CI-P: Collaborative Project: Massively-Parallel Analog Co-Processors for Simulating Complex Systems

CI-P：协作项目：用于模拟复杂系统的大规模并行模拟协处理器

• 批准号: 1629790
• 负责人: Soumyajit Mandal
• 金额: $ 5万
• 依托单位: Case Western Reserve University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-08-01 至 2018-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1629790&HistoricalAwards=false
• 关键词: CI; Collaborative; Project; Massively; Parallel

Digital supercomputers have become exponentially faster in recent years, but these general-purpose machines still leave much to be desired in terms of computational speed for mathematically rigorous simulations of various physics-based systems. Moreover, digital computer performance is beginning to saturate due to fundamental limitations in transistor scaling and switching speed. However, the analog performance of transistors continues to improve, and frequency ranges exceeding 400 GHz are now available for applications such as millimeter wave wireless communications, imaging, and radar. Hence there is a lot of interest in high-speed analog co-processors that can rapidly perform certain specialized simulations by taking advantage of modern nanoscale integrated circuit technology. The overall goal of the project is to discover the theory, design the circuits, and verify the operation of scalable analog computation units with multi-GHz bandwidths that can be seamlessly integrated into modern digital computation architectures, thus enabling novel hybrid (analog/digital) computing paradigms.This "planning" research is designed to obtain the foundational knowledge, derive the requirements and specifications, and determine the teaming needs for an eventual full project to the CRI program that will enable the CISE community to more easily explore this "beyond Moore" computing paradigm. In particular, it will provide shared resources for developing and using radio frequency (RF) analog accelerators for various scientific computing applications. This CI-P planning activities will explore several mathematical models that can be efficiently mapped to massively-parallel integrated analog accelerators. The mathematical models may include simplified versions of difficult scientific computing problems (e.g., the shallow-water version of the Navier-Stokes equations) and important problems in engineering (e.g., Maxwell's equations of electromagnetism). The planning stage will include a one-day "Ideas Workshop" to facilitate informal discussions between CISE researchers belonging to several universities who are either experienced in analog and hybrid computing, or have significant interest in the topic. The project includes significant educational and outreach activities, including outreach activities at local high schools in the Cleveland-Akron area, summer workshops at the lead-PI's laboratory at the University of Akron, and mentorship/research experiences at Case Western Reserve University.

近年来，数字超级计算机的速度呈指数级增长，但这些通用机器在数学上严格模拟各种基于物理的系统的计算速度方面仍有很大的不足。此外，由于晶体管缩放和开关速度的基本限制，数字计算机性能开始饱和。然而，晶体管的模拟性能不断提高，超过400ghz的频率范围现在可用于毫米波无线通信、成像和雷达等应用。因此，人们对高速模拟协处理器非常感兴趣，因为它可以利用现代纳米级集成电路技术快速执行某些特定的模拟。该项目的总体目标是发现理论，设计电路，并验证具有多ghz带宽的可扩展模拟计算单元的操作，这些单元可以无缝集成到现代数字计算架构中，从而实现新的混合（模拟/数字）计算范式。这项“规划”研究旨在获得基础知识，得出需求和规范，并确定最终完整的CRI项目的团队需求，这将使CISE社区能够更轻松地探索这种“超越摩尔”的计算范式。特别是，它将为开发和使用各种科学计算应用的射频（RF）模拟加速器提供共享资源。这个CI-P规划活动将探索几个数学模型，这些模型可以有效地映射到大规模并行集成模拟加速器。数学模型可能包括科学计算难题的简化版本（例如，浅水版的Navier-Stokes方程）和工程中的重要问题（例如，麦克斯韦电磁学方程）。计划阶段将包括一个为期一天的“创意研讨会”，以促进来自几所大学的CISE研究人员之间的非正式讨论，这些研究人员要么在模拟和混合计算方面有经验，要么对该主题有浓厚的兴趣。该项目包括重要的教育和推广活动，包括在克利夫兰-阿克伦地区当地高中的推广活动，在阿克伦大学领导pi实验室的夏季讲习班，以及在凯斯西储大学的指导/研究经验。