An Accurate Coverage Forecasting Model for Behavioral Model Verification

Statistically forecasting potential returns in terms of code coverage for a given set of test cases (patterns) to be applied to a behavioral model can improve the overall effectiveness of behavioral model verification. In this paper, we present a forecasting model for behavioral VHDL model verification. The statistical assumptions of the proposed model are based on experimental evaluation of probability distribution functions and correlation functions. Results show that the forecasting model is of high accuracy. The prediction error of the proposed forecast model in estimating the probability of new coverage is, at most, 2% from the actual probability of having coverage when predicting 1000 simulation cycles into the future. When the prediction window size increases to 10,000 simulation cycles, the expected error in predicting the probability of having coverage is 13%, at most. The marginal error in predicting the waiting time to coverage is less than +-30% in forecasting 1000 simulation cycles, and at most +-22% in forecasting 10,000 simulation cycles.