Evolutionary Design of Complex Approximate Combinational Circuits

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26230%2F16%3APU121566" target="_blank" >RIV/00216305:26230/16:PU121566 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/s10710-015-9257-1" target="_blank" >http://dx.doi.org/10.1007/s10710-015-9257-1</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/s10710-015-9257-1" target="_blank" >10.1007/s10710-015-9257-1</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Evolutionary Design of Complex Approximate Combinational Circuits

Popis výsledku v původním jazyce

Functional approximation is one of the methods allowing designers to approximate circuits at the level of logic behavior. By introducing a suitable functional approximation, power consumption, area or delay of a circuit can be reduced if some errors are acceptable in a particular application. As the error quantification is usually based on an arithmetic error metric in existing approximation methods, these methods are primarily suitable for the approximation of arithmetic and signal processing circuits. This paper deals with the approximation of general logic (such as pattern matching circuits and complex encoders) in which no additional information is usually available to establish a suitable error metric and hence the error of approximation is expressed in terms of Hamming distance between the output values produced by a candidate approximate circuit and the accurate circuit. We propose a circuit approximation method based on Cartesian genetic programming in which gate-level circuits are internally represented using directed acyclic graphs. In order to eliminate the well-known scalability problems of evolutionary circuit design, the error of approximation is determined by binary decision diagrams. The method is analyzed in terms of computational time and quality of approximation. It is able to deliver detailed Pareto fronts showing various compromises between the area, delay and error. Results are presented for 16 circuits (with 27-50 inputs) that are too complex to be approximated by means of existing evolutionary circuit design methods.

Název v anglickém jazyce

Evolutionary Design of Complex Approximate Combinational Circuits

Popis výsledku anglicky

Functional approximation is one of the methods allowing designers to approximate circuits at the level of logic behavior. By introducing a suitable functional approximation, power consumption, area or delay of a circuit can be reduced if some errors are acceptable in a particular application. As the error quantification is usually based on an arithmetic error metric in existing approximation methods, these methods are primarily suitable for the approximation of arithmetic and signal processing circuits. This paper deals with the approximation of general logic (such as pattern matching circuits and complex encoders) in which no additional information is usually available to establish a suitable error metric and hence the error of approximation is expressed in terms of Hamming distance between the output values produced by a candidate approximate circuit and the accurate circuit. We propose a circuit approximation method based on Cartesian genetic programming in which gate-level circuits are internally represented using directed acyclic graphs. In order to eliminate the well-known scalability problems of evolutionary circuit design, the error of approximation is determined by binary decision diagrams. The method is analyzed in terms of computational time and quality of approximation. It is able to deliver detailed Pareto fronts showing various compromises between the area, delay and error. Results are presented for 16 circuits (with 27-50 inputs) that are too complex to be approximated by means of existing evolutionary circuit design methods.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

20206 - Computer hardware and architecture

Projekt

<a href="/cs/project/GA14-04197S" target="_blank" >GA14-04197S: Pokročilé metody evolučního návrhu složitých číslicových obvodů</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2016

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

Genetic Programming and Evolvable Machines

ISSN

1389-2576

e-ISSN

1573-7632

Svazek periodika

17

Číslo periodika v rámci svazku

2

Stát vydavatele periodika

US - Spojené státy americké

Počet stran výsledku

24

Strana od-do

169-192

Kód UT WoS článku

000376876700004

EID výsledku v databázi Scopus

2-s2.0-84949685438