Presentation
A Journey to Optical Computing: From Physics Fundamentals to Hardware-Software Co-Design, Automation, and Application
DescriptionIn the post-Moore era, conventional computing solutions of digital electronics have become a limiting factor in certain domains, most notably intelligent information processing. The proliferation of big data and artificial intelligence (AI) has motivated the investigation of next-generation computing platforms to support low-latency, energy-efficient machine intelligence. With the advances in device, manufacturing, and integration technology, optics/photonics has attracted increasing attention in recent years and is considered to have great potential in key areas of communication, interconnects, sensing, and computing. Due to the ultra-fast speed and high parallelism of optics, computing platforms based on optical interconnects and processing units can make transformative impacts in important use domains, e.g., future datacenters, automotive, smart sensing, and intelligent edge. In recent years, efforts have been made to facilitate the design automation stack of optical computing platforms and push forward the realization and practical application of next-generation optical processors. However, as the complexity of the heterogeneous mixed-signal computing platform escalates, significant design and optimization challenges need to be tackled. In order to have a scalable design closure and unleash the full power of optics, two critical threads need synergistic advancement along this journey, i.e., cross-layer co-design and electronic-photonic design automation.
In this tutorial, we will guide the interested audience on a journey toward next-generation optical processing platforms with a comprehensive introduction to the literature, co-design & automation technologies, hardware implementation, applications, as well as hands-on demos of optical AI computing.
We separate the tutorial into four talks covering integrated photonic computing, optical networks-on-chip and robustness analysis, synthesis for field-programmable photonic arrays (FPPAs), and diffractive optical neural networks (DONNs).
Talk I aims to introduce the literature and basics of optical computing, as well as novel photonic integrated circuit (PIC) designs for both digital logic and analog AI computing. Talk I will also introduce cross-layer co-design methodologies to enable area-efficient, noise-robust, and self-learnable photonic AI engines, followed by a hands-on demo to guide the audience to walk through the basic usage of our developed open-source optical AI library TorchONN. Talk II aims to introduce advanced topology and physical layout (place & route) optimization technologies for scalable, efficient 2D/3D wavelength-routed optical networks-on-chip design. Talk II will also cover a thorough analysis and comprehensive solution to the process variation, noise robustness, and crosstalk issues of photonic integrated circuits. Talk III will introduce a novel field-programmable photonic array (FPPA) platform with its underlining physical theory and automatic synthesis techniques to realize various important functions, such as routing, signal splitting, filtering, wavelength-division multiplexing, followed by a hands-on live demo (SPODE). Talk IV will introduce free-space diffractive optical neural networks (DONN) utilizing the natural physics of light diffraction and phase modulation of coherent light, with successful applications demonstrated in image classification, segmentation, and graph learning. Talk IV will introduce the fundamentals of DONNs from Fourier optics, differentiable physics modeling, co-design algorithms, and hardware realizations, following an end-to-end compilation framework (live demo) LightRidge.
In this tutorial, we will guide the interested audience on a journey toward next-generation optical processing platforms with a comprehensive introduction to the literature, co-design & automation technologies, hardware implementation, applications, as well as hands-on demos of optical AI computing.
We separate the tutorial into four talks covering integrated photonic computing, optical networks-on-chip and robustness analysis, synthesis for field-programmable photonic arrays (FPPAs), and diffractive optical neural networks (DONNs).
Talk I aims to introduce the literature and basics of optical computing, as well as novel photonic integrated circuit (PIC) designs for both digital logic and analog AI computing. Talk I will also introduce cross-layer co-design methodologies to enable area-efficient, noise-robust, and self-learnable photonic AI engines, followed by a hands-on demo to guide the audience to walk through the basic usage of our developed open-source optical AI library TorchONN. Talk II aims to introduce advanced topology and physical layout (place & route) optimization technologies for scalable, efficient 2D/3D wavelength-routed optical networks-on-chip design. Talk II will also cover a thorough analysis and comprehensive solution to the process variation, noise robustness, and crosstalk issues of photonic integrated circuits. Talk III will introduce a novel field-programmable photonic array (FPPA) platform with its underlining physical theory and automatic synthesis techniques to realize various important functions, such as routing, signal splitting, filtering, wavelength-division multiplexing, followed by a hands-on live demo (SPODE). Talk IV will introduce free-space diffractive optical neural networks (DONN) utilizing the natural physics of light diffraction and phase modulation of coherent light, with successful applications demonstrated in image classification, segmentation, and graph learning. Talk IV will introduce the fundamentals of DONNs from Fourier optics, differentiable physics modeling, co-design algorithms, and hardware realizations, following an end-to-end compilation framework (live demo) LightRidge.
Event Type
Tutorial
TimeMonday, July 10th1:30pm - 5:00pm PDT
Location3003, 3rd Floor
AI