Improvement of Standard and Non-Standard Floating-Point Operators

Abstract

This paper presents the design and analysis of a floating-point arithmetic accelerator in compliance with the IEEE standard single precision floatingpoint format. The accelerator can be used to extend a general-purpose processor such as Motorola MC6820, where floating-point execution units are unembedded by default. It implements standard and non-standard mathematic functions, addition/subtraction, multiplication, Product-of-Sum and Sumof- Product through a micro-instruction set supported by both single and multi-processors systems. The architecture of the unit is based on an instruction pipeline which can simultaneously fetch and execute an instruction within one clock cycle. The non-standard operations such as Product-of-Sum and Sum-of-Product are introduced to compute threeinput operands. The algorithm complexity and hardware critical delay are determined for each operator. The synthesis results of the accelerator on a Xilinx FPGA Virtex 5 xc5vlx110t-3ff-1136 and on Faraday 130-nm Silicon technology report that the design respectively achieves 200 MHz and 1 GHz.