Protocol Selection, Implementation, and Analysis for Asynchronous Circuits

Abstract

This dissertation presents new methods for handshaking expansion of asynchronous circuits. Handshaking expansion includes protocol selection, reshuffling, and state variable insertion. The starting point is a channel-level specification of a design. The goal is a signal-level description of the given design that is correct, synthesizable, and efficient. This dissertation studies the impact of protocol selection and implementation on deadlock avoidance, complete state coding (CSC), CPU time required to compile a given example, and the quality of the circuit. This dissertation treats reshuffling and state-variable insertion as special cases of concurrency reduction. Prior work in the field has also taken this approach. However, this dissertation extends this approach and applies it to specifications that contain quantitative timing assumptions. The concurrency reduction algorithms have been implemented within a computer aided design (CAD) tool. Starting from a signal-level specification that contains the constraints of the desired protocol, these algorithms search the concurrency reduction design space, guided by an estimate of the performance of the final circuit. The CAD tool that this dissertation presents also contains a front end that, given a channel-level specification, produces the starting point for concurrency-reduction. This front end currently handles only pure synchronization channels, using one protocol. Finding all possible ways to reduce concurrency of a specification is a fundamentally exponential problem. However, this dissertation presents techniques to dramatically prune the search space. This dissertation demonstrates that these techniques are capable of reducing the search space by several orders of magnitude compared to the theoretical upper bound – and by one order of magnitude beyond existing techniques – without significantly impacting the quality of the solutions.

Type
Eric Peskin
Eric Peskin
Yale University, Director of High-Performance Computing

Related