Large-Area Fabrication-Aware Computational Diffractive Optics

TL;DR

This work addresses the persistent gap between simulation and fabrication in learned large-area diffractive optics. It introduces a fabrication-aware, end-to-end design framework that couples a super-resolved neural lithography model with a differentiable image-formation model and a tensor-parallel computation engine (D^2FFT and GSPMD) to enable centimeter-scale DOEs with sub-micron features. The method is validated through experiments in holographic display, beam shaping, and broadband imaging, demonstrating close agreement between simulated designs and fabricated prototypes and enabling high-quality results with minimal post-processing. By enabling mass-producible, fabrication-aware computational optics, the approach significantly advances practical deployment of learning-based DOE systems in real-world imaging and display applications.

Abstract

Differentiable optics, as an emerging paradigm that jointly optimizes optics and (optional) image processing algorithms, has made innovative optical designs possible across a broad range of applications. Many of these systems utilize diffractive optical components (DOEs) for holography, PSF engineering, or wavefront shaping. Existing approaches have, however, mostly remained limited to laboratory prototypes, owing to a large quality gap between simulation and manufactured devices. We aim at lifting the fundamental technical barriers to the practical use of learned diffractive optical systems. To this end, we propose a fabrication-aware design pipeline for diffractive optics fabricated by direct-write grayscale lithography followed by nano-imprinting replication, which is directly suited for inexpensive mass production of large area designs. We propose a super-resolved neural lithography model that can accurately predict the 3D geometry generated by the fabrication process. This model can be seamlessly integrated into existing differentiable optics frameworks, enabling fabrication-aware, end-to-end optimization of computational optical systems. To tackle the computational challenges, we also devise tensor-parallel compute framework centered on distributing large-scale FFT computation across many GPUs. As such, we demonstrate large scale diffractive optics designs up to 32.16 mm $\times$ 21.44 mm, simulated on grids of up to 128,640 by 85,760 feature points. We find adequate agreement between simulation and fabricated prototypes for applications such as holography and PSF engineering. We also achieve high image quality from an imaging system comprised only of a single DOE, with images processed only by a Wiener filter utilizing the simulation PSF. We believe our findings lift the fabrication limitations for real-world applications of diffractive optics and differentiable optical design.

Figures (15)

• Figure 1: Fabrication-aware Image Formation Model. We illustrate the proposed image formation model (see text) for computational diffractive optics. With this differentiable model in hand, we optimize design layouts (i.e., inputs of the lithography machine) end-to-end informed by the proposed super-resolved neural lithography model, via backpropagation.
• Figure 2: Toy Example of the "Fabrication Interpolation Kernel". We train and evaluate DOEs for computer-generated 2-D hologram under different settings, including 1) the conventional design at 2µm-spacing grid; 2) design with nearest upsampling in DOE plane at 250 nm-spacing grid (i.e., 8$\times$ upsampling) that meets the $\frac{\lambda}{2}$-spacing requirement of the Nyquist sampling theorem; 3) design with Lanczos upsampling in the DOE plane at 250 nm-spacing grid. Resulting PSNRs are shown in the top-left corner for each image. A comprehensive ablation study is provided in the Supplementary Material.
• Figure 3: Contrast Curve Calibration. (a) the initial and calibrated gray value distribution (GVD) for intensity control of the direct-write grayscale lithography machine. A nonlinear mapping between gray values (1024 levels) and the laser intensity is utilized after calibration. (b) the initial and calibrated contrast curve of the photoresist. After calibration, we find the developed resist depth is almost perfectly linearly proportional to the gray values.
• Figure 4: Lithography Model Evaluation. We report here a design pattern and its corresponding AFM measurement from the constructed evalset, along with lithography model predictions---the modulation transfer function-based physical model and the proposed neural lithography model. The error maps (with respect to the AFM measurement ground truth) are also annotated with the associated PSNR values, validating the proposed model predictions.
• Figure 5: GSPMD-based Distributed Computing Framework tailored for large-area fabrication-aware diffractive optics. We illustrate the distributed computation of the proposed D$^2$FFT (top-left) and the spatial-partitioning convolution (bottom-left) leveraging tensor parallelism. The (GPU) processors can be arranged into a multi-dimensional mesh to enable arbitrary combinations of hybrid data and tensor parallelism.