Date of Award

Fall 2014

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Sotirios Ziavras

Second Advisor

Durgamadhab Misra

Third Advisor

Roberto Rojas-Cessa

Abstract

Processors for high-performance computing applications are generally designed with a focus on high clock rates, parallelism of operations and high communication bandwidth, often at the expense of large power consumption. However, the emphasis of many embedded systems and untethered devices is on minimal hardware requirements and reduced power consumption. With the incessant growth of computational needs for embedded applications, which contradict chip power and area needs, the burden is put on the hardware designers to come up with designs that optimize power and area requirements.

This thesis investigates the efficient design of an embedded system for morphological image processing applications on Xilinx FPGAs (Field Programmable Gate Array) by optimizing both area and power usage while delivering high performance. The design leverages a unique capability of FPGAs called dynamic partial reconfiguration (DPR) which allows changing the hardware configuration of silicon pieces at runtime. DPR allows regions of the FPGA to be reprogrammed with new functionality while applications are still running in the remainder of the device.

The main aim of this thesis is to design an embedded system for morphological image processing by accounting for real time and area constraints as compared to a statically configured FPGA. IP (Intellectual Property) cores are synthesized for both static and dynamic time. DPR enables instantiation of more hardware logic over a period of time on an existing device by time-multiplexing the hardware realization of functions. A comparison of power consumption is presented for the statically and dynamically reconfigured designs. Finally, a performance comparison is included for the implementation of the respective algorithms on a hardwired ARM processor as well as on another general-purpose processor. The results prove the viability of DPR for morphological image processing applications.

Share

COinS