Date of Award

5-2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Electrical Engineering

Major Professor

Benjamin J. Blalock

Committee Members

Vasilios Alexiades, Jeremy H. Holleman III, Lynne E. Parker

Abstract

A configurable, low power analog implementation of a multilayer perceptron (MLP) is presented in this work. It features a highly programmable system that allows the user to create a MLP neural network design of their choosing. In addition to the configurability, this neural network provides the ability of low power operation via analog circuitry in its neurons. The main MLP system is made up of 12 neurons that can be configurable to any number of layers and neurons per layer until all available resources are utilized. The MLP network is fabricated in a standard 0.13 μm CMOS process occupying approximately 1 mm2 of on-chip area. The MLP system is analyzed at several different configurations with all achieving a greater than 1 Tera-operations per second per Watt figure of merit. This work offers a high speed, low power, and scalable alternative to digital configurable neural networks.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS