An Optimal Multi-Functional Unit Dynamic Instruction Selection Logic at Submicron Technologies

As the technology scales, reduction in transistor size creates many opportunities for increased circuit capabilities in reduced chip area. In modern wide-issue processors, performance of the processor is directly impacted by the time delay complexity of the dynamic scheduling logic. In this paper, we analyze the scaling of time delay of instruction select logic at the submicron technologies, and also present novel designs that provide a single selection tree for two similar functional units. The designs are based on a tree structure using arbiter cells of two and four inputs which can handle one or two functional units. The effects of technology and design decisions are shown based on simulations using four submicron technologies. The delays in the select logic trees are shown to decrease by an average of 60% from 130 nm technology to 45 nm technology when servicing a single functional unit. The double grant arbiter cells are shown to build a tree that will serve multiple functional units simultaneously with 65% lesser delay as compared to multiple single-grant trees¹.