A new power optimization perspective based on constraint based library access in physical synthesis is presented in this paper. Power constraints are modelled and applied at the terminals of designs or subdesigns based on relative power criticality of terminals. These constraints are then used in Physical Synthesis to selectively mask or expose different drive strength or threshold voltage variants of a given cell type. This constrained library access approach is compared with regular library access approach using a state-of-the-art power optimization engine. Experimental data shows it is possible to claim additional power savings in proposed approach with no or minimal performance impact. Also, it is possible to modulate this methodology to have selective focus on dynamic or leakage power if necessary.