A Knowledge Based Method for Optimization of Decoupling Capacitors in Power Delivery Networks.

For power delivery applications, a method for reducing impedance is through the addition of decoupling capacitors (decaps). In this article, we propose a knowledge-based optimization method to determine decap design in power delivery networks (PDNs). The proposed method provides the optimized PDN with the minimum number of decaps by iteratively assigning the decaps that maximize the frequency range where the target impedance is satisfied. In addition, by integrating the time-domain analysis into the proposed method, a voltage response arising from any arbitrary load current sources is guaranteed within the threshold level while preventing the over-design problem. Unlike the recent random search-based methods (genetic algorithm and machine learning (ML) etc.) requiring a large number of PDN simulations for learning and selection, the proposed method determines effective decaps through multiple steps based only on domain knowledge, significantly reducing the number of PDN simulations and computational cost to obtain the optimized decap solution. We apply the proposed method to three different PDN models. The results show that only 1.3% of the computing time is required to achieve the optimized decap design compared with the ML-based method while having the same number of decaps as the optimal number obtained by the full-search simulation. [ABSTRACT FROM AUTHOR]