145. SYSTEM IDENTIFICATION WITH EIGENVALUE CONSTRAINTS APPLIED TO THE THERMAL ANALYSIS OF AN INTEGRATED CIRCUIT

Department: Mechanical & Aerospace Engineering
Research Institute Affiliation: Center for Control Systems and Dynamics (CCSD)
Faculty Advisor(s): Raymond de Callafon

Primary Student
Name: Daniel N Miller
Email: d6miller@ucsd.edu
Phone: 858-356-7003
Grad Year: 2012

Abstract
A new method of constructing dynamic models from experimental data is presented that constrains the poles of the model to lie within arbitrary regions of the complex plane. The method incorporates the concept of linear-matrix-inequality regions into a realization-based identification procedure to form a semidefinite program, which is convex and easily solvable by modern optimization software. Using measured step-response data, the algorithm generates an accurate model of a linear, time-invariant system with no nonlinear optimization required. The new method is applied to a sample problem in which an estimated model of heat dissipation in an integrated circuit is required to mach prior knowledge of the system dynamics. The poles of the estimated discrete-time model are constrained to lie on the real axis of the complex plane between 0 and 1; the steady-state value of the response is constrained to match a known value; and the step response is constrained to be monotonically increasing. The resulting model is guaranteed to be transformable into a sequence of stable and strictly real time constants in continuous time. The discrete-time state space system is then converted into a time constant spectrum that may be used for defect detection.

Related Links:

  1. http://sites.google.com/site/dnmiller

« Back to Posters or Search Results