Incremental Latin hypercube sampling for lifetime stochastic behavioral modeling of analog circuits

In advanced technology node, not only process variations but also aging effects have critical impacts on circuit performance. Most of existing works consider process variations and aging effects separately while building the corresponding behavior models. Because of the time-varied circuit property, parametric yield need to be reanalyzed in each aging time step. This results in expensive simulation cost for reliability analysis due to the huge number of circuit simulation runs. In this paper, an incremental Latin hypercube sampling (LHS) approach is proposed to build the stochastic behavior models for analog/mixed-signal (AMS) circuits while simultaneously considering process variations and aging effects. By reusing previous sampling information, only a few new samples are incrementally updated to build an accurate stochastic model in different time steps, which significantly reduces the number of simulations for aging analysis. Experiments on an operational amplifier and a DAC circuit achieve 242x speedup over traditional reliability analysis method with similar accuracies.