Architectural partitioning of control memory for application specific programmable processors

Abstract

Because of programmability of Application Specific Programmable Processors (ASPPs), microcode-based control is effectively used to drive ASPP datapaths for different applications. In ASPPs, each application needs a separate microprogram resulting in large microcode memory. This paper proposes a distributed microcode memory model in which only distinct microcodes are stored in each separate memory module to save memory area. A hierarchical clustering approach is also proposed for the design of this distributed microcode memory. Experimental results indicate this approach is especially well suited for ASPP microcode memory design because of the existence of repetitive microcodes across multiple behaviors.