TABLE OF CONTENTS Let us now turn our attention to the main topic of this chapter:
How to decide on the necessary hardware resources for solving a given computational problem and how to best organize them.
Their conceptual differences notwithstanding, many techniques for obtaining high performance at low cost are the same for general- and special-purpose architectures. As a consequence, much of the material presented in this chapter applies to both of them. Yet, the emphasis is on dedicated architectures as the a priori knowledge of a computational problems offers room for a number of ideas that do not apply to instruction-set processor architectures. [14]
Most data and signal processing algorithms would lead to grossly inefficient or even infeasible solutions if they were implemented in hardware as they are. Adapting processing algorithms to the technical and economic conditions of large-scale integration is one of the intellectual challenges in VLSI design.
Basically, there is room for remodelling in two distinct domains, namely in the algorithmic domain and in the architectural domain.
In the algorithmic domain, the focus is on minimizing the number of computational operations weighted by the estimated costs of such operations. A given processing algorithm thus gets replaced by a different one better suited to hardware realization in VLSI. Data structures and number representation schemes are also subject to optimizations such as subsampling and/or changing from floating-point to fixed-point arithmetics.
