FMSG: Cyber: 3D Printing of Holographic Optical Processors

光学处理器是一种使用光而不是电来传感和处理图像的计算设备。多层光栅模拟神经网络的结构，通过绕射引导光通过这些层的路径，实现低延迟和低功耗的计算。这项技术为现有的计算机视觉方法提供了一个很有前途的替代方案，这些方法在高帧速率下受到大量计算(数十亿个参数)造成的延迟的影响。这个未来制造种子资助(FMSG)项目将研究全息光学处理器的加法制造，使计算机视觉具有最少的计算时间或没有计算时间。如果成功，该项目将产生关于高分辨率制造的新知识，并导致更负担得起的下一代计算设备。计算能力的增强可能会改变多个领域，从人工智能到量子计算，从网络安全到下一代通信。该项目将在未来制造业和人工智能的交界处激发人们对STEM教育的兴趣，并为这两个领域的劳动力发展做出贡献。该项目旨在通过将高通量微米级VPP与纳米级全息记录过程相结合，建立一种全息辅助增值税光聚(H-VPP)过程。H-VPP制造了数十亿个体积衍射栅，取代了干涉图案化的折射率，使得具有大量层的高分辨率光学结构能够形成复杂的光路。需要解决的几个基本研究问题包括：(1)阐明透明光固化树脂如何用于印刷全息器件；(2)研究如何通过添加额外的激光来干涉投影掩模图像来实现纳米级的折射率调制；以及(3)利用深度学习中的反向传播算法来反向设计掩模图像图案。这项研究将对全息过程、光学结构和处理器性能之间的关系获得新的认识。虽然这个项目的重点是VPP，但由此产生的科学成果将促进对其他基于聚合物的高分辨率制造工艺的理解。这一未来制造业奖由土木工程、机械和制造业创新部支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Optical processors are computing devices that use light, rather than electricity, for sensing and processing images. Layers of gratings mimic the structure of neural networks, guide the paths of light through these layers via diffraction, and achieve low-latency and low-power computation. This technology provides a promising alternative to the existing computer vision methods, which are subject to delays caused by massive computations (billions of parameters) at high frame rates. This Future Manufacturing Seed Grant (FMSG) project will investigate the additive manufacturing of holographic optical processors and enable computer vision with minimal or no computational time. If successful, this project will generate new knowledge about high-resolution manufacturing, and lead to more affordable next-generation computing devices. The increased computation capability will potentially transform multiple fields, from artificial intelligence to quantum computing, cybersecurity to next-generation communications. This project will stimulate the interest in STEM education at the interface of future manufacturing and artificial intelligence, and contribute to workforce development in both areas.This project aims to establish a holographically-assisted Vat Photopolymerization (H-VPP) process, by combining high-throughput microscale VPP with the nanoscale holographic recording process. H-VPP fabricates billions of volumetric diffractive gratings, in place of interference-patterned refractive indices, allowing for high-resolution optical structures with a large number of layers to form complex optical pathways. Several fundamental research questions to be addressed include: (1) elucidating how transparent photocurable resins can be used for printing holographic devices, (2) investigating how nanoscale refractive index modulation can be achieved by adding extra laser beams to interfere with the projected mask-images, and (3) inversely designing the mask image patterns using the back-propagation algorithm in deep learning. This research will gain new knowledge about the relationship among the holographic process, optical structure, and processor performance. Though this project focuses on VPP, the resulting science will advance the understanding of other polymer-based high-resolution manufacturing processes. This Future Manufacturing award was supported by Division of Civil, Mechanical and Manufacturing Innovation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.