Variation-aware behavioral models of analog circuits using support vector machines with interval parameters

Machine learning algorithms have recently been used successfully to generate behavioral models of analog circuits. We take this approach one step further and include parameter variations directly into models using specialized interval arithmetics. We developed a new support vector machine algorithm which estimates functions with interval-valued parameters. We applied this approach to modeling non-linear, static transfer functions of analog circuits with parameter variations and successfully simulated these models using a custom-built simulator.