Civil Engineering Reference
In-Depth Information
Structural correlations are automatically accounted for‚ since all the requisite
information required to model these correlations is embedded in the principal
components.
The overall flow of the algorithm is shown in Figure 6.10. To further speed
up the process‚ several techniques may be used:
1.
Before running the statistical timing analyzer‚ one run of deterministic STA
is performed to determine loose bounds on the best-case and worst-case
delays for all paths. Any path whose worst-case delay is less than the best-
case delay of the longest path will never be critical‚ and edges that lie only
on such paths can safely be removed.
During the “max” operation of statistical STA‚ if the value of
of one path has a lower delay than the value of of another path‚
the max function can be calculated by ignoring the path with lower delay.
2.
Although the above exposition has focused on handling spatially correlated
variables‚ it is equally easy to incorporate uncorrelated terms in this framework.
Only spatially correlated variables are decomposed into principal components‚
and any uncorrelated variables remain as they are. Therefore‚ in a case where no
variables are spatially correlated‚ the approach reduces to Berkelaar's method
in [Ber97]. However‚ heuristics from the other approaches in Section 6.4 may
be used to improve the modeling of structural correlations.
