Fast, area-efficient 32-bit LNS for computer arithmetic operations

Abstract

The logarithmic number system has been proposed as an alternative to floating-point. Multiplication, division and square-root operations are accomplished with fixedpoint arithmetic, but addition and subtraction are considerably more challenging. Recent work has demonstrated that these operations too can be done with similar speed and accuracy to their floating-point equivalents, but the necessary circuitry is complex. In particular, it is dominated by the need for large lookup tables for the storage of a non-linear function. This thesis describes the architectures required to implement a newly design approach for producing fast and area-efficient 32-bit LNS arithmetic unit. The designs are structured based on two different algorithms. At first, a new cotransformation procedure is introduced in the singularity region whilst performing subtractions in which the technique capable to generate less total storage than the cotransformation method in the previous LNS architecture. Secondly, improvement to an existing interpolation process is proposed, that also reduce the total tables to an extent that allows their easy synthesis in logic. Consequently, the total delays in the system can be significantly reduced. According to the comparison analysis with previous best LNS design and floating-point units, it is shown that the new LNS architecture capable to offer significantly better in speed while sustaining its accuracy within floating-point limit. In addition, its implementation is more economical than previous best LNS system and almost equivalent with existing floating-point arithmetic unit.