The Propagated Instruction Processor

As the dimensions of electronic devices become smaller, they enter the realm of nano-electronics, where the device dimensions lie between the inter-atomic spacing (lnm) and the electron wavelength (lOnm). This offers greatly-enhanced packing density and speed of operation, but introduces the difficulty of maintaining signal quality over large distances. The paper introduces an architectural solution to this problem which is appropriate for the class of highly data-parallel computers [1,2,3]. In the propagated instruction architecture proposed here, instructions sweep across the array (and hence across the data set) in bands, one instruction following another closely. The paper examines the architectural implications of this idea, and assesses the benefits in the application area of image processing.