R.M. Biernacki

Electromagnetic optimization exploiting aggressive space mapping

We propose a significantly improved space mapping (SM) strategy for electromagnetic (EM) optimization. Instead of waiting for upfront EM analyses at several base points, our new approach aggressively exploits every available EM analysis, producing dramatic results right from the first step. We establish a relationship between the novel SM optimization and the quasi-Newton iteration for solving a system of nonlinear equations. Approximations to the matrix of first-order derivatives are updated by the classic Broyden formula. A high-temperature superconducting microstrip filter design solution emerges after only six EM simulations with sparse frequency sweeps. Furthermore, less CPU effort is required to optimize the filter than is required by one single detailed frequency sweep. We also extend the SM concept to the parameter extraction phase, overcoming severely misaligned responses induced by inadequate empirical models. This novel concept should have a significant impact on parameter extraction of devices.

A trust region aggressive space mapping algorithm for EM optimization

A new robust algorithm for EM optimization of microwave circuits is presented. The algorithm integrates a trust region methodology with aggressive space mapping (ASM). A new automated multipoint parameter extraction process is implemented. EM optimization of a double-folded stub filter and of an HTS filter illustrate our new results.

Microstrip filter design using direct EM field simulation

For the first time, we present minimax filter design with electromagnetic (EM) simulations driven directly by a gradient-based optimizer. Challenges of efficiency, discretization of geometrical dimensions, and continuity of optimization variables are overcome by a three-stage attack: 1) efficient on-line response interpolation with respect to geometrical dimensions of microstrip structures simulated with fixed grid sizes; 2) smooth and accurate gradient evaluation for use in conjunction with the proposed interpolation; and 3) storing the results of expensive EM simulations in a dynamically updated database. Simulation of a lowpass microstrip filter illustrates the conventional use of EM simulation for design validation. Design optimization of a double folded stub bandstop filter and of a millimeter-wave 26-40 GHz interdigital capacitor bandpass microstrip filter illustrates the new technique. >

Multiple-fault location of analog circuits

This paper deals with multiple-fault detection for linear analog circuits. The method proposed is based on measurements of voltage using current excitations and has been developed for the location of a number of faults. It utilizes certain algebraic invariants of faulty elements. Computationally, it depends on checking the consistency or inconsistency of suitable sets of linear equations. The equations themselves are formulated via adjoint circuit simulations.

A unified theory for frequency-domain simulation and sensitivity analysis of linear and nonlinear circuits

The harmonic balance technique from nonlinear simulation is extended to nonlinear adjoint sensitivity analysis. This provides an efficient tool for the otherwise expensive but essential gradient calculations in design optimization. The hierarchical approach widely used for circuit simulation, is generalized to sensitivity analysis and to computing responses in any subnetwork at any level of the hierarchy. Important aspects of frequency-domain circuit computer-aided design (CAD) such as simulation and sensitivity analysis, linear and nonlinear circuits, hierarchical and nonhierarchical approaches, voltage and current excitations, or open- and short-circuit terminations are unified in this general framework. The theory provides a basis for the next generation of microwave CAD software. It takes advantage of mature techniques such as syntax-oriented hierarchical analysis, optimization, and yield-driven design to handle nonlinear as well as linear circuits. The sensitivity analysis approach has been verified by a MESFET mixer example, exhibiting a 90% saving of CPU time over the prevailing perturbation method. >

Fully automated space mapping optimization of 3D structures

We present new results of fully automating the aggressive Space Mapping (SM) strategy for electromagnetic optimization. The generic SM update loop and the model-specific parameter extraction loop are automated using a two-level Datapipe architecture. We apply the automated SM strategy to the optimization of waveguide transformers. We introduce a multi-point parameter extraction procedure for sharpening the solution uniqueness and improving the SM convergence. We present, for the first time, automated electromagnetic optimization utilizing the commercial 3D structure simulator HFSS.

Integrated physics-oriented statistical modeling, simulation, and optimization (MESFETs)

Physics-based modeling of MESFETs is addressed from the point of view of efficient simulation, accurate behavior prediction and robust parameter extraction. A novel integration of a large-signal physics-based model into the harmonic balance equations for simulation of nonlinear circuits, involving an efficient Newton update, is presented and exploited in a gradient-based FAST (feasible adjoint sensitivity technique) circuit optimization technique. For yield-driven MMIC design a relevant physics-based statistical modeling methodology is presented. Quadratic approximation of responses and gradients suitable for yield optimization is discussed. The authors verify their theoretical contributions and exemplify their computational results using built-in and user-programmable modeling capabilities of the CAE systems OSA90/hope and HarPE. Results of device modeling using a field-theoretic nonlinear device simulator are reported. >

FAST gradient based yield optimization of nonlinear circuits

Yield optimization of nonlinear microwave circuits operating in the steady state under large-signal periodic excitations is studied. Two novel high-speed methods of gradient calculation, the integrated gradient approximation technique (IGAT) and the feasible adjoint sensitivity technique (FAST) are introduced. IGAT utilizes the Broyden formula with special iterations of Powell to update the approximate gradients. FAST combines the efficiency and accuracy of the adjoint sensitivity technique with the simplicity of the perturbation technique. IGAT and FAST are compared with the simple perturbation approximate sensitivity technique (PAST) on the one extreme and the theoretical exact adjoint sensitivity technique (EAST) on the other. A FET frequency doubler example treats statistics of both linear elements and nonlinear device parameters. This design has six optimizable variables, including input power and bias conditions, and 34 statistical parameters. Using either IGAT or FAST, yield is driven from 40% to 70%. FAST exhibits superior efficiency. >

Huber optimization of circuits: a robust approach

The authors introduce an approach to robust circuit optimization using Huber functions, both two-sided and one-sided. They compare Huber optimization with l/sub 1/, l/sub 2/, and minimax methods in the presence of faults, large and small measurement errors, bad starting points, and statistical uncertainties. They demonstrate FET statistical modeling, multiplexer optimization, analog fault location, and data fitting. They extend the Huber concept by introducing a one-sided Huber function for large-scale optimization. For large-scale problems, the designer often attempts, by intuition, a preliminary optimization by selecting a small number of dominant variables. It is demonstrated, through multiplexer optimization, that the one-sided Huber function can be more effective and efficient than minimax in overcoming a bad starting point. >

Yield-driven electromagnetic optimization via multilevel multidimensional models

The authors present the foundation of a sophisticated hierarchical multidimensional response surface modeling system for efficient yield-driven design. The scheme dynamically integrates models and database updating in real optimization time. The method facilitates a seamless, smart, optimization-ready interface. It has been specially designed to handle circuits containing complex subcircuits or components whose simulation requires significant computational effort. This approach makes it possible, for the first time, to perform direct gradient-based yield optimization of circuits with components or subcircuits simulated by an electromagnetic simulator. The efficiency and accuracy of the technique are demonstrated by yield optimization of a three-stage microstrip transformer and a small-signal microwave amplifier. The authors also perform yield sensitivity analysis for the three-stage microstrip transformer. >