PROJECT TITLE :

A Novel Data Format for Approximate Arithmetic Computing - 2017

ABSTRACT:

Approximate computing has become one in all the foremost popular computing paradigms in the era of the Internet of things and big knowledge. It takes benefits of the error-tolerable feature of the many applications, like Machine Learning and image/Signal Processing, to reduce the resource needed to deliver bound level of computation quality. In this paper, we have a tendency to propose an approximate integer format (AIF) and its associated arithmetic operations for energy minimization with controllable computation accuracy. In AIF, operands are segmented at run time such that the computation is performed solely on half of operands by computing units (like adders and multipliers) of smaller bit-width. The proposed AIF can be used for any arithmetic operation and will be extended to fastened point numbers. It will conjointly be incorporated into higher level design like architectural and programming language to grant user the control of approximate computing. Experimental results show that our AIF based approximation computing approach will achieve high accuracy, incurs very little extra overhead, and save considerable energy.