Over the last few years, scaling in deep submicron technologies has shifted the paradigm from device-dominated to interconnect-dominated design methodology. Consequently, there is an increasing interest towards the miniaturization of the guiding medium in nanoscale integrated circuits by exploring plasmon-based waveguides to alleviate the scaling issues associated with today's copper interconnect. In this thesis, we seek short and long-term solutions of on-chip interconnect by developing accurate compact models of on-chip interconnects and impedance characterization of nanophotonic structures. The developed system models are compact and accurate over the operating frequency range and the adopted approach have provided many critical insights and produced many important results.

This thesis first presents a new modeling strategy that represents the nanostructure by its equivalent impedance. By applying either quasistatic approximation or separately solving for voltage and current for dominant mode, we reduce the field problem to a circuit problem. The impedance expressed in terms of circuit components is dependent on the material constant as well as the operating frequency. The modeling methodology is successfully applied to nanoparticles and oscillating nanosphere. The proposed model characterizes plasmon resonance in these nanostructures, thereby providing basic building block to develop spice models of complex plasmon-based waveguide for sub-wavelength propagation.

We also presented several techniques to develop compact models of on-chip interconnects and passive components for accurate estimation of power, noise and delay of high speed integrated circuits. The automated method generates reduced order models that are accurate across either a narrow or a wide-range of frequencies. The proposed methods are based on Krylov subspace method with interpolation points dynamically selected using either spline based algorithm or discrete wavelet transform. Narrow and wideband frequency projection are also achieved using spectral zeros by applying either a frequency selective scheme or an adaptive wavelet transform to dynamically select spectral zeros. To demonstrate the efficacy of the approach, we simulated complex circuit models of spiral inductors, RLC networks and interconnect busses. The results indicate greater accuracy than techniques that apply other Krylov subspace methods or Singular Value Decomposition (SVD) based methods for model order reduction of on-chip interconnects.