Design considerations for a general purpose systolic processing element using bit serial techniques

The demands of modern high performance numerical applications are such that they must be met in principle by special-purpose computational architectures. Processing architectures fall into two distinct categories, serial-data and parallel-data. For fixed-function, real-time computational requirements, serial data architectures exhibit many advantages over equivalent bit-parallel architectures, especially when size and area are of primary concern. Advantages of serial data hardware include: performance and efficiency, functional parallelism, physical partitioning and testability. Single wire communication, as opposed to parallel busses, ease interconnection problems and decrease layout area. With the new technology available today, serial techniques must be seriously reconsidered.<
>