Research On Performance Improvement Of 3-D QMM-Based Parasitic Capacitance Extraction

The current feature size of integrated circuits (IC) has scaled down intodeep submicron level, which makes electromagnetic effect betweeninterconnects dominating the circuit performance and even determining thecircuit validity. As a result, accurate calculation of interconnect parasiticparameters has become one of the critical tasks in IC design flow.On the basis of QMM-accelerated direct boundary element method, thisthesis improves the speed and stability for 3-D parasitic capacitanceextraction computation. The following two aspects show the main work.1 It improves the speed of full-coupled capacitance matrix computation.This thesis applies refined blocked Gauss method into solving equationsystem with multiple right-hand sides (RHS) and thus accelerates theQMM-based local method for capacitance matrix. Afterwards, we compareblocked Gauss in multiple RHS problems with iterative solver, GMRES, boththeoretically and experimentally. The refined blocked Gauss works well infull-coupled capacitance matrix extraction and improves its efficiency.2 It improves the computational stability. This thesis designscomplicated data structures and algorithms so that the hierarchical h-adaptivecomputation has been implemented for the first time on basis of QMM. At thesame time, the refinement of elements at arbitrary level is permitted. It alsominimizes the primary elements to make element division more reasonable.Computational results show its accuracy, good stability and high efficiency.The algorithms above have been implemented in QBEM, a program forQMM-accelerated direct BEM capacitance extraction. Experiments on someactual structures verify good improvement on both speed and stability.