4. Conclusions

Although the incremental approach with spatiotemporal correlation provide roughly the same gain in accuracy as the global one, the running time is clearly much shorter. The ratio of running times when neglecting one or both of the spatial or temporal correlations is different for the global vs. incremental approach: for the global one, it is more expensive to consider temporal correlations, while in the incremental approach, the spatial correlations are more time consuming. These observations were proved to be consistent in all our experiments on benchmark circuits. In the following, we give the error values only for PR inputs, using the incremental approach. In reporting the error, we compared our switching activity estimates with the results of logic simulation at every internal node and primary output. To conclude, two important observations should be made. First, Markov Chains are useful in modelling input correlations. Second, the degree in which any type of correlations affects the overall quality of estimations, depends on the internal structure of the circuit and the correlations among the primary inputs. The best way to use this framework in practice would be to combine both approaches in a hierarchical manner. We have proposed a novel approach for estimation of the switching activities in combinational logic circuits under pseudorandom or biased inputs. Using the zero-delay hypothesis, we have derived a probabilistic model based on lag-one Markov Chains and conditional probabilities. The main feature of our approach is the systematic way in which we can deal with complex dependencies that may appear in practice; more precisely, our model supports spatiotemporal correlations among the primary inputs and internal lines of the circuit under consideration. A comparative analysis and benchmark evaluations emphasize the superiority of our approach over the current existing techniques and show its practicality on large combinational circuits. Our future work will concentrate on heuristics for improving the accuracy and time/space complexity and also on extensions of this approach beyond the logic level.