Useful-state Encoding: Network Control with Minimal Redundancy

Kris Heyrman and Peter Veelaert

Ghent University College

Abstract

This paper presents useful-state encoding (USE), a novel method to design control circuitry for networks that feature static topological scheduling. These include low-power on-chip networks, for which it was originally developed. The goal of USE is to obtain a small control circuit, consuming little area and power; USE establishes a minimal state-space description. Since the demands on the communication architecture are specified optimally, this description is much smaller than the full combinatorial state-space of the network.

We propose analysis and design methods for the control circuit, deduce rules of thumb from use cases, and reach a conclusion on the scalability of our design method with large networks. We find that control circuits remain feasible as long as the number of network nodes remains reasonably low, independently of other measures of topological complexity. Feasibility is determined mainly by issues of algorithmic and methodological complexity, not by physical on-chip limitations like area, latency or power.