Physics-based microwave device modeling and circuit optimization

Abstract

<p>This thesis addresses physics-based microwave device modeling and circuit optimization, including conventional and statistical device modeling, performance-driven and yield-driven circuit design. Approaches for physics-based device modeling are reviewed. Fundamental techniques of physics-based analytical MESFET modeling are presented. Device performance and parameter extraction with physics-based models (PBMs) are discussed. Nonlinear circuit analysis with PBMs integrated into the harmonic balance (HB) method is presented. A detailed formulation of the HB equations with MESFET PBMs is given. An efficient Newton method for solving the HB equations is discussed. Gradient-based optimization for circuit design is addressed. Physics-based circuit optimization integrates efficient adjoint sensitivity analysis approaches, the HB simulation method and PBMs. The physical (geometrical, material and process-related) parameters can be directly treated as design variables. Simultaneous device-circuit design is facilitated. The features of physics-based circuit optimization are demonstrated by two circuit design examples. Statistical modeling at different levels is discussed. Statistical parameter extraction and postprocessing are used to obtain statistical models to predict parameter statistics. The resulting statistical device models are verified by comparing the statistics of measurements with the corresponding statistics obtained by Monte Carlo simulation. Statistical modeling with equivalent circuit models (ECMs) and PBMs is demonstrated. Yield-driven circuit design is addressed based on a one-sided ℓ₁ optimization algorithm with a generalized ℓｐ function. Yield optimization of MMICs with PBMs for passive and active devices is discussed. Its features are demonstrated by a three stage X-band MMIC amplifier design. A comprehensive approach to predictable yield-driven circuit design exploiting a novel statistical model is presented. For the first time, the yield estimated by Monte Carlo simulation is shown to be consistent with the yield predicted directly from device measurement data. Simultaneous device-circuit yield optimization assisted by yield sensitivity analysis is also demonstrated.</p>