关键词: 先进封装, 供电网络, 去耦电容, 启发式算法, 机器学习

Abstract: As advanced packaging technologies evolve toward 2.5D/3D integration, the power delivery architecture has transitioned from the conventional chip-package-printed circuit board hierarchy to a complex power distribution network (PDN) in which chiplets, interposers, through-silicon vias (TSVs), microbumps, package substrates, and voltage regulator modules are tightly coupled. Under low-voltage, high-current, and high-switching-frequency conditions, impedance resonances, voltage drops, simultaneous switching noise (SSN), and multiport coupling in the PDN have become increasingly prominent, posing critical constraints on the performance and reliability of advanced packaging systems. This paper systematically reviews recent advances in PDN modeling, joint analysis, and decoupling capacitor optimization for 2.5D/3D integrated circuits (ICs). The modeling methods mainly include rapid modeling of power/ground planes based on transfer matrices and distributed equivalent networks, as well as segmented modeling of silicon interposers and TSV structures. By combining resistance, inductance, conductance, and capacitance (RLGC) parameter extraction, matrix-based interconnection, and multiport network cascading, these methods enable joint analysis of frequency-domain impedance, steady-state voltage drop, and time-domain switching noise. Decoupling capacitor optimization focuses on the coordinated configuration of capacitor locations, types, and quantities, and employs heuristic algorithms, reinforcement learning, and multi-objective collaborative optimization to improve the efficiency of design-space exploration. In light of emerging trends in heterogeneous chiplet power delivery architectures, multiphysics collaborative optimization, and artificial intelligence (AI)-driven design automation, this paper further discusses the key challenges facing future PDN design and provides a reference for power integrity analysis and system-level collaborative design of advanced packaging systems.

Key words: advanced packaging, power delivery network, decoupling capacitors, heuristic algorithms, machine learning