Artificial Intelligence (AI) Mode-Space Optical/Quantum Processor Design for Energy-Autonomous AI Applications

适用于能源自主 AI 应用的人工智能 (AI) 模式空间光学/量子处理器设计

• 批准号: RGPIN-2021-03480
• 负责人: LiboironLadouceur, Odile
• 金额: $ 4.66万
• 依托单位: McGill University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=741846
• 关键词: Artificial; Intelligence; AI; Mode; Space

Over the last six years, Dr. Liboiron-Ladouceur discovered and demonstrated how optical modes lead to higher throughput systems, thanks to technology advancements in silicon photonics (SiPh) integrated circuits. Recently, Dr. Liboiron-Ladouceur uncovered powerful design approaches referred as inverse design by developing AI-based algorithms towards more performing and compact devices. The research outcomes reveal how we can surpass conventional design methodology currently relying on human experience and intuition. In parallel, Dr. Liboiron-Ladouceur has recently developed an optical processor for computation enabling mathematical transformation of vectors and matrices. This research outcome reveals the inherent properties of photons for processing and computing data beyond their transmission capabilities in low-loss optical channels. Optical computation can lead to better energy efficiency and be accelerators in modern computer architectures. The initial work demonstrates good prediction accuracy in a practical framework accounting for fabrication and energy usage. However, increasing the optical processor size to the required vector dimensions remains an important challenge. Indeed, the processor sizes needed, such as in autonomous AI applications, are greater by one to two orders of magnitude than what is achievable through conventionally designed photonic integrated circuits. The research objectives of the proposed discovery program investigate optical modes to increase the size of optical processors exploiting recent design methodology in photonic integrated circuits. Mode division multiplexing view modes as orthogonal data channels mainly for transmission purpose. In the discovery program, scaling the vector size from tens to hundreds of elements leverages Dr. Liboiron-Ladouceur's significant contributions in SiPh mode-based device development. Her research methodology exploits machine learning algorithms in new inverse design techniques to develop the required devices manipulating photons for computation. These techniques will allow to reduce the device area from tens of microns to a few microns while maintaining the required performance in terms of insertion loss, crosstalk, and robustness to process variations. The benefits of the proposed research program are excellent with important impact from inverse design techniques leading to a paradigm shift in photonic integrated circuits. The feasibility of a mode-space optical processor will lead to energy-efficient computing platforms for AI applications in autonomous vehicles. The proposed discovery program allows for a multidisciplinary, inclusive, and diverse HQP training platform. Students will gain valuable research skills through industry-approved simulation tools and experimental methodologies, while developing their creative and critical mindset as future world leaders.

在过去的六年中，利博罗·拉杜科尔（Liboiron-Ladouceur）博士发现并证明了光学模式如何导致更高的吞吐量系统，这要归功于硅光子学（SIPH）集成电路的技术进步。最近，Liboiron-Ladouceur博士通过开发基于AI的算法来实现更紧凑和紧凑的设备，发现了强大的设计方法，称为反设计。研究结果揭示了我们如何超越当前依靠人类经验和直觉的常规设计方法。同时，Liboiron-Ladouceur博士最近开发了一种用于计算载体和矩阵数学转换的光学处理器。这项研究结果揭示了在低损失光通道中其传输功能以外的处理和计算数据的光子的固有特性。光学计算可以提高能源效率，并成为现代计算机架构中的加速器。最初的工作证明了在构成制造和能源使用的实用框架中的良好预测准确性。但是，将光处理器的大小增加到所需的矢量尺寸仍然是一个重要的挑战。实际上，所需的处理器尺寸（例如在自主AI应用中）比通过传统设计的光子集成电路可实现的数量级要高一到两个数量级。 拟议的发现计划的研究目标研究光学模式，以增加利用光子综合电路中最新设计方法的光学处理器的大小。模式下多路复用视图模式作为正交数据通道主要用于传输目的。在Discovery计划中，将矢量规模从数十到数百个元素扩展，利用Liboiron-Ladouceur博士在基于SIPH模式的设备开发中的重要贡献。她的研究方法利用了新的逆设计技术中的机器学习算法来开发操纵光子进行计算的所需设备。这些技术将允许将设备区域从数十微米减少到几微米，同时在插入损失，串扰和鲁棒性方面保持所需的性能以使过程变化。 拟议的研究计划的好处是极好的，对逆设计技术的重要影响，导致光子综合电路的范式变化。模式空间光学处理器的可行性将导致自动驾驶汽车中AI应用的节能计算平台。拟议的发现计划允许建立多学科，包容和多样化的HQP培训平台。学生将通过行业批准的模拟工具和实验方法获得宝贵的研究技能，同时发展自己的创造性和批判性思维方式，成为未来的世界领导者。