Data-based Modeling of Analog-Mixed Signal System in Automotive Applications with Support Vector Machines

H. Mielenz, R. Dölling, and W. Rosenstiel (Germany)

Keywords

automotive simulation, support vector machines, auto mated behavioral modeling, simulation speed-up

Abstract

Functional verification by simulation is an important step during the development of present microelectronic solu tions for automotive applications. Its relevance is based on the capability to compare the behavior of a developed circuit with its specification. Since the transient simulation of application specific integrated circuits (ASICs) normally shows long runtimes, the behavior of time-critical compo nents is manually modeled in order to speed up simulation. The present article describes a data-based approach for semi-automated generation of behavioral models for ana log mixed-signal (A/MS) systems. The approach is based on support vector machines and a transformation dictionary for extraction of dynamic properties. The application of this method results in highly accurate pin-compatible be havioral models for A/MS systems with a significant re duction in simulation times. Additionally, the generated models can be easily integrated in description languages like VDHL-AMS, Verilog-AMS, Simulink and MAST. An other benefit of the proposed method consists in its flexi bility to model systems of different physical domains. The emphasized properties will be illustrated by the modeling of two examples belonging to analog-digital and electrome chanic systems respectively.

Important Links:



Go Back


IASTED
Rotating Call For Paper Image