The optimization algorithm module (OAM)

The OAM supports the choice of different optimization algorithms in a predefined set; three kinds of algorithm are currently included:

• a SLOP-like algorithm (§4.2.4, page ), which works increasing at each step the size of a single gate, chosen according to the best possible reduction of the delay along the critical path.
• The Powell algorithm (§4.2.3.2, page ), which is a particular form of the conjugate directions algorithm family ([17]): it does not require the computation of any gradient function and it converges quadratically to the minimum of the cost function.
• The simulated annealing algorithm (§4.2.5, page ): it chooses the transistor dimensions according to an ``annealing'' scheme, converging thus to a global minimum, getting rid of local minima. It is surely much slower than the previous ones, and requires a fine tuning of the annealing parameters.

For all the chosen methods, the analytical knowledge of the objective functions and their derivatives is not required, but just numerical approximations are exploited.
However methods requiring the gradient evaluation (e.g. the Fletcher-Reeves-Polak-Ribiere version of conjugate directions algorithm [17]) can be also supported.