Studies on Hybrid Static Segmentation Based Imprecise Multiplier Designs for Image Processing Applications
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
newline ABSTRACT
newlineApproximate computing (AC) is a transpiring trend in several digital designs,
newlinealleviating the requisites of precise computation, to obtain valuable performance
newlinerefinement in terms of power, speed and area. This perspective is befitting more and
newlinemore important for embedded and mobile systems, characterized by severe constraints
newlineof energy and speed. Moreover, approximate computing can be bounteously
newlineapplicable in multiple error-resilient applications like multimedia processing, machine
newlinelearning, data mining and recognition. Further with approximate computing, greater
newlinespeed applications of multimedia have also provided a way for the completely new
newlinearea of zenith-speed error-tolerant circuits. These implementations dispatch high
newlineperformance at the cost of shrink inexactness and also such enactments reduce the
newlineentanglement of the system architecture, delay and power consumption.
newlineIn general, multipliers are considered to be one of the primary sub systems for
newlinemicroprocessors, digital signal processors, and embedded systems with requisitions
newlinefluctuating from simple filtering to convolutional neural networks (CNN). Also, they
newlineplay a crucial role in DSP applications, namely in arithmetic operations. But it is
newlineunfortunate that, multipliers are characterized by tricky design logics and holds one of
newlinethe prominent energy-consuming digital units. Consequently, inexact multiplier
newlineconfigurations have become a pivotal investigation concern in present generation.
newlineNeoteric developments in the processor models are focused on the multiplier
newlinearchitectures which consume minimal power at circuit level. But such architectures
newlineand algorithms mostly perform their elemental operation as matrix multiplications,
newlineapart from the multiplier being typically an energy-consuming integrant. In any
newlinedesign, the main criteria of interest are to obtain high speed, low delay, usage of
newlineminimal area and low power consumption. The duo significant, as yet often clashing
newlinedesign parameters are consumption of power and speed. By ca