An adjoint-based approach to computing time-domain sensitivity of multi-port systems described by reduced-order models

The present thrust in the electronics industry towards integrating multiple functions on a single chip while operating at very high frequencies has highlighted the need for efficient Electronic Design Automation (EDA) tools to shorten the design cycle and capture market windows. One of the fundamental tasks performed by modern EDA tools is the sensitivity analysis, which is concerned with computing the derivatives of various circuit performance metrics w.r.t. the circuit designable parameters.;However, the increasing complexity of modern circuit design has made simulation and sensitivity analysis a computationally cumbersome task. Furthermore, the presence of non-linearity in virtually all designs in the form of buffers, driving circuitry, and analog blocks necessitates that these tasks be carried out in the time domain.;This thesis describes a new algorithm to compute the time-domain sensitivity using the concept of Model-Order Reduction (MOR). The proposed algorithm enables an interconnected set of large linear systems, that have been abstracted by reduced-order models for the purpose of simulation, to have their sensitivity evaluated in the time domain. Therefore, the key advantage of the proposed work stems from the fact that it brings the computational efficiency of MOR in areas of simulation to bear on the problem of sensitivity analysis.;The large systems are assumed to be interconnected within a larger network using lumped linear or nonlinear components, and the desired sensitivity is to be evaluated w.r.t. their internal designable parameters. The proposed algorithm can thus be applied in computing the sensitivity of circuits constructed from microwave devices, such as multi-conductor transmission lines, which when discretized produce large linear systems.;The proposed technique also employs the concept of adjoint networks in the sensitivity computation. In addition it can also be used to obtain the sensitivity information in the frequency domain.