Hardware Implementation of Basic Arithmetics and Elementary Functions for Unum Computing

The universal number format (unum) is a recently proposed floating point format that solves rounding issues with interval arithmetic and provides variable exponent and mantissa widths. Only few approaches for unum hardware implementations have been published so far. In this paper, we propose architectures for comparisons and basic arithmetic operations, as well as for elementary functions using the CORDIC algorithm with unums. The designs are synthesized and evaluated for a 65 nm process and an Artix-7 FPGA.