Date of Award

2011-01-01

Degree Name

Master of Science

Department

Electrical Engineering

Advisor(s)

Eric MacDonald

Abstract

As Moore's Law continues to hold true and transistor density becomes exponentially larger the need to rely on computer aided design (CAD) simulators has become more relevant and necessary. As we dive into the age of electronic reliance there presents a need to constantly improve methods to reduce time-to-solution for developing technologies. The reliance on CAD simulators for electronic development leaves engineers with an ability to improve the design process by improving these simulators or improving the methods in which these simulators are being utilized. With the advancement of quality fabrication processes and the ability to operate at sub-threshold levels the need for high fidelity simulations is ever present.

The digital circuit design process begins with a behavioral description of the intended circuit and is followed by a CAD simulator testing of the desired behavior. These initial circuit simulations are evaluated using a method of testing that does not account for all of the physics involved in transistor behavior but only take into account the average delays and behavior of generally manufactured transistors for a given manufacturing process. For the use of sub-threshold and related technologies these behavioral simulations cannot capture the needed physics level performance of each transistor to verify the functionality and operation of the design. For these digital circuit designs we must rely on higher fidelity simulators called Simulation Program with Integrated Circuit Emphasis (SPICE).

SPICE simulators are the paramount programs for digital circuit design testing and verification. These high fidelity simulators mimic the behavior of actual circuits to a degree of accuracy that rival prototype testing. The increase in accuracy also increases the simulation time, this increase in simulation time reduces the time-to-solution for the design. The present work generates a proposal for reducing time-to-solution while continuing to use the high fidelity circuit simulators by utilizing high performance computing machines. The increase in parallelization capabilities of modern SPICE simulator software and the ability to use multi-core and multi-processor machines to run multiple simulations concurrently has the potential to further optimize the digital circuit design process.

The overall goal of this process is a set of Integrated Circuit (IC) design tools and methodologies that are scalable and extensible over all high performance computing platforms. This project is focused on design challenges that are outside the mainstream of commercial and open source tools (sub-threshold and extreme low-power designs). This objective will have the most significance on large scale designs and those that the initial modeling and design quality will greatly impact the quality of the final design.

The end objective of Markov Chain Monte Carlo (MCMC) parallel simulation of the Marine Corps TRSS magnetic field detector detection algorithm was preceded by testing of applicable sub-circuits within the detection algorithm digital circuit. Initial testing was conducted with a 64 bit Counter focusing on the energy per operation status over a design space that varied the Vdd of the circuit. Analysis showed that through 5,080 concurrent simulations in the Monte Carlo sampling, 4,469 produced correct operation under an acceptance rate of 75%. Of the simulations that produced correct results, an analysis of energy per operation as a function of Vdd was conducted and it was shown that two optimal VDD values produced the lowest energy per operation while maintaining functionality. This analysis was completed in just under 7 hours utilizing high performance computing machines, had these simulations been conducted on common dual core workstations the run time would exceed 20,000 hours. The 99% reduction in simulation time of this simple circuit illustrates significant advantages of using high performance computing machines in the analysis of more complex systems especially those focused on low power, sub-threshold designs.

Language

en

Provenance

Received from ProQuest

File Size

93 pages

File Format

application/pdf

Rights Holder

Matthew C. Markulik

Download

Share

COinS