Résumé
Decaps can be very effective in reducing power grid noise, but they cannot be inserted on a chip in an ad-hoc manner without considering the grid parasitic impedances, switching frequencies of the blocks, and proximity to the power supply pins. It is possible that a decoupling capacitor inserted into a vacant space left by a placer may not only be inefficient, but can also be detrimental. In this paper, we present a detailed study of the decoupling capacitor's effectiveness in the uniform RLC power and ground grid networks. Based on the analysis of a simple circuit in which the decoupling capacitor amplifies the power grid noise, we explain why a decap can be detrimental. We introduce effectiveness metrics for determining those decoupling capacitor locations that capture the effects of parasitic impedances between the decap and switching circuit, decap and power supply, and switching frequency and magnitude of the switching circuit. Our experimental results demonstrate the effectiveness of the proposed metrics.