Improving the performance of CAD optimization algorithms using on-line meta-level control

We present a profile based meta-reasoning model for parameter control of CAD algorithms working under constrained run-time. We also propose a unified framework, that can take informed decision about the time allocation and parameter adaptation of the algorithm, where there is no hard run-time constraints, instead the quality-time tradeoff is expressed by a utility function. We use the proposed strategy to get an adaptive cooling schedule for the simulated annealing algorithm. Application on two classical NP-hard problems in the VLSI domain, namely, the standard cell placement problem and the circuit partitioning problem shows that significant improvement of quality can be achieved using a profile based control.