The authors describe a sequential strategy for designing manufacturable integrated circuits using available CAD tools. Optimizing the performance of complex designs in the presence of unwanted parameter variations can take a prohibitively large number of computer runs. These methods overcome this complexity by combining sequential experimentation with modeling of the CAD simulator output as realizations of stochastic processes. These models give inexpensive approximations that enable location of subregions on which to continue the search for the optimal design. The multistage experimentation reduces the number of simulation runs required to obtain sufficiently accurate approximations. The strategy is able to treat circuits with many parameters, performances, and criteria. Included are two examples where improved designs have been found by these methods. The first is a zero-temperature-coefficient current reference IC. The second is a gallium arsenide voltage level shifter IC with multiple performances of interest.>

Model selection based on a predictive performance measure is compared to model selection based on maximum likelihood. Both procedures exhibit unstable relative performance when predictive performance is measured over time periods that may contain overly small proportions of total failures. Both procedures may be stabilized by bounding this proportion away from zero. In that case, both procedures exhibit similar predictive performance relative to the other. This provides evidence that model selection based on the commonly used maximum likelihood approach may in fact have good predictive performance.>

Poisson processes with failure intensity functions that are approximately constant are investigated. Maximum likelihood estimation procedures are used to estimate the failure intensity for a fixed level of approximation. The level of approximation is chosen through an adaptation of crossvalidation. Since model performance changes with time, the authors utilize a procedure that adaptively selects models over time. The predictive performance of such an adaptive procedure based on selection from among three nonparametric models is compared to that of the logarithmic Poisson and the exponential Poisson models for two software reliability data sets.>

A method for parametric yield optimization which significantly reduces the simulation cost is proposed. The method assumes that the circuit performances ultimately determining yield can be approximated by computationally inexpensive functions of the inputs to the circuit simulator. These inputs are the designable parameters, the uncontrollable statistical variations, and the operating conditions of interest. The authors fit the approximating functions to data from a statistical experiment incurring relatively few runs of the simulator. For a given set of designable parameters, the fitted models predict the circuit performances as the uncontrollable parameters and operating conditions vary. The predictions lead to estimates of the parametric yield, which is then numerically maximized with respect to the designable parameters. The authors give a CMOS circuit example where sufficiently accurate predictions and good actual yield result from about 100 runs of the circuit simulator.>