An Ultra-Efficient Imprecise Adder for Approximate Computing Based on CNTFET

An Ultra-Efficient Imprecise Adder for Approximate Computing Based on CNTFET

Abstract

Nowadays, mobile multimedia consumer electronic devices are wildly used. These battery-dependent and energy constraint devices require very low power consumption. However, as they are associated with human senses, they can perform complex arithmetic operations with a degree of impreciseness, which is not sensed by a human. In this paper, an ultra-efficient imprecise serial adder is proposed for approximate computing based on carbon nanotube FETs (CNTFETs). Serial addition is suitable for low-energy applications and leads to a considerable design efficiency. It is also well known that the CNTFET device can be a feasible replacement for the CMOS technology. In this design the resemblance of Carry' and Sum and a simple Cout' generator are utilized to provide an imprecise adder. The circuits are simulated using HSPICE with the Stanford CNTFET model at 32 nm feature size. The results indicate that the proposed method leads to considerable improvement in terms of performance and power consumption, while having fewer devices and small power-NED product.

References