A Small Group of Research Projects in Machine Design for Scientific Computation.

Abstract

Few computers to-day have arithmetic units which satisfy numerical analysts and the author expresses an urgent need for research to be undertaken in order to specify the needs and desires of people doing scientific computation and influence the design and architecture of future machines.

Formal definitions of concepts such as abstract representation, finite machine representation, and implementation representation are given and a number of possible research projects for the realizations of arithmetic units for various kinds of arithmetics are outlined.