Analog behavioral modeling flow using statistical learning method

This paper presents a novel behavioral-level analog circuit performance modeling methodology using kernel based support vector machine (SVM). Behavioral modeling for analog circuits is in high demand for architectural exploration and system prototyping of increasingly complex electronic systems. In this paper, we investigate the effectiveness of applying SVM to model analog circuits. Based on the different perspectives of model accuracy, we develop a model performance optimizer which automatically tunes the learning engine to achieve either the lowest worst-case error or the average error percentage. The modeling performance is compared against SPICE simulation result to validate this approach. We also present its advantages in automation and simulation speed.