Multi-Layer Approximate Computing

It appears your Web browser is not configured to display PDF files. Download adobe Acrobat or click here to download the PDF file.

Click here to download the PDF file.


Brar, Jasleen Kaur




Approximate computing can efficiently trade accuracy for power savings, thus optimizing the power for applications that don't require exact solutions. This tradeoff has been studied in recent years across several implementation technologies, but mostly focused on approximations on a specific system layer. This thesis investigates how approximations interact at different layers and affect accuracy and power. We present a hierarchical model that attempts to analytically define levels of approximations, such that we can model the effects of approximations at multiple layers, as a function of a priori knowledge of the approximations' effects at each layer. We evaluate the validity of our model for an image processing application on a RISC-V simulator across two layers. The results indicate while type A model predicts accuracy pessimistically, type B model anticipates both power and accuracy with >98% accuracy. We discuss work towards improving this model and evaluating it across more comprehensive empirical results.


System Science
Engineering - Electronics and Electrical




Carleton University

Thesis Degree Name: 

Master of Applied Science: 

Thesis Degree Level: 


Thesis Degree Discipline: 

Engineering, Electrical and Computer

Parent Collection: 

Theses and Dissertations

Items in CURVE are protected by copyright, with all rights reserved, unless otherwise indicated. They are made available with permission from the author(s).