Design methodology for digital signal processing

Abstract

Improvements in semiconductor integration density and the resulting problem of having to manage designs of increasing complexity is an old one, but still current. The new challenge lies in a new level of architecture heterogeneity, e.g. mixing hard-wired digital circuits with software programmed signal processors on one die. Hence, we are moving by one level of abstraction from semi-custom standard-cells to semi-custom 'block cells'. This results in a new dimension in the gap between algorithm/system design and architecture/circuit design, not addressed by any tools sufficiently yet today. This paper presents a method of analyzing the problem by orthogonalizing algorithms into data transfer and data manipulation, and carrying this over to the control and I/O design as well. This approach might be a promising basis for flexibly mapping the algorithms onto future 'block cell' designs, and furthermore for designing new system simulation tools which allow for tools to be integrated for a flexible mapping of algorithms onto various different hardware architecture domains.