Abstract:
Signoff timing analysis is still considered as a critical element in the SoC design
flow. With the advancements of the leading edge technologies towards
the deep sub-micron realms, the performance of a multi-million transistor
SoC is challenged by the aggravation of process variations. Traditionally,
the performance of an integrated circuit (IC) is characterized by the clock
frequency at which it operates, and is evaluated considering only worst-case
gate delays. This is a pessimistic approach, which highly underestimates
the performance of the design in deep sub-micron. This pessimism drives
the engineers to further optimize the design, eventually increasing the design cost drastically, as analysing such a design would require consumption
of more resources, adding to increase in the number of iterations. With
the aggravation of process variations, the main contributor to the delay
of the circuit is dominated by the interconnect delays. Hence, the number of scenarios required to analyse and x the design is enormous and
ever increasing. Such a pessimism hampers with time-to-market window of
product, which in worst case may be missed too.
In world of large number of multi-scenarios, the engineering change order
(ECO) becomes inevitable to achieve design timing closure with a low respin
cost. This concern is attributed to the fact that, firstly, performing ECOs
is an experimental process requiring an expert for each mode. The involvement of huge manual e ffort utilising their own respective scripts leads to
signi cant amount of iterations to achieve its nal goal. Secondly, timing
violations under multi-scenario design exists even after the detailed routing.
In the absence of such an implementation system, that can handle xing
of violations in multi-scenario design, without the involvement of manual
eff ort, causes non-convergence of timing closure, making the design prone
to the ping-pong e ect. This becomes a bottleneck to the design cycle time.
Additionally, this approach requires large memory usage and licensed CPU
resources which are already very limited.
This work presents a hierarchical approach to target the xing of timing violations, gaining signi cant cycle time in overall timing closure. A complete
solution for true hierarchical timing and crosstalk delay signo is presented
in this work. It explores the gaps in the current approach and provides a
novel solution to address them. It discusses the smart and e cient Timing
Closure Methodology which first identifies the Dominant Scenarios and then
prioritizes the violations for xing, in a manner which eliminates the danger
of ping-pong effect, where xing one set of violations can create hundreds
of new violations. This iterative approach seamlessly integrates the steps
of timing closure thereby reducing 90% of the manual effort.