It is well recognized in the literature that the fully-synchronous design style, once the best choice due especially to the simplicity of its design flow, is not suitable for present-days systems, which contain many more gates compared to their predecessors, and has to be superseded to meet the new needs of the industry. The alternative solution that has enjoyed more success in industry and the literature consists in breaking down a system into several fully-synchronous modules clocked with independent clocks. Such systems go under the name of Globally-non-Synchronous (GnS) and make no assumption on the phase alignment between the clocks in the individual modules. GnS design styles do not require a globally balanced clock tree and employ special synchronizers to achieve latency-insensitivity. The individual modules, whose sizes are relatively small, remain fully-synchronous, thus easy to design andmaintain. Two main classes of GnS systems have been proposed: the GALS (for Globally-Asynchronous, Locally-Synchronous) design style allows each module to be clocked at its own independent clock frequency; the mesochronous design style constrains all modules to run at the same frequency. GALS systems support per-module Dynamic Voltage-Frequency Scaling (DVFS), but GALS interfaces are complex and introduce high performance penalties; mesochronous systems do not support per-module DVFS but support simpler and faster interfaces. It is well recognized that neither of the two design styles can fully satisfy all the contrasting needs of the electronic industry, and often hybrid solutions are deployed as a trade-off. We propose Globally-Ratiochronous, Locally-Synchronous (GRLS) systems, where GRLS is a design style intermediate between the mesochronous and the GALS design paradigms: local frequencies in a GRLS system do not need to be identical, but are required to be rationally-related (such as one being 3/4 or 2/5 of the other). The periodic properties of rationally-related systems allow the deployment of interfaces that do not use any form of handshake and, thanks to this, are much more performant than GALS interfaces; on the other hand, GRLS supports quantized per-module DVFS. In this work we deploy and analyse all the components of the GRLS design style: the frequency regulation system, the voltage regulation system, and the GRLS latency-insensitive interfaces. We perform a theoretical analysis of DVFS efficiency in different GRLS systems, and then study a GRLS NoC-based platform. We also develop a complete GRLS power management system for a GRLS Network-on-Chip (NoC)-based platform. Experimental results show that GRLS performances are close to those of mesochronous systems and GRLS flexibility is close to that of GALS systems, which results in high figures of merit for GRLS systems. As an example, the GRLS NoC-based platform we study in this work has at least ≈ 21% lower latency-power product compared to alternative mesochronous-GALS hybrid platforms, and respectively ≈ 32% and ≈ 48% better latency-power product compared to mesochronous and GALS platforms.

Globally-Ratiochronous, Locally-Synchronous Systems

Pipelining is a key element of high-performance design. Distributed synchronization is at the same time one of the key strengths and one of the major difficulties of asynchronous pipelining. It automatically provides elasticity and on-demand power consumption. This tutorial provides an overview of the best-in-class asynchronous pipelining methods that can be used to fully exploit the advantages of this design style, covering both static and dynamic logic implementations.

http://www.cs.columbia.edu/~nowick/nowick-singh-ieee-dt-11-published.pdf

High-Performance Asynchronous Pipelines: An Overview

An asynchronous design paradigm is capable of addressing the impact of increased process variability, power and thermal bottlenecks, high fault rates, aging, and scalability issues prevalent in emerging densely packed integrated circuits. The first part of the two-part article on asynchronous design presents a chronicle of past and recent commercial advances, as well as technical foundations, and highlights the enabling role of asynchronous design in two application areas: GALS systems and networks-on-chip.

Asynchronous Design—Part 1: Overview and Recent Advances

Asynchronous networks-on-chip (NoCs) are an appealing solution to tackle the synchronization challenge in modern multicore systems through the implementation of a GALS paradigm. However, they have found only limited applicability so far due to two main reasons: the lack of proper design tool flows as well as their significant area footprint over their synchronous counterparts. This paper proposes a largely unexplored design point for asynchronous NoCs, relying on transition-signaling bundled data, which contributes to break the above barriers. Compared to an existing lightweight synchronous switch architecture, xpipesLite, the post-layout asynchronous switch achieved a 71% reduction in area, up to 85% reduction in overall power consumption, and a 44% average reduction in energy-per-flit, while mastering the more stringent timing assumptions of this solution with a semi-automated synthesis flow.

A transition-signaling bundled data NoC switch architecture for cost-effective GALS multicore systems

The growing variability and complexity of advanced CMOS technologies makes the physical design of clocked logic in large Systems-on-Chip more and more challenging. Asynchronous logic has been studied for many years and become an attractive solution for a broad range of applications, from massively parallel multi-media systems to systems with ultra-low power & low-noise constraints, like cryptography, energy autonomous systems, and sensor-network nodes. The objective of this embedded tutorial is to give a comprehensive and recent overview of asynchronous logic. The tutorial will cover the basic principles and advantages of asynchronous logic, some insights on new research challenges, and will present the GALS scheme as an intermediate design style with recent results in asynchronous Network-on-Chip for future Many Core architectures. Regarding industrial acceptance, recent asynchronous logic applications within the microelectronics industry will be presented, with a main focus on the commercial CAD tools available today.

Advances in asynchronous logic: from principles to GALS & NoC, recent industry applications, and commercial CAD tools

A new asynchronous interconnection network is introduced for globally-asynchronous locally-synchronous (GALS) chip multiprocessors. The network eliminates the need for global clock distribution, and can interface multiple synchronous timing domains operating at unrelated clock rates. In particular, two new highly-concurrent asynchronous components are introduced which provide simple routing and arbitration/merge functions. Post-layout simulations in identical commercial 90 nm technology indicate that comparable recent synchronous router nodes have 5.6-10.7 more energy per packet and 2.8-6.4 greater area than the new asynchronous nodes. Under random traffic, the network provides significantly lower latency and identical throughput over the entire operating range of the 800 MHz network and through mid-range traffic rates for the 1.36 GHz network, but with degradation at higher traffic rates. Preliminary evaluations are also presented for a mixed-timing (GALS) network in a shared-memory parallel architecture, running both random traffic and parallel benchmark kernels, as well as directions for further improvement.

A Low-Overhead Asynchronous Interconnection Network for GALS Chip Multiprocessors

A new asynchronous interconnection network is introduced for globally-asynchronous locally-synchronous (GALS)chip multiprocessors. The network eliminates the need for global clock distribution, and can interface multiple synchronous timing domains operating at unrelated clock rates. In particular, two new highly-concurrent asynchronous components are introduced which provide simple routing and arbitration/merge functions. Post-layout simulations in identical commercial 90nm technology indicate that comparable recent synchronous router nodes have 5.6-10.7x more energy per packet and 2.8-6.4x greater area than the new asynchronous nodes. Under random traffic, the network provides significantly lower latency and competitive throughput over the entire operating range of the 800 MHz network and through mid-range traffic rates for the 1.36 GHz network, but with degradation at higher traffic rates. Preliminary evaluations are also presented for a mixed-timing (GALS) network in a shared-memory parallel architecture, running both random traffic and parallel benchmark kernels, as well as directions for further improvement.

/pdf/a-low-overhead-asynchronous-interconnection-network-for-gals-2bcs29yc2j.pdf

Michael N. Horak

Papers

A Low-Overhead Asynchronous Interconnection Network for GALS Chip Multiprocessors

A Low-Overhead Asynchronous Interconnection Network for GALS Chip Multiprocessors