A novel methodology to create generative statistical models of interconnects

This paper addresses the problem of constructing a generative statistical model for an interconnect starting from a limited set of S-parameter samples, which are obtained by simulating or measuring the interconnect for a few random realizations of its stochastic physical properties. These original samples are first converted into a pole-residue representation with common poles. The corresponding residues are modeled as a correlated stochastic process by means of principal component analysis and kernel density estimation. The obtained model allows generating new samples with similar statistics as the original data. A passivity check is performed over the generated samples to retain only passive data. The proposed approach is applied to a representative coupled microstrip line example.