Multiparametric Optimization and Penalty System in Integrated Circuit Component Placement
Vazgen Meliqyan, David Revazyan, Ashot Harutyuyan, Vachagan Davtyan, Elen Ghazaryan · 2024
This paper addresses the complex problem of placement optimization in integrated circuit (IC) design by proposing a novel approach that integrates a penalty system within a multiparametric optimization framework. The primary goal is to improve the placement process by not only optimizing traditional parameters such as delay, power consumption, and area but also ensuring adherence to design rules from the early stages of the algorithm. A matrix-based genetic algorithm is employed, leveraging a quadratic assignment problem (QAP) model to handle the placement of circuit elements. By incorporating design rule checks (DRC) directly into the fitness function, the algorithm can effectively eliminate placements that exhibit significant rule violations, thereby reducing the need for extensive post-processing adjustments. Two distinct methods for DRC integration were explored: one based on rectangular trees with query systems and the other utilizing a quadrant tree with boundary tracking. The impact of these methods on the overall optimization process was analyzed, demonstrating that the integration of penalty systems significantly enhances the efficiency and accuracy of IC placement by minimizing the occurrence of rule violations and subsequently reducing the need for human intervention during the final stages of design.