E2CDA: Type I: Collaborative Research: Energy Efficient Computing with Chip-Based Photonics

在今天的数据爆炸中，对计算能力的需求正在加速，解决科学、工程、商业和智能处理中的关键问题的能源需求正在急剧增加。为了解决这个关键问题，科学和工程界开始探索新的计算方法，例如模拟大脑的结构或耦合粒子的动态行为。然而，从能源和计算的角度来看，用传统的计算机体系结构实现这些方法是非常低效的。因此，迫切需要开发革命性的计算机体系结构来克服这些严重的障碍。在这个项目的框架内，一个雄心勃勃的计划将被追求，在这个计划中，光而不是电子被用来实现具有优越能量可扩展性的新计算范式。具体来说，将探索利用光的波动性质（即其振幅和相位）的计算架构，利用过去十年中具有数千个高性能设备的复杂光子芯片的纳米制造的显着进步。这些光子平台将具有强大的计算能力，以无与伦比的能效运行，并且具有可扩展性和高度可重构性。为了实现这一节能光子计算的愿景，建议的研究工作将集中在以下两种类型的光子处理器上：(1)Ising Machine和(2)Neuromorphic computing Machine。这些光子处理器的统一之处在于它们由耦合光子单元的动态网络组成，并依靠光的波动性质来解决在电子计算系统中没有类似的问题。除了开发新型光学处理器之外，光子学技术还将成为未来计算机系统的基石。将探索一种集成光子处理器并与电子存储器和处理器互连的新架构，以最大限度地发挥每种技术的优势。研究还将进行完整的，具有挑战性的问题映射到光子加速器和电子处理器的组合，并确定如何最好地扩展它们以最大化全系统性能。通过在系统的各个层面进行创新——从设备到架构，包括系统、编译器和算法——该项目旨在实现任何单一领域都无法实现的进步。

Amidst today's data explosion, demand for computing power is accelerating, and the energy requirements for solving critical problems in science, engineering, business, and intelligent processing are increasing dramatically. In order to address this critical issue, the scientific and engineering community is beginning to explore new approaches to computing, such as mimicking the brain's structure or the dynamical behavior of coupled particles. However, implementing these approaches with conventional computer architectures is highly inefficient from both energy and computing standpoints. As a result, there is a pressing need to develop revolutionary computer architectures that overcome these severe roadblocks. An ambitious program is to be pursued within the framework of this project in which light, rather than electrons, is used to realize new computing paradigms with superior energy scalability. Specifically, computing architectures will be explored that exploit the wave nature of light (i.e., its amplitude and phase) harnessing the remarkable advancements over the past decade in nanofabrication of complex photonic chips with thousands of high-performance devices. These photonic platforms would have the potential to be computationally powerful, operate with unparalleled energy efficiency, and are scalable and highly reconfigurable.To fulfill this vision of energy efficient photonic computing, the proposed research efforts will focus on the following two types of photonic processors: (1) Ising Machine and (2) Neuromorphic Computing Machine. The unifying aspect of these photonic processors is that they consist of dynamic networks of coupled photonic units and rely on the wave nature of light to solve problems which has no analogy in electron-based computing systems. Beyond developing new types of optical processors, photonics technology will be developed as the cornerstone of future computer systems. A novel architecture will be explored that integrates photonic processors and interconnects with electronic memory and processors to maximize the benefits of each technology. Research will also be undertaken to map complete, challenging problems to combinations of photonic accelerators and electronic processors, and to determine how to best scale them to maximize full-system performance. By innovating at all levels of the system - from devices to architectures including systems, compilers, and algorithms - the project would aim to achieve advances that cannot be realized within any single field.