| The continuous scaling trends of smaller technologies, higher operating frequencies, lower power supply voltages, and more functionalities for integrated circuits and systems have made it extremely challenging to design a reliable power delivery network (PDN): Design uncertainties, process uncertainties and operation uncertainties have to be considered during the modeling and optimization to guarantee the quality of the PDN.;To model the design uncertainties, this dissertation develops two reduced order modeling techniques for design sign-off and design optimization respectively. The reduced models can be used for fast simulation with design changes. In addition, an incremental random walk algorithm is proposed for incremental and on-demand analysis of the PDN. To model the process and operation uncertainties, this dissertation proposes an stochastic algorithm to extract the process uncertainties and operation uncertainties in the current loads and form parameterized current models.;Geared by those uncertainty models, this dissertation further proposes novel algorithms to suppress two dominant types of the noise present in a PDN, the peak noise and resonance noise, using decoupling capacitance (decap) and frequency actuator respectively. Experimental results show that significant noise reduction is achieved, which proves the effectiveness of the proposed models.;To the best of our knowledge, this dissertation is the first of the kind that provides novel models to simultaneously consider design uncertainties, operation uncertainties and process uncertainties. It is also the first in-depth study on proactively reducing runtime PDN resonance noise. |
