Methodology for the Design of Comtinuous-Dataflow Synchronous Systems

摘要

Increasingly it is the propagation velocity of information that limits processing speed rather than the inherent delays in processing elements. One consequence is that optimal processing throughput is not possible when memory elements are employed to synchronize information parcels. In the paper the design of systems lacking such memory elements is discussed. Such systems are characterized by the need for precise control over data propagation delays. Synchronization in the context of the systems is characterized and related to existing graph theory. An algorithm is presented that allows efficient distribution of lumped delay designs while preserving synchronization. Applications in optical computing are discussed. The techniques described are also applicable to existing data processing architectures, particularly those based on pipelining.