depth and Lateral Resolution: PMMA-based GEBL in Si achieves $e(x,y,z)$5 nm depth and $e(x,y,z)$6 nm lateral resolution, as measured by AFM and SEM (Goldberg et al., 29 Mar 2026). Dynamic range is set by resist thickness and maximum clearable depth; depth steps are $e(x,y,z)$7200–300 nm in 8-bit systems, finer in direct-beam implementations (Borghi et al., 2024).
• Dose Quantization: Practical dose control supports 8–16 grayscale levels per pixel, with local PEC necessary for accurate relief in dense or complex patterns (Li, 2015).
• Resist and Process Sensitivities: Depth control is limited by resist chemistry (contrast, swelling/saturation), process history (bake, development), and e-beam system stability; frequent calibration is required when process parameters change (Borghi et al., 2024).
• Proximity Effects: Proximity effect broadening can limit minimum feature size and reduce depth contrast at pattern edges or in closely spaced features; robust correction schemes are necessary for high-fidelity grayscale (Li, 2015, Borghi et al., 2024).
• Material-Specific Effects: For direct-writing materials like Sb₂S₃, process sensitivity (low sensitivity, proximity, and thickness constraints) can limit large-area patterning; applications favor prototyping and master fabrication (Wang et al., 2024).
• Throughput: Serial e-beam writing remains slow for large areas; static mask projection and “write-once, read-many” approaches substantially mitigate this, at the expense of mask design flexibility (Shiloh et al., 2014).

• (Li, 2015) Optimization of Spatial Dose Distribution for Controlling Sidewall Shape in Electron-beam Lithography
• (Borghi et al., 2024) Rapid Prototyping of 3D Microstructures: A Simplified Grayscale Lithography Encoding Method Using Blender
• (McCoy et al., 2021) Fabrication of astronomical x-ray reflection gratings using thermally activated selective topography equilibration
• (Goldberg et al., 29 Mar 2026) Single-Step Grayscale Lithography of Multi-Depth Mie Void Metasurfaces
• (Wang et al., 2024) Grayscale Electron Beam Lithography Direct Patterned Antimony Sulfide
• (Shiloh et al., 2014) Sculpturing the Electron Wave Function