Hybrid Continuous-Discrete Computers for Cyber-Physical Systems.pptx

Abstract:

We describe collaborative research between Columbia University and the University of Texas at Austin, under CNS-1239134 and 1239136. Our goal is to investigate and demonstrate the capabilities of hybrid computers, combining both discrete and continuous computation, in the context of cyber-physical systems. We combine the continuous and discrete models of computation in order to make possible higher energy efficiency, higher speed, and better numerical convergence in the computation involved in such systems. The continuous model is more consistent with the mathematics of differential equations, which are continuous in time and state variables, and involve the real number system as opposed to discrete numbers. Combining the continuous model with the today-ubiquitous discrete model, which naturally supports sequencing, on a single silicon chip, creates a powerful computing platform. This research uses analog circuits to provide direct hardware support for the continuous primitives, thereby enabling higher speed and energy-efficient execution. While analog computers (and even hybrid analog-digital ones) were popular for several decades starting in the 40s, they were overtaken by digital counterparts due to precision, size, and reproducibility concerns. To make hybrid computing in cyber-physical systems a reality today, this research aims at original contributions in several areas: (1) Hardware: novel circuit techniques in modern silicon chip technology to bypass problems that plagued analog computers in the past, and merges analog computing hardware on the same chip with digital hardware, (2) Microarchitecture: this involves choices on the granularity, type and organization of analog and hybrid functional units, and co-optimization of continuous and discrete cores as a unified computing platform, (3) Architecture: methods are devised for making hybrid computing functionality accessible to the software, and (4) Concrete application to a realistic cyber-physical system, in the context of robotics. We present our rationale and preliminary design choices for integrated hybrid computers, in the areas of hardware, architecture, microarchitecture, and applications. These considerations have culminated in the design of our first text chip. The chip includes analog blocks, digital blocks, arrays of analog-to-digital converters, arrays of digital-to-analog converters, and a flexible interconnect fabric. The chip accepts, and delivers, both analog and digital signals, and is highly programmable. This initial test chip is meant to solve problems of up to 4th order, and is sufficient for us to test all building blocks and the communication between them. At the time of this writing, the design of this chip is nearing completion