Analog circuit design space description based on ordered clustering of feature uniqueness and similarity

This paper presents a symbolic technique to create ordered feature clustering schemes that express the main similarities and differences between analog circuits. Four separation scores, based on entropy, item characteristics, category characteristics, and Bayesian classifiers, were studied to produce clustering schemes that offer insight about the uniqueness and importance of specific design features in setting AC performance as well as the limiting factors of the designs. The experiments consider a set of 50 state-of-the-art amplifier circuits. The paper offers a detailed discussion on using the insight obtained from circuit feature clustering for topology synthesis and refinement.