Hui Li, Senior Software Engineer, Technology Infrastructure Group, National Semiconductor
Rajesh Berigei, Senior Manager, National Semiconductor
Sury Maturi, Director, National Semiconductor
Li Wang, Professor, Electrical Engineering Department, University of California, Santa Barbara

This paper presents a methodology for behavioral modeling of analog chips. Our approach uses kernel-based learning called support vector machine to model how analog circuits respond to different test-bench stimuli. We will demonstrate how we have integrated the methodology to our company internal modeling tool and how the models can be used in an outside marketing tool for outside customers developing prototypes quickly in their specific system. In the following sections, we present in detail a sampling methodology, learning advantage for modeling, and model validation. We have studied a step-down DC–DC converter to demonstrate the feasibility of using the learning approach for analog behavioral modeling with encouraging results.