CPS: Synergy: Collaborative Research: Hybrid Continuous-Discrete Computers for Cyber-Physical Systems

本研究的目的是针对网络物理系统的混合(离散-连续)计算。这项研究用连续(模拟)计算扩充了当今无处不在的离散(数字)计算模型，它非常适合于网络物理系统中涉及的连续自然变量，以及这种系统中计算的容错性质。其结果是在单个硅芯片上形成了一个计算平台，与纯离散计算相比，具有更高的能效、更快的速度和更好的数值收敛。这项研究有几个重点：(1)硬件：使用现代硅芯片技术将同一芯片上的模拟计算硬件与数字硬件合并，后者用于控制和协同计算；(2)体系结构：设计了使混合计算功能可供软件访问的方法；(3)微体系结构：根据模拟和混合模拟-数字功能单元的粒度、类型和组织进行选择；以及(4)具体应用于由机器人团队组成的现实网络物理系统。这项研究将现代计算机体系结构技术和主要在通信背景下开发的模拟/数字混合芯片技术的进展扩展到用于网络物理系统的混合计算。它为容错工作负载带来了更高水平的能效，未来的计算机将不得不处理这些工作负载。开发的技术可以扩展到其他必须进行有效计算的系统，如天气预报和高能物理。这项工作将研究与教育结合起来，并包括广泛传播所取得成果的计划。

This research aims at hybrid (discrete-continuous) computation for cyber-physical systems. The research augments the today-ubiquitous discrete (digital) model of computation with continuous (analog) computing, which is well-suited to the continuous natural variables involved in cyber-physical systems, and to the error-tolerant nature of computation in such systems. The result is a computing platform on a single silicon chip, with higher energy efficiency, higher speed, and better numerical convergence than is possible with purely discrete computation. The research has several thrusts: (1) Hardware: modern silicon chip technology is used to merge analog computing hardware on the same chip with digital hardware, the latter used for control and co-computation, (2) Architecture: methods are devised for making hybrid computing functionality accessible to the software, (3) Microarchitecture: Choices are made on the granularity, type and organization of analog and hybrid analog-digital functional units, and (4) Concrete application to a realistic cyber-physical system consisting of a team of robots. The research extends modern computer architecture techniques, and advances in mixed analog/digital chip technology mainly developed in the context of communications, to hybrid computing for cyber-physical systems. It brings higher levels of energy efficiency to error-tolerant workloads that future computers will have to handle. The techniques developed can be extended to other systems in which efficient computation is a must, such as weather forecasting and high-energy physics. The work integrates research with education and includes plans for broad dissemination of the results obtained.