Nonlinear Model Order Reduction for Behavioral Models of Emerging Technologies

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

ABSTRACT 0541150 Steven Levitan U of Pittsburgh Physical phenomena in nature are inherently nonlinear. However, most techniques used by engineers to model and analyze these complex interactions are based on linear approximations. Until recently, these approximations were acceptable for most engineering applications. However, we are now forced to rethink this approach due the fact that emerging micro- and nano- technologies have lead to systems that are complex, with many degrees of freedom, and highly nonlinear behavior. For example, consider the design of lab-on-a-chip type systems, which use micromechanics and optoelectronics to manipulate and analyze the behavior of complex fluids. These systems could revolutionize the way that bio-chemical synthesis and analysis are performed. However, they are inherently difficult to design because they involve interacting electronic, optoelectronic, fluidic, thermal and mechanical sub-systems. On the one hand, such multi-technology systems can only be accurately modeled with formulations that have many degrees of freedom and also capture the nonlinear characteristics of the underlying physics; and these models are necessarily computationally expensive. On the other hand, to effectively design such systems, it is necessary to simulate and analyze their behavior over a broad range of stimuli in a realistic operating environment. This requires models that can be simulated efficiently in order to explore the range of behaviors that they exhibit, in an engineering product design flow. Consequently, it is essential to have a methodology to reduce the complexity of nonlinear systems of high dimensionality, without recourse to linear approximation, since only such a solution will give an accurate description for ever increasingly complex micro- and nano- multi-technology systems. To address these needs, this research will develop a methodology for extraction of nonlinear behavioral models. The results of this work will lead to a general methodology for nonlinear model order reduction. Such methodologies are essential for reducing design costs and increasing both quality and reliability of multi-technology micro systems. Having accurate compact models that can be efficiently simulated will enable the robust design of complex heterogeneous systems that span multiple energy domains. This methodology will be broadly applicable not only to electronic systems design, but also to emerging technologies at the confluence of engineering and physical sciences, such as nanotechnology based sensors, smart materials and systems biology.