Approximate arithmetic units under voltage under-scaling

Increasingly power hungry processors and the identification of error tolerant applications have made approximate computing techniques one of the researching scopes of the present. The main objective of this work is to study and assess the benefits of certain approximate arithmetic units for their in...

Descripción completa

Detalles Bibliográficos
Autor: Etxezarreta Martinez, Imanol
Tipo de recurso: tesis de maestría
Fecha de publicación:2021
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/344137
Acceso en línea:https://hdl.handle.net/2117/344137
Access Level:acceso abierto
Palabra clave:Electronics
Arithmetic
Approximate
arithmetic
adders
multipliers
synthesis
digital design
integrated circuits
electronic engineering
Electrònica
Aritmètica
Àrees temàtiques de la UPC::Enginyeria electrònica
Descripción
Sumario:Increasingly power hungry processors and the identification of error tolerant applications have made approximate computing techniques one of the researching scopes of the present. The main objective of this work is to study and assess the benefits of certain approximate arithmetic units for their inclusion in the DRAC project?s approximate accelerator. In order to accomplish this objective, different approximate and accurate adders and multipliers are implemented in Verilog, synthesized and simulated using Cadence tools and finally compar- ed between them to evaluate whether they offer any benefits or not. Simulations are performed for both Overclocking and Voltage Underscaling conditions. The results indicated that some of the approximate adders show very competitive precision values at overclocking simulations for 400 ps, with slightly better power results, and other adders also presented very encouraging results for voltage overscaling simulations, offering higher precision values than the accurate adders for 0.4V. In contrary, the multipliers exhibited worse results than the accurate ones in every aspects for the realistic case dataset provided by the SW team.