Configurable logic : a dynamically programmable cellular architecture and its VLSI implementation

Institute for Computing Systems ArchitectureAt present there are two main paradigms for computation : interpretation of a data stream representing a program by a processing unit (software) and an interconnection of active logic elements (hardware). While both systems can (given reasonable definitions) be shown to be equivalent in terms of which functions they can compute they have radically different properties : hardware can potentially provide a much higher performance implementation of a single simple algorithm whereas software can implement a wide variety of extremely complex algorithms. here we will consider a third paradigm for computation - configurable hardware in which the interconnection between the active logic elements is dependant on a control store. This paradigm can potentially offer many of the performance advantages of hardware white retaining much of the flexibility of software. This thesis examines the general properties of configurable systems and examples of previous designs in order to develop a new cellular-array architecture called Configurable Array Logic (CAL). The implementation of the architecture a related statically-programmed system, the Configurable Logic Array (CLA) in VLSI are discussed. The potential of the CAL system for implementation using Wafer Scale Integration is considered. The CAD system which would be required to allow algorithms expressed in normal programming languages to be implemented on the cellular architecture is discussed and the tools developed during the course of the project are covered. Four example designs using the system are presented : a digital stopwatch, a Data Encryption Standard (DES) encryptor, a unit to compute the 3-4 distance transform (which is used in image pattern matching) and a sketch of a system using configurable logic to implement cellular-automation models for fluid-flow simulations. Finally, methods of extending the Configurable Logic architecture to allow more complex computations to be performed and other directions for further research are discussed